[
  {
    "altLabels": [
      "Ad Hoc Search",
      "Adhoc Retrieval"
    ],
    "definition": "A retrieval task in which a one-off query or topic is submitted against a fixed corpus or evaluation dataset to find topically relevant items.",
    "id": "ad-hoc-retrieval",
    "kind": "Task",
    "prefLabel": "Ad Hoc Retrieval",
    "scopeNote": "Ad hoc retrieval is the classic TREC-style retrieval task: the searchable corpus or test-collection dataset is fixed for evaluation, while each query or topic represents a new information need. It is usually framed around topical document or passage relevance rather than ongoing filtering, source selection, entity lookup, or interactive exploration.",
    "uri": "https://id.searchplex.net/ad-hoc-retrieval/"
  },
  {
    "altLabels": [
      "Agentic Retrieval-Augmented Generation"
    ],
    "definition": "A retrieval-augmented generation pattern in which an agent plans, issues, evaluates, and iterates retrieval or tool-use actions before producing a generated answer or result.",
    "id": "agentic-rag",
    "kind": "GeneralConcept",
    "prefLabel": "Agentic RAG",
    "scopeNote": "Agentic RAG combines RAG with agentic control over retrieval, query reformulation, tool use, evidence gathering, or stopping decisions. It is distinct from RAG in general, which may use a fixed retrieval pipeline, and from agentic retrieval, which names the retrieval loop rather than the full generation pattern.",
    "terms": [
      {
        "community": "industry",
        "label": "Agentic RAG System",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Agentic RAG Workflow",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/agentic-rag/"
  },
  {
    "altLabels": [
      "Agentic Search"
    ],
    "definition": "A retrieval pattern in which an AI agent plans, issues, evaluates, and iterates retrieval actions as part of a broader reasoning or task-solving loop.",
    "id": "agentic-retrieval",
    "kind": "Task",
    "prefLabel": "Agentic Retrieval",
    "scopeNote": "Agentic retrieval emphasizes multi-step and adaptive retrieval behavior by an agent, often involving query reformulation, tool use, filtering, verification, or repeated searches. It is not the same as retrieval generally, RAG, semantic search, or multi-stage ranking.",
    "terms": [
      {
        "community": "industry",
        "label": "Agent Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/agentic-retrieval/"
  },
  {
    "altLabels": [
      "ANN Search",
      "ANNS"
    ],
    "definition": "A search method that returns near-neighbor results using an approximation strategy to trade some exactness for lower latency, memory use, or computational cost.",
    "id": "approximate-nearest-neighbor-search",
    "kind": "Method",
    "prefLabel": "Approximate Nearest Neighbor Search",
    "scopeNote": "Approximate nearest neighbor search is widely used in vector search and dense retrieval systems, especially at large scale. It is not the same as vector search itself, because vector search may use exact search, approximate search, or other retrieval mechanisms.",
    "uri": "https://id.searchplex.net/approximate-nearest-neighbor-search/"
  },
  {
    "artifactKind": "Dataset",
    "definition": "An argument-retrieval evaluation dataset used to retrieve counterarguments or related argumentative texts.",
    "id": "arguana",
    "kind": "Artifact",
    "prefLabel": "ArguAna",
    "scopeNote": "ArguAna is included in BEIR as an argument-retrieval dataset. It is a component evaluation dataset rather than a benchmark suite.",
    "uri": "https://id.searchplex.net/arguana/"
  },
  {
    "altLabels": [
      "AP"
    ],
    "definition": "A ranking evaluation metric that averages precision values at the ranks where relevant items are retrieved for a single query or topic.",
    "id": "average-precision",
    "kind": "Metric",
    "prefLabel": "Average Precision",
    "scopeNote": "Average Precision is a per-query or per-topic metric. Mean Average Precision averages AP values across multiple queries or topics.",
    "uri": "https://id.searchplex.net/average-precision/"
  },
  {
    "artifactKind": "Benchmark",
    "definition": "A benchmark suite for evaluating information-retrieval systems across multiple evaluation datasets, domains, and retrieval settings.",
    "id": "beir",
    "kind": "Artifact",
    "prefLabel": "BEIR",
    "scopeNote": "BEIR is not itself a single retrieval dataset. It aggregates multiple test collections or IR evaluation datasets and standardizes their use for evaluating retrieval effectiveness, especially zero-shot generalization.",
    "terms": [
      {
        "community": "academic",
        "label": "BEIR Dataset",
        "usage": "informal"
      },
      {
        "community": "academic",
        "label": "BEIR Benchmark",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/beir/"
  },
  {
    "definition": "A WAND-family dynamic-pruning method that uses score upper bounds computed for blocks of postings to enable more aggressive skipping during top-k retrieval.",
    "id": "block-max-wand",
    "kind": "Method",
    "prefLabel": "Block-Max WAND",
    "scopeNote": "Block-Max WAND is a refinement of WAND-style dynamic pruning rather than a generic synonym for WAND.",
    "uri": "https://id.searchplex.net/block-max-wand/"
  },
  {
    "altLabels": [
      "Okapi BM25"
    ],
    "definition": "A term-based ranking function that scores documents for a query using term frequency, inverse document frequency, and document-length normalization.",
    "id": "bm25",
    "kind": "Method",
    "prefLabel": "BM25",
    "scopeNote": "BM25 is a widely used lexical retrieval and ranking method. It is often used as a baseline or component in keyword, lexical, sparse, and hybrid retrieval systems, but it is not itself the same as keyword search, lexical retrieval, or sparse retrieval.",
    "uri": "https://id.searchplex.net/bm25/"
  },
  {
    "definition": "A field-aware variant of BM25 that combines term evidence from multiple document fields with field-specific weights or normalization.",
    "id": "bm25f",
    "kind": "Method",
    "prefLabel": "BM25F",
    "scopeNote": "BM25F is used for fielded documents such as title, body, metadata, or anchor text. It is a BM25-family method, not a synonym for BM25 generally or for fielded retrieval as a whole.",
    "uri": "https://id.searchplex.net/bm25f/"
  },
  {
    "altLabels": [
      "Boolean Search"
    ],
    "definition": "A retrieval method in which queries are expressed as Boolean combinations of terms or conditions and matching items satisfy the Boolean expression.",
    "id": "boolean-retrieval",
    "kind": "Method",
    "prefLabel": "Boolean Retrieval",
    "scopeNote": "Boolean retrieval returns matches according to logical conditions rather than necessarily producing a relevance-ranked list. It is distinct from keyword search, exact match, and ranked lexical retrieval, although systems may combine these ideas.",
    "uri": "https://id.searchplex.net/boolean-retrieval/"
  },
  {
    "artifactKind": "Benchmark",
    "definition": "A benchmark for evaluating browsing agents on their ability to locate hard-to-find information on the web and produce short, verifiable answers.",
    "id": "browsecomp",
    "kind": "Artifact",
    "prefLabel": "BrowseComp",
    "scopeNote": "BrowseComp evaluates web-browsing agent behavior over live web search and browsing conditions. It is not a fixed IR test collection in the BEIR sense and should be distinguished from BrowseComp-Plus, which introduces a fixed curated corpus for more controlled evaluation.",
    "uri": "https://id.searchplex.net/browsecomp/"
  },
  {
    "altLabels": [
      "BrowseComp Plus"
    ],
    "artifactKind": "Benchmark",
    "definition": "A controlled benchmark for evaluating deep-research agents and retrievers, derived from BrowseComp and built around a fixed curated corpus with supporting documents and hard negatives.",
    "id": "browsecomp-plus",
    "kind": "Artifact",
    "prefLabel": "BrowseComp-Plus",
    "scopeNote": "BrowseComp-Plus is designed to make evaluation of deep-research agents and retrieval methods more fair and transparent by controlling the document corpus and evidence. It is related to BrowseComp but should not be treated as identical to live-web browsing evaluation.",
    "terms": [
      {
        "community": "academic",
        "label": "BCP",
        "usage": "abbreviation"
      }
    ],
    "uri": "https://id.searchplex.net/browsecomp-plus/"
  },
  {
    "altLabels": [
      "Candidate Selection"
    ],
    "definition": "Retrieving an initial set of candidate items for later ranking, filtering, or reranking.",
    "id": "candidate-retrieval",
    "kind": "Task",
    "prefLabel": "Candidate Retrieval",
    "scopeNote": "The boundary with first-stage retrieval is editorially sensitive and should be reviewed per use case.",
    "terms": [
      {
        "community": "industry",
        "label": "Candidate Generation",
        "usage": "communityPreferred"
      },
      {
        "community": "academic",
        "label": "First-Stage Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/candidate-retrieval/"
  },
  {
    "definition": "A method for splitting larger content into smaller units that can be indexed, retrieved, embedded, passed to a model, or otherwise processed.",
    "id": "chunking",
    "kind": "Method",
    "prefLabel": "Chunking",
    "scopeNote": "In RAG and search systems, chunking defines candidate units such as passages, sections, windows, or semantic segments. It influences retrieval quality but is not itself retrieval, ranking, or embedding.",
    "uri": "https://id.searchplex.net/chunking/"
  },
  {
    "altLabels": [
      "Conference and Labs of the Evaluation Forum",
      "Cross-Language Evaluation Forum"
    ],
    "definition": "The Conference and Labs of the Evaluation Forum, an evaluation program and research venue for information-access systems, especially multilingual and multimodal information access.",
    "id": "clef",
    "kind": "EvaluationProgram",
    "prefLabel": "CLEF",
    "scopeNote": "CLEF organizes evaluation labs, shared tasks, workshops, conference publications, and reusable test collections. It is not itself a metric, dataset, qrels file, or a single retrieval task, though CLEF labs may define tasks and publish datasets, runs, and relevance judgments.",
    "terms": [
      {
        "community": "academic",
        "label": "CLEF Initiative",
        "usage": "communityPreferred"
      },
      {
        "community": "academic",
        "label": "CLEF Labs",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/clef/"
  },
  {
    "artifactKind": "Model",
    "definition": "A neural retrieval model that represents queries and documents using contextualized token-level vectors and scores them using late interaction.",
    "id": "colbert",
    "kind": "Artifact",
    "prefLabel": "ColBERT",
    "scopeNote": "ColBERT is a concrete model architecture that introduced a widely used form of contextualized late interaction. It is one realization of the broader late-interaction paradigm, not a synonym for every late-interaction model.",
    "uri": "https://id.searchplex.net/colbert/"
  },
  {
    "artifactKind": "Model",
    "definition": "A refinement of ColBERT that improves the effectiveness and storage efficiency of late-interaction retrieval through improved supervision and compressed representations.",
    "id": "colbertv2",
    "kind": "Artifact",
    "prefLabel": "ColBERTv2",
    "scopeNote": "ColBERTv2 remains a ColBERT-family late-interaction retriever. It is distinct from PLAID, which is a retrieval engine and optimization for serving late-interaction models.",
    "uri": "https://id.searchplex.net/colbertv2/"
  },
  {
    "altLabels": [
      "Conversational Information Retrieval",
      "CIR"
    ],
    "definition": "A search task or interaction pattern in which users seek information through a multi-turn natural-language conversation with a search or retrieval system.",
    "id": "conversational-search",
    "kind": "Task",
    "prefLabel": "Conversational Search",
