Research Article
Improving Recommender System via Personalized Reconstruction of Reviews
@INPROCEEDINGS{10.1007/978-3-030-92635-9_17, author={Zunfu Huang and Bo Wang and Hongtao Liu and Qinxue Jiang and Naixue Xiong and Yuexian Hou}, title={Improving Recommender System via Personalized Reconstruction of Reviews}, proceedings={Collaborative Computing: Networking, Applications and Worksharing. 17th EAI International Conference, CollaborateCom 2021, Virtual Event, October 16-18, 2021, Proceedings, Part I}, proceedings_a={COLLABORATECOM}, year={2022}, month={1}, keywords={Recommender system Personalized reconstruction Cross attention Cross transformer}, doi={10.1007/978-3-030-92635-9_17} }- Zunfu Huang
Bo Wang
Hongtao Liu
Qinxue Jiang
Naixue Xiong
Yuexian Hou
Year: 2022
Improving Recommender System via Personalized Reconstruction of Reviews
COLLABORATECOM
Springer
DOI: 10.1007/978-3-030-92635-9_17
Abstract
Textual reviews of items are a popular resource of online recommendation. The semantic of reviews helps to achieve improved representation of users and items for recommendation. Current review-based recommender systems understand the semantic of reviews from a static view, i.e., independent of the specific user-item pair. However, the semantic of the reviews are personalized and context-aware, i.e., same reviews can have different semantics when they are written by different users or towards different items. Therefore, we propose an improved recommendation model by reconstructing multiple reviews into a personalized document. Given a user-item pair, we design a cross-attention model to build personalized documents by selecting important words in the reviews of the given user towards the given item and vice versa. A semantic encoder of personalized document is then designed using a cross-transformer mechanism to learn document-level representation of users and items. Extensive experiments on three real-world datasets demonstrate the effectiveness of the proposed model.
