Sentiment Analysis of Online Learning Feedback Using Deep Learning
A Project Work Synopsis
Submitted in the partial fulfillment for the award of the degree of
BACHELOR OF ENGINEERING
IN
COMPUTER SCIENCE WITH SPECIALIZATION IN
ARTIFICIAL INTELLIGENCE AND MACHINE LEARNING
Submitted by:
23BAI70309 Kabir
23BAI70433 Tisha
23BAI70460 Divyansh Mandhotra
Under the Supervision of:
Mr Parasdeep
CHANDIGARH UNIVERSITY, GHARUAN, MOHALI - 140413,
PUNJAB
BONA-FIDE CERTIFICATE
This is to certify that the Project Work Synopsis entitled
"Sentiment Analysis of Online Learning Feedback Using Deep Learning"
is a bonafide work carried out by
Kabir (23BAI70309)
Tisha (23BAI70433)
Divyansh Mandhotra (23BAI70460)
students of Bachelor of Engineering in Computer Science with Specialization in Artificial
Intelligence and Machine Learning, Chandigarh University, Gharuan, Mohali, Punjab, in
partial fulfillment of the requirements for the award of the degree of Bachelor of Engineering
during the academic year 2024-2025. The synopsis has been approved as it satisfies the
academic requirements in respect of the project work prescribed for the said degree.
Signature of Supervisor
_______________________
Mr Parasdeep
Department of CSE (AI & ML)
Chandigarh University
Date: ____________________
Place: Chandigarh University, Mohali
TABLE OF CONTENTS
Title Page.......................................................................................................................i
Bona-Fide Certificate...................................................................................................ii
Table of Contents........................................................................................................iii
List of Tables...............................................................................................................iv
List of Figures...............................................................................................................v
Abstract........................................................................................................................vi
1. Introduction...........................................................................................................1
1.1 Problem Definition..............................................................................................1
1.2 Project Overview.................................................................................................1
1.3 Hardware Specification.......................................................................................2
1.4 Software Specification.........................................................................................2
2. Literature Survey...................................................................................................3
2.1 Existing System...................................................................................................3
2.2 Proposed System..................................................................................................3
2.3 Literature Review Summary................................................................................4
3. Problem Formulation.............................................................................................5
4. Research Objectives..............................................................................................5
5. Methodology..........................................................................................................6
6. Experimental Setup................................................................................................7
7. Conclusion.............................................................................................................8
8. Tentative Chapter Plan..........................................................................................9
9. References.............................................................................................................9
LIST OF TABLES
Table No. Title........................................................................................Page No.
Table 1.1 Hardware Specification.............................................................................2
Table 1.2 Software Specification...............................................................................2
Table 2.1 Literature Review Summary......................................................................4
Table 6.1 Hyperparameter Settings for LSTM and BERT........................................7
Table 6.2 Model Performance Comparison...............................................................7
Table 6.3 Teaching Quality Index per Aspect...........................................................8
LIST OF FIGURES
Figure No. Title.......................................................................................Page No.
Figure 2.1 Comparison of Sentiment Analysis Methods...........................................3
Figure 5.1 System Architecture / Flowchart of Proposed Sentiment Analysis Pipeline 6
Figure 6.1 Dataset Class Distribution (Pie Chart).....................................................7
Figure 6.2 Model Performance Comparison (Bar Chart)..........................................8
Figure 6.3 Confusion Matrix for BERT Model.........................................................8
Figure 6.4 Teaching Quality Index per Aspect (Bar Chart)......................................8
ABSTRACT
The rapid growth of online learning platforms has generated massive volumes of student
feedback that are difficult to analyze manually. This project proposes a deep learning-based
system for automated sentiment analysis of student feedback, classifying each response as
Positive, Negative, or Neutral. Two deep learning models, Long Short-Term Memory
(LSTM) and BERT (Bidirectional Encoder Representations from Transformers), are
implemented and evaluated against traditional machine learning methods including Naive
Bayes and Support Vector Machines (SVM). The system employs standard NLP
preprocessing techniques such as tokenization, stop word removal, and word embeddings. A
dataset of 181 manually annotated student feedback samples was used for training and
testing. BERT achieved the highest accuracy of 89.5%, outperforming all other models.
