Harmonizing Large Language Models with Collaborative Behavioral Signals for Conversational Recommendation
Guanrong Li, Kuo Tian, Jinnan Qi, Qinghan Fu, Zhen Wu, Xinyu Dai
TL;DR
This work tackles the gap between collaborative behavioral signals and natural-language preferences in conversational recommendation by introducing LatentCRS, a framework that uses discrete latent intents to bridge the two modalities. It combines a traditional recommender (inference model) with a generative CRS (decoder) and optimizes via a variational EM procedure, employing an ELBO objective and infoNCE-based ranking, plus a hard-filter strategy for multi-turn dialogues. The approach demonstrates consistent improvements over strong baselines across one-turn and multi-turn settings on multiple datasets, while also reducing reliance on costly LLM usage through efficient inference and data augmentation. The results highlight the practical potential of integrating collaborative information with LLMs, offering improved personalization and scalability for real-world CRS deployments.
Abstract
Conversational recommendation frameworks have gained prominence as a dynamic paradigm for delivering personalized suggestions via interactive dialogues. The incorporation of advanced language understanding techniques has substantially improved the dialogue fluency of such systems. However, while modern language models demonstrate strong proficiency in interpreting user preferences articulated through natural conversation, they frequently encounter challenges in effectively utilizing collective behavioral patterns - a crucial element for generating relevant suggestions. To mitigate this limitation, this work presents a novel probabilistic framework that synergizes behavioral patterns with conversational interactions through latent preference modeling. The proposed method establishes a dual-channel alignment mechanism where implicit preference representations learned from collective user interactions serve as a connecting mechanism between behavioral data and linguistic expressions. Specifically, the framework first derives latent preference representations through established collaborative filtering techniques, then employs these representations to jointly refine both the linguistic preference expressions and behavioral patterns through an adaptive fusion process. Comprehensive evaluations across multiple benchmark datasets demonstrate the superior performance of the proposed approach compared to various state-of-the-art baseline methods, particularly in aligning conversational interactions with collaborative behavioral signals.