カンファレンス (国際)Chat Detection in an Intelligent Assistant: Combining Task-oriented and Non-task-oriented Spoken Dialogue Systems

Satoshi Akasaki (Tokyo University) andNobuhiro Kaji

The Annual Meeting of the Association for Computational Linguistics (ACL2017)

2017.7.29

Recently emerged intelligent assistants on smartphones andhome electronics (e.g., Siri and Alexa) can be seen as novel hybrids ofdomain-specific task-oriented spoken dialogue systems andopen-domain non-task-oriented ones. To realize such hybrid dialogue systems,this paper investigates determining whether or not a user is going to have achat with the system. To address the lack of benchmark datasets for thistask, we construct a new dataset consisting of 15, 160 utterances collectedfrom the real log data of a commercial intelligent assistant (and willrelease the dataset to facilitate future research activity). In addition, weinvestigate using tweets and Web search queries for handling open-domainuser utterances, which characterize the task of chat detection. Experimentsdemonstrated that, while simple supervised methods are effective, the use ofthe tweets and search queries further improves the F1-score from 86.21 to87.53.

Paper :Chat Detection in an Intelligent Assistant: Combining Task-oriented and Non-task-oriented Spoken Dialogue Systems（外部サイト）