Synonym extraction device, synonym extraction method, and synonym extraction program

A synonym extraction method performed by a synonym extraction apparatus, the synonym extraction method comprising:

• determining, for compound words included in a document, that types of nouns constituting a compound word of the compound words are each a Sahen-noun or a noun other than a Sahen-noun;
• extracting, based on a determination result of the types of nouns constituting the compound word, a first group of compound words having an identical pattern of a sequence of the determined types of the nouns constituting the compound word from among compound words included in the document;
• extracting a second group of compound words having an identical leading or ending word from the first group of compound words having the identical pattern of the sequence of the determined types of the nouns;
• creating, for each compound word of the second group of compound words, a co-occurrence vector having, as a vector component, a noun co-occurring in a same sentence as a sentence including the compound word;
• calculating, for each compound word of the second group of compound words, a degree of similarity between created co-occurrence vectors;
• outputting, as synonyms, a third group of compound words having the degree of similarity between the created co-occurrence vectors equal to or greater than a predetermined threshold;
• and
• for each compound word of the second group of compound words that have been extracted,
• creating, fora character string obtained by adding a case particle between nouns constituting the compound word, a first co-occurrence vector of the compound word having, as a vector component, a noun co-occurring in a same sentence as a sentence including the character string, and creating a second co-occurrence vector of the compound word by adding the first co-occurrence vector of the compound word created to the co-occurrence vector of the compound word, and
• calculating a degree of similarity between the second co-occurrence vectors of the compound words as the degree of similarity between the created co-occurrence vectors of the compound words.

A non-transitory computer readable medium comprising a synonym extraction program configured to cause a computer to execute a method comprising:

• determining, for compound words included in a document, that types of nouns constituting a compound word of the compound words are each a Sahen-noun or a noun other than a Sahen-noun;
• extracting, based on a determination result of the types of nouns constituting the compound word, a first group of compound words having an identical pattern of a sequence of the determined types of the nouns constituting the compound word from among the compound words included in the document;
• extracting a second group of compound words having an identical leading or ending word from the first group of compound words having the identical pattern of the sequence of the determined types of the nouns;
• creating, for each compound word of the second group of compound words, a co-occurrence vector having, as a vector component, a noun co-occurring in a same sentence as a sentence including the compound word;
• calculating, for each compound word of the second group of compound words, a degree of similarity between the created co-occurrence vectors;
• outputting, as synonyms, a third group of compound words having the degree of similarity between the created co-occurrence vectors equal to or greater than a predetermined threshold;
• and
• for each compound word of the second group of compound words that have been extracted,
• creating, for a character string obtained by adding a case particle between nouns constituting the compound word, a first co-occurrence vector of the compound word having, as a vector component, a noun co-occurring in a same sentence as a sentence including the character string, and creating a second co-occurrence vector of the compound word by adding the first co-occurrence vector of the compound word created to the co-occurrence vector of the compound word, and
• calculating a degree of similarity between the second co-occurrence vectors of the compound words as the degree of similarity between the created co-occurrence vectors of the compound words.