Trust-based User Reliability Prediction with Association Rule Clustering for Mining User Interests in Social Networks

Abstract

Due to the emergence of virtual assistants and online users, the recommendation models driven by users’ preferences have evolved enormously. To mine the user’s preferences from the social networking platforms, many data mining techniques have emerged. Amongst, a hybrid Competitive Swarm Optimizer and Gravitational Search Algorithm (CSO-GSA) was applied to select the most appropriate terms to generate the rules based on the Association Rule Mining (ARM). Also, rule-based clustering was employed to obtain unwanted patterns in the generated rules and extract the user preferences. In contrast, the terms from every user were treated equally; yet, the influence of every user on social media changes based on their trust. Therefore in this paper, a new trust framework is proposed, which systematically computes the implicit reliability of online users. In this framework, a novel user's reliability prediction framework is provided based on the social media variables saliency and relevance to trust. The basic factors of social media, such as user-user interactions and user profile data are precisely mapped to the trust factors: integrity, fidelity and Social Relationship Strength (SRS). This trust framework relies on Homophily, familiarity and behavior. Also, to predict a user's reliability, this paradigm integrates subjective and objective trust criteria. Further, the characteristic vectors for the rules are created according to the user’s reliability to obtain the desired rules. Those obtained rules help to extract the user’s preferences for a recommendation. Finally, the experimental results show that the presented framework accomplishes better effectiveness than the classical frameworks for user preference extraction