A Suggested Algorithm to Improve Passage Retrieval in Arabic Question Answering Systems

Abstract

Users often have specific questions in mind, for which they hope to get answers. Question Answering Systems (QAS) aim at retrieving accurate answers for the user’s questions from a large dataset. Passage retrieval is a crucial component for any QAS, it identifies top-ranked passages that may contain the answer for a given question. Also, it is a longstanding challenge widely studied over the last decades. However, it still requires further efforts in Arabic QAS.  In this research, we focus on the passage retrieval phase to get the most related passages to the correct answer. We suggested a model that measures the similarity between passages and the question and combines the BM25 ranker and Word Embedding approach. We tested our system on the ACRD dataset, the system was able to achieve an accuracy of 92.4% in finding the passages that contain the correct answer for a given question.