الفهرس 
Chapter 1.يوجد فقط 14 صفحة متاحة للعرض العام
 |  | from | 114 |  |  |
| | | from | 114 | | |
المستخلص
The process of assessment is crucial to the educational process. However, assessment time is a difficult task that requires a lot of work, time, and resources. It puts a lot of pressure on the teachers, especially if they educate many students and frequently assign writing assignments. Compared to manual assessments, the automatic assessment technique has several benefits and can solve many manual assessment issues. The essay question is one of the most significant types of questions that can reveal a student’s writing abilities and demonstrate how well a student comprehends the subject matter. Due to varying writing styles, answer lengths, the use of numerous synonyms, misspellings, grammar, and formal structure, developing automatic assessment systems for essay questions is a challenging and complex process. These issues become apparent when attempting to develop automatic evaluation systems for languages like Arabic. An automatic system for assessing ’ responses to essays written in Arabic is created and implemented in this thesis. The grading engine and the adaptive fusion engine make up the system’s two primary components. The degree of similarity between the student’s response and the model response is calculated by employing a number of well-known similarity algorithms in the grading engine. Afterward, the proposed adaptive fusion engine combines the similarity values using the three described fusion methods to improve the performance of the proposed system. The first fusion method is applied using the algorithm of Majority Voting Based Fusion Approach (MVBFA) which is based on the majority voting algorithm. The second fusion method is the Distance Based Fusion Approach (DBFA) which is based on measuring the distance between the degree of similarity suggested by the system and the degree of similarity suggested by the human grader. Finally, the third fusion method is implemented using Feature selection (FS) and Machine Learning (ML) algorithms. The algorithms of FS that are applied are Recursive Feature Elimination (RFE) and Boruta. Moreover, the used ML algorithms are K-Nearest Neighbor (KNN), lasso, Random Forest (RF), Decision Tree (DT), Bagging, and Adaboost. Using two custom-built datasets, the proposed system was implemented and assessed. The acquired experimental findings demonstrated the effectiveness of the proposed fusing text similarity algorithms in terms of Pearson’s Correlation Coefficient (r), Willmot’s Index of Agreement (d), and Root Mean Square Error (RMSE) metrics. The results, using the first dataset, demonstrate that DBFA outperforms MVBFA in the first of two fusion methods with 0.940, 0.950, and 0.127 for each r, d, and RMSE, respectively. Moreover, the RF algorithm outperforms all other ML algorithms on the original dataset in the third fusion method with 0.913, 0.942, and 0.153 for each r, d, and RMSE, respectively. Also, using the second dataset, DBFA outperforms MVBFA in the first of two fusion methods with r = 0.948, d = 0.975, and RMSE=0.120. Moreover, the RF algorithm outperforms all other ML algorithms on the dataset after applying the RFE algorithm in the third fusion method with 0.920, 0.960, and 0.143 for each r, d, and RMSE, respectively.