|
Title:
|
A MODEL-BASED APPROACH FOR WEB USER CLUSTERING AND BEHAVIOUR ANALYSIS |
|
Author(s):
|
Syed Tauhid Zuhori and James Miller |
|
ISBN:
|
978-989-8533-82-1 |
|
Editors:
|
Pedro Isaías and Hans Weghorn |
|
Year:
|
2018 |
|
Edition:
|
Single |
|
Keywords:
|
Website Personalization, Atomic Propositions, Discrete Time Markov Chain Inference Process |
|
Type:
|
Full Paper |
|
First Page:
|
117 |
|
Last Page:
|
124 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Today e-commerce activities are increasing rapidly. Hence, it is very important to understand the behavior of users based on their interactions with a website. In order to do so, web usage mining is needed. This paper proposes and develops an architecture for performing Web usage mining. Firstly, we clean the web server logs by using a traditional clustering approach. After that, we apply a Discrete Time Markov Chain approach to generate a model of the user behavior. For generating the nodes for the model, we use a technique (regular expressions) to find out the atomic propositions. Then we find a directed graph as an output of a Discrete Time Markov Chain inference process. Next, we find the optimal number of clusters from that graph and apply spectral clustering on that. This clustering works on the affinity of the graph nodes and that divides the nodes into clusters. Finally, we produce clusters of unique links and analyzing the behavior of each cluster. To evaluate the approach, we use server log files from the website www.ualberta.ca and get a accuracy of 81.67! This is show to be more accurate than existing algorithms. |
|
|
|
|
|
|
