Semantic Mapping Vocabulary

• Distinguished from single-step LLM-based matching by the agent's autonomy over its own process. Distinguished from composite matching, which combines heterogeneous matching techniques under a fixed pipeline, by the agent's runtime control over what to do next. In current practice, agent-based matchers are predominantly driven by large language models, but this class does not commit to a particular agent technology — a specific process may also be classified as semapv:LLMBasedMatching (or any other matching class) where appropriate.

• A matching process in which an autonomous agent plans and executes a multi-step workflow — typically involving tool use, retrieval, intermediate reasoning, and self-evaluation — to determine matches between entities. The agent decides at runtime what information to gather, which techniques to apply, and when to stop.

An agent is given a pair of candidate entities and autonomously decides to look up their labels, query an external knowledge base, retrieve definitions, compare them with a similarity measure, and revise its judgement before producing a final match decision; the OntoAligner toolkit provides agentic matching of this kind.

A matching process that exploits background knowledge from external resources, commonly referred to as background knowledge resources. This approach is also known as indirect matching, BK-based matching or context-based matching.

A match between a subject A and an object B was established because they appear equivalent under consideration of externally provided background knowledge.

A normalization technique replacing all blank characters, such as space, tabulation, carriage return (or sequences of these) into a single blank character.

The string 'ontology  matching' (two spaces between 'ontology' and 'matching') becomes 'ontology matching' (one space).

• Blocking is conceptually distinct from matching: a blocker is recall-oriented and typically encodes each entity independently of the others, whereas a matcher is precision-oriented and decides about a specific candidate pair. Many modern pipelines combine an embedding-based blocker with a transformer- or LLM-based matcher.

• A mapping activity that, given two sets of input entities, produces a reduced set of candidate entity pairs likely to match — favouring recall over precision — to be passed on to a downstream matching process.

Out of 10^7 possible subject–object pairs, a blocker returns 10^5 candidate pairs by applying inexpensive comparisons such as token overlap, hash buckets, or embedding-space nearest-neighbour search.

• Described as a structural matching technique in Euzenat & Shvaiko, "Ontology Matching" (2nd ed., 2013), and implemented in classical structural OM systems such as Anchor-PROMPT.

• A structural matching process based on the comparison of the matched super- and subclass paths (up to a bounded depth) of two candidate entities.

Two classes are matched because their direct superclasses and subclasses (within a bounded depth) are already paired by other matchers, and the structural agreement among neighbours raises confidence in the match.

A normalization technique converting each alphabetic character in a string to their lower case counterpart.

The string 'Ontology' is normalised to 'ontology' (lower case).

A matching process based on multiple, possibly intertwined, matching approaches.

An ontology matching tool determines that a subject should be mapped to an object by employing a range of techniques, including lexical, semantic and structural.

A normalization technique replacing diactritic signs (e.g. accents, hats) with their most frequent replacements.

The string 'Protégé' is normalised to 'Protege'.

A normalization technique removing all numbers in a string.

The string 'ontology1' becomes 'ontology'.

A blocking process in which entities are encoded independently as vector embeddings and candidate pairs are produced by nearest-neighbour search or similarity-threshold filtering in the embedding space.

A sentence-transformer encodes every subject and object label into a vector; an approximate nearest-neighbour index (e.g. FAISS) returns, for each object, the top-k subjects whose embeddings are closest in cosine distance, as the set of candidate pairs to be matched downstream.

• Where embedding similarity is used only to generate candidate pairs for a downstream matcher (rather than to make the final match decision), the process is more accurately classified as semapv:EmbeddingBasedBlocking.

• A machine learning-based matching process in which matches are identified by comparing learned vector representations (embeddings) of entities under a vector similarity measure.

A sentence-transformer encodes the label of each candidate entity; the pair is reported as a match when the cosine similarity between the two embeddings exceeds a fixed threshold.

• graph representation learning-based matching process ^c
• transformer-based matching process ^c

An embedding-based matching process in which matches are identified by comparing learned vector representations that encode the structural and semantic features of entities and their surrounding graph context.

