Marknadssegmentering med klustringsalgoritmer på data från streaming av filmer och TV-serier

Sammanfattning: Over the last years, the growth and development of video on demand (VOD) services has given new possibilities of performing machine learning on large amounts of video history data. A common usage of machine learning for businesses is market segmentation, which is usually addressed with cluster analysis. Market segmentation with cluster analysis has been performed for the video streaming service company Viaplay. It was found that K-means with cosine measure performed best of the attempted methods and has been shown to facilitate a useful and interpretable market segmentation based on a set of segment criteria: understandability, homogeneityindependence, stability and actionability. The thesis also shows an example of how to evaluate clustering of video streaming users. A version of term frequency-inverse document frequency (tf-idf) was introduced, which is called video importance score (VIS). VIS is used to find videos specifically important to a cluster, and has proven to be helpful in interpreting the resulting clusters. The results were evaluated within a commonly used market segmentation evaluation framework, which was adapted to the problem at hand. Although the market segmentation strongly indicates to be useful, it still has to be in real-word scenario evaluated by the company before any definitive conclusions can be drawn.