Sequential pattern finding: A survey
Naseer Ahmed Sajid, Salman Zafar, Sohail Asghar · 2010
To gain the competitive advantage in today's age of technology, growing data and to bear the competitive pressure, making strong decisions according to customer's need and market trend has become very important. With huge amount of data on internet, web data mining has become very significant. Web Usage helps companies to produce productive information pertaining to the future of their business function ability by analyzing the usage information from their websites. Sequential Patterns allow the collection of web access data for web pages. This usage data provides the paths leading to accessed web pages and their sequences which can give us valuable information about user's behavior and effectiveness of websites. The focus of this survey paper is to review the challenges, issues and techniques for finding sequential patterns in the context of web usage mining. For this, different techniques from the literature have been reviewed. Analysis of the existing techniques related to the sequential pattern finding has been done on the basis of the existing literature and some parameters are proposed. The comparison criteria for the analysis include Graph Structure, Input Data Structure, Input Parameters, Pattern Type, Application, Technique, Method, Algorithm, Scalability and Tool Support. The reviewed techniques are divided into three groups Clustering; Association; Markov Model based techniques. By analyzing all the techniques with the help of above defined parameters, compare and contrast we finally conclude that the two clustering techniques based on Markov Models (Line Prediction and Path Analysis Using Markov Chains & Using Markov Chains for Structural Link Prediction for Adaptive Web Sites) proposed by Sarukkai & Zhu et al. respectively, are the best sequential pattern finding techniques. Analyzing different techniques also show that the hybrids of different techniques can be used to find sequential patterns in web usage mining data. Strength and weaknesses of one technique can complement of the other technique so it is best to use techniques in combination if we want accurate patterns efficiently.