Aspect-Similarity-Aware Historical Influence Modeling for Rating Prediction

Ryo Shimura, Shotaro Misawa, Masahiro Sato, Tomoki Taniguchi, Tomoko Ohkuma


Abstract
Many e-commerce services provide customer review systems. Previous laboratory studies have indicated that the ratings recorded by these systems differ from the actual evaluations of the users, owing to the influence of historical ratings in the system. Some studies have proposed using real-world datasets to model rating prediction. Herein, we propose an aspect-similarity-aware historical influence model for rating prediction using natural language processing techniques. In general, each user provides a rating considering different aspects. Thus, it can be assumed that historical ratings provided considering similar aspects to those of later ones will influence evaluations of users more. By focusing on the review-topic similarities, we show that our method predicts ratings more accurately than the previous historical-inference-aware model. In addition, we examine whether our model can predict “intrinsic rating,” which is given if users were not influenced by historical ratings. We performed an intrinsic rating prediction task, and showed that our model achieved improved performance. Our method can be useful to debias user ratings collected by customer review systems. The debiased ratings help users to make decision properly and systems to provide helpful recommendations. This might improve the user experience of e-commerce services.
Anthology ID:
2020.ecomnlp-1.8
Volume:
Proceedings of Workshop on Natural Language Processing in E-Commerce
Month:
Dec
Year:
2020
Address:
Barcelona, Spain
Editors:
Huasha Zhao, Parikshit Sondhi, Nguyen Bach, Sanjika Hewavitharana, Yifan He, Luo Si, Heng Ji
Venue:
EcomNLP
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
76–86
Language:
URL:
https://aclanthology.org/2020.ecomnlp-1.8
DOI:
Bibkey:
Cite (ACL):
Ryo Shimura, Shotaro Misawa, Masahiro Sato, Tomoki Taniguchi, and Tomoko Ohkuma. 2020. Aspect-Similarity-Aware Historical Influence Modeling for Rating Prediction. In Proceedings of Workshop on Natural Language Processing in E-Commerce, pages 76–86, Barcelona, Spain. Association for Computational Linguistics.
Cite (Informal):
Aspect-Similarity-Aware Historical Influence Modeling for Rating Prediction (Shimura et al., EcomNLP 2020)
Copy Citation:
PDF:
https://aclanthology.org/2020.ecomnlp-1.8.pdf