In recent years, various dialogue systems have been developed. However, there are some problems; needing for the Internet environment, the inability to distinguish multiple meaningful words (word sense disambiguation) and the dialogue system tends to be passive conversation beginning with user’s utterance. Therefore, we developed non-task-oriented dialogue system dealing these problems in this research. The system was realized by applying method of generating responses using word embedding without requiring the Internet. Also, a new algorithm was proposed for the word sense disambiguation by comparing similar words between the topic of the previous conversation and the current conversation. The system can be active talker by adding independent modules for speaking to users. With regards to the proposed system, it is clarified that the proposed system has improved conventional system’s problems and is useful as a dialogue system by evaluating with the co-occurrence network and the Spearman correlation.

Key words:Natural Language Processing, Word2Vec, Word Sense Disambiguation