International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 05 | May 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 409 Automatic Grading of Handwritten Answers Harsh Jain 1 , Mohd SherAli Shaikh 2 , Ravi Shankar 3 , Vinita Mishra 4 1,2,3 Student, Information Technology, VESIT, Mumbai, Maharashtra, India 4 Professor, Information Technology, VESIT, Mumbai, Maharashtra, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - In this digital world most of the activities are transitioning to an online medium which includes conducting exams online, but still pen and paper exams are given more priority when it comes to accreditation. During this pandemic we have seen that the traditional pen and paper exams at the exam centre were not possible and we were forced to use the online mode. In this online mode the answers can be submitted in two ways, first is digital MCQ form, and in second, the answers are written, scanned and submitted using a smart phone. In this paper, a solution to grading of papers of theory based subjects is obtained where Automatic Paper Grading will be performed using Natural Language Processing. We’ll be using the OCR (Optical Character Recognition) algorithm for extracting the handwritten text from the papers and converting them into digital text. It will be graded by comparing the vector embeddings of the written answer and the answer provided by the teacher. This system will grade higher if the distance between the two answers in the vector form is small, i.e. , the similarity is higher. Key Words: Machine Learning, Natural Language Processing, Optical Character Recognition, Vector Embeddings, Sentence Similarity 1. INTRODUCTION For our project, we have tried to identify one of the most pressing problems in the current education system and tried to come up with a solution that will help the professors and other staff of educational institutes in general. Today, with the growing number of online classes and modes of education, there is a shortage of staff that can assess the exams written by students. Speeding up the evaluation remains as the major bottleneck for enhancing the throughput of instructors. Teachers spend a lot of their valuable time on correcting hundreds of answer papers, time which can be better spent on other work like projects, research or generally helping students. This technique is significant since MCQ examinations cannot always be used to assess a student’s grasp of a subject. Our system will automatically grade handwritten papers without manual supervision of any kind and with a lower rate of error than normal. The system also provides a full evaluation of the student’s performance on the test, allowing the t eacher to stay up to speed on the student’s strengths and weaknesses and help them develop. In the current situation of the world, this tool will save a lot of valuable time and effort which can be directed towards something more productive. 2. OBJECTIVES  The primary objective of this system is the extraction of text from a handwritten paper by a student, followed by pre-processing by the system.  This system will quickly generate the result by comparing the student’s answer, with one or more correct answers.  This system will make use of NLP and image processing that will help in high accuracy. [4]  To design a system that will require a minimal amount of time to provide an evaluation while not compromising on the accuracy.  To provide a detailed assessment report of the student’s performance in the test to the respective student.[1] 3. LITERATURE SURVEY We studied multiple papers and their findings are being summarised in this section(Fig 1). This section illustrates papers studied before and during the development of the project. The papers helped in gaining insight into existing solutions, possible ways to optimize algorithms and facilitate the selection of algorithms based on their performance. Figure 1 shows a comparison between all the papers that were referred to get a contrast between existing solutions of similar nature.