A type-2 fuzzy review topic-based model for personalized recommendation

Abstract

Recommender systems are becoming increasingly indispensable for e-commerce. To achieve interpretable recommendation, review topic-based recommendation is an important research area that aims to infer users’ ratings over their unrated items using existing reviews and corresponding ratings simultaneously. However, combining latent factors and review topics can also introduce uncertainties in the model learning process, both in inference and sampling. To address this challenge, we propose a new model called type-2 fuzzy review topic-based recommendation (T2FR). T2 fuzzy membership functions are introduced to represent the topic parameter uncertainties, and a strategy of dual sampling is developed to deal with the topic T2 membership functions and further used for uncertainty handling. Abundant experiments on data collected from real-world e-commerce platforms demonstrate the effectiveness of the proposed model compared with baseline methods in terms of rating prediction accuracy with interpretation. Our proposed T2FR can substantially benefit both e-commerce platforms and consumers.