Extraction of Taxonomic Relations from Arabic Text for Ontology Construction

Abstract

The huge amount of textual information available electronically has made it difficult for many users to search and find the right information within acceptable time. The ontology based techniques can contribute to solve these problems and help users in exploiting these vast resources. Ontology could be an efficient way to improve the process of searching and exploiting information on the web. The benefit of ontology is that it provides a standard for the vocabulary used in a specific domain and relations. This thesis proposes a method to extract taxonomic relations to construct ontology automatically from natural Arabic text on Political News domain using four stages. First perform pre-processing operations in text such as tokenization, normalization and stop-word removing and then morphological information in pre-processing is extracted to detect the part of speech of each word. Second extraction of terms by integration between lexical resources and machine-learning classifier for Arabic named entities recognition. Third extraction of taxonomic relations between terms using rule based domain. Finally constructing a set of transformation rules to identify the appropriate ontological elements from the terms and taxonomic relations that extracted. After constructing the ontology, we build RDF language to represent information about resources on the text and build ontology with class-subclass relations and property relations. Two methods are performed to test and evaluate the accuracy of approach, first using measures calculate precision, recall and f-measure. Second using a reasoner to check the consistency. The results shows satisfactory results for all terms and taxonomic relations extraction, with precision = 92% and recall = 91%.