Toward a New Arabic Question Answering System

Imane Lahbari, Said El Alaoui, and Khalid Zidani

Computer Science and Modeling Laboratory, Sidi Mohammed Ben Abdellah University, Morocco

Abstract: Question Answering Systems (QAS) aim at returning precise answers to user’s questions that are written in natural language. In this paper, we describe our question processing and document retrieval as two components of Arabic QAS. First, we present Arabic question classification method based on SVM classifier and Li and Roth’s [24] taxonomy. Then, we describe our proposed technique to transform an Arabic question, to a query which is available to get information from the Arabic Wikipedia. In this paper, we use a hybrid Arabic Part-of-Speech (POS) tagging and Arabic WordNet (AWN) for query expansion. We have conducted several experiments using Text Retrieval Conference (TREC) and Cross Lingual Evaluation Forum (CLEF) datasets. The obtained results have shown that the proposed method is more effective as compared with the existing methods.

Keywords: Natural language processing, Arabic question answering system, question classification, taxonomy, machine learning approach, SVM, decision-tree, naive bayes, POS tagging, query expansion, AWN.