Detection of Topic Transition in Chat Dialogue System
Moemi Seki, Atsushi HAYASHI, Shino Iwashita · 2022 Joint 12th International Conference on Soft Computing and Intelligent Systems and 23rd International Symposium on Advanced Intelligent Systems (SCIS&ISIS) · 2022
We propose a method for detecting topic transitions in a chat dialogue system to provide a response that takes into account the topic. First, we conducted a questionnaire in which participants categorized transitions in dialogues as a transition, possible transition, or non-transition, and provided their reasoning. Next, for transition detection, we introduce the concept of a slight topic transition, interrogative sentence detection, and an order of priority for detection techniques. These were integrated with the baseline techniques which use the similarity between utterances, the intersection of nouns, and keywords. The results show that the interrogative sentence detection and prioritizing certain detection techniques were effective for detecting topic transitions. Furthermore, transitions which are considered ambiguous by people can be clarified by classifying them into one of the three types.