ABSTRACT

In today's world, currently there are many exam conduction methods, be it online
exams or OMR sheet exams or MCQ type exams. Various examinations are
conducted every day around the world. The most important aspect of any examination
is the checking of the answer sheet of the student . Usually it is done by the teacher
manually, thus making it a very tedious job if the number of students is very large. In
such a case automating the answer checking process would definitely prove to be of
great use. Automating the answer checking process would not only relieve the exam
checker but the checking process would also get way more transparent and fair as
there would not be any chances of biasedness from the teacher's side. Nowadays
various online tools are available for checking multiple choice questions but there are
very few tools to check subjective answer type examinations.
This brings in the need of Artificial Intelligence in our online exam systems. The
automation of scoring of subjective or descriptive answers is a need considered
nowadays. This project focuses on designing an efficient algorithm that will
automatically evaluate the answers given by students and assign a score based on the
AI technologies which are as good as scores given by a human being. Herein the
concept of automatic text similarity has been used to establish an application that
would check the answers of provided questions automatically and decide the
correctness of the answers.

KEYWORDS:E-assessment, Online Subjective Answer Checker, Automatic text Similarity,

Grammar, Descriptive.

1
INTRODUCTION

2
Examination is a test of a person’s knowledge in a particular area which is either subjective
or objective or both. Usually, competitive examinations consist of multiple-choice questions
or MCQs. Automatic evaluation of the objective exams is beneficial as it saves time, provides
efficiency and reduces usage of resources. However, this automated evaluation technique is
only for the objective exams and not for the subjective ones. Subjective answer sheet
checking is one of the huge administrative tasks for any education institute. In this
examination process, candidates need to write answers, an examiner collects those answer
sheets and submits them to authority for further checking process. This process involves 3
levels of paper checking: -
● First Level Paper Checker
● First Level Moderation
● Second Level Moderation
So, the amount of pressure education systems and teachers hold is understandable as the
number of answer sheets to evaluate is too large. So, there is a necessity for an approach
which will reduce the usage of resources and will automatically evaluate the answers given
by students and provide results.
This project is made exactly for this purpose. We have developed a system which takes
answers from student for the question paper set by the teacher and compares them with the
reference answers present in the system and awards marks by checking them automatically. It
takes certain parameters into consideration while evaluating, such as keywords, grammar,
similarity between sentences.
This application can be used in various educational institutes for checking subjective answer
type examinations. Further, on improving the application, it can even be extended for
conducting online subjective answer type examinations.

On running the application , the main window of the application will give two options to the
user , whether to login as an admin or as a student.

After selecting one of the options , the user will get to see a login window where he will be
asked to login using his/her credentials .

The admin will have the options like uploading the question paper and seeing the responses
of the students .

The student will have the option to upload the answer sheet and see the marks allotted to
them then and there.

3
 WHY DID WE CHOOSE ONLINE DESCRIPTIVE ANSWER
CHECKER?
There are a lot of reasons why introduction of Artificial intelligence into the online exam
systems would prove to be of great use.
Firstly , as currently exams are marked by examiners so this leads to fatigue and boredom as
they have to check large number of answer sheets, but with the online system, this problem
automatically gets solved . Moreover, the accuracy and speed with which a computer can
generate results , it is something which a human would take hours to do. The proposed
system would also produce unbiased results which would further make everything more
transparent.

EASE OF USE :

The proposed system is very easy and efficient to use. A student would be able
to easily interact with the interface without any confusion.
The application has been designed in such a way that anyone could make the
best use out of it without facing any major problems.

BALANCED LOAD :

Here the system would only be accessed by the admin and this would result in
lower load on the server.
Also the Automatic Answer Checker has been designed in such a way that its
maintenance can be done very easily.

EFFICIENCY AND RELIABILITY OF THE SYSTEM :

The system has high efficiency as it is made with negligible errors so it is even
more reliable.

4
System Requirements

5
HARDWARE REQUIREMENTS :

❖ PROCESSOR : i3 8th generation or above.

❖ RAM : Minimum 4 GB.

