A Graph Based Approach to Automate Essay Evaluation

Despite studies of over six decades, research on automated essay scoring continues to grab ample attention in the Natural Language Processing (NLP) community in part because of its commercial and educational value. However, evaluating such writing compositions or essays in terms of reliability and t...

Full description

Saved in:
Bibliographic Details
Main Authors: Bhatt, Reecha, Patel, Malvik, Srivastava, Gautam, Mago, Vijay
Format: Conference Proceeding
Language:English
Subjects:
automated
essays
evaluation
Grammar
NLP
Reliability
Roads
Semantics
Sentiment analysis
statistics
Syntactics
syntax
Task analysis
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Despite studies of over six decades, research on automated essay scoring continues to grab ample attention in the Natural Language Processing (NLP) community in part because of its commercial and educational value. However, evaluating such writing compositions or essays in terms of reliability and time is a very challenging process. The need for reliable and rapid scores has elevated the need for a computer system that can answer essay questions that fit precise prompts automatically. NLP and machine learning strategies use Automated Essay Scoring (AES) systems to solve the difficulty of scoring writing tasks. In this paper, we suggest an AES approach that involves not only rule-based grammar and consistency tests, but also the semantic similarity of sentences, thus giving priority to question prompts. Similarity vectors are used obtained after applying semantic algorithms and calculated statistical features. Our system uses 22 features with high predicting power, which is less than current systems, while considering every aspect a human grader may focus on.Predicting scores is achieved using the data provided by Kaggle's ASAP competition using Random Forest. The resulting agreement between the score of the human grader and the prediction of the system is compared with promising results through experimental evaluation.
ISSN:2577-1655
DOI:10.1109/SMC42975.2020.9282902