Thesis (M.Sc.) - Cairo University - Faculty of Graduate Studies for Statistical Research - Department of Computer and Information Science

Classification is an important technique used in information retrieval. Supervised classification suffers from certain limitations concerning the collection and the labeling of the training dataset. The problem gets more complicated when facing Multi-domain classification where multiple training datasets and classifiers are needed which is typically difficult. This thesis proposes a training-less multi-domain classification approach where each domain is represented by an ontology. A document is mapped on each ontology based on the weights of the mutual tokens between them. A mapping degree for the document with each domain is then determined with the help of fuzzy sets. A Multi-Domain Document Classification information retrieval system (MDDC) is built as an implementation of the proposed approach. A fuzzy matching approach, based on fuzzy triangular numbers, has also been used as another way in determining the mapping degree. The system was tested on a dataset of 180 journal articles of different domains where it succeeded in classifying them with an accuracy of 92.22%. The fuzzy triangular numbers approach succeeded in obtaining comparable results with the original approach. A number of evaluations have also been performed including comparing the system{u2019}s results with those of other algorithms using WEKA and RapidMiner as two of the top machine learning tools nowadays. The evaluation results were highly comparable and promising

Issued also as CD