Neural Attentional Rating Regression with Review-level Explanations

被引：425

作者：

Chen, Chong ^{[1
]}

Zhang, Min ^{[1
]}

Liu, Yiqun ^{[1
]}

Ma, Shaoping ^{[1
]}

机构：

[1] Tsinghua Univ, DCST, Beijing, Peoples R China

来源：

WEB CONFERENCE 2018: PROCEEDINGS OF THE WORLD WIDE WEB CONFERENCE (WWW2018) | 2018年

关键词：

Recommender Systems; Neural Attention Network; Explainable Recommendation; Review Usefulness; AGREEMENT;

D O I：

10.1145/3178876.3186070

中图分类号：

TP39 [计算机的应用];

学科分类号：

081203 ; 0835 ;

摘要：

Reviews information is dominant for users to make online purchasing decisions in e-commerces. However, the usefulness of reviews is varied. We argue that less-useful reviews hurt model's performance, and are also less meaningful for user's reference. While some existing models utilize reviews for improving the performance of recommender systems, few of them consider the usefulness of reviews for recommendation quality. In this paper, we introduce a novel attention meclunism to explore the usefulness of reviews, and propose a Neural Attentional Regression model with Review-level Explanations (NARRE) for recommendation. Specifically, NARRE can not only predict precise ratings, but also learn the usefulness of each review simultaneously. Therefore, the Ifighly-useful reviews are obtained which provide review-level explanations to help users make better arid faster decisions. Extensive experiments on benchmark datasets of Amazon and Yelp on different domains show that the. proposed NARRE model consistently outperforms the stale-of-the-art recommendation approaches, including PME, NMP, SW++, HFT, and DeepCoNN in terms of rating prediction, by the proposed attention model that takes review usefulness into consideration. Furthermore, the selected reviews are shown to be effective when taking existing review-usefulness ratings in the system as ground truth. Besides, crowd-sourcing based evaluations reveal that in most cases, NARRE achieves equal or even better performances than systenfs usefulness rating method in selecting reviews. And it is flexible to offer great help on the dominant cases in real e-commerce scenarios when the ratings on review-usefulness are not available in the system.

引用

页码：1583 / 1592

页数：10

共 45 条

[31] Nair V., 2010, P 27 INT C MACH LEAR, P807
[32] Pennington Jeffrey, 2014, C EMP METH NAT LANG, P1532
[33] Social Collaborative Viewpoint Regression with Explainable Recommendations
Ren, Zhaochun
Liang, Shangsong
Li, Piji
Wang, Shuaiqiang
de Rijke, Maarten
[J]. WSDM'17: PROCEEDINGS OF THE TENTH ACM INTERNATIONAL CONFERENCE ON WEB SEARCH AND DATA MINING, 2017, : 485 - 494
[34] Rendle Steffen, 2010, Proceedings 2010 10th IEEE International Conference on Data Mining (ICDM 2010), P995, DOI 10.1109/ICDM.2010.127
[35] Ribeiro M. T., 2016, P 22 ACM SIGKDD INT, DOI DOI 10.1145/2939672.2939778
[36] Ricci F, 2011, RECOMMENDER SYSTEMS HANDBOOK, P1, DOI 10.1007/978-0-387-85820-3_1
[37] Salakhutdinov R., 2008, Advances in Neural Information Processing Systems, P1257
[38] Srivastava N, 2014, J MACH LEARN RES, V15, P1929
[39] Su X., 2009, ADV ARTIF INTELL, V2009, P4, DOI [10.1155/2009/421425, DOI 10.1155/2009/421425]
[40] Vig J., 2009, P 14 INT C INT US IN, P47, DOI 10.1145/1502650.1502661

← 1 2 3 4 5 →