Identifying web sessions with simulated annealing

Abstract

Delivery of efficient service through a web site makes it compulsory in the redesigning stage to take into account the behavior of the users, which can be studied by means of a web log file that partially records information about user visits. The reconstruction of all of the sequences of pages that are visited by users who browse a web site is known as the web sessionization problem, and it has been formulated by means of an integer programming model; however, because a web log can accumulate a large amount of information, it is necessary to reconstruct the sessions over a period of weeks or months, thus the solution to this problem requires a long computational processing time. This paper presents a heuristic approach based on simulated annealing for the sessionization problem. Using this approach, it has been possible to reduce the processing time up to 166 times compared to the time that is required for the integer programming model. Furthermore, the metaheuristic solution finds new optimum values, which achieve increases on the order of 17% in the best cases.