Neural News Recommendation with Long- and Short-term User Representations

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 336–345 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 336 Neural News Recommendation with Long- and Short-term User Representations Mingxiao An1,*, Fangzhao Wu2, Chuhan Wu3, Kun Zhang1, Zheng Liu2, Xing Xie2 1University of Science and Technology of China, Hefei 230026, China 2Microsoft Research Asia, Beijing 100080, China 3Department of Electronic Engineering, Tsinghua University, Beijing 100084, China {anmx,zhkun}@mail.ustc.edu.cn, wufangzhao@gmail.com wuch15@mails.tsinghua.edu.cn, {zhengliu,xingx}@microsoft.com Abstract Personalized news recommendation is impor- tant to help users find their interested news and improve reading experience. A key problem in news recommendation is learning accurate user representations to capture their interests. Users usually have both long-term preferences and short-term interests. However, existing news recommendation methods usually learn single representations of users, which may be insufficient. In this paper, we propose a neu- ral news recommendation approach which can learn both long- and short-term user represen- tations. The core of our approach is a news encoder and a user encoder. In the news en- coder, we learn representations of news from their titles and topic categories, and use atten- tion network to select important words. In the user encoder, we propose to learn long-term user representations from the embeddings of their IDs. In addition, we propose to learn short-term user representations from their re- cently browsed news via GRU network. Be- sides, we propose two methods to combine long-term and short-term user representations. The first one is using the long-term user repre- sentation to initialize the hidden state of the GRU network in short-term user representa- tion. The second one is concatenating both long- and short-term user representations as a unified user vector. Extensive experiments on a real-world dataset show our approach can effectively improve the performance of neural news recommendation. 1 Introduction Online news platforms such as MSN News1 and Google News2 which aggregate news from various sources and distribute them to users have gained *This work was done when the first author was an intern in Microsoft Research Asia. 1https://www.msn.com/news 2https://news.google.com/ 2017 NBA Championship Celebration From Warriors Rami Malek Wins the 2019 Oscar Oklahoma City Thunder vs. Golden State Warriors Bohemian Rhapsody Is Highest- Grossing Musician Biopic Ever 𝑡𝑡1 𝑡𝑡𝑖𝑖 𝑡𝑡𝑖𝑖+1 𝑡𝑡𝑗𝑗 … … Figure 1: An illustrative example of long-term and short-term interests in news reading. huge popularity and attracted hundreds of millions of users (Das et al., 2007; Wang et al., 2018). However, massive news are generated everyday, making it impossible for users to read through all news (Lian et al., 2018). Thus, personalized news recommendation is very important for online news platforms to help users find their interested contents and alleviate information overload (Lavie et al., 2010; Zheng et al., 2018). Learning accurate user representations is criti- cal for news recommendation (Okura et al., 2017). Existing news recommendation methods usually learn a single representation for each user (Okura et al., 2017; Lian et al., 2018; Wu et al., 2019). For example, Okura et al. (2017) proposed …

Content truncated to 500 words