❖ STORAGE : Minimum 256 GB.

SOFTWARE REQUIREMENTS :

❖ [Link]

❖ Atom IDE

❖ Visual Studio Code

❖ MySQL

6
Technologies

7
Technologies

Our automated descriptive answer checker uses a combination

of natural language processing (NLP) algorithms and machine
learning techniques to evaluate student responses.
Here are the technical details of how our tool works:

❖ Tokenizer :
The first step in the process is to break down the student's response into individual
words or tokens using a tokenizer. This allows the tool to analyze the structure of the
response and identify important features, such as nouns, verbs, and adjectives.

❖ Parser :
The tool then uses a parser to analyze the syntax and grammar of the response. This
step helps to identify the relationships between words and phrases and understand the
overall meaning of the response.

❖ Semantic Analyzer :
Once the syntax and structure of the response have been analyzed, the tool uses a
semantic analyzer to understand the meaning of the response. This involves analyzing
the context and identifying the key concepts and ideas in the response.

❖ Machine Learning Models :

To evaluate and grade student responses, our tool uses a supervised machine learning
model. The supervised learning models are trained on a large corpus of text data,
allowing the tool to recognize patterns and make predictions about the quality of the
8
 response. The unsupervised learning models help to identify important features and
relationships in the data, improving the accuracy and performance of the tool. We also
use Large Language Model to improve our predictions.

❖ Database Schema and Models :

Our tool uses a database to store student responses, user data, and teacher’s Questions
and answers.

❖ Programming Languages and Frameworks :

Our tool was developed using Python, a powerful and flexible programming
language for data analysis and machine learning. We used popular deep learning
frameworks such as TensorFlow and PyTorch to build and train our models.

❖ User Interface :
The user interface of our tool was designed to be intuitive and easy to use. Teachers
can upload student responses in a image format(i.e. jpg, png, etc.) or else student can
take test from student’s portal. The tool then evaluates the responses and provides
responses to the teacher so the teacher can cross check the student’s answer and then
publish the results.

9
Architecture of Project

10
Modules Developed

11
Modules Developed

1. FORMATION OF WEBSITE :
❖ The website is designed using technologies like flask which is a python based
framework .
❖ The website has a main page which provides the user to login as an admin or as a
student .
❖ The admin window provides the option to upload the question paper.

❖ The student window has the facilities to upload his/her answer sheet.

2. FORMATION OF DATABASE :
❖ Database is designed using MySQL to store the original answers.

3. ANALYSIS OF DATA :
❖ The data set is collected in the very first step which consists of answers to the
questions in the question paper. Upon collecting the data, all the text in the data is
converted to lowercase.

❖ After the conversion to lowercase , word tokenization is performed on the text. Word
tokenization is the process of splitting a large sample of texts into words. This is a
requirement in natural language processing tasks where each word needs to be
captured and subjected to further analysis like classifying and counting them for a
particular sentiment etc. The Natural Language Toolkit (NLTK) is a library used to
achieve this.

❖ Moving forward , the next important step that is performed is removal of stop words
and punctuations . A stop word is a commonly used word (such as “the”, “a”, “an”,
“in”) that a search engine has been programmed to ignore, both when indexing entries
for searching and when retrieving them as the result of a search query. We would not
want these words to take up space in our database, or taking up valuable processing
time.

12
❖ At the end , stemming is applied to the dataset leading to a separate set of words .
Stemming is the process of producing morphological variants of a root/base word.
Stemming programs are commonly referred to as stemming algorithms or stemmers.

❖ A stemming algorithm reduces the words “chocolates”, “chocolatey”, “choco” to the

root word, “chocolate” and “retrieval”, “retrieved”, “retrieves” reduce to the stem
“retrieve”. Stemming is an important part of the pipelining process in Natural
language processing. The input to the stemmer are tokenized words.

13
Graphical User Interface

TEACHER PORTAL :

Purpose : Define the endpoints for teachers to create, update, or delete questions.

