Extracting and Evaluating Personal Interests with Dialogue Agent

In order to realize a continuous conversation between a dialog agent and a user, it is important to generate topics that are close to the user. In this paper, we propose a method for extracting and evaluating the concept "Personal Interests" that users are interested in through dialogue with a virtual agent (VA). In the proposed method, the VA first talks about a certain genre, asks the user for interests in that genre, and asks the user to talk about episodes about each interest. In addition, the system extracts the personal knowledge "Personal Ontology" from the user's answer, and built and stores that in the form of Linked Data. The extracted interests are evaluated based on three criteria P1 (length of the episode), P2 (the number of concepts included in the episode), and P3 (the number of reference sources of the concepts), and calculate the level of interest in each concept. We implemented a prototype of the proposed method and performed a subject experiment to extract and evaluate personal interests. As a result, it was found that the scores calculated by the criterion P1 and the criterion P2 correlated with the actual interest level of the user. In addition, it was suggested that the criterion P3 could discover new concerns from other concepts included in the episode.