Clustering users in an electronic business reference system

Abstract

When making strategic decisions in a business setting it is advantageous to know as much about the users of your products as possible. Information about what interest segments exist can be used for search optimization, product improvement and custom tailored marketing. This project belongs to the field of knowledge discovery in databases and concerns the discovery of user interest clusters in an electronic business reference system using an implicit voting scheme based on the sytem s web logs. A literature review is conducted to explore recent efforts in the field, experiments are conducted to apply the theory from the literature review and a qualitative analysis is conducted on the results of the experiments. The main contributions of this thesis are a comparison of Spearman Rank correlation and Frequency-Weighted Pearson correlation in terms of scalability and the application of Blondel s algorithm on a previously unexplored data set generated by users in a professional work setting. The results show that FrequencyWeighted Pearson correlation is the more scalable alternative, and that clusters do exist in the data set. Furthermore it is shown that there is seasonal variations in the data set and the discovered interest groups.