[4M3-J-9] Natural language processing, information retrieval: inferring emotion and intension

[4M3-J-9-01] ENOVA RNN: Dialogue Act Classification Considering Frequency of Occurrence

〇Haruno Izumi¹, Shohei Kato^1,2 (1. Dept. of Computer Science and Engineering, Graduate School of Engineering, Nagoya Institute of Technology, 2. Frontier Research Institute for Information Science, Nagoya Institute of Technology)

Keywords:Dialogue act classification, RNN

This paper discusses the method to classify dialogue act of an utterance in chat dialogue. The content of past utterances in dialog is an important feature for dialogue act classification. In previous studies on dialogue act classification, they classified dialogue acts using limited number of utterances (context length) as a context, but the number of utterances effective to classify is not certain. Also, the classification tends to be biased towards dialogue act with a large rate. This paper investigates the effect of context length on classification accuracy. From the results, the paper proposes a method to capture the characteristics of dialogue act with low rate and discusses the effectiveness. In this study, it was confirmed that ENOVA RNN can classify dialogue acts using contexts less than 9 sentences and reduce the deviation at the classification while maintaining the accuracy.