Web-Based Arabic Question Answering System using Machine Learning Approach

Abstract

Question Answering (QA) systems are complex software capable of answering a question in natural language. The source of information for these systems is a given corpus or, as assumed here, the Web. To generate the exact answer, these systems carry out several subtasks among which the question classification and answer extraction. The main goal of this paper is to employ machine learning techniques for question classification and answer selection tasks. This study presents a supervised support vector machine (SVM) classifier for question classification and answer selection. The question classifier is trained to identify the answer type of the submitted question and directs the answer extraction module to re-rank the answer candidates retrieved from the initial retrievers. A set of flexible features are used , such as lexical features, and syntactic features. Similarly, the answer selection stage is using SVM classifier with a set of extracted features and trained on a set of question-answer pairs. We assess the performance of answer selection task using the Mean Reciprocal Rank(MRR).

Keywords: Question Answering System, Information Retrieval, Information Extraction, Machine Learning, Natural Language Processing