A graph embedding method such as node2vec, GraphSAGE, or OWL2Vec* encodes each entity together with its neighbourhood in the ontology graph; matches are produced where the resulting embeddings are close under a similarity measure.

A matching process based on individual representations (or instances).

A match between a subject A and an object B was established because they share the same instances.

• Where an LLM is used only to produce embeddings that are then compared by similarity, the process is an embedding-based (transformer-based) matching process rather than an LLM-based one. The classifier is how the model is used, not which model. Multi-step, tool-using, autonomous workflows driven by an LLM are captured by semapv:AgentBasedMatching.

• A machine learning-based matching process in which a large language model determines matches between entities, typically through single-step or few-shot prompting, or through fine-tuning the model to judge mapping candidates or generate mappings directly.

OLaLa prompts a large language model with the labels and surrounding context of two candidate entities and uses the model's yes/no decision as the match outcome.

The process of reducing the different forms of a word to one single form.

A matching process based on a lexical comparison between one or more syntactic features of the subject with one or more syntactic features of the object.

The label of a subject entity matches to an exact synonym of an object entity.

A lexical matching process based on a minimum threshold of a score from a comparison based on a lexical similarity algorithm.

A match between a subject and an object was established because they had a Levenshtein score higher than 0.8.

A normalization technique replacing specific links between words, such as apostrophes, dashes, underscore, etc into dashes or blanks.

The string 'alzheimer's' becomes 'alzheimers'.

A matching process based on the inferences made by a logical reasoner.

Two classes across ontologies are determined equivalent by an OWL reasoner such as ELK.

A matching process in which matches between entities are predicted by machine learning models trained on labelled data or learned representations.

A trained machine learning model classifies a pair of candidate entities as matching or not based on learned features; the OntoAligner toolkit packages a number of such models.

• embedding-based matching process ^c
• LLM-based matching process ^c

A matching process that is performed by a human agent and is based on human judgement and domain knowledge.

A human curator determines that a subject should be mapped to an object by virtue of their domain expertise.

A triple comprising a subject entity s, an object entity o and a mapping predicate p.

The subject entity NCI:C9305 is mapped to the object entity ICD10:C80.9 using the skos:relatedMatch mapping predicate.

A process that relates to the creation, confirmation, rejection or curation of a mapping.

Matching is a mapping activity that results in the creating of a mapping; mapping review is an activity that results in the confirmation of a mapping.

• blocking ^c
• mapping review ^c
• matching process ^c
• matching process post-processing ^c
• matching process pre-processing ^c

A matching process based on the traversing of multiple mappings.

A match between a subject A and an object B was established because A was mapped to C, C was mapped to D and D was mapped to B.

A matching process based on the inverting or flipping of the subject with the object of a mapping in accordance with the semantics of the mapping predicate.

A broad match between a subject A and an object B using the skos:narrowMatch mapping relation was established because B was mapped to A using the skos:broadMatch mapping relation.

A process that is concerned with determining if a mapping “candidate” (otherwise determined) is reasonable/correct.

A match between a subject A and an object B was established elsewhere, and a human reviewer determined that the mapping is true (or false) based on an independent evaluation.

An process that results in a mapping between a subject and an object entity.

The label of a subject entity matches to an exact synonym of an object entity.

• agent-based matching process ^c
• background knowledge-based matching process ^c
• composite matching process ^c
• instance-based matching process ^c
• lexical matching process ^c
• lexical similarity threshold-based matching process ^c
• logical reasoning matching process ^c
• machine learning-based matching process ^c
• manual mapping curation ^c
• mapping chaining-based matching process ^c
• mapping inversion-based matching process ^c
• semantic similarity threshold-based matching process ^c
• structural matching process ^c
• unspecified matching process ^c

A process applied to a set of mappings, usually with the intention of changing it.

• cardinality filtering ^c
• logical consistency filtering ^c
• stable marriage filtering ^c
• threshold filtering ^c

A preliminary processing of inputs prior to performing matching.

A family of preprocessing techniques for reducing strings to be compared to a common format.