    "scopeNote": "Conversational search emphasizes dialogue, context carryover, clarification, and multi-turn interaction. It overlaps with conversational information retrieval and interactive search, but is not simply a single query or a RAG system.",
    "uri": "https://id.searchplex.net/conversational-search/"
  },
  {
    "altLabels": [
      "Document Corpus",
      "Document Collection"
    ],
    "artifactKind": "Dataset",
    "definition": "A collection of documents, passages, web pages, items, or other retrievable units used for search, retrieval, training, or evaluation.",
    "id": "corpus",
    "kind": "Artifact",
    "prefLabel": "Corpus",
    "scopeNote": "In IR evaluation, the corpus is the searchable collection component of a test collection. The bare term collection is ambiguous: it may mean a corpus or searchable document collection, while test collection usually means the full evaluation package with corpus, topics or queries, and relevance judgments.",
    "terms": [
      {
        "community": "academic",
        "label": "Collection",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Search Collection",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/corpus/"
  },
  {
    "definition": "A neural architecture that jointly processes both sides of a matching problem, such as a query and a document, allowing direct interaction between them before producing a relevance, similarity, or classification score.",
    "id": "cross-encoder",
    "kind": "Method",
    "prefLabel": "Cross-Encoder",
    "scopeNote": "Because each query-candidate pair must be processed jointly, Cross-Encoders are usually much more expensive for large-scale retrieval than independently encoded representations. In search systems they are therefore commonly used to rerank a bounded candidate set.",
    "terms": [
      {
        "community": "industry",
        "label": "Reranker",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/cross-encoder/"
  },
  {
    "altLabels": [
      "Cross-Language Information Retrieval",
      "Cross-Lingual Retrieval",
      "Cross-Lingual Information Retrieval",
      "CLIR"
    ],
    "definition": "Retrieval in which the user's query or information need is expressed in one language and relevant retrieved items may be written in another language.",
    "id": "cross-language-retrieval",
    "kind": "Task",
    "prefLabel": "Cross-Language Retrieval",
    "scopeNote": "Cross-language retrieval is often called cross-lingual information retrieval. It is narrower than multilingual retrieval and commonly involves query translation, document translation, multilingual representations, or cross-lingual embeddings.",
    "uri": "https://id.searchplex.net/cross-language-retrieval/"
  },
  {
    "altLabels": [
      "Cross-Modality Retrieval",
      "Cross-Modal Search"
    ],
    "definition": "A retrieval task in which the query and retrieved items may be expressed in different modalities, such as retrieving images from text queries or text from image queries.",
    "id": "cross-modal-retrieval",
    "kind": "Task",
    "prefLabel": "Cross-Modal Retrieval",
    "scopeNote": "Cross-modal retrieval is a narrower case within multimodal retrieval. It emphasizes matching across a modality boundary rather than searching within one modality or combining multiple modalities in a single representation.",
    "terms": [
      {
        "community": "academic",
        "label": "Image-Text Retrieval",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Text-to-Image Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Image-to-Text Search",
        "usage": "nearSynonym"
      },
      {
        "community": "vendor",
        "label": "Cross-Type Search",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/cross-modal-retrieval/"
  },
  {
    "altLabels": [
      "At-K Metric",
      "Top-K Metric",
      "Metric@K"
    ],
    "definition": "A family of evaluation metrics computed over only the top K retrieved or ranked results.",
    "id": "cutoff-metric",
    "kind": "Metric",
    "prefLabel": "Cutoff Metric",
    "scopeNote": "Cutoff metrics are commonly written with an @K suffix, such as Precision@K, Recall@K, Success@K, or nDCG@K. They evaluate effectiveness at a user- or system-relevant rank cutoff rather than over an unbounded result list.",
    "uri": "https://id.searchplex.net/cutoff-metric/"
  },
  {
    "altLabels": [
      "Agentic Deep Research"
    ],
    "definition": "An agentic information-seeking pattern in which a system performs multi-step research by planning searches, browsing or retrieving sources, reasoning over evidence, and synthesizing a cited answer or report.",
    "id": "deep-research",
    "kind": "GeneralConcept",
    "prefLabel": "Deep Research",
    "scopeNote": "Deep research usually combines agentic retrieval, web browsing, source evaluation, citation, and synthesis. It is related to RAG and question answering, but it emphasizes multi-step research behavior and report-style synthesis rather than a single retrieval or generation step.",
    "terms": [
      {
        "community": "industry",
        "label": "Deep Research Agent",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "DRA",
        "usage": "abbreviation"
      }
    ],
    "uri": "https://id.searchplex.net/deep-research/"
  },
  {
    "definition": "An encoder that maps an input, such as a query, document, image, or other item, to a dense vector representation.",
    "id": "dense-encoder",
    "kind": "Method",
    "prefLabel": "Dense Encoder",
    "scopeNote": "In dense retrieval, dense encoders commonly produce query and document vectors for similarity-based retrieval. A dense encoder is an encoding component, not by itself a complete retrieval architecture. A dual encoder typically contains or uses independently applied encoders, but not every dense encoder is part of a dual-encoder retrieval system.",
    "uri": "https://id.searchplex.net/dense-encoder/"
  },
  {
    "altLabels": [
      "DPR"
    ],
    "definition": "A neural passage-retrieval method that represents questions and passages with dense vectors, commonly using a dual-encoder architecture, to retrieve candidate passages.",
    "id": "dense-passage-retrieval",
    "kind": "Method",
    "prefLabel": "Dense Passage Retrieval",
    "scopeNote": "Dense Passage Retrieval is often abbreviated DPR and is associated with the open-domain question-answering method introduced under that name. It is not a synonym for dense retrieval generally, vector search, or semantic search.",
    "uri": "https://id.searchplex.net/dense-passage-retrieval/"
  },
  {
    "definition": "Retrieval using dense vector representations of queries and items, typically compared using a vector similarity or distance function.",
    "id": "dense-retrieval",
    "kind": "Representation",
    "prefLabel": "Dense Retrieval",
    "scopeNote": "Dense retrieval normally refers to learned low- or moderate-dimensional dense embeddings. It is narrower than the broad practitioner term vector search.",
    "terms": [
      {
        "community": "industry",
        "label": "Embedding Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/dense-retrieval/"
  },
  {
    "altLabels": [
      "Diversity Retrieval",
      "Search Result Diversification",
      "Diversified Search"
    ],
    "definition": "A retrieval task or objective in which a result set should cover multiple aspects, intents, interpretations, or subtopics of an information need.",
    "id": "diversified-retrieval",
    "kind": "Task",
    "prefLabel": "Diversified Retrieval",
    "scopeNote": "Diversified retrieval is especially important for ambiguous or multifaceted queries, where returning near-duplicate results for one interpretation may be less useful than covering several plausible intents. It is distinct from ordinary ranking by relevance alone.",
    "uri": "https://id.searchplex.net/diversified-retrieval/"
  },
  {
    "altLabels": [
      "Document Expansion by Query Prediction"
    ],
    "definition": "Enriching a document representation with additional generated or selected text, terms, queries, or signals intended to improve retrieval matching.",
    "id": "document-expansion",
    "kind": "Task",
    "prefLabel": "Document Expansion",
    "scopeNote": "Document expansion modifies or augments the document side, often before indexing. It is distinct from query expansion, which modifies or augments the query side at search time or query-processing time.",
    "terms": [
      {
        "community": "academic",
        "label": "Doc2Query",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "docT5query",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/document-expansion/"
  },
  {
    "altLabels": [
      "Document Search"
    ],
    "definition": "A retrieval task in which the retrievable units are documents rather than passages, entities, answers, or other smaller or more structured objects.",
    "id": "document-retrieval",
    "kind": "Task",
    "prefLabel": "Document Retrieval",
    "scopeNote": "Document retrieval is a core IR task formulation and is often used in ad hoc retrieval and web-search evaluation. It is distinct from passage retrieval, which retrieves passage-like units, and from question answering, which aims to return answers rather than documents.",
    "terms": [
      {
        "community": "industry",
        "label": "Enterprise Document Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Knowledge Base Search",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/document-retrieval/"
  },
  {
    "altLabels": [
      "Bi-Encoder"
    ],
    "definition": "A neural architecture that encodes the two sides of a matching problem independently, such as a query and a document, and compares their resulting representations using a similarity or scoring function.",
    "id": "dual-encoder",
    "kind": "Method",
    "prefLabel": "Dual Encoder",
    "scopeNote": "In information retrieval, a dual encoder commonly uses separate query and document encoding paths so document representations can be computed ahead of time and efficiently searched. The two encoders may use separate parameters or share some or all parameters. Bi-encoder is commonly used as an equivalent term.",
    "terms": [
      {
        "community": "industry",
        "label": "Two-Tower Model",
        "usage": "communityPreferred"
      },
      {
        "community": "industry",
        "label": "Two Tower",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/dual-encoder/"
  },
  {
    "definition": "A family of query-evaluation techniques that avoid fully evaluating candidates that cannot affect the desired top-ranked results.",
    "id": "dynamic-pruning",
    "kind": "Method",
    "prefLabel": "Dynamic Pruning",
    "scopeNote": "Dynamic pruning uses score bounds or related statistics to skip unnecessary scoring work while preserving or approximating top-k effectiveness.",
    "uri": "https://id.searchplex.net/dynamic-pruning/"
  },
  {
    "altLabels": [
      "Vector Embedding"
    ],
    "definition": "A vector representation of an item, token, query, document, or other object in a continuous or otherwise structured representation space.",
    "id": "embedding",
    "kind": "Representation",
    "prefLabel": "Embedding",
    "scopeNote": "In retrieval systems, embeddings are commonly used as vector representations for similarity search, dense retrieval, reranking, clustering, or generation support. An embedding is the representation object, not the model that produced it and not the retrieval system that searches over it.",
    "uri": "https://id.searchplex.net/embedding/"
  },
  {
    "altLabels": [
      "Entity Disambiguation"
    ],
    "definition": "Mapping an entity mention in a query, document, or other text to a canonical entity in a knowledge base, catalog, graph, or other authority.",
    "id": "entity-linking",
    "kind": "Task",
    "prefLabel": "Entity Linking",
    "scopeNote": "Entity linking is the canonical-resolution step. Named entity recognition identifies and types mentions; entity linking resolves those mentions to specific entities.",
