Grouping Users Through Pair Wise Sequence Alignment and Graph Traversal Based on Web Page Navigation Behaviour

Abstract

The increasing number of online users endow with various opportunities for research in web mining. User grouping plays a major role in web personalization. Finding the similar user consortium gains more interest among the web researchers so that a customized and better environment users can be provided. This paper attempts on finding user consortiums based on their web page navigation pattern. The methodology first and foremost incorporates interpreting the input navigation sequence and then investigates the influence on graph traversal and the level of thresholding in user grouping. For each interpretation, a global pair-wise sequence alignment is carried out by considering the alignment between any two users web navigation sequences. Subsequently, based on the number of aligned and the unique number of pages between the users a similarity matrix is formulated. Then, based on the maximum value at each column and at each row as column thresholding and row thresholding similarity matrix is thresholded at different levels. After that, graph traversal is performed to identify the user groups. To assess the proposed methodology MSNBC dataset, a publicly available data is used. Jaccard similarity co-efficient is used to find the inter-group similarity. Then, the influence of thresholding and the threshold level was investigated. The results revealed that with the sorted input navigation sequence without redundancy the column thresholding at 0.75 level yielded the highest possible outcome in forming the user groups.KeywordsWeb user groupingBreadth first searchMSNBCPair wise alignment