To overcome the limitations of existing conversational recommender systems (CRSs), which fail to account for users' heterogeneous decision-making styles and knowledge levels, this paper proposes a framework called the Consumer Type-Enhanced Conversational Recommender System (CT-CRS), which integrates consumer type modeling into conversational recommendations. Based on consumer type theory, we define four user types—Dependent, Efficient, Cautious, and Expert—based on two dimensions: decision-making style (Maximizer vs. Satisfyer) and knowledge level (High vs. Low). CT-CRS automatically infers user types in real time by leveraging interaction history and fine-tuning a large-scale language model, eliminating the need for static questionnaires. We integrate user types into state representations and design a type-adaptive policy that dynamically adjusts recommendation granularity, diversity, and attribute query complexity. To further optimize the conversational policy, we employ inverse reinforcement learning (IRL) to enable the agent to approximate expert-like strategies based on consumer types. Experimental results on LastFM, Amazon Books, and Yelp demonstrate that CT-CRS increases recommendation success rates and reduces the number of interaction turns compared to existing methods. Additional experiments confirm that both consumer type modeling and IRL significantly contribute to improved performance. In conclusion, CT-CRS provides a scalable and interpretable solution for enhancing CRS personalization by integrating psychological modeling and advanced policy optimization.
