Customer Segmentation basedon Behavioural Data in E-marketplace

Abstract: In the past years, research in the fields of big data analysis, machine learning anddata mining techniques is getting more frequent. This thesis describes a customersegmentation approach in a second hand vintage clothing E-marketplace Plick.These customer groups are based on user interactions with items in themarketplace such as views and "likes". A major goal of this thesis was to constructa personal feed for each user where the items are derived from the user groups.The customer segmentation method discussed in this paper is based on theclustering algorithm K-means using cosine similarity as the similarity measure. Theinput matrix used by the K-means algorithm is a User-Brand ratings matrix whereeach brand is given a rating by each user. A visualization tool was also constructedin order to get a better picture of the data and the resulting clusters. In order tovisualize the highly dimensional User-Brand matrix, Principal Component Analysis isused as a dimensionality reduction algorithm.