This paper proposes to grade various essays and literary materials automatically using the feature extraction techniques
from Natural Language Processing (NLP) and Support Vector Machine (SVM), a powerful machine learning algorithm for
classification, modelled around Education Testing Service’s GRE Analytical Writing scoring guidelines. We extracted various
features like word count, TF-IDF score, number of paragraphs, part of speech tagging and number of spelling mistakes on the essay
dataset sourced from Kaggle [1]. After extracting the features using NLP, there were two possible approaches to tackle the problem;
a regression model or a model based on classification. We used a classification-based approach to train our model with training
essays, normalized to a scale of 1 to 6. Upon predicting the grades for the essays in testing set, we found that the accuracy of our
model stood at 0.52, and 0.89 with a tolerance of one point, as permissible by ETS that uses automatic essay grading [2]. Individual
essay sets can be graded with an automatic grading framework, a lot of human effort could be saved and literary pieces can be
graded with transparency.
Abhishek Suresh : received his BTech in Mechanical
Engineering from Manipal Institute of Technology, Karnataka,
India. He worked as a Trainee Decision Scientist at Mu Sigma
Business Solutions Pvt. Ltd., Bengaluru, India after graduation.
Currently he is pursuing Master’s in Computational Linguistics,
Analytics, Search and Informatics at the University of Colorado
Boulder. His interests include computational linguistics, NLP,
machine learning and text analytics.
Manuj Jha : received his BE in Telecommunication from R.V.
College of Engineering, Bengaluru, India. He worked as a
Trainee Decision Scientist at Mu Sigma Business Solutions Pvt.
Ltd., Bengaluru, India after graduation. Currently he is pursuing
Master’s in Data Science at Texas Tech University. His interests
include NLP, machine learning and descriptive analytics.
Natural Language Processing, Essay Grading, Machine Learning, Support Vector Machine
In this paper, we identified a classification-based
approach to solve the problem of grading literary
materials manually. We used Natural Language
Processing to extract various features which are
characteristic of a good writing. The accuracy of model
could be further improved if a metric for similarity
between the essay topic/ problem statement were added.
The topic on which essay was written was not described
in the dataset we used, but this is mostly known in most
of the exams/ standardized tests. The model we designed
performed reasonably well with an allowance of one
point in marking and could definitely be used to grade
written essays/ literary materials.
[1] Kaggle. The Hewlett Foundation: Automated Essay
Scoring. Available from:
https://www.kaggle.com/c/asap-aes.
[2] Valenti, S., F. Neri, and A. Cucchiarelli, An overview
of current research on automated essay grading.
Journal of Information Technology Education:
Research, 2003. 2(1): p. 319-330.
[3] Shermis, M.D., et al., Automated essay scoring:
Writing assessment and instruction. International
encyclopedia of education, 2010. 4(1): p. 20-26. [4] Drolia, S., et al., Automated Essay Rater using
Natural Language Processing. International Journal
of Computer Applications, 2017. 163(10).
[5] Ramineni, C., et al., Evaluation of the eirater®
Scoring Engine for the GRE® Issue and Argument
Prompts. ETS Research Report Series, 2012.
2012(1).
[6] Drucker, H., D. Wu, and V.N. Vapnik, Support
vector machines for spam categorization. IEEE
Transactions on Neural networks, 1999. 10(5): p.
1048-1054.
[7] Abhishek Suresh and Manuj Jha, Automated Essay
Grading Python Code. Available from:
https://github.com/absu5530/AES/blob/master/AES.
py
[8] Durgesh, K.S. and B. Lekha, Data classification
using support vector machine. Journal of Theoretical
and Applied Information Technology, 2010. 12(1): p.
1-7.
[9] Li, Y., K. Bontcheva, and H. Cunningham, Adapting
SVM for data sparseness and imbalance: a case
study in information extraction. Natural Language
Engineering, 2009. 15(2): p. 241-271.