Additionally, a Teaching Quality Index (TQI) was computed to provide educators with
actionable, aspect-level insights into instructional quality. The proposed system reduces
manual effort, eliminates human bias, and enables data-driven decision making in education.
Keywords:
Sentiment Analysis, Deep Learning, LSTM, BERT, NLP, Online Learning, Student
Feedback, Teaching Quality Index, Text Classification.
1. INTRODUCTION
1.1 Problem Definition
Online learning platforms generate thousands of student feedback responses every
semester. Manually reading and analyzing all this feedback to assess teaching quality is time-
consuming, inconsistent, and subject to human bias. There is a need for an automated system
that can accurately classify the sentiment of student feedback and provide meaningful
insights to educators without any manual intervention.
1.2 Project Overview
This project develops a deep learning-based sentiment analysis system for student
feedback collected from online learning platforms. The system takes raw feedback text as
input, preprocesses it using NLP techniques, and classifies it as Positive, Negative, or Neutral
using LSTM and BERT models. It further computes a Teaching Quality Index (TQI) for
different teaching aspects such as clarity, content, examination, and engagement, giving
educators a clear picture of where improvements are needed.
The system was trained and tested on a dataset of 181 student feedback samples collected
from Chandigarh University. BERT achieved the best accuracy of 89.5%, proving the
effectiveness of transformer-based deep learning models for this task.
1.3 Hardware Specification
Table 1.1: Hardware Specification
Component Specification
Processor Intel Core i5 or higher
RAM Minimum 8 GB (16 GB recommended)
Storage 50 GB free disk space
GPU NVIDIA GPU with 4 GB VRAM (optional)
Operating System Windows 10 / Ubuntu 20.04 or later
1.4 Software Specification
Table 1.2: Software Specification
Software / Library Version / Details
Programming Language Python 3.10
Deep Learning Framework TensorFlow 2.x / PyTorch
Transformer Library HuggingFace Transformers
NLP Library NLTK, spaCy
Data Processing NumPy, Pandas
Visualization Matplotlib, Seaborn
Development Environment Jupyter Notebook / VS Code
2. LITERATURE SURVEY
2.1 Existing System
Existing approaches to sentiment analysis of student feedback rely mainly on traditional
machine learning techniques. Naive Bayes classifiers use word frequency counts and
probabilistic rules to classify text but cannot understand sentence context. Support Vector
Machines (SVM) improved on this by learning a decision boundary in high-dimensional
feature space, but still required manual feature extraction using Bag-of-Words or TF-IDF.
These methods perform reasonably well on simple feedback but fail to handle negation,
sarcasm, and complex sentence structures.
2.2 Proposed System
The proposed system overcomes the limitations of existing approaches by using deep
learning models that automatically learn features from raw text. LSTM captures long-term
sequential dependencies using memory gates, while BERT uses bidirectional self-attention to
understand the full context of each word. The system includes a complete NLP preprocessing
pipeline, pre-trained word embeddings, and a Teaching Quality Index (TQI) for aspect-level
insight generation.
2.3 Literature Review Summary
The following table summarizes the key research articles reviewed for this project:
Table 2.1: Literature Review Summary
Year Author / Article Tools Technique Source Evaluation
Parameter
2022 Shaik et al. - Sentiment analysis Python, NLTK Survey / J. King Saud Coverage,
on educational data Review Univ. Accuracy
2021 Kastrati et al. - NLP and Deep TensorFlow, LSTM, BERT IEEE Access Accuracy, F1-
Learning for feedback BERT Score
2023 Grimalt-Alvaro et al. - SA for Python, spaCy Systematic Education & Precision, Recall
formative assessment Review IT
2022 Dake et al. - SA for student scikit-learn SVM, Naive Computers in Accuracy, Kappa
responses Bayes Edu.
