Incorporating global and local social networks for group recommendations. (July 2022)

• Record Type: Journal Article
• Authors: Leng, Youfang; Yu, Li
• Is Part Of: Pattern recognition. Volume 127(2022)
• Publication Date: 2022-07
• Subjects: Group recommendation -- Recommendation systems -- Graph neural network -- Social network analysis -- Graph-based method; Pattern perception -- Periodicals; Perception des structures -- Périodiques; Patroonherkenning; 006.4
• DOI: 10.1016/j.patcog.2022.108601
• Languages: English
• ISSNs: 0031-3203

Abstract

Highlights: We propose GLOW for group recommendations from global and local social networks, fully exploiting social interaction at macro and micro levels. Multi-layer attentive GCN based Global Network Diffusion module is proposed to model the social influence diffusion in social networks. We propose a multi-channel attention based LNF module to model the group decision-making process, capturing multiple types of interaction among group members and dynamically assigning different weights to each member. Experiments on two real-world datasets prove that our model achieves state-of-the-art in group recommendation. Abstract: Due to the social nature of human beings, group activities have become an integral part of daily life. This creates the need for an in-depth study of the group-recommendation task: recommending items to a group of users. Unlike individual decision-making, which relies primarily on personal preferences, group decision-making is a process of negotiation and agreement among group members, in which social characteristics are a critical factor in achieving positive recommendation results. Therefore, in this paper, we propose a new model to solve the group recommendation problem from both global and local social networks. In a global network, a user's social influence spreads through social connections and affects the preferences of others. In a local network, group members may contribute differently to the final decision, forming a dynamic negotiation and consensus process. We propose to model global and local networks with two components: 1) an attentive graph convolutional network based global network diffusion (GND) module to simulate the spread of social influence and capture the social gate of each user, and 2) a multi-channel attention-based local network fusion (LNF) module to learn the complex decision-making process among group members and integrate them into a final representation of the group. Finally, two separate neural collaborative filtering (NCF) modules are presented to model group-item and user-item interactions, respectively, to enhance each other. Extensive experimental results from two real-world datasets show the effectiveness of our proposed model.