Reciprocal ranking: A hybrid ranking algorithm for reciprocal recommendation

Abstract

Reciprocal recommendation is an important class of recommendation. It is the core of many social websites like online dating, online recruitment and so on. Different from item-to-people recommenders which only need to satisfy the preference of users, reciprocal recommenders match people and people while trying to satisfy the preferences of both parties. For each user, we provide a ranking list while trying to increase the click rate as well as the probability that clicks receiving positive replies (reciprocal interactions). Most existing methods only consider either unilateral clicks or reciprocal interactions to make recommendation. Few methods consider both of these kinds of information. In this paper, we propose a novel reciprocal recommendation method called Reciprocal-Ranking (RRK), which combines the prediction of unilateral clicks and reciprocal interactions. Experimental results on both a real-world dataset and a synthetic dataset show that RRK performs better than several state-of-the-art methods.