DFGR: Diversity and Fairness Awareness of Group Recommendation in an Event-based Social Network

Abstract

An event-based social network is a new type of social network that combines online and offline networks, and one of its important problems is recommending suitable activities to users. However, the current research seldom considers balancing the accuracy, diversity and fairness of group activity recommendations. To solve this problem, we propose a group activity recommendation approach that considers fairness and diversity perception. Firstly, we calculate activity similarity based on the context and construct an activity similarity graph. We define the weighted coverage on the similarity graph as a submodular function and transform the problem of fair and diverse group activity recommendation into maximizing the weighted coverage on the similarity graph while considering accuracy, fairness, and diversity. Secondly, we employ a greedy algorithm to find an approximate solution that maximizes the weighted coverage with an approximation ratio. Finally, we conducted experiments on two real datasets and demonstrate the superiority of our method compared to existing approaches. Specifically, in the domain of diversity-based recommendation algorithms, our method achieves a remarkable 0.02% increase in recall rate. Furthermore, in the domain of fairness-based recommendation algorithms, our proposed method outperforms the latest approach by 0.05% in terms of overall metrics. These results highlight the effectiveness of our method in achieving a better balance among accuracy, fairness, and diversity.