WORD SENSE EMBEDDING APPARATUS AND METHOD USING LEXICAL SEMANTIC NETWORK, AND HOMOGRAPH DISCRIMINATION APPARATUS AND METHOD USING LEXICAL SEMANTIC NETWORK AND WORD EMBEDDING

摘要

Examples of the present invention provide a word sense embedding apparatus and a method using a lexical semantic network, the apparatus and method being capable of: generating processing data by using a word list to be learned in a lexical semantic network and word sense data of words to be learned (for example, dictionary definitions, hyperonyms, antonyms, and the like); and performing learning by using the generated processing data through a negative-sampling and a feature mirror model, which is a modified skip-gram model, of word sense embedding, thereby enabling a semantic relationship and correlation between words to be expressed as a vector. In addition, examples of the present invention provide a homograph discrimination apparatus and method using a lexical semantic network and word embedding, the apparatus and method being capable of: learning through word embedding learning using a word list to be learned from various resources (for example, a corpus, a standard unabridged dictionary, and a lexical semantic network), a converted corpus, and the word sense data; and accurately discriminating homographs with respect to a non-learning pattern by comparing the similarity between the homograph and an adjacent syntactic word so as to distinguish the homographs.