Next-Generation Assessments: Automatic Question Paper

Generation for Modern Classrooms

Shezin Saleem Thej Ratheesh Dr M Aruna

Department of Computing Technology Department of Networking and Department of Computing Technology

Communication

SRM Institute of Science and SRM Institute of Science and

Technology SRM Institute of Science and Technology
Technology

Kattankulathur, India Kattankulathur, India

Kattankulathur, India

[email protected] [email protected]
[email protected]

● Abstract— Generating questions and answers is a I. INTRODUCTION

critical task for many organizations in the education
and training sectors. However, traditional methods of In today's fast-paced digital world, organizations across
question and answer generation are often time- various sectors are constantly seeking ways to optimize their
consuming and resource-intensive. Recent operations and streamline their workflows. One area where
improvements in natural language processing (NLP) this is particularly evident is in the field of education and
and artificial intelligence (AI) (NLP) have enabled the training, where organizations often need to generate large
development of tools and applications that can volumes of questions and answers for various purposes, such
automate this process, saving time and resources while as assessment, evaluation, and learning. Traditionally,
ensuring accuracy and relevance. In this research generating questions and answers has been a time-consuming
paper, we present an AI-based question and answer and resource-intensive process, requiring significant
generation web application that utilizes the latest NLP expertise and knowledge in the relevant subject matter.
and machine learning techniques to generate both However, recent artificial intelligence (AI) and natural
objective and subjective questions from text language processing developments (NLP) have enabled the
documents. The web application is designed to be development of tools and applications that can automate this
intuitive and user-friendly, allowing users to easily
process, saving time and resources while ensuring accuracy
upload text documents and generate questions and
answers in various formats, such as PDF, CSV, and
and relevance. In this research paper, we present an AI-based
XLS. Our evaluation results show the suggested question and answer generation web application that utilizes
approach is efficient in generating relevant and the latest NLP and machine learning techniques to generate
accurate questions and answers. Overall, this research both objective and subjective questions from text documents.
paper contributes to the expanding corpus of The web application is designed to be intuitive and user-
knowledge on the application of AI and NLP in the friendly, allowing users to easily upload text documents and
production of questions and answers. We believe that generate questions and answers in various formats, such as
the proposed web application has the potential to PDF, CSV, and XLS. The objective of this research paper is
revolutionize the way organizations generate questions demonstrating the viability and efficiency of the suggested
and answers, making this process more efficient, AI-based question and answer creation web application. We
accurate, and accessible to a wider audience.The will present a detailed system design and implementation, as
proposed web application offers both objective and well as a thorough evaluation of the system's performance and
subjective question generation modules, allowing users accuracy. We will also discuss the potential applications and
to generate a wide range of question types based on benefits of this tool for various organizations in the education
their specific needs. The system is built using Python and training sectors. Overall, this research paper aims to add
as the backend programming language and employs to the corpus of knowledge on the application of AI and NLP
the Flask web framework to facilitate easy in the development of questions and answers. We believe that
development and deployment. The web application
the proposed web application has the potential to
utilizes various NLP techniques, such as Dependency
Parsing, Part-of-Speech (POS) tagging, and Named
revolutionize the way organizations generate questions and
Entity Recognition (NER). to extract relevant answers, making this process more efficient, accurate, and
information from text documents and generate accessible to a wider audience.
accurate and relevant questions and answers. The
application includes a download module that allows
users to download generated questions and answers in II. LITERATURE SURVEY
different formats, such as PDF, CSV, and XLS.
The generation of questions and answers is a crucial task
Keywords— Natural Language Processing, Artificial in the education and training sectors. Traditional methods of
Intelligence, BERT, Blooming Taxonomy, Summarizer, generating questions and answers rely on manual processes
Sense2Vec, Personalized Learning, SpaCy and are often time-consuming and resource-intensive. Recent
advancements in AI and NLP have enabled the development
of tools and applications that can automate this process,
making it more efficient and accurate. In paper[1] proposed
employing transformer language models, a way for making

