News Recommendation Model Based on Long-Term and Short-Term Interests.

Lei Ren,Dehao Chen,Zexin Miao,Huisheng Zhang

DOI: https://doi.org/10.1145/3565291.3565321

2022-01-01

Abstract:Nowadays internet users can receive massive amounts of information every day. Providing users with personalized news that they are concerned about together with improving their reading interest has become a hot and difficult research topic. Long-term interests of users are usually relatively stable, but short-term interests are easy to change, and usually related to hot events in a certain period of time. Existing research on news recommendation methods usually focuses on learning long-term interest representations of users and omits the short-term one. In this paper, a news recommendation system based on user's long-term and short-term interests is proposed. The proposed method learns the long-term interest representation from all the articles that the user has browsed, and at the same time, by finding the users whose long-term interests are most similar to the user in history, it learns the news type representation they have recently read together as the short-term interest representation of the current user. Finally, the long-term interest representation vector and the short-term interest representation vector are jointly considered to obtain the final long-term and short-term interest representation. Extensive experiments are conducted to show the efficiency of the proposed model in improving the accuracy of the news recommendation system.