Target-guided conversation systems conduct dialogues to achieve predefined conversation targets, such as recommending target goods or talking about target topics. In such systems, it is important to transition topics naturally toward the target without letting the user notice the intention behind the topic induction. In this study, we implement a surprisal-based framework that quantifies the sense of induction, target awareness, and naturalness of system utterances by computing surprisal using an external language model. Experimental results from dialogue sessions demonstrate that utterance selection based on the proposed surprisal-based evaluation reduces the perceived induction of system utterances. Furthermore, correlation analysis reveals that the proposed metric aligns with human perception of induction. We also observe that surprisal values with respect to the target gradually decrease as the conversation progresses, indicating that the model implicitly learns to approach the target more naturally over time.

In our research, we aim to achieve SDS capable of generating responses considering user preferences. While users have individual topic preferences, existing SDSs do not adequately consider such information. With the development of LLMs, SDSs are expected to be implemented in various tasks, including coexisting with humans in robotic applications. To become better partners with humans, systems are anticipated to memorize user preferences and utilize them in their response generation. Our future reserarch aim to realize SDSs that can remember and complement user information through dialogue, enabling personalized interactions. In YRRSDS, The author would like to propose the following topics for discussion. 1. What is the necessity of SDSs aimed specifically at dialogue rather than being just user interfaces? What do general users need from SDSs through conversation? 2. The relationship between SDSs and users: Should SDSs act just as agents, or should they aim to become like friends or family? 3. Privacy in conversational content. Nowadays, many SDS applications operate online via APIs, but is this preferable from a privacy perspective? If it is not preferable, how can this issue be resolved?