501.txt

International Journal of Information Technology and Language Studies(IJITLS)Vol. 5, Issue. 1, (2021). pp. 9-17International Journal of Information Technology and Language Studies (IJITLS).http://journals.sfu.ca/ijitlsA Framework for Building an Educational Question Answering SystemHeba Samy1, Ehab E. Hassanein1, and Khaled Shaalan2hbsamy@gmail.com; e.ezat@fci-cu.edu.eg; khaled.shaalan@buid.ac.ae1Faculty of Computers & Information,Cairo University, Egypt2Faculty of Engineering & IT, The British University in Dubai, UAEAbstract.A question answering unit in any educational system can be very helpful and effective for enhancing the learning process. This paper introduces a design for the building blocks and workflow for an educational question answering system that can enhance thelearning process. This system uses an Arabic ontology as a prototype to prove its efficiency. It depends mainly on natural language processing techniques that exist in the research area. A rule-based question classification method is created as well as a querying process depending on named entity recognition and pre-made templates.Keywords: Question Answering, Question Answering systems, Ontology, linked data, e-learning, online education, ontology-based Arabic Question Answering1.Introduction“Asking good questions is a central process in learning and sometimes it's even more important than finding the answers, especially when the questions foster the learner's critical thinking. That's one can say that questioning is a basic skill for good education” (Betts, 1910, p. 55). Asking questions is also good for educators as it helps them to investigate the knowledge gaps of learners by digging into their thinking, evaluate their high-order thinking, and adding or adapt part of the rest of the learning topic so that it enables students to influence the curriculum under study and to get more motivated to learn (Chin et al., 2008). But despite its importance, not all types of people/ages can get the benefit. For example,for higher grades, researchers have found that in most educational environments whether it's offline or online, higher grades students tend to ask fewer questions than those of lower grades. That's maybe because they don’t like to draw attention or their instructors are not welcomingmore questions (Chin et al., 2008). For this type of people, that are teenagers or shy people, an online educational question answering system can serve as a safe place for learners to ask questions without being noticed by their colleagues and givethe instructor more time to check learners' questions, analyze, and answer them as well. This question answering system should be able to answer students instantly from an underlined database containing the curriculum understudy. Then the instructor wouldbe able to check these questions and answers for more educational enrichment.Question answering as a term in information retrieval is defined as the process of a computer answering a human question that is being asked in his/her natural language, depending on either an underlined structured database or a repository of text documents in natural language (Samy et al., 2019). In comparison to other Latin languages, Arabic is somehow a challenging language when building a question answering system. This is because of the lack of tools and the specialty of the Arabic language itself (Lopez et al., 2011).Question Answering systems can be classified according to four dimensions based on the input question and answer type, data sources format, the scope of the domain being asked, and problems the system is trying to solve (Al Chalabi, Hani., 200).This paper presents the build process of a question answering system that answers mainly Arabic factual questions  that areclosed  to  the  prophet  Muhamed  pbuh  biography  domain  with  an  educational perspective. As in figure 1, the system workflow begins by a student asking a question in his natural Arabic language. The system receives the question via a GUI and sends it to the question-answering module. This 
10A Framework for Building an Educational Question Answering System1.Isri Stemmer: http://oujda-nlp-team.net/en/programms/stemmer/2.Alkhalil: http://oujda-nlp-team.net/en/programms/stemmer/3.Farasa: https://farasa.qcri.org/4.PY4J: https://www.py4j.org/module is responsible for answering the question instantly for the learner,which he receives with a remark that it is an unvalidated answer by the educator. Then the QA module saves the question with the generated answer in a database. This question and answer are saved with a pending validation status. The instructor then receives these pending question/answer pairs to be validated. The instructor then has the ability to validate the answers generated by the system or correct them and send them back to the database,which saves them under a validated question/answer status. This pair is then sent to the learner that asked the question as your validated question/answer pair.Figure 1.Edu. QAS Architecture2.MethodologyThe methodology presented here consists of 5 stages,as shown in Figure 2, data processing, question analysis, ontology mapping, query generation, and answer processing. The first stage is building theknowledge base underlying the system. The knowledge base is made up of Arabic ontology built on theProtégé tool as an RDF/OWL file. The second stage is classifying and analyzing the question generating question class, expected answer type,and question keywords. The third stage is mapping the question keywords with the underlying ontology elements. Thefourth stage is generating the question pattern according to the mapped keywords with the expected answer type then generating the query templates according to the question templates. The fifth stage is ranking the answer received and presenting the selected answer. The following sections explain each step in detail.Figure 2.The question answering unit stages3.OntologyThis ontology is designed with the help of an Arabic linguist to adjust its classes, properties hierarchy,and linguistic issues. It's written in OWL. The domain understudy is represented as RDF triples (i.e.subject, predicate, object) for our QAS to answer factoid, yes/no, and definition questions in this domain. This ontology is not restricted to our system but can be used by other applications serving this special domain. Data MatchingQuestion ProcessingQuery ConstructionAnswer Processing
Heba Samy, Ehab E. Hassanein, and Khaled Shaalan11That's because the design of the ontology is made to be easily edited by adding ordeleting Sera facts without defecting the ontology structure.  The ontology domain covers a small part of the prophet Muhammad pbuh Biography ”Sera”. It presents his family members' names, some of his friends, where and when he was born and died, his message and for whom this message, his “ghazawat”, and the mosques he prayed in. It also presents info about theprevious prophets and messengers from an Islamic perspective.3.1.Building the ontologyStrict and clear steps are followed to define, design,and build our ontology from scratch. It started by determining the domain and the scope of the ontology. This ontology is built to serve under a question answering system. The questions that are expected to be asked are factoid as well as yes/no questions. The following questions are examples:●ɸلتزوجالنۗܣمن أمسلمة؟●إڲʄأين ɸاجرمحمد؟ ●من ɸوالܶݰاȌيالذيɠانزوجفاطمةبɴتمحمد؟ After defining the domain and the scope of the ontology, comes the design steps which are designing classes and subclasses. After that,we defined the relations ofthe object properties between these classes. Then triples are created by creating instances for these classes and asserting the object properties.3.2.Ontology classesFigure 3 presents the whole ontology classes. Most of the classes in the ontology are made to serve mainly in the answer ranking process where we can detect the type of the answer from the ontology and compare it with the answer type expected in the question classification process. For example,a question like: من ɸمقوممحمد؟“who are Muhammad people?”. The expected answer should be of type “قوم”(a group of people or simply people) according to the question focus (I.e. first noun in the question as shown later).Figure 3.QAS classes tree3.3.Classes propertiesThe properties serve as the predicates in RDF triples. A predicate connects between named individuals(instances) of the ontology. Figure 4 presents the ontology properties.مجمشرقومA subclass