Author(s)

B. J. Park, W. Choi & S. H. Noh

Abstract

Recommending Web pages based on the access patterns by previous visitors to a user visiting some Web site can be helpful if they are is not familiar with the site. To provide a proper recommendation service for users, it is necessary to efficiently mine the Web page access patterns from a huge amount of Web server log data. This paper presents an efficient Web page recommendation scheme using a stochastic process model based on the time-homogeneous DTMC (Discrete-Time Markov Chain). It views a series of Web page access as a discrete-time stochastic process and constructs a state transition matrix based on the relative frequency of moving from one page to another in the Web site. Then it uses the initial state transition matrix to compute the probability for each Web page to be accessed after some number of page references from the starting Web page and selects a page with the largest probability value for recommendation. Although this process involves a simple matrix multiplication, its computational overhead could be expensive if the number of pages in the Web site is very large. Our approach employs the PageGather algorithm to find clusters of closely related ones among all the Web pages and uses each of these clusters as a state in the state transition matrix. As a result, the size of the transition matrix can be considerably reduced and so is the computation time without sacrificing the effectiveness of recommendation too much. We demonstrate the effectiveness and efficiency of our recommendation scheme with a series of experiments. Keywords: Web mining, Web page recommendation, stochastic process model.

Keywords

Web mining, Web page recommendation, stochastic process model.