Generative News Recommendation
Shen Gao, Jiabao Fang, Quan Tu, Zhitao Yao, Zhumin Chen, Pengjie Ren, Zhaochun Ren
TL;DR
The paper addresses the disconnect between traditional news recommendations and user-centric, narrative-driven understanding by introducing Generative News Recommendation (GNR), which leverages Large Language Models to create theme-level representations and to generate coherent multi-news narratives. The approach comprises three modules: Generative Dual-level Representation, Personalized Related News Exploration, and Interest-aware Multi-news Narrative Fusion, with an additional training mechanism called User Interest Alignment Fine-Tuning (UIFT). Empirical results on the MIND dataset show that dual-level representations improve standard recommendation metrics and that the generated narratives are more personalized and factually consistent, especially when guided by UIFT and a carefully selected reference news set. The work demonstrates a practical integration of retrieval and generation, translating user interests into concise, narrative overviews that accompany recommended news. Overall, GNR advances personalized, narrative-rich news delivery and offers a scalable framework for combining implicit article relationships with user preferences in real time.
Abstract
Most existing news recommendation methods tackle this task by conducting semantic matching between candidate news and user representation produced by historical clicked news. However, they overlook the high-level connections among different news articles and also ignore the profound relationship between these news articles and users. And the definition of these methods dictates that they can only deliver news articles as-is. On the contrary, integrating several relevant news articles into a coherent narrative would assist users in gaining a quicker and more comprehensive understanding of events. In this paper, we propose a novel generative news recommendation paradigm that includes two steps: (1) Leveraging the internal knowledge and reasoning capabilities of the Large Language Model (LLM) to perform high-level matching between candidate news and user representation; (2) Generating a coherent and logically structured narrative based on the associations between related news and user interests, thus engaging users in further reading of the news. Specifically, we propose GNR to implement the generative news recommendation paradigm. First, we compose the dual-level representation of news and users by leveraging LLM to generate theme-level representations and combine them with semantic-level representations. Next, in order to generate a coherent narrative, we explore the news relation and filter the related news according to the user preference. Finally, we propose a novel training method named UIFT to train the LLM to fuse multiple news articles in a coherent narrative. Extensive experiments show that GNR can improve recommendation accuracy and eventually generate more personalized and factually consistent narratives.