Machine Learning Based Essay Grading System

Machine Learning and automation has progressed immensely over the years and has tend to make human lives simpler with reducing human effort and time on tasks by enabling a machine to perform them. One such task is to grade essays. Essay writing is an integral part for anyone willing to learn a language or skill or to simply exhibit one’s thoughts and ideas on a topic. This leads us to the reason why essay grading is important. When a work is scored against some parameters, a scope of improvement is possible. Hence, when essays are graded and feedbacks are provided, it guides the writer to analyse the work and to have a better understanding of the topic in general. Although, manual grading of essays could create discrepancy because of being graded by different individuals having different perceptions of the same content. It also consumes a lot of human time and effort. Therefore, automatic grading of essays could prove to be the saviour. In this project, we build a machine learning model which grades essays based on various features extracted using Natural Language Processing. We also test the model’s performance using several regression models like Linear, Lasso, and Ridge, and methods like Artificial Neural Network to find the best fit giving the maximum correlation with human grades.

SURVEY PAPER ON SMART ESSAY GRADER

In order to optimize Human-Machine agreement for automatic evaluation of textual summaries or essays, automated essay grading has been a research field. With a growing number of people taking multiple exams such as the GRE, TOEFL, and IELTS, grading each paper would become more challenging, not to mention the challenge for humans to maintain a consistent mindset. In this situation, it is extremely difficult to rate a large number of essays in a short amount of time. This project aims to address this issue by developing a stable interface that will aid humans in grading essays. This study served as a medium for us to extract features such as the Bag of Words, numerical features such as the count of sentences and words, as well as their average lengths, structure, and organization, in order to rate the essay with the highest level of accuracy. This algorithm was chosen because it works well for small datasets.

Kemampuan Menulis Mahasiswa PBJ UNNES Semester 3 Tahun 2019/2020 (Penilaian Kemampuan Berdasarkan JF Standard)

The application of JF Standard in learning Japanese especially in the Sakubun Shochukyu course is new and needs to be evaluated how the achievement of it’s learning targets. This study aims to find out how the semester 3 students' abilities are in writing. The research uses a quantitative approach and the research data are discussed with descriptive techniques. To assess the results of student essays used essay grading rubric based on JF Standard with 4 grading levels namely ganbare, mousukoshi, dekimashita, and subarashii for each item assessed. The results showed that in general, the students' writing ability was in the "dekimashita" or “very good” category with the note that there were still some errors in using grammar. Statistically, 1 student has a very high ability, 4 students have high ability, 18 students have the medium ability, and the remaining 8 are students who have low ability.

Intelligent Essay Grading System using Hybrid Text Processing Techniques

Educational Institutions are facing enormous tasks of marking and grading students at the end of every examination within the shortest possible time. Marking theoretical essay questions which involves thousands of examinees can be biased, subjective and time-consuming, leading to variation in grades awarded by different human assessors. This study presents an Essay Grading System called Intelligent Natural Language Processing Essay Grading System (iNLPEGS) with high accuracy percentage and minimal loss function for scoring assessment that can accommodate more robust questions. Secondary dataset collected from Kaggle provided by The Hewlett Foundation was used to aid semantic analysis and Part of Speech tagging. Assemblage of Computer Science questions and answers were collected from Babcock University Computer Science Department to create a more robust dataset to ensure high reliability. An Intelligent Natural Language Processing Essay Grading Model was designed based on Enhanced Latent Semantic Analysis using Part of Speech n-gram Inverse Document Frequency. Web based application was developed using Django, Gensim, Jupyter Notebook and Anaconda as the development tools due to availability of several Python libraries with SQLite as the database. Results of performance evaluation of iNLPEGS showed accuracy of 89.03% and error of 10.97% connoting that there is very little difference between scores from the developed intelligent essay grading system and a human grader. Also, the loss function from Root Mean Square Error (RSME) showed value of 0.620 which is very small and thus signifies closeness to the line of best fit from the regression equation.