Sentiment Analysis Dashboard

A production-style Sentiment Analysis Dashboard built using pretrained Transformer models and Streamlit. The project demonstrates how to go from raw text data → cleaned dataset → model inference → interactive visual analytics.

This repository is suitable for:

• Machine Learning / NLP internships
• Data Science portfolios
• Demonstrating real-world use of pretrained models

---

Live Demo

• Streamlit App: https://sentimental-dashboard.streamlit.app/
• GitHub Repository: https://github.com/maroofiums/Sentiment-Dashboard
• Kaggle Notebook: https://www.kaggle.com/code/maroofiums/sentiment-analysis/

---

Project Overview

This dashboard allows users to:

• Analyze sentiment of a single text input in real time
• Upload a CSV file and run batch sentiment analysis
• Use demo data for instant visualization
• Explore results through multiple interactive graphs

The model is not trained from scratch. Instead, it uses pretrained Transformer models from Hugging Face, which reflects how NLP systems are commonly built in industry.

---

Tech Stack

• Python
• Streamlit
• Hugging Face Transformers
• PyTorch (backend for transformers)
• Pandas
• Matplotlib

---

Folder Structure

Sentiment-Dashboard/
├── Data
│   ├── Clean
│   │   └── comments.csv
│   └── Raw
│       └── imdb_dataset.csv
├── Frontend
│   ├── main.py
│   └── requirements.txt
├── Notebook
│   ├── Clean_Demo_Data_For_Sentimental_Dashboard_Project.ipynb
│   └── sentiment_analysis.ipynb
└── requirements.txt

---

Features

1. Live Text Sentiment Analysis

• Enter any text

• Instantly see:

  • Predicted sentiment (Positive / Negative)
  • Confidence score

2. Demo Data Mode

• One-click loading of demo reviews
• Useful for quick testing and visualization

3. CSV Upload & Column Selection

• Upload your own dataset
• Select the column that contains text
• Limit number of rows for faster analysis

4. Batch Sentiment Analysis

• Efficient batch inference using pretrained models
• Cached model loading for performance

5. Visual Analytics (5 Graphs)

• Sentiment Count (Bar Chart)
• Sentiment Distribution (Pie Chart)
• Confidence Score Histogram
• Score by Sentiment (Box Plot)
• Sentiment Trend (Line Chart)

Layout is optimized as:

• First row: 3 plots
• Second row: 2 centered plots

---

Dataset Details

• Raw data sourced from IMDB-style reviews
• Cleaned using Pandas in Jupyter notebooks
• Dataset is intentionally noisy and slightly imbalanced to simulate real-world user reviews

This helps make trends, histograms, and distributions more realistic.

---

How to Run Locally

1. Create Conda Environment

conda create -n sentiment-dashboard python=3.10 -y
conda activate sentiment-dashboard

2. Install Dependencies

pip install -r requirements.txt

3. Run Streamlit App

cd Frontend
streamlit run main.py

---

Model Used

• Default Hugging Face sentiment-analysis pipeline
• Transformer-based pretrained model
• No custom training required

This reflects real-world usage where pretrained models are fine-tuned or directly deployed.

---

Limitations

• Binary sentiment only (Positive / Negative)
• Very long texts are truncated for performance
• Confidence score reflects model probability, not human certainty

---

Future Improvements

• Neutral sentiment support
• Time-based sentiment aggregation
• Export analyzed results as CSV
• Replace Matplotlib with Plotly for richer interactivity
• Authentication and user-specific dashboards

---

This project is designed to demonstrate practical NLP skills, clean code structure, and real-world ML workflows rather than just model accuracy.