    "terms": [
      {
        "community": "general",
        "label": "Entity Resolution",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/entity-linking/"
  },
  {
    "altLabels": [
      "Entity-Oriented Search",
      "Entity Retrieval"
    ],
    "definition": "A retrieval task focused on finding entities, entity lists, or entity-related information that satisfies an information need.",
    "id": "entity-search",
    "kind": "Task",
    "prefLabel": "Entity Search",
    "scopeNote": "Entity search is broader than retrieving documents that mention an entity. TREC Entity Track tasks included entity-oriented search on Web data, such as finding entities and properties of entities.",
    "uri": "https://id.searchplex.net/entity-search/"
  },
  {
    "altLabels": [
      "Exact-Match Search",
      "Exact Phrase Search",
      "Phrase Search"
    ],
    "definition": "A matching method or query constraint that requires specified terms, values, or phrases to appear exactly as requested.",
    "id": "exact-match",
    "kind": "Method",
    "prefLabel": "Exact Match",
    "scopeNote": "Exact match may refer to exact field-value matching, exact term matching, or exact phrase matching depending on the system. It is narrower than lexical retrieval generally and should not be used as a synonym for BM25 or keyword search.",
    "terms": [
      {
        "community": "industry",
        "label": "Quoted Search",
        "usage": "informal"
      },
      {
        "community": "industry",
        "label": "Phrase Match",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/exact-match/"
  },
  {
    "altLabels": [
      "Faceted Navigation",
      "Faceted Browsing"
    ],
    "definition": "A search and navigation method that lets users narrow or explore a result set through multiple structured dimensions, called facets.",
    "id": "faceted-search",
    "kind": "Method",
    "prefLabel": "Faceted Search",
    "scopeNote": "Faceted search, also called faceted navigation or faceted browsing, uses facets such as author, topic, brand, language, date, or format to expose available refinements. It commonly applies filters, but the concept includes the faceted organization and interactive navigation model rather than filtering alone.",
    "terms": [
      {
        "community": "industry",
        "label": "Faceting",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/faceted-search/"
  },
  {
    "altLabels": [
      "Federated Information Retrieval",
      "Distributed Information Retrieval"
    ],
    "definition": "A retrieval task or system pattern in which a query is sent to multiple search services, collections, indexes, or databases and their results are combined for the user.",
    "id": "federated-search",
    "kind": "Task",
    "prefLabel": "Federated Search",
    "scopeNote": "Federated search commonly involves source selection, query dispatch or translation, result merging, and presentation of a unified result set. It is distinct from ad hoc retrieval over a single known collection.",
    "uri": "https://id.searchplex.net/federated-search/"
  },
  {
    "altLabels": [
      "Fact Extraction and VERification"
    ],
    "artifactKind": "Dataset",
    "definition": "A fact-verification dataset used for claim verification and evidence retrieval evaluation.",
    "id": "fever",
    "kind": "Artifact",
    "prefLabel": "FEVER",
    "scopeNote": "FEVER is included in BEIR as a fact-checking retrieval dataset. It is a dataset/test collection, not the BEIR benchmark suite.",
    "uri": "https://id.searchplex.net/fever/"
  },
  {
    "altLabels": [
      "FiQA"
    ],
    "artifactKind": "Dataset",
    "definition": "A financial question-answering and retrieval evaluation dataset used for information-retrieval benchmarking.",
    "id": "fiqa-2018",
    "kind": "Artifact",
    "prefLabel": "FiQA-2018",
    "scopeNote": "FiQA-2018 appears in BEIR as a financial-domain retrieval dataset. It is an evaluation dataset, not the broader BEIR benchmark suite.",
    "uri": "https://id.searchplex.net/fiqa-2018/"
  },
  {
    "altLabels": [
      "Forum for Information Retrieval Evaluation"
    ],
    "definition": "The Forum for Information Retrieval Evaluation, a South Asian evaluation program and meeting for information-retrieval and information-access systems.",
    "id": "fire",
    "kind": "EvaluationProgram",
    "prefLabel": "FIRE",
    "scopeNote": "FIRE was started in 2008 and is commonly described as a South Asian counterpart to TREC, CLEF, and NTCIR. It has emphasized multilingual information access, Indian languages, mixed-script retrieval, legal information access, and related evaluation tasks.",
    "uri": "https://id.searchplex.net/fire/"
  },
  {
    "altLabels": [
      "Initial Ranking"
    ],
    "definition": "Producing an initial ranked ordering of items before later-stage reranking or fusion.",
    "id": "first-stage-ranking",
    "kind": "Task",
    "prefLabel": "First-Stage Ranking",
    "scopeNote": "Use this when the first stage produces an ordered list, not merely an unordered candidate set.",
    "terms": [
      {
        "community": "industry",
        "label": "Fast Ranking",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/first-stage-ranking/"
  },
  {
    "altLabels": [
      "Graded Relevance Metric",
      "Graded Metric"
    ],
    "definition": "A family of evaluation metrics designed to use multi-level relevance judgments rather than only binary relevant/not-relevant labels.",
    "id": "graded-relevance-metric",
    "kind": "Metric",
    "prefLabel": "Graded-Relevance Metric",
    "scopeNote": "Graded-relevance metrics are useful when some results are more relevant than others. nDCG is the common example in IR evaluation.",
    "uri": "https://id.searchplex.net/graded-relevance-metric/"
  },
  {
    "altLabels": [
      "AI Grounding"
    ],
    "definition": "The practice or goal of anchoring generated or system output in supplied, retrieved, or otherwise authoritative source information.",
    "id": "grounding",
    "kind": "GeneralConcept",
    "prefLabel": "Grounding",
    "scopeNote": "In AI and RAG systems, grounding is often discussed as a way to reduce unsupported or fabricated claims. Retrieval can support grounding, but grounding may also use provided context, tools, databases, citations, or other evidence sources.",
    "uri": "https://id.searchplex.net/grounding/"
  },
  {
    "altLabels": [
      "Hierarchical Navigable Small World"
    ],
    "definition": "A graph-based approximate nearest neighbor search method using hierarchical navigable small-world graphs.",
    "id": "hnsw",
    "kind": "Method",
    "prefLabel": "HNSW",
    "scopeNote": "HNSW is commonly used as a vector-search index method in production systems. It is one ANN method, not a synonym for vector search, dense retrieval, or nearest neighbor search generally.",
    "uri": "https://id.searchplex.net/hnsw/"
  },
  {
    "artifactKind": "Dataset",
    "definition": "A multi-hop question-answering dataset used in retrieval and open-domain QA evaluation.",
    "id": "hotpotqa",
    "kind": "Artifact",
    "prefLabel": "HotpotQA",
    "scopeNote": "HotpotQA appears in BEIR as a question-answering retrieval dataset. It is a dataset, not the broader question-answering task or the BEIR benchmark suite.",
    "uri": "https://id.searchplex.net/hotpotqa/"
  },
  {
    "altLabels": [
      "Hybrid Search"
    ],
    "definition": "Retrieval that combines evidence, candidates, or scores from multiple retrieval approaches or representations.",
    "id": "hybrid-retrieval",
    "kind": "GeneralConcept",
    "prefLabel": "Hybrid Retrieval",
    "scopeNote": "In current production search and RAG usage, hybrid retrieval most often means combining BM25 or lexical retrieval with dense vector retrieval, commonly single-vector retrieval. The concept is broader than that default pairing and does not imply one particular fusion algorithm.",
    "uri": "https://id.searchplex.net/hybrid-retrieval/"
  },
  {
    "altLabels": [
      "Image Search",
      "Visual Search"
    ],
    "definition": "A retrieval task in which the target items are images or image-like visual objects.",
    "id": "image-retrieval",
    "kind": "Task",
    "prefLabel": "Image Retrieval",
    "scopeNote": "Image retrieval may use text queries, metadata, visual features, embeddings, or example images. In industry it is often called image search or visual search. It is distinct from reverse image search, where the query itself is an image, and from cross-modal retrieval, where query and target modalities differ.",
    "terms": [
      {
        "community": "commerce-search",
        "label": "Visual Product Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Vision Search",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/image-retrieval/"
  },
  {
    "altLabels": [
      "Initiative for the Evaluation of XML Retrieval",
      "INitiative for the Evaluation of XML Retrieval"
    ],
    "definition": "The Initiative for the Evaluation of XML Retrieval, an evaluation program for retrieval from structured XML documents.",
    "id": "inex",
    "kind": "EvaluationProgram",
    "prefLabel": "INEX",
    "scopeNote": "INEX focused on structured-document and focused retrieval, including evaluation of systems that retrieve document elements or passages rather than only whole documents. Since 2012, INEX has been run as part of CLEF.",
    "uri": "https://id.searchplex.net/inex/"
  },
  {
    "altLabels": [
      "Filtering"
    ],
    "definition": "An IR task in which a system identifies items from a stream or changing collection that satisfy a relatively stable information need, profile, or topic.",
    "id": "information-filtering",
    "kind": "Task",
    "prefLabel": "Information Filtering",
    "scopeNote": "Information filtering emphasizes ongoing selection of relevant incoming items, often using persistent user profiles or topics. It is distinct from result filtering, which restricts already retrieved or retrievable results by constraints, and from ad hoc retrieval over a fixed collection.",
    "terms": [
      {
        "community": "academic",
        "label": "Adaptive Filtering",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/information-filtering/"
  },
  {
    "altLabels": [
      "Interactive Retrieval",
      "Interactive Information Retrieval"
    ],
    "definition": "A search task or setting in which users interact with the system over time by reformulating queries, examining results, applying controls, or providing feedback.",
    "id": "interactive-search",
    "kind": "Task",
    "prefLabel": "Interactive Search",
    "scopeNote": "Interactive search emphasizes the search session and user-system interaction rather than a single batch retrieval run. It may involve browsing, faceting, relevance feedback, query reformulation, and exploratory behavior. It is not the same as iterative retrieval: interaction is about user-system exchange, while iteration is about repeated retrieval attempts.",
    "uri": "https://id.searchplex.net/interactive-search/"
  },
  {
    "definition": "An index structure that maps terms or other lexical features to postings lists of documents or items in which those features occur.",
    "id": "inverted-index",
    "kind": "Method",
    "prefLabel": "Inverted Index",
    "scopeNote": "Inverted indexes are foundational infrastructure for lexical retrieval and keyword search. They enable efficient lookup of lexical evidence, but they are not the retrieval task, a ranking function, or a search engine by themselves.",
    "uri": "https://id.searchplex.net/inverted-index/"
  },
  {
    "altLabels": [
      "ir-datasets"
    ],
    "artifactKind": "Software",
    "definition": "A Python package and catalog that provides a common interface for accessing, documenting, and working with many information-retrieval datasets and benchmarks.",
    "id": "ir-datasets",
    "kind": "Artifact",
    "prefLabel": "ir_datasets",
    "scopeNote": "ir_datasets helps acquire and iterate over IR dataset components such as documents, queries, and relevance judgments. It is not itself a dataset, benchmark suite, qrels file, or evaluation metric.",
