Search

In real-life classification problems, prior information about the problem and expert knowledge about the domain are often used to obtain reliable and consistent solutions. This is especially true in fields where the data is ambiguous, such as text, in which the same words can be used in seemingly similar texts but have a different meaning. Many of the proposed approaches rely on the bag-of-words representation, which loses the information about the structure of the text. In this thesis, a literature review of related works in text classification is provided which includes an overview of text classification methods. In addition, detailed review of related works of two text classification domains; search engines and question answering systems. The core contribution is divided into three main parts. The first contribution is the Customizable Grammar Framework for user-intent text classification (CGF) which employs a formal grammar approach and exploits domain-related information in a new way to represent text as a series of syntactic categories forming syntactic patterns. In addition, the proposed framework has been applied to different domains which resulted in the second and third contribution. The second contribution is the Grammar-Based Framework for Query Classification (GQC) which helped in the improvement of query identification and classification. The third contribution is the Grammar-Based Framework for Question Categorization and Classification (GQCC) which helped in the enhancement of question identification and classification. In addition, using different machine learning algorithms the overall results show that the proposed approach outperforms previous ones in terms of classification performance for query and question classifications. Finally, comparison of the classification performance with the state-of-the-art approaches has been conducted, results validate that the proposed approach improves the classification accuracy and the identification of the different types of queries and questions.

The ancient technique of information hiding known as text steganography has enjoyed much research in recent years due to the rising popularity of social media platforms, and the abundant availability of online literature and other text as cover media for steganography. Whilst the majority of the research approaches have focused on manipulating or replacing text, in some form or another, to embed secret information, the utilisation of the structure of the document itself for such embedding has rarely been researched. Subsequently, research in the field of written English paragraph structure and the related error analysis is outdated. Therefore, a new approach in embedding secret messages in textual documents based on the splitting, merging, and resizing of paragraph text is proposed. The size comparison between adjacent paragraphs embeds one bit of information. We outline a basic and advanced model, and define the syntax and semantics of the embedding language. We also propose and applied two analysis applications, one using Machine Learning by classifying text based on their attributes, such as; words per paragraph, paragraph proportion based on sentences, and other written English data points. The other analysis technique utilised is the Chi-squared method, here the distribution of paragraphs sizes is analysed to see if it can statistically detect between clean and embedded text when and if the manipulation is applied. The embedding model showed to be resilient against the analysis techniques. It detected around 50% of the bad corpus. However, we concluded that it extremely difficult to detect an embedding model that manipulates paragraphs and structure of novel texts.