Along with the fast growth of Internet, the issue of Web Mining has drawn so much attention for recent years. Web content mining can be viewed as a sub-field of Text Mining. There are two main steps used by a Pattern Taxonomy Model (PTM)-based Web Mining system. First, it uses indexing technique to build a feature space. Then the features are transformed into patterns by using pattern evolving method for the purpose of implementing classification or information filtering works. However, we encounter the problem of how to find a sufficient amount of useful patterns for dealing with document indexing. The other problem is how to generate powerful patterns from as less features as possible during the step of pattern evolving. The traditional way to form a feature space is to adopt either information-retrieval or information-theory feature weighting methods. Then the features with higher weighs are selected into the feature vector. However, these methods ignore some important information, such as term dependency and term correlation. Therefore, we have to analyze the impact of pattern dependency to solve the problem. The main goal of this project is to develop an effective feature selection method and to find an efficient way to evolve the patterns by analyzing the feature dependency. Furthermore, a PTM-based Web Mining system will be developed and experiments will be performed on several real datasets. The experimental results then will be compared to other methods based on standard measures for the purpose of system evaluation.
