Development of novel fuzzy clustering techniques in the context of e-learning

This thesis investigates the performance of fuzzy clustering for dynamically discovering content relationships in e-Learning material based on document metadata descriptions. This form of knowledge representation is exploited to enable flexible content navigation in eLearning environments. However, the methods and tools developed in this thesis have wider applicability. The purpose of clustering techniques is to determine underlying structures and relations in data sets usually based on distance or proximity measures. A number of clustering methods to suit particular applications have been developed throughout the years. This thesis specifically considers the well-known Fuzzy c-Means (FCM) clustering technique as the basis for document clustering. Initially, novel expressions are developed to extend the FCM algorithm, which is based on the Euclidean metric, to an algorithm based on other proximity measures more appropriate for quantifying document relationships. These include the cosine, Jaccard and overlap similarity coefficients. This novel algorithm works with normalised k-dimensional data vectors that lie in hyper-sphere of unit radius and hence has been named Hyper-Spherical Fuzzy c-Means (H-FCN). Subsequently, the performance of the H-FCM algorithm is compared to that of the FCM as well as conventional hard (ie non-fuzzy) clustering algorithms with respect to four test document collections. Both the impact of different proximity measures as well as the impact of pre-processing the document vector representations for dimensionality reduction are thoroughly investigated. Results demonstrate that the H-FCM clustering method outperforms both the conventional FCM method as well as hard clustering techniques. This thesis also considers the integration of fuzzy clustering techniques in an end-to- end e-Leaming system. In particular, a tool to convert the H-FCM document clustering outcome into a knowledge representation, based on the Topic Maps standard, suitable for Web-based environments is developed. Moreover, a tool to enable flexible navigation of e-Learning material based on the fuzzy knowledge space is also developed. This tool is deployed in a real e-Learning environment where user trials are carried out. Finally, this thesis considers the important problem of defining a suitable number of clusters for appropriately capturing the concepts of the knowledge space. In particular, an hierarchical H-FCM algorithm is developed where the sought granularity level defines the number of clusters. In this algorithm, a novel heuristic based on asymmetric similarity measures is exploited to link document clusters hierarchically and to form a topic hierarchy