Automatic Subjective Answer Grading Software Using Machine Learning

Abstract

One of the major challenges during online examinations is the assessment of answers, particularly of the subjective type. Subjective answers test a student's ability to retain information and express it in natural language. While objective questions have a correct fixed answer, subjective questions can have multiple correct answers. These answers can convey the same information while using a completely different language and grammatical syntax. This makes it difficult to automate the process of grading subjective questions and requires a lot of manual work hours. This study intends to automate the process of grading subjective questions using Machine Learning (ML) and Natural Language Processing (NLP). The study has compared the subjective answer with an ideal answer that is provided by the authority that creates the question. Based on the similarity between the two answers, a score is generated which can be mapped to an appropriate grade. The authors have provided a web application made using the Django framework for people to give online examinations and be automatically graded in near real-time. No machine learning model can be 100% accurate, so there is a functionality for admins to edit the grades.