Nada Ahmed, Mariam Yasser, Rawan Ahmed, Philip George
9 November 2021
Lately, short answer question grading has been a hassle. So The main objective of the system is to build an Automatic correction model for the subject question that will ease out the process of checking of answer essay questions. Students can take exams online, and the system will automatically calculate the results and provide a record for the administrator. The project will focus on correcting responses based on the presence of specific keywords in each answer and awarding grades to students based on the presence of the keywords in the replies. This system will assist in the reduction of all human errors, resulting in a more efficient system. Machine learning trading approaches, on the other hand, necessitate a far larger collection of labeled student responses than similarity-based methods in order to develop automatic grading models from the data. Whichever categorization algorithm is employed, the results will be the same.
After the pandemic education became online. this made doctors and students find difficulty in learning and exams became a problem. There are two types of questions multiple choice and short answers(essay) questions. Short answer questions ask students to write answers in several phases and sentences. Although the examination system is based on physical labor, from printing to delivering the paper to the test hall, invigilation, and the most time-consuming task of reviewing the answer sheets, which are a tremendous mess for any examiner and can result in data loss. As a result, we decided to create a short-answer auto-correction system. Our method aims to raise the bar for online examinations by allowing students to write subjective responses that will be reviewed on their own, compared to the model answer, and given a reasonable mark depending on specific keywords determined by the examiner. Using data mining, NLP, the system would be able to detect spelling problems and keywords given by the teacher, which will be compared to the student’s answer and graded accordingly. We shall use the BERT model to make the computer understand the context like (people=human).
BERT is a natural language processing machine learning framework that is open-source (NLP). BERT is a computer programmer that uses surrounding text to establish context to help computers interpret ambiguous words in a text. The BERT framework was trained on Wikipedia text and can be fine-tuned using question and answer datasets.
1-Solve problems of an instructor who instructor theoretical subjects more (requires essay questions).
2-An Awareness of most of the problems and mistakes that face professors and teaching assistants while correcting the students’ work
1.3 Problem Statement
The approach will eliminate the time and effort involved in manually correcting online tests, as well as the risk of materials being lost. The main challenge is making the program understand the context. The system shall be able to compare the two answers and understand the synonym words such as (good-better). Also, it will give a grade according to the comparison made before.