Efficient Arabic Query Auto-Completion for Question Answering at a University

In this paper we describe an implementation of an Arabic query auto-completion system for student question-answering at a University. University students make many inquiries concerning academic life: about majors, concentrations, dates, instructors, courses, rooms, exams and more. Auto-completion (AC) has recently been part of many user interfaces, such as search bars on web pages, social media sites and mobile applications. We investigate multiple approaches to completion candidate generation and ranking and the role Arabic NLP may play in that. After experimenting with other options, we collected the data used in our system directly from students at the University. This data can be expanded to account for more types of queries and ways to express information needs. We describe our dataset, give an evaluation of individual system components and of the system results in general. The goal of this work is to improve the answer search experience by reducing the time of entering the query and biasing the completed query towards unambiguous and easily answerable questions. We divided the implementation into several stages and report on the results for each stage separately. Despite the difficulties of Arabic NLP, the results of our system were encouraging and compared well with other QAC systems described in the literature. Finally, we give an informal evaluation of overall results and the improvement resulting from using QAC for our QA system.