Description

The expansion of web has led to the design and development of methodologies to address the user's need in searching the web. The need to retrieve relevant information and classify the web documents into a desired set of categories is a key research issue. In this regard, ontology based classification plays an important role to classify and organize the information.With this motivation, the objective of the research work presented in this thesis is web information retrieval based on ontology. The multimedia contents of web documents are represented by the associated metadata which is in the text form. Finally, the aim is to efficiently categorize text using semantic relevance based on ontology. This thesis addresses the following research objectives: 1. A term weighting method based on semantic similarity of documents using ontology In the literature review which is presented in Chapter 2, it is observed that several approaches for document classification do not reflect the semantic relevance of the terms which is essential to address the problems posed by polysemous words and the associated categories. Hence, a novel term weighting method named semTfidf based on the semantic similarity among documents is proposed in the Chapter 3. The proposed semTf-idf method is compared to other methods in the literature, namely, ICF-based and Tf-idf in terms of performance metrics, namely, Micro-F1 and Macro-F1 values. Experimental results of the proposed semTf-idf method with kNN, SVM and Fuzzy kNN classifiers are obtained for benchmark datasets, namely, WebKB, Reuters 21578 and 20 Newsgroups. In general, it is observed that the proposed semTf-idf method yields higher Micro-F1 and Macro-F1 values using Fuzzy kNN classifier in comparison with the other algorithms, namely, kNN and SVM. Reuters 21578 and 20 Newsgroups dataset contain plain text documents while the WebKB dataset contains the metadata of the multimedia contents of web documents.