Dissertation > Excellent graduate degree dissertation topics show

Study of Text Feature Selection and Weighting Based on Ontology

Author: ChenGuangHui
Tutor: XuJianMin；YangXiuDan
School: Hebei University
Course: Management Science and Engineering
Keywords: Text feature selection Body Mapping Concept Feature Weighting
CLC: C931.6
Type: Master's thesis
Year: 2010
Downloads: 77
Quote: 3
Read: Download Dissertation

Abstract

Traditional text feature selection and weighting system is mainly based on statistical theory and machine learning methods to solve the data sparseness flawed, based on text classification accuracy often can not obtain satisfactory results, and a lot of research The results show that, using the traditional text feature, select the feature vectors obtained by the method, there are semantic association between a large number of feature items. The ontology is a formal and explicit description of the objective existence of shared concepts and their relationships, has a good concept of hierarchy and the support of logical reasoning. Body good feature selection from the word level to rise to the level of the concept, digging a deeper level of relationship between the terms introduced in the text feature selection process. Based on this, this paper proposes an ontology-based text feature selection and weighting method. First preprocess the text by removing stop words and stemming text preprocessing methods, and then use the vector space model text feature, the initial feature vector. Then introduced into the body, will feature items mapped to the concept. Ontology-based text feature selection methods include three aspects: constructing text concept tree feature items to the concept of mapping and calculating the initial weight. Protégé has a good visualization capabilities in the process of ontology construction can automatically generate information on the concept of the relationship between the structure of Figure the text concept tree. Text concept tree clearly showed a hierarchical relationship between terms to feature items can be mapped to the concept. According to different terminology, the mapping process there will be a one-to-one, many-to-one and many-to-many, these three cases, the use of multiple feature item to map the same concept matching method. TF IDF method is the most widely used feature weighting method, we use the method to calculate the initial weights of the feature items, but the the TF IDF method in the computing term weights not take into account the semantic relationship between the feature item. In order to make it better feature empowerment, according to the method of the mapping of the weighted improvement. This article uses the OWL language and ontology building tool Protégé ontology model built from a small field of educational technology, the model used in the experiment. The experimental results show that the proposed method can effectively improve the text classification precision and reduce the number of dimensions of the feature vector.

Study of Text Feature Selection and Weighting Based on Ontology

Abstract

Related Dissertations