Translate 
Abstract
Recommendation systems nowadays are deployed everywhere to solve the information overload problem. Thousands of news items that appear on online news platforms also need recommendations. However, news recommendation differs from product recommendation because of its focus on public interest. In this paper, we propose a multi-role social behavior model that includes a chaser model, which describes behavior patterns of chasing popular news, and a sider model, which describes behavior patterns of seeking news from a similar ideology standpoint. We propose a framework that integrates this model into a graph-based recommendation system. Unlike other intent disentangling techniques, our model explicit models social behavior patterns originated from public interest. We test our framework with two real-world news recommendation datasets. Compared to state-of-the-art baseline news recommendation models, our method achieves significantly higher recommendation accuracy. By analyzing the model outputs, we also gain a better understanding of news-seeking social behavior.
Highlights
Recommendation systems are designed to solve the information overload problem by providing personalized item recommendations
News is one major type of information that nowadays faces the information overload problem, as thousands of news items appear in online news platforms each day
We propose a method called Chaser-Sider Multi-role Embedding Learning (CS-MREL) for graph-based news recommendation
Summary
Recommendation systems are designed to solve the information overload problem by providing personalized item recommendations. Recommendation systems have been used to recommend e-commerce products [1], online videos [2], location-based services [3], and social network friends [4]. There are thousands or even millions of candidate items, and the task is to find what could be interesting for millions of users. News is one major type of information that nowadays faces the information overload problem, as thousands of news items appear in online news platforms each day. There is a need to design recommendation systems for news [5]. The mainstream recommendation systems nowadays are mostly based on collaborative filtering, in which a user’s preference is learned by comparing this user with similar users [6]. Matrix factorization is a neat solution in collaborative filtering.
Sign-in/Register to view highlights & summary 
Published Version
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