    "terms": [
      {
        "community": "academic",
        "label": "IR datasets catalog",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "IR dataset loader",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/ir-datasets/"
  },
  {
    "altLabels": [
      "Iterative Search"
    ],
    "definition": "A retrieval pattern in which retrieval is performed repeatedly, with later retrieval attempts informed by earlier results, feedback, reasoning, reformulation, or intermediate state.",
    "id": "iterative-retrieval",
    "kind": "Task",
    "prefLabel": "Iterative Retrieval",
    "scopeNote": "Iterative retrieval is broader than multi-hop retrieval and broader than agentic retrieval. The iteration may be user-driven, system-driven, feedback-driven, or agent-driven. It is distinct from interactive search, which emphasizes user-system interaction, and from single-shot retrieval, which performs one non-iterative retrieval attempt.",
    "terms": [
      {
        "community": "industry",
        "label": "Repeated Retrieval",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Retrieval Loop",
        "usage": "nearSynonym"
      },
      {
        "community": "ml-nlp",
        "label": "Iterative RAG",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/iterative-retrieval/"
  },
  {
    "altLabels": [
      "Keyword-Based Search",
      "Keyword Retrieval"
    ],
    "definition": "Search that retrieves results primarily by matching explicit query keywords, terms, or phrases against indexed text or metadata.",
    "id": "keyword-search",
    "kind": "GeneralConcept",
    "prefLabel": "Keyword Search",
    "scopeNote": "Keyword search is often used interchangeably with lexical or full-text search in industry. In Searchplex ID, keyword search names the practitioner- and user-facing search style centered on explicit query words or phrases, while lexical retrieval names the broader technical retrieval family based on lexical features.",
    "uri": "https://id.searchplex.net/keyword-search/"
  },
  {
    "altLabels": [
      "Known Item Search",
      "Known-Item Retrieval"
    ],
    "definition": "A search task in which the user is trying to find a particular known item rather than explore a topic or discover unknown relevant material.",
    "id": "known-item-search",
    "kind": "Task",
    "prefLabel": "Known-Item Search",
    "scopeNote": "Known-item search is common in library, web, enterprise, and catalog search. It is distinct from ad hoc topical retrieval and exploratory or interactive search, because the user has a specific target item in mind.",
    "terms": [
      {
        "community": "industry",
        "label": "Navigational Search",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/known-item-search/"
  },
  {
    "definition": "A retrieval approach in which queries and documents are independently encoded into fine-grained representations and their interactions are computed at query time rather than fully collapsed into a single vector beforehand.",
    "id": "late-interaction",
    "kind": "Representation",
    "prefLabel": "Late Interaction",
    "scopeNote": "ColBERT popularized token-level late interaction for neural retrieval. Late interaction is a retrieval paradigm rather than a specific model.",
    "uri": "https://id.searchplex.net/late-interaction/"
  },
  {
    "altLabels": [
      "Neural Sparse Retrieval"
    ],
    "definition": "Retrieval using sparse query and document representations whose feature activations or weights are learned from data.",
    "id": "learned-sparse-retrieval",
    "kind": "Representation",
    "prefLabel": "Learned Sparse Retrieval",
    "scopeNote": "Learned sparse systems commonly retain compatibility with inverted-index-style retrieval while learning lexical or vocabulary-space representations. Expansion-like activations may occur, but the representation family is not itself the query-expansion task.",
    "terms": [
      {
        "community": "academic",
        "label": "Sparse Neural Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/learned-sparse-retrieval/"
  },
  {
    "altLabels": [
      "LTR"
    ],
    "definition": "A family of supervised machine-learning methods for learning ranking functions from training data such as relevance judgments, preferences, clicks, or other labels.",
    "id": "learning-to-rank",
    "kind": "Method",
    "prefLabel": "Learning to Rank",
    "scopeNote": "Learning to Rank is a method family used to produce or improve rankings. It is not the ranking task itself, and it is not limited to any one model family or stage of a retrieval pipeline.",
    "uri": "https://id.searchplex.net/learning-to-rank/"
  },
  {
    "altLabels": [
      "Lexical Search"
    ],
    "definition": "Retrieval based primarily on explicit lexical features such as terms, tokens, phrases, or other symbolic features shared between queries and indexed items.",
    "id": "lexical-retrieval",
    "kind": "Representation",
    "prefLabel": "Lexical Retrieval",
    "scopeNote": "Lexical retrieval commonly uses inverted indexes and term-based scoring, and is often called keyword or full-text search in practitioner material. Searchplex ID treats it as the broader technical retrieval family, not as only the user-facing keyword-search style.",
    "uri": "https://id.searchplex.net/lexical-retrieval/"
  },
  {
    "altLabels": [
      "LLM Judge",
      "LLM-as-a-Judge"
    ],
    "definition": "A method that uses a large language model to evaluate, label, score, or compare outputs or query-item pairs.",
    "id": "llm-as-judge",
    "kind": "Method",
    "prefLabel": "LLM-as-Judge",
    "scopeNote": "In retrieval evaluation, LLM-as-judge methods may be used to generate or assist relevance assessments. They are not the same as human relevance judgments, qrels, or the relevance relationship being judged.",
    "uri": "https://id.searchplex.net/llm-as-judge/"
  },
  {
    "definition": "A late-interaction scoring operation that, for each query representation, selects the maximum similarity to document representations and aggregates those maxima into a relevance score.",
    "id": "maxsim",
    "kind": "Method",
    "prefLabel": "MaxSim",
    "scopeNote": "MaxSim is a scoring operation used by ColBERT-style late interaction. It is not itself a retrieval model.",
    "uri": "https://id.searchplex.net/maxsim/"
  },
  {
    "altLabels": [
      "MAP",
      "mAP"
    ],
    "definition": "A ranking evaluation metric that averages Average Precision values across a set of queries or topics.",
    "id": "mean-average-precision",
    "kind": "Metric",
    "prefLabel": "Mean Average Precision",
    "scopeNote": "MAP summarizes ranked retrieval effectiveness across topics using AP as the per-topic measure. It should not be confused with Average Precision for a single topic or with Precision@K at a fixed cutoff.",
    "uri": "https://id.searchplex.net/mean-average-precision/"
  },
  {
    "altLabels": [
      "Mean Reciprocal Rank"
    ],
    "definition": "A ranking evaluation metric that averages the reciprocal rank of the first relevant result across queries.",
    "id": "mrr",
    "kind": "Metric",
    "prefLabel": "MRR",
    "scopeNote": "MRR is most appropriate when the position of the first relevant result is the main objective.",
    "uri": "https://id.searchplex.net/mrr/"
  },
  {
    "artifactKind": "Benchmark",
    "definition": "A large-scale passage-retrieval and ranking benchmark derived from real search queries, a corpus of passages, and relevance or answer-based supervision.",
    "id": "ms-marco-passage-ranking",
    "kind": "Artifact",
    "prefLabel": "MS MARCO Passage Ranking",
    "scopeNote": "MS MARCO includes multiple datasets and tasks. This identity refers specifically to the passage-ranking resource or task family and should remain distinct from MS MARCO document ranking and other MS MARCO datasets.",
    "terms": [
      {
        "community": "academic",
        "label": "MS MARCO Passage",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "MS MARCO Passage Ranking Dataset",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/ms-marco-passage-ranking/"
  },
  {
    "altLabels": [
      "Massive Text Embedding Benchmark"
    ],
    "artifactKind": "Benchmark",
    "definition": "The Massive Text Embedding Benchmark, a benchmark suite for evaluating text embedding models across multiple tasks, datasets, and languages.",
    "id": "mteb",
    "kind": "Artifact",
    "prefLabel": "MTEB",
    "scopeNote": "MTEB includes retrieval and reranking tasks, but it is broader than IR evaluation alone because it also covers tasks such as classification, clustering, semantic textual similarity, summarization, pair classification, and bitext mining. It should not be treated as the same kind of benchmark suite as BEIR, which is focused on information retrieval.",
    "terms": [
      {
        "community": "ml-nlp",
        "label": "MTEB Leaderboard",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Embedding Benchmark",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/mteb/"
  },
  {
    "altLabels": [
      "Multihop Retrieval"
    ],
    "definition": "A retrieval task or pattern in which satisfying an information need requires multiple dependent retrieval steps or evidence gathered across several linked pieces of information.",
    "id": "multi-hop-retrieval",
    "kind": "Task",
    "prefLabel": "Multi-Hop Retrieval",
    "scopeNote": "Multi-hop retrieval is common in complex question answering and agentic RAG, where one retrieved result can inform the next retrieval step. It is distinct from multi-stage ranking, which describes a staged ranking architecture, and from retrieve-and-rerank, which is a particular pipeline pattern.",
    "terms": [
      {
        "community": "industry",
        "label": "Multi-Step Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/multi-hop-retrieval/"
  },
  {
    "altLabels": [
      "Cascaded Ranking"
    ],
    "definition": "Ranking architecture in which successive stages apply different retrieval or ranking methods to progressively refine results.",
    "id": "multi-stage-ranking",
    "kind": "Task",
    "prefLabel": "Multi-Stage Ranking",
    "scopeNote": "Use for staged ranking pipelines rather than a single ranking model.",
    "terms": [
      {
        "community": "industry",
        "label": "Ranking Cascade",
        "usage": "communityPreferred"
      }
    ],
    "uri": "https://id.searchplex.net/multi-stage-ranking/"
  },
  {
    "definition": "Retrieval in which a query, an indexed item, or both are represented by multiple vectors rather than a single vector.",
    "id": "multi-vector-retrieval",
    "kind": "Representation",
    "prefLabel": "Multi-Vector Retrieval",
    "scopeNote": "Late interaction commonly uses multi-vector representations, but multi-vector retrieval is a broader representational pattern and does not by itself imply ColBERT-style interaction.",
    "uri": "https://id.searchplex.net/multi-vector-retrieval/"
  },
  {
    "altLabels": [
      "Multilingual Information Retrieval",
      "MLIR"
    ],
    "definition": "Retrieval over collections, queries, users, or results involving more than one language.",
    "id": "multilingual-retrieval",
    "kind": "Task",
    "prefLabel": "Multilingual Retrieval",
    "scopeNote": "Multilingual retrieval is a broader setting than cross-language retrieval. It may include multiple monolingual indexes, mixed-language collections, multilingual result lists, language-specific processing, or cross-language retrieval where query and result languages differ.",
    "uri": "https://id.searchplex.net/multilingual-retrieval/"
  },
  {
    "altLabels": [
      "Multimodal Search",
      "Multimedia Retrieval",
      "Multimedia Information Retrieval"
    ],
    "definition": "Retrieval involving information needs, indexed items, representations, or results that span more than one data modality, such as text, images, audio, video, tables, or structured metadata.",
    "id": "multimodal-retrieval",
