Neural News Recommendation with Heterogeneous User Behavior

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 4874–4883, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 4874 Neural News Recommendation with Heterogeneous User Behavior Chuhan Wu1, Fangzhao Wu2, Mingxiao An3, Tao Qi1, Jianqiang Huang4, Yongfeng Huang1, and Xing Xie2 1Department of Electronic Engineering, Tsinghua University, Beijing 100084, China 2Microsoft Research Asia, Beijing 100080, China 3University of Science and Technology of China, Hefei 230026, China 4Peking University, Beijing 100871, China {wu-ch19, qit16, yfhuang}@mails.tsinghua.edu.cn, {fangzwu, xing.xie}@microsoft.com, anmx@mail.ustc.edu.cn, 1701210864@pku.edu.cn Abstract News recommendation is important for online news platforms to help users find interested news and alleviate information overload. Ex- isting news recommendation methods usually rely on the news click history to model user interest. However, these methods may suf- fer from the data sparsity problem, since the news click behaviors of many users in online news platforms are usually very limited. For- tunately, some other kinds of user behaviors such as webpage browsing and search queries can also provide useful clues of users’ news reading interest. In this paper, we propose a neural news recommendation approach which can exploit heterogeneous user behaviors. Our approach contains two major modules, i.e., news representation and user representation. In the news representation module, we learn representations of news from their titles via CNN networks, and apply attention networks to select important words. In the user rep- resentation module, we propose an attentive multi-view learning framework to learn uni- fied representations of users from their het- erogeneous behaviors such as search queries, clicked news and browsed webpages. In ad- dition, we use word- and record-level atten- tions to select informative words and behavior records. Experiments on a real-world dataset validate the effectiveness of our approach. 1 Introduction Online news platforms such as Google News1 and MSN2 News have gained huge popularity for on- line digital news reading (Das et al., 2007). Tens of thousands of news articles are streamed from various sources every day, making it very difficult for users to read all news to find their interested content (Phelan et al., 2011). Thus, personalized 1https://news.google.com/ 2https://www.msn.com/en-us/news whether google map safe cars toyota crown Twitter - Login Free and Safe Emails How to Buy a Safe Car Pros and Cons of Toyota Search Queries Browsed Webpages Browse Snow : Winter Weather This Week The 2020 Toyota Supra Is Here and It Looks Glorious NFL Winners and Losers Today Will Click? Clicked News Click Figure 1: An illustrative example of news recommen- dation based on heterogeneous user behaviors. news recommendation is critical for online news platforms to target user interests and alleviate in- formation overload (IJntema et al., 2010). News and user representation learning is criti- cal for news recommendation. Many deep learn- ing based methods have been proposed for this task (Okura et al., 2017; Wang et al., 2018). For example, Okura et al. (2017) proposed to learn news representations from news bodies via auto- encoders, and learn user …

Content truncated to 500 words