A Web-Based Recommendation System To Predict User Movements Through Web Usage Mining

• Main Author: Jalali, Mehrdad
• Format: Thesis
• Language: English; English
• Published: 2009
• Subjects: Web usage mining
• Online Access: http://psasir.upm.edu.my/id/eprint/11931/1/FSKTM_2009_12.pdf; http://psasir.upm.edu.my/id/eprint/11931/
• Institution: Universiti Putra Malaysia

Web usage mining has become the subject of exhaustive research, as its potential for Web based personalized services, prediction user near future intentions, adaptive Web sites and customer profiling is recognized. Recently, a variety of the recommendation systems to predict user future movements through web usage mining have been proposed. However, the quality of the recommendations in the current systems to predict users‘ future requests can not still satisfy users in the particular web sites. The accuracy of prediction in a recommendation system is a main factor which is measured as quality of the system. The latest contribution in this area achieves about 50% for the accuracy of the recommendations. To provide online prediction effectively, this study has developed a Web based recommendation system to Predict User Movements, named as WebPUM, for online prediction through web usage mining system and proposed a novel approach for classifying user navigation patterns to predict users‘ future intentions. There are two main phases in WebPUM; offline phase and online phase. The approach in the offline phase is based on the new graph partitioning algorithm to model user navigation patterns for the navigation patterns mining. In this phase, an undirected graph based on the Web pages as graph vertices and degree of connectivity between web pages as weight of the graph is created by proposing new formula for weight of the each edge in the graph. Moreover, navigation pattern mining has been done by finding connected components in the graph. In the online phase, the longest common subsequence algorithm is used as a new approach in recommendation system for classifying current user activities to predict user next movements. The longest common subsequence is a well-known string matching algorithm that we have utilized to find the most similar pattern between a set of navigation patterns and current user activities for creating the recommendations.