Multi-Attribute BERT for Preferences Completion in Multi-Criteria Recommender System

Abstract

For a multi-criteria recommender system (MCRS), a complete set of criteria ratings is necessary to produce an accurate recommendation. Incomplete preferences, known as the "partial preferences problem," is one of the problems in MCRS. This issue affects the performance of MCRS due to an increase in data sparsity. Criteria rating prediction is one method for completing the preferences. Therefore, this study proposes a new method for preferences completion, that is a multi-attribute Bidirectional Encoder Representations from Transformers (BERT). The proposed method incorporates reviews and overall ratings to predict incomplete criteria ratings. Rule-based adjustment is also performed to enhance the performance of the proposed method in predicting the worst rating. This study shows that the proposed method outperforms the baseline method. The proposed method is also evaluated on MCRS using a user-based multi-criteria collaborative filtering approach. The result is that it has a positive impact on the recommendation system.