2024 Huang et al. - Deep Learning for TensorFlow, LSTM, CNN IEEE Access Accuracy, F1-
course reviews Keras Score
2019 Onan - Mining instructor Python, Keras Deep Learning Expert Accuracy, AUC
evaluation reviews Systems
2022 Ren et al. - Automatic scoring PyTorch, BERT Aspect-Based Education & Precision, Recall,
using ABSA SA IT F1
3. PROBLEM FORMULATION
Due to the rapid growth of internet-based learning systems, vast amounts of unstructured
student feedback are being generated every day. Processing such data manually is laborious,
slow, and subject to human bias. Existing machine learning-based sentiment analysis
methods also tend to ignore the context and semantics of the text, resulting in poor
classification performance on complex or nuanced feedback.
Therefore, there is a need for a robust and automated system that can accurately analyze
student feedback using state-of-the-art deep learning techniques. The problem is formally
defined as follows: given a dataset of student feedback texts, the system must automatically
classify each feedback as Positive, Negative, or Neutral, and further extract teaching quality
insights at the aspect level. The system must outperform traditional machine learning
baselines and be capable of understanding the contextual meaning of feedback text.
4. RESEARCH OBJECTIVES
The main objective of this project is to develop an automated sentiment analysis system
that classifies student feedback from online learning platforms into Positive, Negative, and
Neutral categories using deep learning models such as LSTM and BERT. The system aims to
outperform traditional machine learning methods by better understanding the context of
feedback, and to provide educators with meaningful teaching quality insights through a
computed Teaching Quality Index (TQI).
The specific objectives are:
1. To build an automated sentiment analysis system for classifying student feedback
from online learning platforms.
2. To implement and compare LSTM and BERT deep learning models against
traditional baselines, Naive Bayes and SVM.
3. To preprocess raw student feedback text using NLP techniques including
tokenization, stop word removal, and normalization.
4. To evaluate model performance using Accuracy, Precision, Recall, and F1-Score.
5. To compute a Teaching Quality Index (TQI) providing aspect-level insights into
instructional quality.
5. METHODOLOGY
The proposed methodology follows a five-stage pipeline to process raw student feedback
and classify it into sentiment categories. The overall flow is shown in Figure 5.1.
Stage 1: Data Collection and Annotation
A dataset of 181 student feedback samples was collected through end-of-semester course
evaluation forms and online questionnaires at Chandigarh University. Each sample was
manually labeled as Positive, Negative, or Neutral by two independent annotators.
Stage 2: Text Preprocessing
Each feedback sample was cleaned by converting text to lowercase, removing
punctuation and special characters, eliminating stop words using NLTK, and tokenizing the
text. Whitespace tokenization was used for LSTM and WordPiece tokenization for BERT.
Stage 3: Feature Extraction
For the LSTM model, pre-trained GloVe word embeddings (100-dimensional) were used
to convert words into dense numerical vectors.
Stage 4: Model Training
Two deep learning models were trained. The LSTM model processes text as a sequence
using forget, input, and output gates to retain important information. BERT uses bidirectional
self-attention and was fine-tuned on the student dataset. Both models output sentiment class
probabilities via a softmax layer.
Stage 5: Teaching Quality Insight Generation
After classification, a Teaching Quality Index (TQI) was computed per aspect: TQI =
((Npos - Nneg) / N) x 100. A positive TQI indicates good performance; a negative score flags
areas needing improvement.
Figure 5.1: Flowchart of Proposed Sentiment Analysis System
6. EXPERIMENTAL SETUP
6.1 Software and Hardware Environment
All experiments were conducted in Python 3.10. LSTM was built using TensorFlow 2.x
with Keras. BERT was implemented using the HuggingFace Transformers library with bert-
base-uncased. Text preprocessing used NLTK. A fixed random seed (42) was used for
reproducibility.
6.2 Dataset Split
The 181 samples were split into 80% training (145 samples) and 20% testing (36
samples) using stratified sampling to maintain class distribution. Figure 6.1 shows the dataset
class distribution.
Figure 6.1: Dataset Class Distribution
6.3 Hyperparameter Settings
Table 6.1: Hyperparameter Settings
Parameter LSTM BERT
Epochs 20 5
Learning Rate 0.001 2 x 10-5
Batch Size 32 16
Optimiser Adam AdamW
Loss Function Cross-Entropy Cross-Entropy
6.4 Results
Table 6.2: Model Performance Comparison
Model Accuracy F1-Score Recall
Naive Bayes 70.5% 69.6% 69.5%
SVM 75.1% 74.2% 74.0%
LSTM 82.0% 81.2% 81.0%
BERT 89.5% 88.7% 88.5%
Figure 6.2: Model Performance Comparison
Figure 6.3: Confusion Matrix for BERT Model
BERT achieved the highest accuracy of 89.5%, outperforming LSTM by 7.5% and both
traditional baselines by a significant margin.
