Data Augmentation for Automated Essay Scoring using Transformer Models Kshitij Gupta

2025-05-06 0 0 341.91KB 5 页 10玖币

侵权投诉

Data Augmentation for Automated Essay Scoring

using Transformer Models

Kshitij Gupta

Department of Electrical and Electronics Engineering

BITS Pilani, Pilani Campus

Pilani, India

mailguptakshitij@gmail.com

Abstract—Automated essay scoring is one of the most im-

portant problem in Natural Language Processing. It has been

explored for a number of years, and it remains partially solved.

In addition to its economic and educational usefulness, it presents

research problems. Transfer learning has proved to be beneﬁcial

in NLP. Data augmentation techniques have also helped build

state-of-the-art models for automated essay scoring. Many works

in the past have attempted to solve this problem by using

RNNs, LSTMs, etc. This work examines the transformer models

like BERT, RoBERTa, etc. We empirically demonstrate the

effectiveness of transformer models and data augmentation for

automated essay grading across many topics using a single model.

Index Terms—Automated System, Transformers, BERT

I. INTRODUCTION

As a result of the COVID-19 pandemic, online schooling

system became necessary. From elementary schools to col-

leges, almost all educational institutions have adopted the on-

line education system. The majority of automated evaluations

are accessible for multiple-choice questions, but evaluating

short and essay type responses remains unsolved since, unlike

multiple-choice questions, there is no one correct solution

for these kind of questions. It is an essential education-

related application that employs NLP and machine learning

methodologies. It is difﬁcult to evaluate essays using basic

computer languages and methods such as pattern matching

and language processing.

Among the most important pedagogical uses of NLP is

automated essay scoring (AES), the technique of using a

system to score short and essay type questions without manual

assistance. Initiated by Page’s [1966] groundbreaking work

on the Project Essay Grader system, this area of study has

seen continuous activity ever since. The bulk of AES research

has been on holistic scoring, which provides a quantitative

summary of an essay’s quality in a single number. At least

two factors contribute to this concentration of effort. To begin

with, learning-based holistic scoring systems may make use

of publically accessible corpora that have been manually

annotated with holistic scores. Second, there is a market for

holistic scoring algorithms because they may streamline the

arduous process of manually evaluating the millions of essays

for tests like GRE, IELTS, SAT.

Past research on automated essay grading has included

training models for essays for which training data is available

and those models are topic speciﬁc. This model is trained on

all the topics thus could be used for assessment of essays of

all those topics without training model speciﬁc for each topic.

This would be useful in the scenario where we did not have

enough data to train a model that is speciﬁc to a particular

topic, but we still needed to evaluate essays on that topic.

Therefore, in order to assess them, We may utilize a model

that has been trained on essays on a variety of topics and a

tiny amount of data on the topic for which we need to develop

a model, which will then be ﬁne-tuned using the limited data

available on the subject being assessed.

This paper is organized as follows: In Section II, we

explore pertinent prior research on automated essay scoring;

in Section III, we cover experimental setup; and in Section IV,

we describe our methodology for augmenting essay data. In

Section V, we give the results and analysis of the automated

essay grading model. Section VI comprises of conclusion and

future work for Automated Essay Scoring.

II. RELATED WORKS

Project Essay Grader (PEG) by [1] started the research on

Automated essay scoring. Shermis (2001) [2] improved the

PEG system by incorporating the grammatical features as well

in the evaluation. Around the turn of century, great majority of

essay scoring systems used conventional methods like latent

semantic analysis by Foltz (1999) [3], as pattern matching and

statistical analysis like Bayesian Essay Test Scoring System by

[4]. These systems employ natural language processing (NLP)

approaches that concentrate on grammar, content to determine

an essay’s score.

Multiple studies studied AES systems, from the earliest

to the most recent. Blood (2011) [6] reviewed the PEG

literature from 1984 to 2010, it has discussed just broad

features of AES systems, such as ethical considerations and

system performance. However, they have not addressed the

implementation aspect, nor has a comparison research been

conducted, nor have the real problems of AES systems been

highlighted.

After 2014, Automated grading systems like as those by

[5] and others, employed deep learning approaches to induce

syntactic and semantic characteristics, producing greater out-

comes than previous systems. Burrows (2015) [7] reviewed on

arXiv:2210.12809v5 [cs.CL] 6 Feb 2023

文档加载中……请稍候！
如果长时间未打开，您也可以点击刷新试试。

下载文档到电脑，查找使用更方便

10 玖币 0人已下载

立即下载

摘要：

DataAugmentationforAutomatedEssayScoringusingTransformerModelsKshitijGuptaDepartmentofElectricalandElectronicsEngineeringBITSPilani,PilaniCampusPilani,Indiamailguptakshitij@gmail.comAbstractAutomatedessayscoringisoneofthemostim-portantprobleminNaturalLanguageProcessing.Ithasbeenexploredforanumberof...

展开>> 收起<<

Data Augmentation for Automated Essay Scoring using Transformer Models Kshitij Gupta.pdf

共5页,预览1页

还剩页未读，继续阅读

声明：本站为文档C2C交易模式，即用户上传的文档直接被用户下载，本站只是中间服务平台，本站所有文档下载所得的收益归上传人(含作者)所有。玖贝云文库仅提供信息存储空间，仅对用户上传内容的表现方式做保护处理，对上载内容本身不做任何修改或编辑。若文档所含内容侵犯了您的版权或隐私，请立即通知玖贝云文库，我们立即给予删除！

Data Augmentation for Automated Essay Scoring using Transformer Models Kshitij Gupta

相关推荐

开通VIP享超值会员特权

作者详情

相关内容

热门标签

举报选择: