Sequential Neural Networks for Sentiment Analysis

[ Live Demo] (https://sequentialneuralnetworks-sentimentanalysis.streamlit.app/)

Overview

This project implements Sequential Deep Learning Models such as RNN, LSTM, and GRU to perform Sentiment Analysis on textual data.
The goal is to classify a given text (such as a movie review or news article) as either positive or negative by understanding the contextual meaning of words using neural sequence models.

---

Objectives

• To preprocess and clean textual data effectively for NLP tasks.
• To represent words as numerical vectors for deep learning models.
• To build and compare the performance of RNN, LSTM, and GRU architectures.
• To evaluate model accuracy, training loss, and generalization ability.
• To deploy the best-performing model using an interactive Streamlit web interface.

---

Dataset Description

The dataset consists of two categories of text documents — real and fake news articles (or positive/negative reviews).
Each record contains the following fields:

Column	Description
title	Headline or title of the article/review
text	Main content of the text
subject	Category or topic
date	Published date
label	Target variable (1 = real/positive, 0 = fake/negative)

Initially, both fake and real datasets were combined into a single DataFrame, followed by data cleaning and text preprocessing.

---

Methodology

1. Data Preprocessing

Steps performed:

• Removal of punctuation, special characters, and stopwords
• Conversion to lowercase
• Tokenization and lemmatization
• Sequence padding and truncation to a fixed length
• Train–test split for model evaluation

This ensures that all text inputs are of uniform length and numerical form before feeding into sequential models.

---

2. Text Vectorization

Since neural networks cannot process raw text directly, words were converted into numeric sequences using:

• Tokenizer (Keras): Converts text to integer sequences
• Padding: Ensures equal input length using pad_sequences
• Embedding Layer: Maps each word index to a dense vector representation capturing semantic meaning

---

3. Model Architectures

(a) Recurrent Neural Network (RNN)

• Sequentially processes words, capturing limited contextual information.
• Struggles with long-term dependencies due to vanishing gradients.

(b) Long Short-Term Memory (LSTM)

• Overcomes RNN’s vanishing gradient problem with memory cells and gates.
• Efficient in remembering long-term dependencies and context flow.

(c) Gated Recurrent Unit (GRU)

• A simplified version of LSTM with fewer parameters.
• Faster to train while achieving comparable performance.

---

4. Model Training

All models were compiled using:

loss = 'binary_crossentropy'
---
optimizer = 'adam'
metrics = ['accuracy']