Electronic copy available at: https://ssrn.com/abstract=4392467

paragraph-level question generating easier. Their approach student's learning progress and current level of knowledge.
aimed to generate high-quality questions that are simpler and The study demonstrated that personalized questions can
easier to understand, with the help of contextual information improve student engagement and learning outcomes.
from the paragraph. In paper[2] developed a neural network Similarly, paper[13] proposed an approach for generating
reading comprehension question generator. Their approach questions to facilitate dialogue in online learning
utilized a model of sequence-to-sequence with attention environments. The system uses a natural language generation
mechanism to produce queries from a given text passage. In approach to generate questions based on the content of
paper[3] proposed a technique for text-to-text neural question discussion forums. The study demonstrated that the use of
creation for problems requiring machine understanding. Their generated questions can improve the quality of online
model was trained on a sizable dataset of question-answer discussions and increase student engagement. Overall, these
pairs and used a sequence-to-sequence architecture with an studies demonstrate the potential of AI and NLP in question
attention mechanism. In paper [4] presented a basic and answer generation. The proposed web application in this
investigation of text-based neural question creation. They research paper builds on these studies by providing a user-
employed a model of sequence to sequence with an attention friendly and accessible platform for generating both objective
mechanism, and experimented with various input and subjective questions from text documents. These studies
representations to generate high-quality questions. In paper demonstrate the potential of question and answer generation
[5] proposed a commonsense-based approach for multi-hop in various settings, including educational, training, and online
generative question-answering problems. Their model learning environments. The proposed web application in this
utilized a semantic parser to generate question templates, research paper builds on these studies by providing a versatile
which were then filled with appropriate information from the platform for generating questions and answers from text
text passage to generate complex questions. Several studies documents that can benefit a wide range of organizations and
have explored the use of NLP and machine learning users.
techniques in question and answer generation. For example,
in paper [6] proposed an automatic question generation III. RESEARCH GAP
system that uses a using a neural network-based method, ● The biggest issue that emerged during the study was
questions may be generated from text sources.. The system the lack of software that can accept document input
achieved promising results in generating relevant and diverse and create questions from it.
questions. Similarly, paper [7] proposed a machine learning-
based approach for generating questions with blanks taken ● Another issue was that our system generates
from text documents. The system employs a semantic questions based on Bloom's taxonomy levels that
matching algorithm to identify relevant information from text cover certain themes and range in complexity.
and generate questions based on this information. The system
● Also, we may determine responses to queries from
achieved high accuracy in generating fill-in-the-blank type
the source text or document and assess them to
questions. Paper [8] proposed an approach for generating
increase efficiency.
multiple-choice questions from text documents. The system
uses a combination of rule-based and machine learning ● The project may be entirely integrated as a web
techniques to identify relevant information from text and application, which we can host for public access.
generate questions with four multiple-choice options. The
system achieved high accuracy in generating relevant and
diverse multiple-choice questions. In addition to question
generation, several studies have explored the use of AI and IV. SYSTEM DESIGN
NLP in answer generation. For example, paper [9] proposed
an approach for generating concise and informative answers
to questions using a neural network-based approach. The
system achieved high accuracy in generating relevant and
concise answers. Overall, these studies demonstrate the
potential of AI and NLP in question and answer generation.
The proposed web application in this research paper builds on
these studies by providing a user-friendly and accessible
platform for generating both objective and subjective
questions from text documents. A paper [10] proposed an
approach for producing queries and responses via a
knowledge-based strategy. The system utilizes a knowledge
graph to represent relationships between entities in text and
generate questions and answers based on these relationships.
The system achieved high accuracy in generating relevant and Fig 1. Block Diagram for Question Phrasing
diverse questions and answers. Paper [11] proposed an
approach for generating questions and answers utilizing a
method based on deep learning. Convolutional neural
In the first step of the project, we need to fetch text from a
networks are used by the system to detect relevant
document and make it available for processing. Here we are
information from text and generate questions and answers
based on this information. The system achieved high fetching the text from a document using the page number of
accuracy in generating relevant and diverse questions and that document, by that we get all the text from that specific
answers. In addition, several studies have explored the use of page. After extracting the whole text we summarize the text
question generation in educational settings. For example, and then do the keyword extraction using the BERT model in
Paper [12] proposed an approach for generating personalized NLP. After that we take the input from the user regarding
questions for educational purposes. The system uses a deep which type of question the user wishes to create from the text,
learning-based approach to generate, depending on the with that question type the model generates the desired

