Japanese vocabulary corpus is rich and semantically complex, and traditional methods are more limited in dealing with its semantic relations, which makes it difficult to deal with large-scale Japanese vocabulary corpus effectively. In this paper, we first establish the conceptual semantic network of Japanese vocabulary by means of co-occurrence analysis statistics and similarity computation, and construct a vocabulary semantic similarity measure model based on Japanese-specific corpus by extracting the contextual features of Japanese vocabulary. Then, we use machine translation to construct word-vector relations between Japanese and English, and then introduce LSTMs network to learn the sentence sequences co-occurring between Japanese word pairs, so as to complete the modeling of lexical relations between Japanese word pairs. The Japanese vocabulary conceptual semantic network constructed in this paper clearly identifies the center nodes and non-center nodes of the Japanese vocabulary semantic network, and the similarity of Japanese vocabulary pairs under this computation and closer to the fact, compared with the JC algorithm, the algorithm in this paper reduces the gap between the computation results and the manual algorithm to 0.00. In general, the method is conducive to the improvement of the efficiency and the intelligence of the processing of large-scale Japanese vocabulary corpus.
