User adaptation of input-response pairs in an example-based dialog system using distributed representation of words

There is a need to develop spoken dialog systems, such as chatbots, which are capable of engaging in natural conversations with people, especially the socially isolated elderly. We propose an example-based dialog system featuring an adaptation method which customizes the dialog for each user. After retrieving profile information about the user from the Web, morpheme analysis is applied to the retrieval results and only nouns and proper nouns are extracted. Those which appear with high frequency are considered to be related to the user. We then calculate the similarity between the selected words related to the user and words in the existing example phrases of the dialog system. Cosine similarity between the distributed representations of the nouns is calculated using the word2vec program. We then generate phrases adapted to the user by substituting user-related words for highly similar words in the original example phrases. We manually evaluated the naturalness of the generated phrases and found that the system could generate natural-sounding phrases. Word2vec has a special property which allows arithmetic operations such as plus and minus to be applied to distributed word representations. Using this feature, we were able to derive replacement of words for the example phrases by applying operations to the words used in the original phrases, improving the success rate for generating natural-sounding phrases.