Comment recommendation based on graph bidirectional aggregation networks. (30th December 2022)

• Record Type: Journal Article
• Authors: Wang, Muchen; Chen, Chao; Li, Zhao; Shu, Minglei
• Is Part Of: Expert systems with applications. Volume 210(2022)
• Publication Date: 2022-12-30
• Subjects: Recommender systems -- Graph neural network -- Online comments -- Node embedding -- Link prediction; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2022.118506
• Languages: English
• ISSNs: 0957-4174

Abstract

Abstract: The function of "Comment" is widely used on the Internet with high efficiency and convenience. However, the quality of comments may have different effects depending on the preferences of the viewers. High-quality comments that users like will promote communication between users and bring positive effects, while low-quality comments that users dislike will hinder communication between users, even spread negative emotions and affect user experience. To avoid the influence, a comment recommendation by link prediction is implemented. Meanwhile, the quality of input features affects the recommendation results. For this reason, a link prediction model based on graph bidirectional aggregation networks (GBAN) is proposed. It constructs user features by text information of comment, adds attention mechanism to consider the impact of different followers in social networks, and modifies the feature representation of user nodes by the reverse aggregation mechanism. Finally, the link prediction method is applied to make recommendations based on the scores of edges among nodes. Experimental results show that the proposed method can make the recommendation performance better. Highlights: A graph bidirectional aggregation network is proposed for strengthening node representation capabilities. Use multi-task learning to fuse node embedding and link prediction models. Alleviate the impact of online trolls on the daily life of netizens based on the proposed method.