    "kind": "Task",
    "prefLabel": "Multimodal Retrieval",
    "scopeNote": "Multimodal retrieval is a broad family. It includes settings where multiple modalities are searched together, represented jointly, or used in query and result interpretation. It is distinct from cross-modal retrieval, where the query and target results are in different modalities, and from image retrieval, where the retrieved objects are images.",
    "terms": [
      {
        "community": "vendor",
        "label": "Multimodal Vector Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Neural Multimodal Search",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/multimodal-retrieval/"
  },
  {
    "altLabels": [
      "NER"
    ],
    "definition": "Identifying spans in text that denote named entities or entity-like mentions and assigning them entity types.",
    "id": "named-entity-recognition",
    "kind": "Task",
    "prefLabel": "Named Entity Recognition",
    "scopeNote": "In search systems, named entity recognition may be applied to queries or documents as part of query understanding, entity extraction, routing, or catalog matching. It identifies and types mentions; it does not by itself link them to canonical entities.",
    "uri": "https://id.searchplex.net/named-entity-recognition/"
  },
  {
    "altLabels": [
      "NQ"
    ],
    "artifactKind": "Dataset",
    "definition": "A question-answering dataset based on natural user questions, used in retrieval and open-domain question-answering evaluation.",
    "id": "natural-questions",
    "kind": "Artifact",
    "prefLabel": "Natural Questions",
    "scopeNote": "Natural Questions appears in BEIR as an open-domain QA retrieval dataset. It should be distinguished from question answering as a task and from BEIR as a benchmark suite.",
    "uri": "https://id.searchplex.net/natural-questions/"
  },
  {
    "altLabels": [
      "nDCG",
      "NDCG",
      "Normalized Discounted Cumulative Gain"
    ],
    "definition": "A ranking evaluation metric that discounts graded relevance by result position and normalizes by the ideal ranking.",
    "id": "ndcg",
    "kind": "Metric",
    "prefLabel": "nDCG@K",
    "scopeNote": "nDCG is commonly used when relevance is graded and rank position matters. It should not be confused with recall-oriented cutoff metrics.",
    "uri": "https://id.searchplex.net/ndcg/"
  },
  {
    "altLabels": [
      "NF Corpus"
    ],
    "artifactKind": "Dataset",
    "definition": "A biomedical and nutrition-focused information-retrieval test collection used for evaluating retrieval systems.",
    "id": "nfcorpus",
    "kind": "Artifact",
    "prefLabel": "NFCorpus",
    "scopeNote": "NFCorpus is one of the component evaluation datasets included in BEIR. It should be distinguished from BEIR as a benchmark suite and from qrels as the relevance-judgment component of the dataset.",
    "uri": "https://id.searchplex.net/nfcorpus/"
  },
  {
    "altLabels": [
      "NII Testbeds and Community for Information Access Research"
    ],
    "definition": "The NII Testbeds and Community for Information Access Research, an evaluation program and conference series for information-access technologies.",
    "id": "ntcir",
    "kind": "EvaluationProgram",
    "prefLabel": "NTCIR",
    "scopeNote": "NTCIR has supported shared evaluation tasks for information retrieval, cross-language information access, question answering, summarization, patent retrieval, and related information-access problems, with strong historical focus on East Asian languages and contexts.",
    "uri": "https://id.searchplex.net/ntcir/"
  },
  {
    "definition": "A retrieval task in which the retrievable units are passages or passage-like text segments rather than whole documents or other larger items.",
    "id": "passage-retrieval",
    "kind": "Task",
    "prefLabel": "Passage Retrieval",
    "scopeNote": "Passage retrieval is common in open-domain question answering, RAG, and passage-ranking benchmarks. It is distinct from chunking, which creates or selects the units, and from passage ranking, which emphasizes ordering a set of passage candidates.",
    "terms": [
      {
        "community": "academic",
        "label": "Passage Ranking",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Chunk Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/passage-retrieval/"
  },
  {
    "artifactKind": "Software",
    "definition": "Performance-optimized Late Interaction Driver, an efficient retrieval engine for accelerating late-interaction search, particularly with ColBERTv2-style representations.",
    "id": "plaid",
    "kind": "Artifact",
    "prefLabel": "PLAID",
    "scopeNote": "PLAID is a serving and retrieval engine and optimization strategy, not a successor model to ColBERTv2.",
    "uri": "https://id.searchplex.net/plaid/"
  },
  {
    "definition": "An evaluation-collection construction method in which documents retrieved by multiple systems or runs are combined into a pool for relevance assessment.",
    "id": "pooling",
    "kind": "Method",
    "prefLabel": "Pooling",
    "scopeNote": "Pooling is central to many TREC-style test collections. It is a way to choose items to judge, not the qrels file itself and not a relevance metric.",
    "uri": "https://id.searchplex.net/pooling/"
  },
  {
    "definition": "A bias in retrieval, ranking, or recommendation where already popular items receive disproportionate exposure or preference.",
    "id": "popularity-bias",
    "kind": "Phenomenon",
    "prefLabel": "Popularity Bias",
    "scopeNote": "Popularity bias can affect recommendations, search ranking, training data, and implicit feedback loops. It is not the same as relevance, although popularity can be used as one relevance or utility signal.",
    "uri": "https://id.searchplex.net/popularity-bias/"
  },
  {
    "altLabels": [
      "Rank Position Bias"
    ],
    "definition": "A user-behavior bias in which items shown at higher positions receive more attention, clicks, or interaction because of their position, independent of their true relevance.",
    "id": "position-bias",
    "kind": "Phenomenon",
    "prefLabel": "Position Bias",
    "scopeNote": "Position bias affects click logs, implicit feedback, learning-to-rank data, and evaluation based on user interactions. It is closely related to presentation bias but should not be treated as relevance itself.",
    "terms": [
      {
        "community": "academic",
        "label": "Presentation Bias",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Examination Bias",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/position-bias/"
  },
  {
    "altLabels": [
      "Precision-Oriented Metric",
      "Precision-Based Metric"
    ],
    "definition": "A family of retrieval or ranking evaluation metrics that emphasize the proportion of retrieved or ranked results that are relevant.",
    "id": "precision-metric",
    "kind": "Metric",
    "prefLabel": "Precision Metric",
    "scopeNote": "Precision-oriented metrics focus on result quality among returned items, often at a cutoff or across relevant ranks. They are distinct from recall-oriented metrics, which emphasize coverage of the relevant set.",
    "uri": "https://id.searchplex.net/precision-metric/"
  },
  {
    "definition": "An evaluation metric measuring the fraction of the top K retrieved or ranked results that are relevant.",
    "id": "precision-at-k",
    "kind": "Metric",
    "prefLabel": "Precision@K",
    "scopeNote": "Precision@K emphasizes the quality of the first K results. It is distinct from Recall@K, which measures how much of the relevant set appears within the top K.",
    "uri": "https://id.searchplex.net/precision-at-k/"
  },
  {
    "altLabels": [
      "PRF"
    ],
    "definition": "A relevance-feedback method that treats top-ranked results from an initial retrieval as if they were relevant and uses them to modify a subsequent retrieval query or model.",
    "id": "pseudo-relevance-feedback",
    "kind": "Method",
    "prefLabel": "Pseudo-Relevance Feedback",
    "scopeNote": "Pseudo-relevance feedback removes the requirement for explicit user relevance judgments. It is commonly used to derive query expansion terms or updated query models.",
    "uri": "https://id.searchplex.net/pseudo-relevance-feedback/"
  },
  {
    "altLabels": [
      "Qrel",
      "Query Relevance Judgments"
    ],
    "artifactKind": "Dataset",
    "definition": "A structured set or file of query-document relevance judgments used to evaluate retrieval runs.",
    "id": "qrels",
    "kind": "Artifact",
    "prefLabel": "Qrels",
    "scopeNote": "Qrels is TREC-style terminology for query relevance judgments. A qrels file usually records relevance labels for query-item pairs and is consumed by evaluation tools such as trec_eval. Qrels are not evaluation metrics and are not the same as the full test collection.",
    "uri": "https://id.searchplex.net/qrels/"
  },
  {
    "altLabels": [
      "Query Auto Completion",
      "QAC"
    ],
    "definition": "Predicting and ranking complete search queries from a partially typed query prefix.",
    "id": "query-autocomplete",
    "kind": "Task",
    "prefLabel": "Query Autocomplete",
    "scopeNote": "Query autocomplete is prefix-conditioned: the user has begun typing a query and the system proposes likely completions. It is distinct from query suggestion, which may recommend related or subsequent queries without being limited to completing a current prefix.",
    "terms": [
      {
        "community": "industry",
        "label": "Typeahead",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Search Autocomplete",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-autocomplete/"
  },
  {
    "altLabels": [
      "Web Query Classification",
      "Query Categorization"
    ],
    "definition": "Assigning a search query to one or more predefined classes, categories, intents, verticals, or other labels.",
    "id": "query-classification",
    "kind": "Task",
    "prefLabel": "Query Classification",
    "scopeNote": "Query classification is broader than query intent classification. Depending on the system, the target classes may represent topic, product category, vertical, navigational intent, commercial intent, routing class, or another operational taxonomy.",
    "terms": [
      {
        "community": "academic",
        "label": "Query Intent Classification",
        "usage": "nearSynonym"
      },
      {
        "community": "commerce-search",
        "label": "Category Prediction",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-classification/"
  },
  {
    "altLabels": [
      "Question Decomposition"
    ],
    "definition": "Breaking a complex query or information need into smaller subqueries or subproblems that can be retrieved or answered separately.",
    "id": "query-decomposition",
    "kind": "Task",
    "prefLabel": "Query Decomposition",
    "scopeNote": "Query decomposition is common in multi-hop question answering, agentic retrieval, and complex search tasks. It is distinct from query reformulation, which changes a query, and from query routing, which decides where or how a query should be handled.",
    "terms": [
      {
        "community": "industry",
        "label": "Subquery Generation",
        "usage": "nearSynonym"
      },
      {
        "community": "ml-nlp",
        "label": "Sub-Question Generation",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-decomposition/"
  },
  {
    "definition": "A failure mode in which query modification or feedback moves the retrieval representation away from the user's intended information need.",
    "id": "query-drift",
    "kind": "Phenomenon",
    "prefLabel": "Query Drift",
    "scopeNote": "Query drift is often discussed around query expansion and pseudo-relevance feedback, where added evidence can change the effective query meaning.",
    "uri": "https://id.searchplex.net/query-drift/"
  },
  {
    "altLabels": [
      "Automatic Query Expansion",
      "AQE"
    ],
    "definition": "Enriching a query with additional retrieval evidence intended to improve retrieval effectiveness.",
    "id": "query-expansion",
    "kind": "Task",
