Modeling User Exposure with Explicit and Implicit Social Relations for Recommendation
Sun Can, Chongyang Shi · 2019
Social recommender systems have been well studied in both academia and industry. Social information helps to solve the data sparsity and cold start problems in traditional recommender systems, while most existing works in social recommendation assume that social friends have similar preferences. This assumption is too strict and not accord with real world situations, because of the diversity of social relations. We tend to share item information with our socially connected friends. We don't know whether they will like the items, while we help them be exposed to the items. So we model the social information for exposure rather than preferences. In this paper, we propose a novel social exposure-based recommendation model by integrating social information into the basic ExpoMF model [5]. In order to address the sparse issue in social network, we exploit implicit social relations. To the author's knowledge, the work reported is the first to extend exposure model with explicit and implicit social relations for recommendation. Experimental results on the two public datasets demonstrate that our approach SoEx++ performs the best comparing to other three models.