Analysis of Algorithms for Detecting Users’ Behavioral Models based on Sessions Data

Analysis of user behavior models based on user session data can be conducted using clustering (or community detecting) algorithms that do not require a predefined number of clusters. An unknown number and quality of potential behavioral models, non-efficient utilization of memory and processing units, and a large amount of data are the main difficulties developers could meet. Besides, the suitability of clustering algorithms for the task highly depends on the characteristics of datasets and still requires additional research. The first step and the goal of this research is to analyze the capabilities of the clustering algorithms in order to determine more promising ones and techniques able to minimize computing resources while solving the task. During the research, the properties of algorithms were analyzed through a literature review and also experimentally. It was found that, although the algorithm’s suitability is theoretically proved, experiments could show unsatisfactory results. Therefore, a certain trade-off analysis and problem-specific modifications of original processing must be done. As a result, the clickstream clustering method of the Louvain clustering algorithm and the modified Longest Common Subsequence algorithm showed more appropriate results in detecting the well-clustered user behavior models based on user sessions data. Thus, these algorithms are suitable for the further development of the detecting method.