    "prefLabel": "Query Expansion",
    "scopeNote": "Covers expansion of the query representation rather than ranking-stage relevance estimation.",
    "uri": "https://id.searchplex.net/query-expansion/"
  },
  {
    "definition": "The set of operations that analyze, interpret, transform, or otherwise prepare a query for execution by a retrieval system.",
    "id": "query-processing",
    "kind": "GeneralConcept",
    "prefLabel": "Query Processing",
    "scopeNote": "Query processing is an umbrella term. Depending on the system, it may include normalization, query understanding, classification, expansion, rewriting, field assignment, routing, and retrieval planning.",
    "uri": "https://id.searchplex.net/query-processing/"
  },
  {
    "definition": "Changing an existing query in order to better express or satisfy an information need.",
    "id": "query-reformulation",
    "kind": "GeneralConcept",
    "prefLabel": "Query Reformulation",
    "scopeNote": "In interactive IR, query reformulation often describes a user modifying a query during a search session. The term is also used for automatic system-driven transformations, where it overlaps with query rewriting.",
    "uri": "https://id.searchplex.net/query-reformulation/"
  },
  {
    "altLabels": [
      "Automatic Query Relaxation"
    ],
    "definition": "Transforming a query or its matching requirements to make retrieval less restrictive, often to avoid zero or low-result searches.",
    "id": "query-relaxation",
    "kind": "Task",
    "prefLabel": "Query Relaxation",
    "scopeNote": "Query relaxation may drop, down-weight, or loosen terms, constraints, exact matches, or matching criteria. It is often discussed as the counterpart of query expansion: expansion adds evidence, while relaxation weakens requirements.",
    "terms": [
      {
        "community": "academic",
        "label": "AQR",
        "usage": "abbreviation"
      },
      {
        "community": "industry",
        "label": "Query Broadening",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Term Dropping",
        "usage": "nearSynonym"
      },
      {
        "community": "commerce-search",
        "label": "Partial Matching",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-relaxation/"
  },
  {
    "definition": "Automatically transforming an input query into another query or query representation intended to improve retrieval or execution.",
    "id": "query-rewriting",
    "kind": "Task",
    "prefLabel": "Query Rewriting",
    "scopeNote": "Depending on the system, rewriting may include spelling correction, normalization, expansion, reduction, phrase construction, structural changes, filters, or field-aware transformations. Usage is broader in some search-engine documentation than in academic IR.",
    "uri": "https://id.searchplex.net/query-rewriting/"
  },
  {
    "definition": "Choosing where or how a query should be handled, such as selecting a source, index, retriever, tool, model, strategy, or no-retrieval path.",
    "id": "query-routing",
    "kind": "Task",
    "prefLabel": "Query Routing",
    "scopeNote": "Query routing uses information about the query to make a control-flow decision. In distributed IR and federated search, it often means selecting sources or servers. In RAG and agentic systems, it may mean selecting among retrievers, tools, indexes, models, or retrieval strategies.",
    "terms": [
      {
        "community": "academic",
        "label": "Source Selection",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Resource Selection",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Router",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/query-routing/"
  },
  {
    "definition": "Dividing a search query into meaningful single-token or multi-token spans, optionally with broad semantic or operational types.",
    "id": "query-segmentation",
    "kind": "Task",
    "prefLabel": "Query Segmentation",
    "scopeNote": "Query segmentation identifies meaningful spans in a query. It is not the same as named entity recognition, because not every useful query segment is a named entity or entity-like mention.",
    "uri": "https://id.searchplex.net/query-segmentation/"
  },
  {
    "altLabels": [
      "Search Query Suggestion"
    ],
    "definition": "Generating and ranking related, alternative, or subsequent search queries intended to help a user continue, refine, or redirect a search.",
    "id": "query-suggestion",
    "kind": "Task",
    "prefLabel": "Query Suggestion",
    "scopeNote": "Query suggestion may use query logs, session context, click behavior, current results, or user context. It is distinct from query autocomplete, which completes a partially typed prefix.",
    "terms": [
      {
        "community": "industry",
        "label": "Recommended Searches",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-suggestion/"
  },
  {
    "definition": "Changing a query representation before retrieval or ranking while preserving or clarifying the user's information need.",
    "id": "query-transformation",
    "kind": "Task",
    "prefLabel": "Query Transformation",
    "scopeNote": "Broader than query expansion; includes rewriting, normalization, decomposition, and related transformations.",
    "terms": [
      {
        "community": "industry",
        "label": "Query Rewrite",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/query-transformation/"
  },
  {
    "altLabels": [
      "Query Interpretation"
    ],
    "definition": "Interpreting a user's query to infer intent, structure, entities, constraints, and useful retrieval signals.",
    "id": "query-understanding",
    "kind": "GeneralConcept",
    "prefLabel": "Query Understanding",
    "scopeNote": "Covers analysis before retrieval rather than the retrieval or ranking stages themselves.",
    "terms": [
      {
        "community": "industry",
        "label": "Query Analysis",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/query-understanding/"
  },
  {
    "altLabels": [
      "QA",
      "Question-Answering"
    ],
    "definition": "A task in which a system returns an answer, answer-bearing snippet, or response to a natural-language question rather than only a ranked list of documents.",
    "id": "question-answering",
    "kind": "Task",
    "prefLabel": "Question Answering",
    "scopeNote": "Question answering may use retrieval internally, especially in open-domain QA and RAG systems, but the task objective is answer production. TREC question-answering evaluations historically emphasized returning short answers or snippets for fact-based questions.",
    "terms": [
      {
        "community": "industry",
        "label": "Answer Search",
        "usage": "nearSynonym"
      },
      {
        "community": "ml-nlp",
        "label": "Answer Generation",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/question-answering/"
  },
  {
    "definition": "A family of fusion methods that combine multiple ranked lists using rank positions or rank-derived evidence.",
    "id": "rank-fusion",
    "kind": "Method",
    "prefLabel": "Rank Fusion",
    "scopeNote": "Rank fusion does not necessarily require comparable raw scores across systems.",
    "uri": "https://id.searchplex.net/rank-fusion/"
  },
  {
    "altLabels": [
      "Rank-Aware Metric",
      "Ranking Metric"
    ],
    "definition": "A family of evaluation metrics whose value depends on where relevant items appear in the ranked result list.",
    "id": "rank-sensitive-metric",
    "kind": "Metric",
    "prefLabel": "Rank-Sensitive Metric",
    "scopeNote": "Rank-sensitive metrics distinguish systems that place relevant results earlier from systems that retrieve the same relevant items lower in the ranking. Examples include MRR, Average Precision, MAP, and nDCG.",
    "uri": "https://id.searchplex.net/rank-sensitive-metric/"
  },
  {
    "altLabels": [
      "Relevance Ranking"
    ],
    "definition": "Ordering retrieved or candidate items according to estimated relevance, utility, or another objective.",
    "id": "ranking",
    "kind": "Task",
    "prefLabel": "Ranking",
    "scopeNote": "Ranking is about ordering; it is distinct from candidate generation, filtering, or pure matching.",
    "uri": "https://id.searchplex.net/ranking/"
  },
  {
    "altLabels": [
      "Recall-Oriented Metric",
      "Recall-Based Metric"
    ],
    "definition": "A family of retrieval or ranking evaluation metrics that emphasize coverage of relevant items.",
    "id": "recall-metric",
    "kind": "Metric",
    "prefLabel": "Recall Metric",
    "scopeNote": "Recall-oriented metrics are especially important for candidate retrieval, high-recall retrieval, and first-stage retrieval, where the goal is to avoid losing relevant items before later ranking stages.",
    "uri": "https://id.searchplex.net/recall-metric/"
  },
  {
    "definition": "An evaluation metric measuring the fraction of relevant items retrieved within the top K results.",
    "id": "recall-at-k",
    "kind": "Metric",
    "prefLabel": "Recall@K",
    "scopeNote": "Recall@K emphasizes coverage of relevant items at a cutoff and is often important when evaluating candidate retrieval or first-stage retrieval.",
    "uri": "https://id.searchplex.net/recall-at-k/"
  },
  {
    "altLabels": [
      "RRF"
    ],
    "definition": "A rank-fusion method that combines multiple ranked result lists using reciprocal functions of item rank.",
    "id": "reciprocal-rank-fusion",
    "kind": "Method",
    "prefLabel": "Reciprocal Rank Fusion",
    "scopeNote": "RRF is one specific rank-fusion method. It should not be used as a synonym for rank fusion or hybrid retrieval generally.",
    "uri": "https://id.searchplex.net/reciprocal-rank-fusion/"
  },
  {
    "definition": "The degree to which a retrieved item satisfies an information need, query, topic, or evaluation criterion.",
    "id": "relevance",
    "kind": "GeneralConcept",
    "prefLabel": "Relevance",
    "scopeNote": "In information retrieval evaluation, relevance may be binary or graded and is usually operationalized through relevance judgments. Relevance is the evaluation target, not itself a metric or a qrels file.",
    "uri": "https://id.searchplex.net/relevance/"
  },
  {
    "altLabels": [
      "Relevance Assessment",
      "Relevance Judgments",
      "Relevance Judgements"
    ],
    "definition": "An assessment assigning a relevance label or level to an item with respect to a query, topic, or information need.",
    "id": "relevance-judgment",
    "kind": "GeneralConcept",
    "prefLabel": "Relevance Judgment",
    "scopeNote": "Relevance judgments are commonly produced by human assessors in Cranfield-style and TREC-style evaluation, though automatic or synthetic judgments may also be studied. Judgments may be binary or graded and are often collected into qrels.",
    "uri": "https://id.searchplex.net/relevance-judgment/"
  },
  {
    "altLabels": [
      "Re-ranking"
    ],
    "definition": "Re-scoring or reordering a bounded candidate set using a subsequent ranking model or additional evidence.",
    "id": "reranking",
    "kind": "Task",
    "prefLabel": "Reranking",
    "scopeNote": "Reranking normally operates over candidates already selected by an earlier retrieval or ranking stage and can therefore use more expensive models.",
    "terms": [
      {
        "community": "industry",
        "label": "Second-Stage Ranking",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/reranking/"
  },
  {
    "altLabels": [
      "Search Filtering"
    ],
    "definition": "A method for restricting search or retrieval results to items that satisfy explicit constraints, predicates, or selected attribute values.",
    "id": "result-filtering",
    "kind": "Method",
    "prefLabel": "Result Filtering",
    "scopeNote": "Result filtering may be applied before retrieval, during retrieval, after retrieval, or through a user interface. It is distinct from faceted search, which presents structured dimensions for exploratory narrowing, and from information filtering, which is a standing-query task over incoming information.",
