AUTOMATIC ESSAY GRADING SYSTEM FOR SHORT ANSWERS IN ENGLISH LANGUAGE | Science Publications

Journal of Computer Science

Mohd Juzaiddin Ab Aziz¹  Ali Muftah Ben Omran¹ 

关键词: Short Answer;    COW;    LCS;    SD;    Semantic Similarity;    Synonym;   

DOI  :  10.3844/jcssp.2013.1369.1382

学科分类：计算机科学（综合）

来源: Science Publications

【 摘 要 】

Automatic Essay Grading (AEG) system is defined as the computer technology that evaluates and grades written prose. The short essay answer, where the essay is written in short sentences where it has two types the open ended short answer and the close ended short answer where it is our research domain based on the computer subject. The Marking of short essay answers automatically is one of the most complicated domains because it is relying heavily on the semantic similarity in meaning refers to the degree to which two sentences are similar in the meaning where both used similar words in the meaning, in this case Humans are able to easily judge if a concepts are related to each other, there for is a problem when Student use a synonym words during the answer in case they forget the target answer and they use their alternative words in the answer which will be different from the Model answer that prepared by the structure. The Standard text similarity measures perform poorly on such tasks. Short answer only provides a limited content, because the length of the text is typically short, ranging from a single word to a dozen words. This research has two propose; the first propose is Alternative Sentence Generator Method in order to generate the alternative model answer by connecting the method with the synonym dictionary. The second proposed three algorithms combined together in matching phase, Commons Words (COW), Longest Common Subsequence (LCS) and Semantic Distance (SD), these algorithms have been successfully used in many Natural Language Processing systems and have yielded efficient results. The system was manually tested on 40 questions answered by three students and evaluated by teacher in class. The proposed system has yielded %82 correlation-style with human grading, which has made the system significantly better than the other state of the art systems.

【 授权许可】

Unknown