Analyzing clusters of web application user sessions

Abstract

User sessions provide valuable insight into the dynamic behavior of web applications. They also play a key role in user-session-based testing, which gathers user sessions in the field and replays selected sessions to test an evolving application. To decrease the testing and analysis effort, testers reduce the set of collected user sessions by either clustering user sessions by their shared URL attributes or by program coverage requirements-based reduction techniques. Clustering URL attributes can be considerably less expensive; however, the tradeoff may be that clustering is not representative of dynamic behavior similarities. This paper describes our analysis of user session data to reveal correlations between sessions clustered on the sessions' attributes and the relative dynamic behavior of the program for those sessions. The results of our analysis also motivate other clustering and test suite reduction techniques. Our results can also be used to learn more about how clusters of web application use cases are related in terms of the underlying user session attributes, program coverage, and fault detection.