Sentiment Analysis Using BERT

This repository contains a sentiment analysis project where a fine-tuned BERT model is used to classify text data into positive or negative sentiments. The model has been deployed as a Flask web application for real-time predictions. This architecture achieved an accuracy of 90.3 percentile.

Table of Contents

• Introduction
• Features
• Dataset
• Installation
• Usage
• Sample Predictions
• Future Work

---

Introduction

Sentiment analysis is a key natural language processing (NLP) task that helps in understanding emotions conveyed in text data. In this project, I have fine-tuned a BERT model to achieve high accuracy in sentiment prediction. The model is served using a Flask backend and can process real-time user inputs via a simple web interface.

---

Features

• Real-Time Sentiment Prediction: Enter a sentence and instantly receive the sentiment score.
• Flask Deployment: A lightweight and easy-to-deploy backend framework.
• Interactive Web Interface: Built for ease of use, allowing users to input text and view predictions directly in the browser.

---

Dataset

The project supports any labeled dataset with columns for text and label. We can make use of any datasets mentioned below on this model:

• IMDb Reviews
• Twitter Sentiment140
• Custom datasets

Ensure the dataset is in a .csv or .tsv format before training and try to do changes in the data format and entities.

---

Installation

1. Clone this repository:

   git clone https://github.com/your-username/sentiment-analysis-bert-flask.git  
   cd sentiment-analysis-bert-flask  

2. Create and activate a virtual environment or install miniconda (https://docs.anaconda.com/miniconda/install/) and create a conda environment and install the dependencies:

   python3 -m venv env  
   source env/bin/activate  # For Linux/MacOS  
   .\env\Scripts\activate  # For Windows  

   conda init
   conda create -n <name> python==<version>
   conda activate <name>

3. Install dependencies:

   pip install -r requirements.txt  

4. Download a pre-trained BERT model for finetuning from huggingface, you can choose whichever model you want to try. But I used bert-base-uncased from huggingface to finetune the model.

---

Usage

1. Prepare Dataset: Create a folder named input and place your dataset in that. And do the necessary changes in dataset and preprocess file for training phase.

2. Train the Model: You can change the parameters and path details which are mentioned in config file however you want.Run the script to fine-tune the BERT model:

   python train.py  

3. Start Flask App: After training the model. Model file model.bin will be generated in the default folder path. Next, Launch the web application:

   python app.py  

   The app will run locally at http://127.0.0.1:5000.

4. Access the Web Interface: Open your browser and input text for prediction.

---

Sample Predictions

Input Screen

Output Screen

---

Future Work

• To Enhance this web application into a fully functional end-to-end pipeline for production use, including deployment on cloud platforms, monitoring, and scalability improvements.

---