EMNLP Findings2025
LLM-based Conversational Recommendation Agents with Collaborative Verbalized Experience
Yaochen Zhu, Harald Steck, Dawen Liang, Yinhan He, Nathan Kallus, Jundong Li
Abstract
Large language models (LLM) have shown impressive zero-shot capabilities in conversational recommender systems (CRS). However, effectively utilizing historical conversations remains a significant challenge. Current approaches either retrieve few-shot examples or extract global rules to augment the prompt, which fail to capture the implicit and preferenceoriented knowledge. To address the above challenge, we propose LLM-based Conversational Recommendation Agents with Collaborative Verbalized Experience (CRAVE). CRAVE starts by sampling trajectories of LLM-based CRS agents on historical queries and establishing verbalized experience banks by reflecting the agents' actions on user feedback. Additionally, a collaborative retriever network finetuned with item content-parameterized multinomial likelihood on query-item pairs is introduced to retrieve preference-oriented verbal experiences for new queries. Furthermore, we developed a debater-critic agent (DCA) system where each agent maintains an independent collaborative experience bank and works together to enhance the CRS recommendations. We demonstrate that the open-end debate and critique nature of DCA benefits significantly from the collaborative experience augmentation with CRAVE. The code is available at https://github.com/yaochenzhu/CRAVE.
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Reflexion: language agents with verbal reinforcement learning cites Language Models can Solve Computer Tasks
- Reflexion: language agents with verbal reinforcement learning cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
- Language Models can Solve Computer Tasks cites Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Builds on16
- Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
- Retrieval-Augmented Generation for Knowledge-Intensive NLP TasksPatrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni et al.NeurIPS 2020 · 19,162 citations
- Reflexion: language agents with verbal reinforcement learningNoah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan et al.NeurIPS 2023 · 5,828 citations
- ChatEval: Towards Better LLM-based Evaluators through Multi-Agent DebateChi-Min Chan, Weize Chen, Yusheng Su, Jianxuan Yu et al.ICLR 2024 · 871 citations
- Language Models can Solve Computer TasksGeunwoo Kim, Pierre Baldi, Stephen McAleerNeurIPS 2023 · 539 citations
Related papers
- Collaborative Retrieval for Large Language Model-based Conversational Recommender SystemsYaochen Zhu, Chao Wan, Harald Steck, Dawen Liang et al.WWW 2025 · 15 citations
- ReLLa: Retrieval-enhanced Large Language Models for Lifelong Sequential Behavior Comprehension in RecommendationJianghao Lin, Rong Shan, Chenxu Zhu, Kounianhua Du et al.WWW 2024 · 151 citations
- CoRAL: Collaborative Retrieval-Augmented Large Language Models Improve Long-tail RecommendationJunda Wu, Cheng-Chun Chang, Tong Yu, Zhankui He et al.KDD 2024 · 32 citations
- Refining Text Generation for Realistic Conversational Recommendation via Direct Preference OptimizationManato Tajiri, Michimasa InabaEMNLP 2025
- LatentCRS: A Variational EM Framework for Bridging Semantics and Behavior in LLM-based Conversational RecommendationGuanrong Li, Kuo Tian, Jinnan Qi, Qinghan Fu et al.KDD 2026 · 1 citation
