Abstract
Chat-oriented dialogue systems designed to provide tangible benefits, such assharing the latest news or preventing frailty in senior citizens, often requireProactive acquisition of specific user Information via chats on user-faVOredTopics (PIVOT). This study proposes the PIVOT task, designed to advance thetechnical foundation for these systems. In this task, a system needs to acquirethe answers of a user to predefined questions without making the user feelabrupt while engaging in a chat on a predefined topic. We found that evenrecent large language models (LLMs) show a low success rate in the PIVOT task.We constructed a dataset suitable for the analysis to develop more effectivesystems. Finally, we developed a simple but effective system for this task byincorporating insights obtained through the analysis of this dataset.