An Interpretable Preference Learning Model Admitting Dynamic and Context-Dependent Preferences

We introduce a novel Bayesian preference learning approach for online user rating scenarios. Leveraging user-generated textual reviews, our model automatically extracts feature-opinion-polarity triplets based on phrase-level sentiment analysis. They are summarized as positive and negative topics, providing a low-dimensional feature-based representation of items. Moreover, our approach models the dynamic nature of user behavior and context-dependent preferences by inferring latent motivations represented as discrete distributions over items and contextual preference models. We apply the proposed model to three real-world data sets. Also, we demonstrate its scalability, interpretability, and superior predictive performance against seven benchmark algorithms in the extensive computational experiment. Our research contributes to advancing preference learning in domains such as recommender systems, decision making, and marketing, offering both actionable insights and transparent recommendations.

History: Accepted by Ram Ramesh, Area Editor for Data Science and Machine Learning.

Funding: J. Liu acknowledges support from the National Natural Science Foundation of China [Grants 72471184 and 72071155]. M. Kadziński acknowledges support from the Polish National Science Center under the SONATA BIS project [Grant DEC-2019/34/E/HS4/00045]. X. Liao acknowledges support from the National Natural Science Foundation of China [Grant 71872144].

Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information (https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2023.0372) as well as from the IJOC GitHub software repository (https://github.com/INFORMSJoC/2023.0372). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/.