.png){kind=icon}
.png){kind=icon}
Return
Interpretable preference learning and prediction: A data-driven method based on multiplicative multiattribute utility function with reference effects
.png){kind=icon}
PRE
AI
DOI：10.1016/j.inffus.2025.103311.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The data-driven preference learning provides a powerful tool to learn preferences of decision makers from data, which is also faced with challenges in a multiattribute context from complex interactions among attributes and interpretability of human behaviors. In order to deal with these challenges, we propose a data-driven method to learn multiattribute utility function based on both attribute-specific evaluation data and ordinal classification data. Specifically, a reference dependent multiplicative multiattribute utility function (RMAU) is formulated as the preference model to be learned based on data. The multiplicative function form is employed to model complex attributes interactions in a general way with reference effects incorporated to improve the model interpretability. Based on the RMAU, we propose a data-driven preference learning model and a preference prediction model. It can be shown that both the preference learning model and prediction model have some desirable properties from theoretical point of view. A comprehensive study on comparisons of the RMAU based data-driven learning and prediction models with some other popular models in the extant literature shows that both our preference learning and prediction models have competitive advantages over these models in 14 available public data sets. Furthermore, using the real-world customers' data on booking.com, we show that using the RMAU based models to learn customers' preferences provides better and clearer interpretations of model results.
Keywords:
Data-driven preference learning
Multiple attribute decision making
Reference effect
Utility model
Interpretability