TY - JOUR
T1 - Stochastic-control multi-attribute preference learning with sparse reviews
AU - Zhu, Bin
AU - Xu, Yuanzhen
AU - Luo, Shucheng
N1 - Publisher Copyright:
© 2025 Elsevier Ltd
PY - 2025/9/15
Y1 - 2025/9/15
N2 - Analyzing consumer preferences with their reviews, including ratings and text comments, is critical for marketing, recommendation, and other applications. Focusing on a target consumer's preferences represented by multi-attribute weights, the existing approaches face the difficulties when analyzing with sparse rating information. To address this issue, we propose a stochastic control multi-attribute preference learning approach. This approach can utilize both the reviews of the target consumer and reference consumers to learn this target consumer's preferred multi-attribute weights, thereby deals with the sparse issue. Specifically, this approach can efficiently learn the weights using rating information extracted from reviews. This includes a control variable that controls how to utilize other consumers’ reviews that aid the learning, which balances the computational cost and its performance. We verify the effectiveness of our approach with generated data. In addition, with real review data, we apply our approach to a recommendation scenario, and show its advantages compared with some state-of-the-art recommendation methods.
AB - Analyzing consumer preferences with their reviews, including ratings and text comments, is critical for marketing, recommendation, and other applications. Focusing on a target consumer's preferences represented by multi-attribute weights, the existing approaches face the difficulties when analyzing with sparse rating information. To address this issue, we propose a stochastic control multi-attribute preference learning approach. This approach can utilize both the reviews of the target consumer and reference consumers to learn this target consumer's preferred multi-attribute weights, thereby deals with the sparse issue. Specifically, this approach can efficiently learn the weights using rating information extracted from reviews. This includes a control variable that controls how to utilize other consumers’ reviews that aid the learning, which balances the computational cost and its performance. We verify the effectiveness of our approach with generated data. In addition, with real review data, we apply our approach to a recommendation scenario, and show its advantages compared with some state-of-the-art recommendation methods.
KW - Choice decision
KW - Multi-attribute decision making
KW - Online consumer
KW - Preference learning
UR - https://www.scopus.com/pages/publications/105007027788
U2 - 10.1016/j.eswa.2025.128271
DO - 10.1016/j.eswa.2025.128271
M3 - Article
AN - SCOPUS:105007027788
SN - 0957-4174
VL - 289
JO - Expert Systems with Applications
JF - Expert Systems with Applications
M1 - 128271
ER -