You are on page 1of 15

PANIMALAR ENGINEERING COLLEGE

(An Autonomous Institution)


Bangalore Trunk Road, Varadharajapuram,
Poonamallee, Chennai-600 123

DEPARTMENT OF COMPUTER SCIENCE AND BUSINESS SYSTEMS

MULTILINGUAL SHORT ANSWER EVALUATION USING NLP AND ML


TECHNIQUES
DOMAIN: NATURAL LANGUAGE PROCESSING

TEAM 23 R ASHWIN 211420244006


GUIDE: Mrs. K. Kiruthika. M.Tech. Ph.D.,
Associate Professor
OVERVIEW

1. Objective
2. Abstract
3. Literature Survey
4. Existing System
5. Proposed System
6. Architecture Diagram
7. System Specifications
8. List Of Modules
9. Learning Curve
10. Confusion Matrix
11. Model Interface
OBJECTIVE

The main objective of the system leverages deep learning models, specifically
BERT and mBART, to understand and evaluate short answers across multiple
languages efficiently. It addresses the challenges faced in evaluating open-ended
questions and enhances accuracy by integrating semantic textual similarity
techniques.

The primary objective of this project is to explore and implement a robust solution
for the automatic evaluation of short answers across multiple languages using state-
of-the-art NLP and machine learning techniques.
ABSTRACT

This research introduces an innovative approach to Multilingual Automatic Short


Answer Evaluation employing Natural Language Processing (NLP) and machine
learning techniques. The system leverages advanced deep learning models,
specifically BERT and mBART, to understand and evaluate short answers across
multiple languages efficiently. It addresses the challenges faced in evaluating open-
ended questions and enhances accuracy by integrating semantic textual similarity
techniques. The system's effectiveness is demonstrated through its application to
diverse datasets, including a novel approach to convert textbooks into a text corpus
for training. The report details the exploration of NLP fundamentals, the
development of a model combining BERT and unsupervised learning algorithms,
and the overcoming of challenges such as skewed datasets and model training
issues.
LITERATURE SURVEY
TITLE OF PAPER AUTHOR YEAR
PUBLISHED

Fooling Automatic Short Answer Grading Systems Anna Filighera, Tim Steuer & 2020
(vol. 11, pp. 37203-37215)
Christoph Rensing

Text-to-text Semantic Similarity for Automatic Short Boštjan Šumak, Saša Brdnik, 2021
Answer Grading Maja Pušnik
(vol. 11, pp. 29541-29553)

Handwritten Short Answer Evaluation System Sijimol P J, Surekha Mariam 2018


(HSAES). Varghese
(vol. 11, pp. 1-4)
EXISTING SYSTEM
1. Keyword Matching Techniques: This method is straightforward and effective for
questions with expected specific terms or phrases. However, its rigidity fails to
account for synonyms, paraphrasing, or conceptually correct answers that do not
include the exact keywords.
2. Rule-based Analysis: These systems employ a set of linguistic rules to evaluate
answers. Rule-based analysis might involve grammar checking, sentence
structure analysis, and the presence of certain key phrases indicative of correct
responses.
3. Corpus-based Similarity Measures: Some systems attempt to evaluate answers
by measuring their similarity to a corpus of correct responses. Techniques such
as cosine similarity or Jaccard index are used to quantify how closely a student's
answer resembles known correct answers.
EXISTING SYSTEM

LIMITATIONS
• High Computational Demands: Training MLP Models multiple
times is very computationally intensive task requiring huge
amounts of GPU and CPU time.

• Limited Accessibility: As it is costlier to develop this model, it is


not widely accessible.

• Underfitting Risk: Due to limited set of datasets, the model might


not get trained and learn the patterns in the data.

• Challenges in Deployment
PROPOSED SYSTEM

The proposed model aims to enhance the speed and effectiveness of short answer
grading, facilitating quicker and more accessible grading system. Through
experimentation and evaluation on student answers datasets, this research
contributes to advancing efficient and accurate classification methods for student
descriptive answer evaluation analysis, ultimately aiding in faster and more
accessible grading, enhancing the experience of online learning.
PROPOSED SYSTEM

ADVANTAGES
• Computational Efficiency
• Improved Speed
• Resource Optimization
• Maintained Accuracy
• Scalability
• Enhanced Accessibility
HARDWARE AND SOFTWARE SPECIFICATIONS

Hardware Requirements
PROCESSOR Intel Core i5, AMD Ryzen 5
RAM 32 GB
ROM 256 GB
GPU NVIDIA RTX (4GB+ RAM)

Software Requirements
OS Windows, macOS, Linux
IDE Visual Studio Code
Programming language Python
Model Training
MODULES

1. Deep Learning Framework (TensorFlow)


2. Data Handling (Pandas)
3. Visualization (Matplotlib, Plotly Express)
4. Model Evaluation Metrics (Scikit-learn)
LEARNING CURVE
MODEL INTERFACE
THANK YOU

You might also like