Association Rule Clustering Technique For User Interest Mining From Social Networking Sites

Abstract

Recommendation systems based on customers’ interests have been broadly emerged because of a growing number of web users and online services. Over the past decades, Association Rule Mining (ARM) was applied to extract user interests from online user profiles. In addition, hybrid Competitive Swarm Optimizer and Gravitational Search Algorithm (CSO-GSA) were utilized to choose the most relevant terms for generating the rules. However, extracting the undesirable pattern of user interest needs a prior knowledge and traditional belief model which was a time-consuming process. Hence in this article, a new framework is designed that can able to automatically identify beliefs from data and expose undesirable patterns. In this framework, a clustering technique is introduced to discover undesirable patterns in the association rules. To cluster the association rules, the non-redundant association rules between user interests are defined as the numerical feature vectors before clustering. This representation is extended to summarize and visualize the association rules and their relationships. Also, a method is proposed for pruning redundant rules depending on the association among items. According to this, the users are able to analyze and choose remarkable rules interactively. Then, the non-redundant rules are classified into 2 classes: (i) beliefs and (ii) possible candidates for undesirable rules. Further, a contradiction check is applied to expose the accurate undesirable rules from the candidates. Finally, the experimental results show that the presented method achieves higher efficiency compared to the state-of-the-art methods for user interest mining.