Table 1. The Research Story.

This part introduces the research to be done and should be limited only to a specific area of research.
Maximum of 20 sentences only. The story should be an inverted pyramid, meaning from general to specific.

The Storyline
Customer feedback is a vital source of information for businesses to improve products, services, and customer experience. With the
rapid growth of online platforms, businesses receive massive volumes of feedback daily through social media, reviews, and surveys.
Analyzing this feedback manually is time-consuming, prone to error, and often fails to extract actionable insights promptly. This
makes the development of AI-based feedback analysis systems an exciting and important research area. Studies from 2021 to 2025
have explored sentiment analysis, text classification, and opinion mining to process customer feedback automatically. Previous
approaches include lexicon-based sentiment models such as VADER and machine learning classifiers like Naïve Bayes, Logistic
Regression, and SVM. More recent studies have demonstrated the superiority of transformer-based models, such as BERT and
DistilBERT, for natural language understanding. However, large-scale transformer models are computationally intensive and difficult
to deploy on low-budget PCs or small-scale businesses. The key problem that remains is providing accurate sentiment analysis and
topic classification without requiring high-end hardware. Our research addresses this gap by proposing an Intelligent Customer
Feedback Analyzer using DistilBERT, a lightweight transformer that retains most of BERT’s accuracy while being faster and more
efficient. The system will automatically classify customer feedback into positive, negative, or neutral sentiments. In addition, it will
extract key topics or themes mentioned by customers, such as service, product quality, or pricing, using keyword extraction
techniques. A user-friendly web interface will allow businesses to upload customer reviews, view sentiment dashboards, and monitor
trends over time. The system will also generate summary reports highlighting major concerns and positive feedback trends for quick
decision-making. DistilBERT’s lightweight architecture allows the system to run efficiently on standard PCs without GPUs, making it
accessible to small and medium enterprises. The combination of sentiment classification and topic extraction ensures that businesses
gain both an overview and detailed insights. Preliminary tests on publicly available review datasets show that DistilBERT achieves
high accuracy while maintaining low memory usage and processing time. By integrating this AI feature into a web-based platform,
businesses can automate their feedback analysis process, reducing manual effort significantly. The proposed system also allows
scalability, as additional datasets can be fed into the model to improve accuracy over time. Overall, this project contributes to the field
of business intelligence by providing a practical, efficient, and accessible tool for customer feedback analysis, leveraging AI
algorithms within a feasible computing environment.

Title of Research
Intelligent Customer Feedback Analyzer Using DistilBERT for Small and Medium Enterprises

Research Area
Artificial Intelligence in Business Intelligence and Natural Language Processing (NLP)

The Storyline

Selection of Algorithm

1 DistilBERT: Lightweight transformer model for sentiment analysis and text classification. Core algorithm for analyzing
customer feedback.

2 Keyword Extraction: Identifies main topics or themes in customer feedback to complement sentiment classification.

3 Optional Baseline Classifiers (Naïve Bayes, Logistic Regression): Used for comparison to validate the performance of
DistilBERT.

Software Application of the algorithm

4 Python: Main programming language to integrate all algorithms and libraries.

5 Hugging Face Transformers: Implements DistilBERT for sentiment analysis..

6 Scikit-learn: For building/testing baseline classifiers and auxiliary tasks.

7 Pandas / NumPy: Data preprocessing and manipulation.

8 Matplotlib / Seaborn: Visualization of sentiment trends, topic frequencies, and summary charts.
Table 2. Compilation of RRL (Matrix). In the research area identified above, list the previous works being used. Add more rows to
the table, if necessary.
Algorithm

Algorithm and Description Strength Weakness

DistilBERT [1] A distilled transformer model for High accuracy (~95% of BERT) May miss subtle nuances in
NLP tasks. Used for sentiment with lower computational extremely complex texts
analysis and text classification. resources; suitable for compared to full BERT.
deployment on standard PCs.

Keyword Extracts main topics from text Provides interpretable insight into Performance depends on text
Extraction [2] using statistical or graph-based dominant themes; lightweight. quality; may miss implicit
methods (TF-IDF, RAKE, context.
TextRank).

System

Reference Description Strength Weakness

Intelligent Customer Web-based system for analyzing Automates feedback Limited by size and
Feedback Analyzer and visualizing customer analysis; lightweight; quality of uploaded
(Proposed, based on feedback. Users upload reviews, scalable; does not require datasets; advanced
studies 2021–2025) [3] receive sentiment scores, topic high-end hardware; contextual
insights, and summary reports. provides actionable understanding may be
insights quickly. limited.