6.5 Teaching Quality Index
Table 6.3: Teaching Quality Index per Aspect
Teaching Aspect TQI Status
Clarity of Instruction +66.3 Good
Course Content +53.0 Good
Examination +58.6 Good
Lab Work +50.3 Good
Library Facilities +52.5 Good
Extracurricular +76.2 Good
Figure 6.4: Teaching Quality Index per Aspect
7. CONCLUSION
This project successfully developed a deep learning-based sentiment analysis system for
classifying student feedback from online learning platforms. BERT achieved the highest
accuracy of 89.5%, demonstrating its superior ability to understand the context of student
feedback compared to LSTM, Naive Bayes, and SVM. The Teaching Quality Index (TQI)
revealed that while content delivery and clarity were well received, student engagement and
assessment methods need improvement. The proposed system automates the feedback
analysis process, reduces manual effort, and enables educators to make data-driven decisions
to improve the quality of online education.
8. TENTATIVE CHAPTER PLAN FOR THE PROPOSED WORK
Chapter 1: Introduction
Introduction to online learning, student feedback, sentiment analysis, limitations of
traditional approaches, and the role of deep learning in solving this problem.
Chapter 2: Literature Review
Review of existing research on sentiment analysis in education, comparison of traditional
ML and deep learning methods, and identification of gaps addressed by this project.
Chapter 3: Objective
Clear statement of the research objectives and the goals of the proposed system.
Chapter 4: Methodology
Detailed description of the five-stage pipeline: data collection, preprocessing, feature
extraction, model training, and teaching quality insight generation.
Chapter 5: Experimental Setup
Description of the software environment, dataset split, hyperparameter configurations,
baseline models, and evaluation metrics used.
Chapter 6: Conclusion and Future Scope
Summary of findings, conclusions drawn from results, and future directions including
real-time deployment, multilingual support, and aspect-based sentiment analysis.
9. REFERENCES
[1] T. Shaik et al., "Sentiment analysis and opinion mining on educational data: A survey,"
Journal of King Saud University - Computer and Information Sciences, 2022.
[2] Z. Kastrati et al., "Sentiment Analysis of Students' Feedback with NLP and Deep Learning: A
Systematic Mapping Study," IEEE Access, 2021.
[3] C. Grimalt-Alvaro et al., "Sentiment analysis for formative assessment in higher education: a
systematic literature review," Education and Information Technologies, 2023.
[4] D. K. Dake et al., "Using sentiment analysis to evaluate qualitative students' responses,"
Computers in Education, 2022.
[5] M. M. F. Fahima et al., "Sentiment Analysis and its Application in the Education Sector: A
Comprehensive Review," 2025.
[6] T. Kasana et al., "Evaluating Educational Commentaries: Sentiment Analysis for Student
Testimonials Using Machine Learning," 2024.
[7] S. B. Deshpande et al., "Elevating educational insights: sentiment analysis of faculty feedback
using advanced machine learning models," 2025.
[8] S. Huang et al., "A Deep Learning Model for Student Sentiment Analysis on Course
Reviews," IEEE Access, 2024.
[9] A. Onan, "Mining opinions from instructor evaluation reviews: A deep learning approach,"
Expert Systems with Applications, 2019.
[10] S. Kumar et al., "Using Machine Learning Sentiment Analysis to Evaluate Students Learning
Impact," 2025.
[11] J. Zhou et al., "Sentiment analysis in education research: a review of journal publications,"
Interactive Learning Environments, 2020.
[12] P. Ren et al., "Automatic scoring of student feedback for teaching evaluation based on
aspect-level sentiment analysis," Education and Information Technologies, 2022.
[13] M. Misuraca et al., "Using Opinion Mining as an educational analytic," Studies in
Educational Evaluation, 2021.