Dual Preference Distribution Learning for Explainable Item Recommendation

Recommender systems can automatically recommend users with items that they probably like. The goal of them is to represent the user and item and model their interaction. Existing methods have primarily learned the user's preferences and item's features with vectorized representations, and modeled the user-item interaction by the similarity of them. In fact, the user's preferences to items can be traced to his/her preferences to item attributes, and the user's different preferences are also related. Thus, exploring such fine-grained preferences and modeling their relationships could help better understand the user's preferences. Toward this end, we propose a dual preference distribution learning framework, which jointly captures the user's preferences to both the items and attributes by a Gaussian distribution, termed the general and specific preference distributions, respectively. In this manner, the mean vector of the Gaussian distribution can capture the user's preferences, while its covariance matrix can learn their relationships. The proposed DUPLE is a generative method that can produce the preference distribution for a given user. Thus, by tracking the user's specific preference distribution, w can summarize a preferred attribute profile for each user, depicting his/her preferred item attributes. Based on that, we can provide the explanation of recommending an item with the overlap between the item attributes of the user prefers and the item has. Extensive quantitative and qualitative experiments on six public datasets demonstrate the effectiveness and explainability of the DUPLE method.