Bridging Conversational and Collaborative Signals for Conversational Recommendation

TL;DR

This work tackles the sparsity and modality gap in conversational recommendation by linking Reddit conversations with MovieLens interactions through the Reddit-ML32M dataset. It introduces BridgeCRS, a zero-shot prompting framework that couples LLM-generated recommendations with CF-derived item embeddings learned from sequential recommender models, yielding refined rankings. The key contributions include constructing Reddit-ML32M with substantial density gains and demonstrating consistent improvements over CF-based, CRS, and LLM-based baselines (e.g., a 12.32% gain in Hit Rate at 5 and a 9.9% gain in NDCG at 5). The approach shows the viability and practical impact of integrating collaborative signals with conversational context in CRS, with clear pathways for extending to other domains.

Abstract

Conversational recommendation systems (CRS) leverage contextual information from conversations to generate recommendations but often struggle due to a lack of collaborative filtering (CF) signals, which capture user-item interaction patterns essential for accurate recommendations. We introduce Reddit-ML32M, a dataset that links Reddit conversations with interactions on MovieLens 32M, to enrich item representations by leveraging collaborative knowledge and addressing interaction sparsity in conversational datasets. We propose an LLM-based framework that uses Reddit-ML32M to align LLM-generated recommendations with CF embeddings, refining rankings for better performance. We evaluate our framework against three sets of baselines: CF-based recommenders using only interactions from CRS tasks, traditional CRS models, and LLM-based methods relying on conversational context without item representations. Our approach achieves consistent improvements, including a 12.32% increase in Hit Rate and a 9.9% improvement in NDCG, outperforming the best-performing baseline that relies on conversational context but lacks collaborative item representations.

Figures (1)

• Figure 1: Illustration of the proposed CRS framework, which integrates LLM-based conversational recommendations with CF signals. The predictor uses these signals to generate ranked recommendations.