SEMANTIC SIMILARITY LEVEL COMPUTATION METHOD, SYSTEM AND PROGRAM

摘要

Provided is a semantic similarity level computation system that is capable of computing a similarity level in accordance with context. Said semantic similarity level computation system is provided with a topic vector creation means (1), a node reference probability computation means (3) and a similarity level computation means (4). Said topic vector creation means (1) inputs two texts, holds a topic word list length that was prepared beforehand on the basis of a word contained in a text, and creates a topic vector that by means of elements, represents which words are contained in the two texts. Said node reference probability computation means (3) inputs a node topic matrix and the topic vector, said node topic matrix representing the contribution level between: each node contained in a thesaurus; and each topic word. For the two input texts represented by the topic vector, said node reference probability computation means (3) computes a reference probability that represents whether or not to reference a node contained in the thesaurus. Using the reference probability of each node, said similarity level computation means (4) computes the similarity level between words contained in each of the two texts.