Pacos: Modeling Users' Interpretable and Context-Dependent Choices in Preference Reversals

Choice problems refer to the problem of selecting the best choices from several available items, and learning users' preferences in choice problems is of great importance in understanding users' decision making mechanisms and providing personalized services. Existing works typically assume that people evaluate items independently. In practice, however, users' preferences depend on the market in which items are placed, which is known as the context effects; and the order of users' preferences for two items may even be reversed, which is called to preference reversals. In this work, we identify three factors contributing to the context effects: users' adaptive weights, the inter-item comparison, and display positions. We propose a context-dependent preference model named Pacos as a unified framework to address three factors simultaneously, and consider two design methods including an additive method with high interpretability and an ANN-based method with high accuracy. We study the conditions for preference reversals to occur and provide a theoretical proof of the effectiveness of Pacos in predicting when preference reversals would occur. Experimental results show that the proposed method has better performance than prior works in predicting users' choices, and has great interpretability to help understand the cause of preference reversals. Choice problems, such as purchasing a festival gift or picking a restaurant, involve comparing several available items. Previous works on preference modeling and analysis typically assume that people evaluate items independently, and the relative preference between two items is fixed regardless of other competing options [1]. However, numerous studies show that the above independence assumption is frequently violated in reality [2], [3]. It is essential to model how the relative preference is influenced by competing options and figure out how people select their best choices. This study can help understand users' decision making mechanisms and offer personalized services, and provide important guidelines on pricing strategies and sales forecasts. To show this independence violation, we conduct a real user test. In our test, we set two markets of Xiaomi scale, as shown in Figure 1 (a) and (b). In these two markets, we consider sellers described by two attributes: price (¥) and seller reputation (REP).