A novel trust recommendation model in online social networks using soft computing methods

• Published in: Concurrency and computation Vol. 34; no. 22
• Main Authors: Sirisala, NageswaraRao, Yarava, Anitha, Reddy, Y. C. A. Padmanabha, Poola, Veeresh
• Format: Journal Article
• Language: English
• Published: Hoboken, USA John Wiley & Sons, Inc 10.10.2022; Wiley Subscription Services, Inc
• Subjects: Asymptotic methods; biased recommendations; Clustering; Dempster Shafer theory; Recommender systems; Scientific papers; Social networks; Soft computing; trust; unsupervised learning
• ISSN: 1532-0626; 1532-0634
• DOI: 10.1002/cpe.7153

Summary Social network (OSN) is an emerging platform through which people can connect with their friends, relatives, and other like‐minded people. On the other hand, users' personal information might be misused because of other users' biased and malicious behavior. Establishing a trusted environment in social networks is one of the current research problems. Some of the research papers proposed to trust computational methods, but still, there is a lack of methods to handle biased recommendations and loss of trust accuracy towards the target user. In this article, to address these open issues, “a novel trust recommendation model in online social networks using soft computing methods (TRMSC)” is proposed for the Twitter social networks. Here direct and indirect trust is computed for known and unknown users, respectively. The direct trust of a user is computed using clustering methods based on his social activities (posts, retweets received, mentions received, listed count, and follower count) with other users. In the computation of indirect trust, the impact of biased recommendations is suppressed using the Dempster Shafer theory(DST) method, and loss of trust is minimized using trust transitive matrices. The performance of the proposed method is analyzed theoretically and experimentally. Time and space complexities are measured using asymptotic notations. In experimental results, TRMSC is evaluated for different network sizes and for target users at different distances (2 to 4‐hops), where it could perform better than existing methods.