User Community Detection From Web Server Log Using Between User Similarity Metric

Abstract

Identifying users with similar interest plays a vital role in building the recommendation model. Web server log acts as a repository from which the information needed for identifying the users and sessions (pagesets) are extracted. Sparse ID list and Vertical ID list are used for identifying the closed frequent pagesets which is beneficial in terms of memory and processing. The browsing behavior of a user is identified by computing similarity among the pageset that belongs to the user. A new metric for measuring within user similarity is proposed. The novelty in this approach is, only the users having consistent behavior over the time are taken into consideration for clustering. Consistent users are then clustered by different clustering techniques such as Agglomerative, Clustering Large Applications Using RANdomized Search (CLARANS) and proposed Density-Based Community Detection (DBCD). The quality of the clusters formed by DBCD is found to do better for clustering the users. The outcomes show significant improvements in terms of quality and speed of the clustering.

Keywords

• Session identification
• Sequential pattern mining
• Clustering
• Community detection