Bridging Conversational and Collaborative Signals for Conversational Recommendation

2024-12-09

arXiv.org

AI Summary

Conversational recommendation systems (CRS) suffer from limited user-item interaction modeling due to the absence of collaborative filtering (CF) signals, constraining recommendation accuracy. To address this, we propose a dialogue–collaborative dual-signal alignment paradigm. First, we construct Reddit-ML32M—the first open-source cross-source dataset integrating Reddit conversational texts with MovieLens explicit behavioral data—to mitigate item representation sparsity and cold-start challenges. Second, we design LLM-CF Alignment, a framework jointly optimizing ranking via LLM fine-tuning, prompt engineering, and CF embedding injection. Experiments on standard benchmarks demonstrate significant improvements: +12.32% in Hit Rate and +9.9% in NDCG over state-of-the-art baselines, outperforming both pure-dialogue and pure-CF approaches. These results empirically validate that infusing collaborative knowledge meaningfully enhances large language model–based recommendation performance.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Conversational AI/Dialog Systems

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

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.

Problem

• Integrates collaborative filtering into conversational recommendation systems.
• Addresses interaction sparsity in conversational datasets.
• Improves recommendation accuracy using LLM and collaborative knowledge.

Innovation

• LLM-based framework integration
• Reddit-ML32M dataset utilization
• CF embeddings alignment enhancement