❖ ADD QUESTIONS

14
❖ VIEW QUESTIONS

15
❖ VIEW RESPONSES

❖ UPLOAD ANSWER SHEETS

❖ Application extracts text from uploaded images using the Google Cloud Vision API.

❖ Google Cloud Vision API: The Google Cloud Vision API uses optical character
recognition (OCR) to recognize text in images and convert it into a machine-readable
format. Our tool integrates with the Google Cloud Vision API to extract text from

16
 uploaded images and convert it into a readable format that can be evaluated by our
tool.

❖ Text Extraction: Our tool uses the Google Cloud Vision API to extract text from
uploaded images, including handwritten responses, diagrams, and other visual
representations of text. The extracted text is then converted into a readable format that
can be evaluated by our tool.

17
Student Portal

● Take Test

View Score Card

View Score Card

18
View Score Card

Database

19
Output

The output of your automated descriptive answer checker

project is the evaluation and scoring of the descriptive
answers provided by students.
The system generates scores for each answer based on the
evaluation criteria and algorithms implemented. These
scores represent the quality and correctness of the answers
provided by the students.

20
Advantages
● Examiners get bored by checking many answer sheets, hence the system reduces
their workload by automating the manual checking the process accurately.
● The system calculates the score and provides results instantly.
● It removes human errors that commonly occur during manual checking.
● The system provides an unbiased result.
● Thus the system excludes human efforts
● and saves time and resources.

Disadvantages
● Limited to specific subjects: The effectiveness of the automated descriptive answer
checker may vary depending on the subject or topic being evaluated. It may not be
equally accurate or suitable for all types of questions or subjects.
●
● Difficulty with complex answers: The tool may struggle to accurately evaluate
complex and nuanced answers that require human judgment or contextual
understanding beyond the capabilities of the machine learning algorithms.
●
● Language limitations: The accuracy of the tool may be influenced by variations in
language, including regional dialects, slang, or complex sentence structures, which
could affect the evaluation and scoring process.
●
● Lack of personalization: While the automated descriptive answer checker provides
objective evaluation, it may not offer personalized feedback tailored to individual
student needs, which can be valuable for targeted improvement.

21
Conclusion

Our automated descriptive answer checker tool is a powerful tool for

improving student learning outcomes. By streamlining the grading
process, providing detailed feedback, and enabling teachers to track
student performance over time, our tool helps students to improve their
understanding of the subject matter and achieve better academic
outcomes. Thank you for considering our automated descriptive answer
checker tool, and we look forward to working with you to improve
student learning outcomes.

22
Bibliography

Burstein, J., & Deane, P. (2018). Handbook of Automated Essay Evaluation: Current
Applications and New Directions. Routledge.
Coniam, D. (2014). Automated essay scoring. ELT Journal, 68(2), 204-206. Dikli, S. (2006). An
overview of automated scoring of essays.
The Journal of Technology, Learning, and Assessment, 5(1), 1-30.
Leacock, C., & Chodorow, M. (2003). Automated grammatical error detection. Journal of
Writing Research, 1(3), 1-15.
Landauer, T. K., Laham, D., & Foltz, P. (2003). Automated essay assessment. Assessment in
Education: Principles, Policy & Practice, 10(3), 295-308.
McCurry, K. (2018). Automated grading of descriptive writing: how machine learning can
improve the writing process. Educational Technology Research and Development, 66(3),
617-633.
Page, E. B., & Petersen, C. C. (1995). The imminence of grading essays by computer—25
years later. Journal of Educational Computing Research, 12(3), 265-278.
Rudner, L. M., & Liang, S. (2002). Automated essay scoring using bayesian networks. Journal
of Technology, Learning, and Assessment, 1(2), 1-14.

Shermis, M. D., & Burstein, J. (Eds.). (2013). Handbook of Automated Essay Evaluation :
Current Applications and New Directions. Routledge.

Wang, J., & Brown, M. (2008). Automated grading of short answer questions. In Proceedings
of the 2008 Conference on Empirical Methods in Natural Language Processing (pp. 1123-
1131). Association for Computational Linguistics.

23