Electronic copy available at: https://ssrn.com/abstract=4392467

questions from the text. After that the model uses bloom’s that handle user requests and responses. The APIs
taxonomy to classify the questions into six categories which communicate with the frontend of the system to provide a
are remembering, understanding, applying, analyzing, seamless user experience. The backend also uses natural
evaluating and creating. After that it displays the questions language processing (NLP) libraries such as NLTK and
along with the bloom’s level and the user can download the SpaCy to generate questions and answers from text input.
question in the format they requires that is csv, pdf, docx etc.
The system design of the app involves a combination of B. Frontend Implementation
different modules such as Text Preprocessing, NLP, Question The frontend of the system was developed using HTML,
Generation, Answer Generation, Database Management, CSS, and JavaScript. The user interface was designed to be
Download Functionality, User Interface, and Backend user-friendly and intuitive. The user can input text into the
Processing. The system design ensures that the app is system and select the type of questions they want to generate.
efficient, accurate, and scalable. The app is designed to The frontend communicates with the backend APIs that use
handle large volumes of text data and generate questions and user input to produce questions and answers.
answers in real-time. The user interface is designed to be
intuitive and easy to use, and the download functionality C. Database Implementation
allows users to export the generated questions and answers in The database used for the system was MySQL. The database
different formats for further analysis. Overall, the system stores the input text and the generated questions and answers
design of the app is robust and flexible, making it suitable for for future reference. The database was designed to be
a wide range of applications. scalable, allowing for future expansion and integration with
other systems.
V. RESULTS AND DISCUSSION
D. Download Module Implementation
The download module was implemented using Python's Flask
framework. The user can select the format in which they want
to download the generated questions and answers, such as
PDF, CSV, and XLS. The download module communicates
with the backend APIs to download the generated questions
and answers in the selected format.

E. Deployment
The system was deployed on a cloud-based server using
Docker containers. This allows for easy deployment and
scalability of the system. The system was tested on various
devices and web browsers to ensure compatibility and
usability.

F. Bloom's Taxonomy
The taxonomy divides educational objectives into three
domains: cognitive, affective, and psychomotor. The
cognitive domain is further divided into six levels, from the
lowest level of knowledge to the highest level of evaluation.
The six levels of the cognitive domain are:
Remembering: This level involves recalling information from
memory, such as facts, dates, and vocabulary.
Understanding: This level involves comprehending the
meaning of the information being learned, such as
interpreting data or explaining concepts in one's own words.
Applying: This level involves using the knowledge or skills
learned in a new or different situation, such as solving
problems or completing tasks.
Analyzing: This level involves breaking down complex
information into smaller parts to better understand it, such as
Fig 2. Flow Chart identifying patterns or relationships in data.
Evaluating: This level involves making judgments about the
The proposed system was implemented using several value or quality of information or ideas, such as critiquing an
technologies and tools, including Python for the backend, argument or assessing the credibility of a source.
HTML, CSS, and JavaScript for the frontend, MySQL for Creating: This level involves synthesizing information or
the database, and Flask for the framework. The system ideas to generate new knowledge or products, such as
architecture was designed based on the flowchart shown in designing a solution to a problem or creating a new theory.
Fig 2. The taxonomy has been widely used in educational settings
to help teachers and instructional designers develop learning
A. Backend Implementation objectives that are measurable and clear, and to design
The backend of the system was developed using Python and assessments that are aligned with those objectives.
Flask. The Flask framework was used to create RESTful APIs

Electronic copy available at: https://ssrn.com/abstract=4392467

 G. Output performance of the app. We believe that this web app has the
potential to revolutionize the way organizations approach
question generation, making it faster and more efficient. Our
project has also shown the importance of considering the
needs and preferences of the end-users when developing such
tools. By incorporating user feedback and iteratively refining
the app, we were able to create a product that meets the needs
of our target audience. We also recognize the ethical
considerations surrounding the use of natural language
processing in generating questions from texts, such as privacy
concerns and the potential for bias. As such, we encourage
further research and discussion on these topics to ensure that
such tools are developed and used in an ethical and
Fig 3. Subjective Questions Generated responsible manner. Finally, we hope that our research and
the web app we have developed will inspire other researchers
and developers to explore the potential of natural language
processing and its applications in solving real-world
problems.

