Rrcn: a reinforced random convolutional network-based reciprocal recommendation approach for online dating

Linhao Luo, Liqi Yang, Ju Xin, Yixiang Fang, Xiaofeng Zhang, Kai Liu

Abstract: Recently, reciprocal recommendation, especially for online dating applications, has attracted increasing research attention. Different from conventional recommendation problems, reciprocal recommendation aims to simultaneously best match users’ mutual interests. Lots of existing methods adopt user attributes and consider the interactions between attributes to capture such interest. However, these methods failed to notice that the interactions might imply both users’ preference and repulsiveness to other users, which contributes opposite to the mutual interest. Moreover, the interactions grow exponentially with user attributes, posing a great challenge for distinguishing them. To address aforementioned issues, in this paper, we propose a novel reinforced random convolutional network (RRCN) approach for reciprocal recommendation. In particular, we first separately consider the preferred and repulsive interactions, where their contributions can be modeled individually. Then, we technically propose a novel random CNN component that can randomly convolute nonadjacent features and learn representations of different interaction subsets. Furthermore, we design a reinforcement learning-based strategy to integrate with the random CNN component to select important interactions for recommendation, so that the number of attribute interactions can be reduced. We evaluate the proposed RRCN against a number of both baselines and state-of-the-art approaches on two real-world datasets, and the promising results have demonstrated the superiority of RRCN against the compared approaches in terms of a number of evaluation criteria.

External IDs:dblp:journals/kais/LuoYXFZL25