Data-driven preference learning methods for sorting problems with multiple temporal criteria

We present novel preference learning approaches for sorting problems with multiple temporal criteria. They leverage an additive value function as the basic preference model, adapted for accommodating time series data. Given assignment examples concerning reference alternatives, we learn such a model using convex quadratic programming. It is characterized by fixed-time discount factors and operates within a regularization framework. This approach enables the consideration of temporal interdependencies between timestamps while mitigating the risk of overfitting. To enhance scalability and accommodate learnable time discount factors, we introduce a novel monotonic Recurrent Neural Network (mRNN). It captures the evolving dynamics of preferences over time, while upholding critical properties inherent to multiple criteria sorting problems. These include criteria monotonicity, preference independence, and the natural ordering of classes. The proposed mRNN can describe the preference dynamics by depicting piecewise linear marginal value functions and personalized time discount factors along with time. Thus, it effectively combines the interpretability of traditional sorting methods with the predictive potential offered by deep preference learning models. We comprehensively assess the proposed models on synthetic data scenarios and a real-case study centered on classifying valuable users within a mobile gaming app based on their historical in-game behavioral sequences. Empirical findings underscore the notable performance improvements achieved by the proposed models when compared to a spectrum of baseline methods, spanning machine learning, deep learning, and conventional multiple criteria sorting approaches.