Explorative analysis and evaluation of commercial web information systems

This paper describes a research framework and methodology for explorative analysis and quantitative evaluation of commercial, business-to-consumer Web information systems (WIS). The suggested approach is suitable for both longitudinal studies and industry-specific comparisons. As such, it may provide the basis for constructing reference models, for optimizing the allocation of marketing resources, or for improving customer support. The presented architecture facilitates automated data gathering, which is a prerequisite for obtaining a training sample of appropriate size. Combined with parameters derived from content analysis, the information is subsequently used as input vector for two neuronal network architectures. The suggested approach uses both, unsupervised learning for WIS clustering (Kohonen self-organizing map) and supervised learning for comparing the results with a classification scheme manually specified in advance (NeuroFuzzy feedforward network).