Parameter-Efficient Conversational Recommender System as a Language Processing Task

Parameter-Efficient Conversational Recommender System as a Language Processing Task

25 Feb 2024 | Mathieu Ravaut, Hao Zhang, Lu Xu, Aixin Sun, Yong Liu
This paper presents a Parameter-Efficient Conversational Recommender System (PECRS) that formulates conversational recommendation as a natural language processing task. PECRS leverages pre-trained language models to encode items, understand user intent through conversations, perform item recommendation via semantic matching, and generate dialogues. The system is designed to be optimized in a single stage without relying on external knowledge graphs or additional item encoders. PECRS uses parameter-efficient fine-tuning techniques, such as LoRA, to adapt the language model to the downstream task with minimal computational overhead. Experiments on benchmark datasets, ReDial and INSPIRED, demonstrate the effectiveness of PECRS in both recommendation and conversation tasks, achieving competitive performance with state-of-the-art methods. The paper also discusses the limitations of current approaches and suggests future directions for improving conversational recommendation systems.This paper presents a Parameter-Efficient Conversational Recommender System (PECRS) that formulates conversational recommendation as a natural language processing task. PECRS leverages pre-trained language models to encode items, understand user intent through conversations, perform item recommendation via semantic matching, and generate dialogues. The system is designed to be optimized in a single stage without relying on external knowledge graphs or additional item encoders. PECRS uses parameter-efficient fine-tuning techniques, such as LoRA, to adapt the language model to the downstream task with minimal computational overhead. Experiments on benchmark datasets, ReDial and INSPIRED, demonstrate the effectiveness of PECRS in both recommendation and conversation tasks, achieving competitive performance with state-of-the-art methods. The paper also discusses the limitations of current approaches and suggests future directions for improving conversational recommendation systems.
Reach us at info@study.space