Extraction des règles d’association à haute utilité basée sur les représentations concises appliquée aux données géolocalisées de Gowalla

Abstract

Our research focuses on the extraction of high-utility association rules from data originating from Location-Based Social Networks (LBSNs), a task that poses significant challenges due to the complexity and scale of the data. In this study, we concentrate on the Gowalla dataset, a social network centered around users’ check-in locations. Unlike traditional approaches that rely solely on frequency, our methodology is based on the uti lity value of itemsets, considering their economic or contextual importance. We propose an efficient approach based on concise representations, specifically HUCI (High Utility Closed Itemsets), to reduce redundancy and the volume of generated rules. The HUCI-Miner algorithm is employed to extract high-utility closed itemsets, followed by the HGBalgorithm to generate complete and non-redundant high-utility association rules. These rules are evaluated using metrics such as : UTIL (total utility), AUTIL (antecedent utility), and UCONF (utility based confidence). The experiments conducted on the Gowalla dataset demonstrate that this approach enables the ex traction of concise, meaningful, and knowledge-rich rules, making it suitable for applications such as location recommendation and user behavior analysis in urban environments.