Bridging the Information Gap Between Domain-Specific Model and General LLM for Personalized Recommendation

Wenxuan Zhang, Hongzhi Liu, Zhijin Dong, Yingpeng Du, Chen Zhu, Yang Song, Hengshu Zhu, Zhonghai Wu

Abstract: Personalized recommendation is widely applicable in various domains like e-commerce and social media. Few recent research efforts have attempted to design general large language model (LLM) based recommenders to alleviate the issue of data sparsity. However, these methods struggle to make use of task-related information which is difficult to be expressed in natural language. On the other hand, the high latency of LLM inference limits their practical application in industrial scenarios. To address these issues, we propose a method to bridge the information gap between the domain-specific models and the general LLMs. Specifically, we propose an information sharing module acting as a bridge for collaborative training between the LLMs and domain-specific models. On the other hand, the inference of LLMs and domain-specific models can be performed independently, offering distinct recommendation paradigms to meet the varied usage habits of different users, as well as ensuring the high-efficiency advantage of domain-specific models. Experimental results on four real-world datasets have demonstrated the effectiveness of the proposed method.