• blank normalisation ^c
• case normalization ^c
• diacritics suppression ^c
• digit suppression ^c
• lemmatization ^c
• link stripping ^c
• punctuation elimination ^c
• regex removal ^c
• regex replacement ^c
• stemming ^c
• stop-word removal ^c
• term extraction ^c
• tokenization ^c

A normalization technique removing all punctation characters from a string.

The string 'e.g.' becomes 'eg'.

A preprocessing method transforming a string by matching a regular expression and then removing that match.

The regex match “[ -]phenotype” is removed from the label field of the subject entity in the mapping.

A preprocessing method transforming a string by matching a regular expression and then replacing that match with a specified substitution string.

The regex match “[ -]phenotype” is replaced by “-disease” for the label field of the subject entity in the mapping.

A matching process based on a minimum threshold of a score from a comparison based on a semantic similarity algorithm.

A match between a subject and an object entity was established because they had a Jaccard score higher than 0.8 based on the set of (common) superclasses.

A technique for determining a score that characterises the similarity between two entities.

• language-based similarity measure ^c
• string-based similarity measure ^c

The process of reducing a word to its word stem.

A preprocessing method transforming a string by removing a list of stop words.

The stop-words “of, and, is, a” are removed from the label field of the subject entity in the mapping.

A similarity measure based on the comparison of strings.

• edit distance ^c
• hamming distance ^c
• string equality ^c
• substring similarity ^c
• substring test ^c
• token-based distance ^c
• ngram similarity ^c

A string-based similarity measure which involves determining if two strings associated with mapping entities are equal.

Structural matching does not involve looking at "values" of properties.

A match between a subject and an object was established because of the similarity of their structural features, e.g., the number of direct property of a class.

• Transformer models are a deep learning architecture designed to handle sequential data, particularly for natural language processing tasks.

• An embedding-based matching process that uses transformer models to produce contextual vector representations of entity features for similarity comparison.

BERTMap fine-tunes BERT on synonym pairs derived from the input ontologies and matches classes whose fine-tuned contextual embeddings have high cosine similarity.

A matching process based on an unspecified comparison.

A mapping between a subject and an object was established, but it is no longer clear how or why.

A match where the object is a broader concept than the subject.

A match where the subject and object are sufficiently similar that they can be used interchangeably in some information retrieval applications.

• has exact match ^ap
• cross-species close match ^ap

A match where the subject and object are sufficiently similar, with a high degree of confidence, that they can be used interchangeably across a wide range of information retrieval applications.

• has close match ^ap
• isomorphic match ^ap

A mapping (alignment) link between two concepts.

• has close match ^ap
• isomorphic match ^ap
• non-isomorphic match ^ap

A match where the object is a narrower concept that the subject.

A match where the subject and the object are inherently related, but where none is in any way more general than the other.

A match where the subject is considered analogous to the subject in a different taxonomic grouping, but the object refers to a broader concept.

FBbt:00007091 (D. melanogaster “subperineurial glial sheath”) and UBERON:0000202 (taxon-neutral “glial blood brain barrier”) are a cross-species broad match.

A match where the subject and the object belong in different taxonomic groupings, but refer to concepts similar enough that they can be used interchangeably.

• has close match ^ap
• non-isomorphic match ^ap

A match where the subject is considered analogous to the object in a different taxonomic grouping.

FBbt:00005074 (D. melanogaster “muscle cell”) and CL:0000187 (taxon-neutral “muscle cell”) are a cross-species exact match.

• has related match ^ap
• isomorphic match ^ap

A match where the subject is considered analogous to the subject in a different taxonomic grouping, but the object refers to a narrower concept.

A match where the subject is isomorphic to the object, i.e. considered of identical or similar form, shape, or structure.

• has exact match ^ap
• cross-species exact match ^ap

A match where the subject cannot be considered isomorphic to the object, i.e. considered of identical or similar form, shape, or structure. The object corresponds to exactly one subject in the subject_source.

• has broader match ^ap
• has narrower match ^ap
• has related match ^ap
• cross-species close match ^ap