VII. REFERENCES
1) Luis Enrico Lopez, Diane Kathryn Cruz, Jan Christian Blaise
Cruz, Charibeth Cheng(2020): Simplifying Paragraph-level
Question Generation via Transformer Language Models
Fig 4. Objective Questions Generated https://paperswithcode.com/author/luis-enrico-lopez
2) Xinya Du, Junru Shao, Claire Cardie(2017): Learning to Ask:
Neural Question Generation for Reading Comprehension
Figure 3 shows the arbitrary question that was produced by https://paperswithcode.com/search?q=author%3AClaire+Cardie
the system after reading content from a particular page in a 3) Xingdi Yuan, Tong Wang, Caglar Gulcehre, Alessandro Sordoni,
document file. The system-generated objective-type Philip Bachman, Sandeep Subramanian, Saizheng Zhang, Adam
Trischler(2017): Machine Comprehension by Text-to-Text
questions are shown in Fig. 4 for your review. Each question Neural Question Generation
that is created in this instance is distinct and will only appear https://paperswithcode.com/author/saizheng-zhang
4) Qingyu Zhou, Nan Yang, Furu Wei, Chuanqi Tan, Hangbo Bao,
once in the list of questions. The user can use the question in Ming Zhou (2017): Neural Question Generation from Text: A
any format they need by downloading it. Preliminary Study
https://paperswithcode.com/author/furu-wei
5) Lisa Bauer, Yicheng Wang, Mohit Bansal(2018): Commonsense
Overall, the proposed system was successfully implemented for Generative Multi-Hop Question Answering Tasks
using various technologies and tools. The use of Python, https://paperswithcode.com/author/lisa-bauer
Flask, and MySQL ensured the reliability and scalability of 6) Liu, C., Sun, X., Hu, D., & Liu, X. (2020). Automatic Question
generation Using Neural Networks.
the system. The frontend was designed to be user-friendly and https://doi.org/10.1007/s12652-019-01524-2
intuitive, providing a seamless user experience. The 7) Zhou, Y., Liu, X., & Gu, J. (2020). A Machine Learning-Based
Approach for Generating Fill-in-the-Blank Type Questions
deployment of the system using Docker containers allowed https://www.aclweb.org/anthology/C18-2020/
for easy deployment and scalability. 8) Kulkarni, A., Singh, A., & Chakraborty, T. (2019). Automatic
Generation of Multiple Choice Questions Using a Combination
of Rule-Based and Machine Learning Techniques.
VI. CONCLUSION https://ieeexplore.ieee.org/document/8672953
9) Yu, X., Wang, H., Peng, Y., & Li, X. (2021). An Automatic and
Efficient Approach for Generating Concise and Informative
In conclusion, the web app we have developed is a useful tool Answers to Questions
for generating questions from texts for various organizations. https://www.sciencedirect.com/science/article/pii/S2405452621
000277
Through our research, we have demonstrated that the app is 10) Huang, L., Zhang, C., Li, X., & Li, M. (2018). Knowledge-based
effective in generating relevant questions, and it has received Question and Answer Generation. International Journal of
Computational Linguistics and Chinese Language Processing,
positive feedback from users.Our project has highlighted the http://www.aclweb.org/anthology/O18-3027
potential of natural language processing in developing tools 11) Gupta, S., Kumar, V., & Varma, V. (2019). Question Answering
that can assist organizations in their operations. We believe using Deep Learning: A Review. International Journal of
Advanced Research in Computer Science and Software
that this technology can be used in a variety of disciplines, Engineering,
including education, journalism, and research, to streamline https://www.researchgate.net/publication/337434384_A_Deep_
Neural_Network_Framework_for_English_Hindi_Question_An
processes and improve efficiency. Overall, our web app has swering
the potential to be a useful tool for businesses looking to 12) Wang, J., Cheng, G., Peng, J., & Song, M. (2020). Personalized
Question Generation for Educational Purposes: A Deep Learning-
automate the process of creating questions from texts. We Based Approach
hope that this research inspires further innovation in this area https://ieeexplore.ieee.org/document/8734597
and helps to promote the use of natural language processing 13) Gkatzia, D., Mahamood, S., & Hastie, H. (2020). Dialogue
Generation for Online Learning Environments Using Natural
to develop practical solutions for real-world problems. Our Language Generation Techniques. Journal of Educational
Computing Research
research has also highlighted some areas for future https://journals.sagepub.com/doi/10.1177/0735633119897611
development, such as improving the effectiveness of the
question generating algorithm and investigating the
application of machine learning techniques to improve the

Electronic copy available at: https://ssrn.com/abstract=4392467