Example-based spoken chat system which can be customized for each user

Abstract

There is a need to develop automated dialog systems, such as chatbots, which are capable of engaging in natural conversations with elderly users. We propose an example-based dialog system featuring an adaptation method which customizes the dialog for each user. After retrieving information about the user from the Web using the user’s profile, morpheme analysis is applied to the retrieval results and only proper nouns are extracted. Words which appear with high frequencies are considered to be related to the user. We then calculate the similarity between words related to the user and words in example phrases in the dialog system. Cosine similarity between distributed representations of words is calculated using Word2vec. We then generate a phrase which has been adapted to include a topic related to the user by substituting user-related words for highly similar words in the original example phrase. We manually evaluated the naturalness of the generated phrases and found that the system did in fact generate natural-sounding phrases. However, Word2vec sometimes mistook words which were grammatically different parts of speech (POS) for similar words, so we added a POS constraint to the substitution process. As a result, the system achieved higher phrase generation precision.