Distributed Representations for Building Profiles of Users and Items from Text Reviews

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2143–2153, Osaka, Japan, December 11-17 2016. Distributed Representations for Building Profiles of Users and Items from Text Reviews Wenliang Chen†, Zhenjie Zhang‡, Zhenghua Li†, Min Zhang† †School of Computer Science and Technology, Soochow University, China ‡Advanced Digital Sciences Center, Illinois at Singapore Pte. Ltd., Singapore {wlchen, zhli13, minzhang}@suda.edu.cn zhenjie@adsc.com.sg Abstract In this paper, we propose an approach to learn distributed representations of users and items from text comments for recommendation systems. Traditional recommendation algorithms, e.g. collaborative filtering and matrix completion, are not designed to exploit the key information hidden in the text comments, while existing opinion mining methods do not provide direct sup- port to recommendation systems with useful features on users and items. Our approach attempts to construct vectors to represent profiles of users and items under a unified framework to max- imize word appearance likelihood. Then, the vector representations are used for a recommen- dation task in which we predict scores on unobserved user-item pairs without given texts. The recommendation-aware distributed representation approach is fully supported by effective and ef- ficient learning algorithms over massive text archive. Our empirical evaluations on real datasets show that our system outperforms the state-of-the-art baseline systems. 1 Introduction With the prosperity of Internet-based e-commerce in the last decade, millions of item (e.g., prod- uct/serivce) comments are now available online and even more are flooding in an explosive manner. Yelp (www.yelp.com), as the largest restaurant comment web site, for example, attracts more than 140 million users to contribute new text comments every month, covering almost all restaurants in USA. It is widely believed that text comments contain much richer information on users as well as items than dull scores, with far more precise descriptions on personal preferences and item features. To fully unleash the potential value underneath the text comments, huge research efforts are now devoted to the design and development of new technologies to capture and understand the semantics in the comments, particularly those associated with users and items. The text analytical outcomes are expected to support decision making and provide new business opportunities in e-commerce. Recommendation is recognized as one of the most important components in e-commerce systems, driving the growth of profits. Traditional recommendation systems rely on scores over user-item pairs under a bipartite graph model, such that accurate score prediction over unobserved user-item pair helps the system to identify potential interested buyers. A huge bulk of prediction and recommendation strate- gies are proposed in this domain, e.g. collaborative filtering and matrix completion (Sarwar et al., 2001; Koren et al., 2009). Because of the limited information included in each individual score, such recom- mendation strategies commonly suffer from the cold-start problem (Zhou et al., 2011), which returns poor recommendations to newcoming users. While text mining on comment archive may provide im- portant information to the recommendation engine, even for fresh users, the integration of such infor- mation adds new complexity and challenges to the system, since …

Content truncated to 500 words