[1N2-J-9] Natural language processing, information retrieval: dialogue

[1N2-J-9-04] Response Dialogue-Act Prediction based on Conversational History

〇Koji Tanaka¹, Junya Takayama¹, Yuki Arase¹ (1. Graduate School of Information Science and Technology, Osaka University)

Keywords:Conversation, Text Classification, Natural Language Processing

Sequence-to-sequence models are widely used to implement a chatbot. One of their advantages is that a chatbot

can be trained in an end-to-end manner. On the other hand, its disadvantage is that a process of response generation

is completely black-box. To solve this problem, interpretable response generation mechanism is desired. As a step

forward in this direction, we focus on dialogue-acts and propose a method to predict a dialogue-act of the next

response based on conversational history. Specically, we consider both histories of utterances and their dialogue

acts. Experiment results using the Switch Board Dialogue Act corpus show that our method achieves 8:6%and

1:2% higher F-score and accuracy on predicting responses ’dialogue-acts, respectively, compared to a previous

method that only considers the utterance history.