الفهرس 
ch.1 IntroductionOnly 14 pages are availabe for public view
 |  | from | 100 |  |  |
| | | from | 100 | | |
Abstract
Assessment is indispensable part of the educational process. However, manual assessment is a tedious task that needs a lot of efforts, time and resources. It makes a huge pressure on the teachers particularly if they teach large numbers of students and if they usually assign frequent writing assignments. On the other hand, essay question is one of the most important types of question that can improve the writing skills of students. Also, it can reflect to what extent the students understand the content of the subject. However, in manual assessment of essay question, different human graders can assign different scores for the same essay which is viewed by many students as an unfairness grading. On other side, automatic assessment technology has many advantages over manual assessment and can overcome many of the problems associated with it. However, building automatic grading systems for essay questions is hard and complex because there is heterogeneity in writing methods, answer lengths, multiple synonyms, spelling errors, grammar and morphological structure. These problems become apparent when attempts are made to build automatic grading systems for languages characterized by high ambiguity, rich morphology, complex morpho-syntactic agreement rules and a large number of irregular forms such as Arabic language. In this thesis, an automatic Arabic essay grading system for Arabic Language has been designed and implemented. It consists of two main components, grading engine and adaptive fusion engine. The grading engine pre-processes the student answer and the model using a number of Natural Language Processing (NLP) techniques. Then, it employs a number of well-known similarity algorithms including string based and corpus-based similarity algorithms such as N-gram, Matching Coefficient, DISCO1, Latent Semantic Analysis (LSA), etc. Then, the adaptive fusion engine employs two proposed fusion approaches. The first approach depends on the majority voting algorithm and it is called Majority Voting Based Fusion Approach (MVBFA). The second approach depends on measuring the distance between the similarity score suggested by the system and the similarity score suggested by human grader and it is called Distance Based Fusion Approach (DBFA). The proposed system has been implemented and evaluated using a number of datasets in terms of correlation metric. Also, it has been compared to a number of well-known similarity algorithms. The obtained experimental results have shown the superiority of the proposed fusion approaches compared to using the similarity algorithms separately. Also, the results have shown that DBFA is better than MVBFA with achieved correlation of 0.94.