In the context of natural language processing words, phrases or documents are regarded as having semantic similarity if they refer to the same topic or concept, even if they do not share any literal words. For example, 'knife' and 'fork' have semantic similarity as they both connect to eating food. Semantic similarity may be used to help disambiguation, for example, if 'bow' and 'arrow' are in the same sentecne then it is most likely that 'bow' refers to the weapon rather than the front of a ship. Semantic similarity may be assessed using databases or lexicons such as WordNet or inferred based on large corpora as is the case with analysis using word co-occurrence frequencies or the generation of LLMs.
Used on page 541