[Abstract]: Web page access prediction gained its importance from the ever increasing number of e-commerce Web information systems and e-businesses. Web page prediction, that involves personalising the Web users’ browsing experiences, assists Web masters in the improvement of the Web site structure and helps Web users in navigating the site and accessing the information they need. The most widely used approach for this purpose is the pattern discovery process of Web usage mining that entails many techniques like Markov model, association rules and clustering. Implementing pattern discovery techniques as such helps predict the next page to
be accessed by theWeb user based on the user’s previous browsing patterns. However, each of the aforementioned techniques has its own limitations, especially
when it comes to accuracy and space complexity. This dissertation achieves better accuracy as well as less state space complexity and rules generated by performing
the following combinations. First, we combine low-order Markov model and association rules. Markov model analysis are performed on the data sets. If the Markov model prediction results in a tie or no state, association rules are used for prediction. The outcome of this integration is better accuracy, less Markov model state space complexity and less number of generated rules than using each of the methods individually. Second, we integrate low-order Markov model and clustering. The data sets are clustered and Markov model analysis are performed on
each cluster instead of the whole data sets. The outcome of the integration is better accuracy than the first combination with less state space complexity than higher
order Markov model. The last integration model involves combining all three techniques together: clustering, association rules and low-order Markov model. The data sets are clustered and Markov model analysis are performed on each cluster. If the Markov model prediction results in close accuracies for the same item, association rules are used for prediction. This integration model achieves
better Web page access prediction accuracy, less Markov model state space complexity and less number of rules generated than the previous two models.
[1]
Chengqi Zhang,et al.
Mining for combined association rules on multiple datasets
,
2007,
DDDM '07.
[2]
Jun Hong,et al.
Using Markov Chains for Link Prediction in Adaptive Web Sites
,
2002,
Soft-Ware.
[3]
Sourav S. Bhowmick,et al.
WAM-Miner: in the search of web access motifs from historical web log data
,
2005,
CIKM '05.
[4]
Hongyuan Zha,et al.
Learning User Clicks in Web Search
,
2007,
IJCAI.
[5]
Jun Hong,et al.
Using Markov models for web site link prediction
,
2002,
HYPERTEXT '02.
[6]
Jiawei Han,et al.
Discovering Web access patterns and trends by applying OLAP and data mining technology on Web logs
,
1998,
Proceedings IEEE International Forum on Research and Technology Advances in Digital Libraries -ADL'98-.