On Modeling User-Aware Proactive Communication in Conversational Recommendation Systems^*

Despite tremendous advances in large language models and conversational recommendation systems, these system are unable to handle many complex recommendation tasks. We present a new recommendation task called business ideation and show that the baseline performance of LLM-only recommendation is far from satisfactory. A part of the reason is neither the LLM not the recommendation has a proper user model and does not have the mechanisms to effective elicit a user model via interactions. We propose borrowing insights from decades of human communication research may bring in new ideas into the development of LLM-centric recommenders and present a research agenda to advance new generation systems.