Research And Application Of Group Recommendation Algorithm With Enhanced Fairness

With the rapid development of internet technology and social networks,traditional personalized recommendation algorithms have been unable to meet the recommendation needs of increasingly closely connected user groups.Group recommendation algorithms aim to provide recommendation services for groups of multiple users and have become a hot research topic in recommendation systems.Currently,most studies on group recommendation have focused on mitigating the preference conflicts among members to improve the recommendation accuracy,and the fairness of group recommendation has rarely been studied.On the one hand,a group is composed of multiple members,coordinating the fairness of group users,and reducing unreasonable feedback from users within the group,which plays a crucial role in the long-term survival of group recommendations.On the other hand,the items in the recommendation system show a long-tail distribution,and items with lower popularity and less user feedback lack recommendation opportunities in the group and are flooded in the recommendation system,giving items a fair chance to be selected is a problem that must be considered for group recommendation.In this thesis,we conduct an in-depth study around the fairness of group recommendations,and the main work is as follows:(1)In order to meet the needs of group users,the current group recommendation algorithm mainly uses users’ consensus preferences instead of individual preferences for group recommendation,which can only meet the needs of most users in the group and ignore the needs of a small number of users who have large differences with the group preferences.In order to improve the user fairness of group recommendation,this thesis proposes a fairness group recommendation algorithm based on dividing subgroups,the algorithm introduces the comprehensive trust value among users to divide the group into different subgroups,obtains the initial subgroup recommendation list according to the Average strategy,sets the subgroup weights using Page Rank idea so as to assign the number of recommendation items to the subgroups,and obtains the group recommendation list by combining the subgroup recommendation list and the number of items assigned to the subgroups.The experiments show that the proposed algorithm improves 1.1%,1.8%,etc.in recommendation accuracy,1.7%,4.1%,etc.in recommendation satisfaction,and 4%,7.5%,etc.in comprehensive evaluation index,respectively,compared with the Average strategy and the Least Misery strategy,etc.The algorithm in this thesis can satisfy the user’s personalized recommendation demand to the greatest extent while improving the group fair recommendation.(2)Removing popularity bias of items and improving item fairness have achieved a series of research results in traditional recommendation algorithms,but little research has been done on item fairness in group recommendation.To address the fairness of items in group recommendation systems,this thesis proposes a fairness group recommendation algorithm based on item popularity and quality.The algorithm filters users with low similarity at the time of group discovery and sets weights for the items by fusing their popularity and quality.Also considering that user preferences change over time,a time decay function is set to recalculate the item prediction scores and obtain the group recommendation list.The experimental results show that the proposed algorithm improves the recommendation accuracy by 1%,3.7%,etc.,the recommendation list diversity by 4.6%,0.9%,etc.,and the recommendation list coverage by 3.5%,4.1%,etc.,compared with the Average strategy and the Least Misery strategy,respectively.(3)A group recommendation prototype system with movie recommendation as an example is designed and implemented for the fairness of group members.The system can not only provide group recommendation for users,but also provide recommendation service according to users’ individual needs.Comparing this system with the group recommendation system under the mean value strategy shows that the group recommendation prototype system designed in this thesis can effectively improve the fairness of group members,which confirms the theoretical value and practical significance of group recommendation.