    "uri": "https://id.searchplex.net/result-filtering/"
  },
  {
    "definition": "Combining outputs from multiple retrieval or ranking processes into a single result set or ranking.",
    "id": "result-fusion",
    "kind": "GeneralConcept",
    "prefLabel": "Result Fusion",
    "scopeNote": "Result fusion is an umbrella concept that may operate on ranks, scores, probabilities, or other evidence.",
    "uri": "https://id.searchplex.net/result-fusion/"
  },
  {
    "definition": "Selecting items from a collection that are estimated to be relevant or useful for a query or information need.",
    "id": "retrieval",
    "kind": "Task",
    "prefLabel": "Retrieval",
    "scopeNote": "Retrieval may itself produce scores or an initial ordering. In multi-stage systems, it is often distinguished from more expensive downstream ranking or reranking.",
    "uri": "https://id.searchplex.net/retrieval/"
  },
  {
    "definition": "Choosing a sequence or strategy of retrieval actions for satisfying an information need, often before or during an agentic retrieval loop.",
    "id": "retrieval-planning",
    "kind": "Task",
    "prefLabel": "Retrieval Planning",
    "scopeNote": "Retrieval planning may include deciding whether retrieval is needed, decomposing an information need, choosing tools or indexes, ordering retrieval steps, and deciding when to stop. It is distinct from query routing, which chooses where or how a query should be handled, and from query decomposition, which breaks an information need into subqueries.",
    "terms": [
      {
        "community": "industry",
        "label": "Search Planning",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Retrieval Strategy Planning",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/retrieval-planning/"
  },
  {
    "altLabels": [
      "RAG"
    ],
    "definition": "A generation pattern in which retrieved external information is supplied to a generative model to support, condition, or ground its output.",
    "id": "rag",
    "kind": "GeneralConcept",
    "prefLabel": "Retrieval-Augmented Generation",
    "scopeNote": "RAG systems commonly combine indexing, retrieval, reranking, prompt or context construction, and generation. RAG uses retrieval, but it is not identical to retrieval, dense retrieval, semantic search, or grounding.",
    "uri": "https://id.searchplex.net/rag/"
  },
  {
    "altLabels": [
      "Retrieve-Then-Rerank"
    ],
    "definition": "A multi-stage search pattern in which an efficient retrieval stage first produces a bounded candidate set and a subsequent, typically more expensive model rescoring or reorders those candidates.",
    "id": "retrieve-and-rerank",
    "kind": "Task",
    "prefLabel": "Retrieve-and-Rerank",
    "scopeNote": "A common neural implementation uses a bi- or dual-encoder for candidate retrieval followed by a Cross-Encoder for reranking, but retrieve-and-rerank does not require those particular model architectures. The first stage may instead be lexical, sparse, hybrid, or another retrieval method.",
    "uri": "https://id.searchplex.net/retrieve-and-rerank/"
  },
  {
    "altLabels": [
      "Reverse Image Retrieval",
      "Search by Image",
      "Image-to-Image Search",
      "Similar Image Search"
    ],
    "definition": "An image-retrieval task in which the user provides an image, image URL, or visual example as the query to find visually similar, related, source, duplicate, or derivative images.",
    "id": "reverse-image-search",
    "kind": "Task",
    "prefLabel": "Reverse Image Search",
    "scopeNote": "Reverse image search is often implemented with content-based image retrieval techniques, visual embeddings, nearest-neighbor search, or perceptual similarity methods. It is distinct from image retrieval in general and from text-to-image cross-modal retrieval.",
    "terms": [
      {
        "community": "academic",
        "label": "Content-Based Image Retrieval",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "CBIR",
        "usage": "abbreviation"
      },
      {
        "community": "industry",
        "label": "Query by Image",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Image Query Search",
        "usage": "informal"
      },
      {
        "community": "academic",
        "label": "Query by Example",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/reverse-image-search/"
  },
  {
    "altLabels": [
      "Information Routing"
    ],
    "definition": "An IR task in which standing queries or profiles are used to retrieve relevant items from a stream or changing supply of new documents.",
    "id": "routing",
    "kind": "Task",
    "prefLabel": "Routing",
    "scopeNote": "Routing is a classic TREC task formulation associated with standing information needs and incoming documents. It is distinct from ad hoc retrieval, where new topics are run against a known collection, and from query routing, where a live query is directed to a source, model, retriever, or handling strategy.",
    "terms": [
      {
        "community": "academic",
        "label": "Selective Dissemination of Information",
        "usage": "historical"
      },
      {
        "community": "academic",
        "label": "SDI",
        "usage": "abbreviation"
      }
    ],
    "uri": "https://id.searchplex.net/routing/"
  },
  {
    "artifactKind": "Dataset",
    "definition": "A scientific claim verification and evidence-retrieval dataset used to evaluate retrieval and fact-verification systems.",
    "id": "scifact",
    "kind": "Artifact",
    "prefLabel": "SciFact",
    "scopeNote": "SciFact is commonly used as a BEIR component dataset for scientific fact-checking retrieval. It is an evaluation dataset, not the BEIR benchmark suite itself.",
    "uri": "https://id.searchplex.net/scifact/"
  },
  {
    "definition": "A family of fusion methods that combine numerical scores produced by multiple retrieval or ranking systems.",
    "id": "score-fusion",
    "kind": "Method",
    "prefLabel": "Score Fusion",
    "scopeNote": "Score fusion commonly requires normalization, calibration, weighting, or another way to make scores from different systems meaningfully combinable.",
    "uri": "https://id.searchplex.net/score-fusion/"
  },
  {
    "altLabels": [
      "Score Normalisation",
      "Score Scaling"
    ],
    "definition": "A method for transforming scores from one or more retrieval or ranking systems onto a scale intended to make them comparable or combinable.",
    "id": "score-normalization",
    "kind": "Method",
    "prefLabel": "Score Normalization",
    "scopeNote": "Score normalization is commonly used before weighted score fusion in hybrid retrieval. It is a score-preparation step, not the fusion method itself, and it is distinct from rank-based methods such as Reciprocal Rank Fusion.",
    "terms": [
      {
        "community": "industry",
        "label": "Linear Normalization",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Min-Max Normalization",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Z-Score Normalization",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/score-normalization/"
  },
  {
    "altLabels": [
      "Search As You Type",
      "Typeahead Search"
    ],
    "definition": "A search interaction pattern in which retrieval is executed and results are updated continuously as the user types a query.",
    "id": "search-as-you-type",
    "kind": "Task",
    "prefLabel": "Search-as-You-Type",
    "scopeNote": "Search-as-you-type returns or updates actual search results during query entry. It is often combined with autocomplete and query suggestions, but it is distinct from query autocomplete, which proposes query completions, and from query suggestion, which recommends related or alternative queries.",
    "terms": [
      {
        "community": "industry",
        "label": "Instant Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Live Search",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "As-You-Type Search",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/search-as-you-type/"
  },
  {
    "definition": "A broad term for search intended to retrieve according to meaning or semantic relatedness rather than relying only on literal lexical overlap.",
    "id": "semantic-search",
    "kind": "GeneralConcept",
    "prefLabel": "Semantic Search",
    "scopeNote": "Semantic search does not denote one retrieval algorithm or representation. Dense retrieval, learned sparse retrieval, entity-based methods, query expansion, and hybrid retrieval may all contribute to systems described as semantic search.",
    "terms": [
      {
        "community": "industry",
        "label": "Semantic Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/semantic-search/"
  },
  {
    "altLabels": [
      "Search Engine Results Page",
      "Search Results Page"
    ],
    "definition": "A search engine results page displayed in response to a user query.",
    "id": "serp",
    "kind": "GeneralConcept",
    "prefLabel": "SERP",
    "scopeNote": "SERP refers to the result presentation surface, not the retrieval or ranking task itself. A SERP may include organic results, sponsored results, snippets, facets, answer boxes, entity panels, or other search features.",
    "uri": "https://id.searchplex.net/serp/"
  },
  {
    "altLabels": [
      "Search Session"
    ],
    "definition": "A search task or setting that considers a sequence of related user interactions within a search session rather than a single isolated query.",
    "id": "session-search",
    "kind": "Task",
    "prefLabel": "Session Search",
    "scopeNote": "Session search may use earlier queries, clicks, reformulations, dwell behavior, or other session context to support later retrieval. It is broader than conversational search because a session need not be dialogue-based.",
    "uri": "https://id.searchplex.net/session-search/"
  },
  {
    "altLabels": [
      "One-Shot Retrieval",
      "Single Shot Retrieval",
      "One Shot Retrieval"
    ],
    "definition": "A non-iterative retrieval setup in which a single query, topic, or information need is submitted and a result set or ranked list is produced without dependent follow-up retrieval steps.",
    "id": "single-shot-retrieval",
    "kind": "Task",
    "prefLabel": "Single-Shot Retrieval",
    "scopeNote": "Single-shot retrieval is useful as a contrast with multi-hop retrieval, interactive search, session search, and agentic retrieval. The phrase one-shot retrieval is sometimes used informally, but single-shot avoids confusion with one-shot learning in machine learning.",
    "terms": [
      {
        "community": "industry",
        "label": "Single-Turn Retrieval",
        "usage": "nearSynonym"
      },
      {
        "community": "academic",
        "label": "Single-Query Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/single-shot-retrieval/"
  },
  {
    "definition": "Retrieval in which each query, indexed item, passage, document, or chunk is represented by one vector for similarity comparison.",
    "id": "single-vector-retrieval",
    "kind": "Representation",
    "prefLabel": "Single-Vector Retrieval",
    "scopeNote": "Single-vector retrieval is often the default format in dense retrieval and vector-search systems, but it is not identical to dense retrieval, vector search, semantic search, or the dual-encoder architecture that commonly produces such vectors.",
    "terms": [
      {
        "community": "academic",
        "label": "Single-Vector Dense Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/single-vector-retrieval/"
  },
  {
    "definition": "Retrieval using high-dimensional representations in which most feature dimensions have zero weight.",
    "id": "sparse-retrieval",
    "kind": "Representation",
    "prefLabel": "Sparse Retrieval",
    "scopeNote": "Traditional lexical representations and learned sparse representations can both be sparse. Sparse retrieval therefore should not be used as a synonym for learned sparse retrieval.",
    "uri": "https://id.searchplex.net/sparse-retrieval/"
  },
  {
    "altLabels": [
      "Query Spelling Correction",
      "Spell Correction"
    ],
