Text Classification with TensorFlow

Overview

This project implements a binary sentiment classification model using TensorFlow and TensorFlow Hub.

The model is trained on the IMDB movie reviews dataset and predicts whether a review is positive or negative.

Natural Language Processing (NLP) techniques are used to convert text into embeddings before feeding them into a neural network classifier.

---

Dataset

The project uses the IMDB Reviews dataset provided by TensorFlow Datasets.

Dataset characteristics:

• 50,000 movie reviews
• Binary sentiment labels
• Balanced dataset

The dataset is automatically downloaded using tensorflow_datasets.

---

Model Architecture

The model consists of:

1. Input Layer

   • Accepts raw text input

2. Embedding Layer

   • Pretrained text embedding from TensorFlow Hub
   • Converts text into numerical vectors

3. Dense Hidden Layer

   • 16 neurons with ReLU activation

4. Output Layer

   • Sigmoid activation for binary classification

---

Technologies Used

• Python
• TensorFlow
• TensorFlow Hub
• TensorFlow Datasets
• NumPy

---

Installation

Create a virtual environment:

python -m venv .venv

Activate it:

Windows:

.venv\Scripts\activate

Install dependencies:

pip install tensorflow tensorflow-hub tensorflow-datasets

---

Running the Project

Run the training script:

python main.py

The script will:

1. Download the IMDB dataset
2. Preprocess the text data
3. Train a neural network classifier
4. Evaluate model performance

---

Example Training Output

The model will output metrics such as:

Epoch 1/5
accuracy: 0.84
val_accuracy: 0.86

---

Possible Improvements

Future improvements could include:

• Using advanced models such as BERT
• Adding dropout layers for regularization
• Hyperparameter tuning
• Saving and exporting the trained model
• Deploying the model as an API

---

Author

Arul Gupta BITS Goa – Electronics and Communication Engineering