Profiling television viewing using data mining

Abstract

This study conducted a critical review of data-mining techniques used to extract meaningful information from very large databases. The study aimed to determine cluster analysis methods suitable for the analysis of binary television-viewing data. Television-viewing data from the South African Broadcasting Corporation was used for the analysis. Partitioning and hierarchical clustering methods are compared in the dissertation. The study also examines distance measures used in the clustering of binary data. Particular consideration was given to methods for determining the most appropriate number of clusters to extract. Based on the results of the cluster analysis, four television-viewer profiles were determined. These viewer profiles will enable the South African Broadcasting Corporation to provide viewer-targeted programming.