Identify similar terms

Semantics/similar uses text vectors to identify semantically similar terms. By calculating the text vector for a term, the endpoint can find similar terms in the semantic space in multiple supported languages.

To find semantically similar terms, provide an input term and one or more result languages. By default, the endpoint will return 10 similar terms in each requested language. You can request up to 50 terms for each language. For each returned term, a similarity value between -1 and 1 is returned, where a value closer to 1 indicates a higher degree of similarity.

Text vectors provide a mechanism for comparing documents or words based on their semantic similarity. For any given term (which may be a document or a single word), a location in semantic space, as represented by a vector of floating point numbers, is calculated. This vector can be mathematically compared with other term or document vectors. Words with similar meanings have similar contexts, so they are mapped close to each other. The terms being compared can be in the same or different languages, providing cross-lingual semantic similarity evaluation without the need for any translation.

In the semantic space, corresponding words in different languages have similar meanings and therefore similar mappings. Take for example, “Washington was the first president of the United States” and” Washington fue el primer presidente de los Estados Unidos“. These sentences have roughly equivalent meanings in both Spanish and English, which will be reflected in the proximity of their text vectors.