Management of contextual information to enhance mobile commerce applications : issues, framework, and approaches

Abstract

We are witnessing the tremendous growth in the adoption of mobile commerce applications that strive to provide various services to mobile users. However, these applications are plagued by the inherent hardware limitations of the mobile platforms. In view of this problem, researchers and practitioners are now in attempt to utilize the widely studied personalization techniques in mobile service recommendation because personalization, which arranges products in coherence to individual user’s preference, is acknowledged to be a prospective remedy. One of the biggest challenges in such task is that conventional personalization techniques generally make decisions and conduct reasoning based on historical demographic and preference information. Therefore, nowadays’ personalization techniques have difficulties in reflecting the dynamic aspects of a user on the move, and this situation motivates researchers to incorporate user’s contextual information into mobile commerce applications. However, primitive contextual information is hard to be directly used because of its characteristics of being inaccurate and temporal. This dissertation addresses the issues of managing contextual information by proposing a conceptual framework and indicating necessary processes for transferring primitive contextual information into usable knowledge. Towards this goal the framework adopts the process view of Knowledge Management and proposes major processes of managing contextual information, including collection, preprocessing, integration, modeling and representation. The framework intends to enable the procedure of transforming data and information to knowledge. In addition, the dissertation identifies three important research problems in terms of the preprocessing, modeling and utilizing contextual information, including: (i) similarity search (ii) repetitive pattern discovery (iii) extracting rules and using for prediction. For the first problem this dissertation presents a time series similarity measuring approach for quantifying the relevance of two contextual scenarios characterized by time series. Meanwhile, this approach effectively tackles the innate inaccuracy and irregularity in primitive contextual time series. For the second problem, we propose a pattern mining approach which outperforms conventional k-motif in that it does not rely on the precise predefining of pattern length and it can aggregate underlying patterns from extracted motifs. This method can be used to identify repetitive patterns from user’s contextual time series. Regarding the third problem, we elaborate an approach for predicting mobile user’s activity via the discovering, selecting and matching of multidimensional sequential rules, which show much stronger prediction ability than conventional single dimensional rules. The associated processes (discovering, selecting and matching) can be viewed as a way of conducting knowledge modeling and matching in the Context Information Management Framework. These three proposed approaches are evaluated using both synthetic and real-world data. The proposed framework serves as a guideline for managing contextual information. Furthermore, the proposed approaches constitute contributions to problems involving temporal data, such as those arise in the enabling of context-awareness for real-world mobile commerce application.