    "definition": "Detecting and correcting misspelled or mistyped query terms in order to improve query interpretation or retrieval.",
    "id": "spelling-correction",
    "kind": "Task",
    "prefLabel": "Spelling Correction",
    "scopeNote": "In search, spelling correction may automatically rewrite a query, offer a 'did you mean' suggestion, or decide to keep the original query. It is distinct from autocomplete and query suggestion because it responds to a completed or submitted query rather than predicting a prefix or recommending related searches.",
    "terms": [
      {
        "community": "industry",
        "label": "Did You Mean",
        "usage": "informal"
      },
      {
        "community": "commerce-search",
        "label": "Typo Correction",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/spelling-correction/"
  },
  {
    "altLabels": [
      "Sparse Lexical and Expansion Model"
    ],
    "artifactKind": "Model",
    "definition": "A family of learned sparse neural retrieval models that produce sparse vocabulary-space representations for queries and documents.",
    "id": "splade",
    "kind": "Artifact",
    "prefLabel": "SPLADE",
    "scopeNote": "SPLADE learns sparse expansion-like activations and is commonly used for first-stage retrieval or ranking. It is an implementation/model family within learned sparse retrieval, not a synonym for either learned sparse retrieval generally or query expansion.",
    "terms": [
      {
        "community": "academic",
        "label": "Sparse Expansion Model",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/splade/"
  },
  {
    "altLabels": [
      "Hit@K",
      "Hits@K",
      "Hit Rate@K"
    ],
    "definition": "A cutoff evaluation metric indicating whether at least one relevant item appears within the top K results.",
    "id": "success-at-k",
    "kind": "Metric",
    "prefLabel": "Success@K",
    "scopeNote": "Success@K is a binary per-query metric often averaged across queries. It is commonly called Hit@K or Hits@K in recommendation, RAG, and retrieval evaluation settings.",
    "uri": "https://id.searchplex.net/success-at-k/"
  },
  {
    "altLabels": [
      "IR Evaluation Dataset",
      "Retrieval Evaluation Dataset",
      "Search Evaluation Dataset",
      "Evaluation Dataset"
    ],
    "artifactKind": "Benchmark",
    "definition": "An information-retrieval evaluation resource consisting of a document or item collection, a set of topics or queries, and relevance judgments.",
    "id": "test-collection",
    "kind": "Artifact",
    "prefLabel": "Test Collection",
    "scopeNote": "In Cranfield-style and TREC-style evaluation, a test collection provides the fixed materials needed for offline retrieval evaluation. It is broader than qrels alone and may be packaged as part of a benchmark suite.",
    "terms": [
      {
        "community": "industry",
        "label": "Benchmark Dataset",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Eval Dataset",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/test-collection/"
  },
  {
    "altLabels": [
      "Term Frequency-Inverse Document Frequency",
      "Term Frequency–Inverse Document Frequency"
    ],
    "definition": "A term-weighting method that combines term frequency with inverse document frequency to weight terms in documents, queries, or vector-space representations.",
    "id": "tf-idf",
    "kind": "Method",
    "prefLabel": "TF-IDF",
    "scopeNote": "TF-IDF is commonly associated with vector space retrieval and classical lexical baselines. It is not the same as BM25, lexical retrieval generally, or sparse retrieval generally.",
    "uri": "https://id.searchplex.net/tf-idf/"
  },
  {
    "altLabels": [
      "Tip-of-the-Tongue Retrieval",
      "Tip-of-the-Tongue Known-Item Retrieval",
      "ToT Search",
      "ToT Retrieval",
      "ToT-KIR"
    ],
    "definition": "A known-item search task in which the searcher is trying to re-find or identify a previously encountered item but cannot recall a precise name, title, identifier, or other direct lookup key.",
    "id": "tip-of-the-tongue-search",
    "kind": "Task",
    "prefLabel": "Tip-of-the-Tongue Search",
    "scopeNote": "Tip-of-the-tongue search queries often contain partial memories, contextual clues, uncertain descriptions, comparisons, or remembered attributes rather than exact identifiers. It is a specialized form of known-item search and is distinct from exact match search and ordinary ad hoc topical retrieval.",
    "terms": [
      {
        "community": "industry",
        "label": "Forgotten Item Search",
        "usage": "informal"
      },
      {
        "community": "general",
        "label": "Find the Thing I Can't Name",
        "usage": "informal"
      }
    ],
    "uri": "https://id.searchplex.net/tip-of-the-tongue-search/"
  },
  {
    "altLabels": [
      "Text REtrieval Conference"
    ],
    "definition": "The Text REtrieval Conference, an evaluation program and workshop series supporting large-scale information-retrieval evaluation.",
    "id": "trec",
    "kind": "EvaluationProgram",
    "prefLabel": "TREC",
    "scopeNote": "TREC organizes tracks, topics, collections, runs, and evaluation resources. It is not itself a metric, a qrels file, or a single test collection, though many test collections and qrels are produced through TREC tracks.",
    "uri": "https://id.searchplex.net/trec/"
  },
  {
    "artifactKind": "Dataset",
    "definition": "A COVID-19 biomedical literature retrieval test collection created through the TREC-COVID challenge.",
    "id": "trec-covid",
    "kind": "Artifact",
    "prefLabel": "TREC-COVID",
    "scopeNote": "TREC-COVID is a specific evaluation dataset/test collection and is also used as a component dataset in BEIR. It should not be confused with TREC as an evaluation program or BEIR as a benchmark suite.",
    "uri": "https://id.searchplex.net/trec-covid/"
  },
  {
    "altLabels": [
      "TREC Video Retrieval Evaluation",
      "TREC Video Track"
    ],
    "definition": "The TREC Video Retrieval Evaluation, an evaluation campaign for content-based video analysis and retrieval.",
    "id": "trecvid",
    "kind": "EvaluationProgram",
    "prefLabel": "TRECVID",
    "scopeNote": "TRECVID began as a TREC video track and became an independent annual evaluation campaign. It has supported tasks such as video search, shot-boundary detection, concept or feature extraction, instance search, event detection, video summarization, and video-to-text description.",
    "uri": "https://id.searchplex.net/trecvid/"
  },
  {
    "definition": "A broad practitioner term for retrieving items using vector representations and vector similarity or distance operations.",
    "id": "vector-search",
    "kind": "GeneralConcept",
    "prefLabel": "Vector Search",
    "scopeNote": "In current industry usage, vector search often refers specifically to dense or embedding-based nearest-neighbor retrieval, but the term itself can also encompass sparse, binary, multi-vector, or other vector representations. It should not be collapsed with the classical Vector Space Model merely because both use vector representations.",
    "terms": [
      {
        "community": "industry",
        "label": "Vector Retrieval",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/vector-search/"
  },
  {
    "altLabels": [
      "VSM"
    ],
    "definition": "A classical information-retrieval model that represents documents and queries as vectors in a term space and compares them using vector similarity.",
    "id": "vector-space-model",
    "kind": "Representation",
    "prefLabel": "Vector Space Model",
    "scopeNote": "This ID uses Vector Space Model in the classical IR sense: sparse term-weighted query and document vectors, often using TF-IDF-style weights. It should not be read as a generic label for modern embedding search, vector databases, approximate nearest-neighbor search, or dense vector retrieval merely because those systems also operate over vectors.",
    "uri": "https://id.searchplex.net/vector-space-model/"
  },
  {
    "altLabels": [
      "Vocabulary Gap",
      "Term Mismatch"
    ],
    "definition": "A retrieval failure mode or challenge in which a query and relevant documents use different terms or expressions for the same or related concepts.",
    "id": "vocabulary-mismatch",
    "kind": "Phenomenon",
    "prefLabel": "Vocabulary Mismatch",
    "scopeNote": "Vocabulary mismatch is one motivation for query expansion, semantic search, learned sparse retrieval, dense retrieval, and controlled vocabularies. It is distinct from query drift, where expansion or transformation moves the query away from the intended information need.",
    "uri": "https://id.searchplex.net/vocabulary-mismatch/"
  },
  {
    "definition": "A dynamic-pruning query-evaluation method that uses upper bounds on potential scores to avoid fully evaluating candidates that cannot enter the current top-k result set.",
    "id": "wand",
    "kind": "Method",
    "prefLabel": "WAND",
    "scopeNote": "WAND is an algorithmic concept. Product operators named after WAND may implement specific variants or expose different guarantees.",
    "uri": "https://id.searchplex.net/wand/"
  },
  {
    "artifactKind": "Software",
    "definition": "A Vespa query operator that performs accelerated OR-like retrieval using a WAND-style dynamic-pruning algorithm while integrating with text matching and linguistic processing.",
    "id": "weakand",
    "kind": "Artifact",
    "prefLabel": "weakAnd",
    "scopeNote": "weakAnd is product-specific terminology. Vespa also exposes a distinct wand operator with different input and retrieval characteristics. weakAnd should therefore not be used as a general synonym for WAND.",
    "terms": [
      {
        "community": "vendor",
        "label": "Weak AND",
        "usage": "vendorUsage"
      }
    ],
    "uri": "https://id.searchplex.net/weakand/"
  },
  {
    "altLabels": [
      "Browsing Agent",
      "Web Agent"
    ],
    "definition": "An AI agent that uses a web browser or web-search tools to navigate pages, issue searches, inspect sources, and gather information for a task.",
    "id": "web-browsing-agent",
    "kind": "GeneralConcept",
    "prefLabel": "Web Browsing Agent",
    "scopeNote": "A web browsing agent may support deep research, question answering, fact finding, form filling, or broader web automation. It is not the same as a search engine, web search task, or the evaluation benchmark used to measure browsing agents.",
    "terms": [
      {
        "community": "industry",
        "label": "Browser Agent",
        "usage": "nearSynonym"
      },
      {
        "community": "industry",
        "label": "Web Research Agent",
        "usage": "nearSynonym"
      }
    ],
    "uri": "https://id.searchplex.net/web-browsing-agent/"
  },
  {
    "altLabels": [
      "Browsing Agent Evaluation",
      "Web Agent Evaluation"
    ],
    "definition": "Evaluation of agents that search, browse, inspect web sources, and synthesize answers or results from information found on the web.",
    "id": "web-browsing-agent-evaluation",
    "kind": "GeneralConcept",
    "prefLabel": "Web Browsing Agent Evaluation",
    "scopeNote": "Web browsing agent evaluation may measure source finding, persistence, answer accuracy, evidence use, citation fidelity, and multi-step browsing behavior. It is distinct from ordinary web search evaluation and from retrieval-only benchmark suites.",
    "uri": "https://id.searchplex.net/web-browsing-agent-evaluation/"
  },
  {
    "altLabels": [
      "Web Retrieval"
    ],
    "definition": "A search task or system setting focused on retrieving and ranking information from web-scale or web-like collections for user queries.",
    "id": "web-search",
    "kind": "Task",
    "prefLabel": "Web Search",
    "scopeNote": "Web search includes information needs such as informational, navigational, and transactional queries and is commonly studied through large-scale web collections and tracks. It is not equivalent to semantic search, vector search, or a particular retrieval method.",
    "uri": "https://id.searchplex.net/web-search/"
  }
]