Community-Based Recommendation for Cold-Start Problem: A Case Study of Reciprocal Online Dating Recommendation

Mo Yu, Xiaolong Luke Zhang, Dongwon Lee, Derek Kreager

Abstract: Online dating services often use recommender systems to help people find their dates. When recommending dates to existing users who have already interacted with other users, such recommender systems tend to work well. However, recommending dates to new users who have made few interactions with others yet, the so-called “cold start” problem, still poses a problem. To address this challenge, in this paper, we propose a novel community-based recommendation framework (CBR) that can recommend dates for new users better. By detecting communities to which existing users belong and matching new users to these communities, our method is able to recommend existing users who are more likely to reply a date request from new users. Empirical validation using real data from a popular US online dating site reveals that our reciprocal online dating recommendations are significantly better than other traditional methods, achieving 5–100% improvements on average in different evaluation metrics.