An efficient info-gain algorithm for finding frequent sequential traversal patterns from web logs based on dynamic weight constraint

Abstract

Many frequent sequential traversal pattern mining algorithms have been developed which mine the set of frequent subsequences traversal pattern satisfying a minimum support constraint in a session database. However, previous frequent sequential traversal pattern mining algorithms give equal weightage to sequential traversal patterns while the pages in sequential traversal patterns have different importance and have different weightage. Another main problem in most of the frequent sequential traversal pattern mining algorithms is that they produce a large number of sequential traversal patterns when a minimum support is lowered and they do not provide alternative ways to adjust the number of sequential traversal patterns other than increasing the minimum support. In this paper, we propose a frequent sequential traversal pattern mining algorithm with dynamic weight constraint. Our main approach is to add the weight constraints into the sequential traversal pattern while maintaining the downward closure property. A weight range is defined to maintain the downward closure property and pages are given different weights and traversal sequences assign a minimum and maximum weight. In scanning a session database, a maximum and minimum weight in the session database is used to prune infrequent sequential traversal subsequence by doing downward closure property can be maintained. Our method produces a few but important sequential traversal patterns in session databases with a low minimum support, by adjusting a weight range of pages and sequence. The support and confidence are the most popular measures for sequential patterns. The support evaluates frequencies of the patterns and the confidence evaluates frequencies of patterns in the case that sub-patterns are given. These parameters are meaningful and important for some applications. The information gain metric which is widely used in the information theory field, may be useful to evaluate the degree of surprise of the pattern. Target is finding set of patterns that have information gain higher than minimum information gain threshold.