International Journal of Knowledge-based and Intelligent Engineering Systems 19 (2015) 109–116 109 DOI 10.3233/KES-150312 IOS Press From concrete to inferred knowledge: Enhanced mining constraint-based cyclic association rules from medical social network Wafa Tebourski a,∗ , Wahiba Ben Abdessalem Kar a and Henda Ben Ghezala b a Computer Science Department, High Institute of Management, Tunis, Tunisie b Computer Science Department, National School of Computer Sciences, Tunis, Tunisie Abstract. In recent years, social network has been given much interest. The explosion of social network activity has lead to generation of massive volumes of user-related data and has given birth to a new area of data analysis. In parallel, the last decade witnessed a fastidious interest in the data mining which efficiently find hidden knowledge that can be extracted from applied information, namely, social data. Among the most used data mining techniques, we particularly focus on cyclic constraint-based association rules. In this paper, we aim to derive significant cyclic constraint-based association rules from social data. Thus, we introduce a new approach EMC 2 for social mining through constraint-based cyclic association rules extraction. The encouraging experimental results carried out prove the usefulness of our approach. Keywords: Social network, Twitter, medicine, data mining, data warehouse, constraint, cyclic association rule, multidimensional association rules 1. Introduction A social network describes the social structure be- tween actors associated through various relationships ranging from informal acquaintance to close familiar relations. For instance, notable examples of social net- works are Twitter, Facebook, LinkedIn, Viadeo, etc. Such a structure may be modeled using multidimen- sional modeling. Indeed, the managed information in social network context may be described using varied dimensions and aggregated through numerous hierar- chies. In this paper, we focus on data warehouse as a tool for the representation and investigation of multidi- mensional social networking data. Performing On-line analytical processing (OLAP) on built data warehouse provides means to explore data cubes in order to extract attractive information. The ∗ Corresponding author: Wafa Tebourski, Computer Science De- partment, High Institute of Management, Tunis, Tunisie. E-mail: wafatebourskiisg@yahoo.com. combination of OLAP and data mining can bring ex- planations of any correlation that may exist between the multidimensional data. In this context, the associa- tion rules were performed on data cubes. An excessive patterns number may be generated due to the highly spare data. Thus, we focus on a partic- ular class of association rules which is the constraint- based patterns [20]. Aiming to improve the quality of generated association rules, several researches were devoted to integrate constraints on association rules. These constraints may concern both the content of frequent item sets, and the form of association rule, namely the premise and/or the consequence. Moreover, such extracted rules may be repeated reg- ularly every given cycle which is, known as cyclic as- sociation rules [3]. The latter emphasizes the time fea- ture. Several studies have been dedicated to mining cyclic association rules from datasets [3–5,7,10,15]. These researches consider only a single dimension to generate cyclic rules: the product dimension. More ad- vanced proposals deal with the combination of several dimensions. ISSN 1327-2314/15/$35.00 c  2015 – IOS Press and the authors. All rights reserved AUTHOR COPY