Sentiment Analysis Using Recurrent Neural Networks

Introduction

• 👨‍💻This project implements a sentiment analysis model for movie reviews using RNNs built with Python's libraries.

• 🔭 Sentiment analysis, also known as opinion mining, is a subfield of Natural Language Processing (NLP) that aims to automatically detect the emotional tone of a piece of text.

• ⚡This project demonstrates proficiency in deep learning, data manipulation, text preprocessing, NLP, model building, evaluation and handling sequential data.

Requirements

• Python 3.x
• pandas
• numpy
• Regular expressions (re)
• NLTK
• Scikit-learn
• Tensorflow
• Scipy

Text Preprocessing

• After loading the data, perform preprocessing techniques like HTML tag removal, punctuation removal, lowercasing, stop word removal using libraries nltk, re and spacy.

• Make Tokens for the preprocessed text using Tokenizer. The Tokenizer class from Keras tokenizes the text, converting words into sequences of integer indices.

• Sequences are padded with zeros to ensure a fixed length for all reviews using pad_sequences.

• Sentiment labels ("positive" and "negative") are converted to numerical labels (0 and 1) using LabelEncoder.

Model Building and Training

• Complex Bidirectional LSTM-GRU has been used in model building. This is a more complex model with multiple stacked Bidirectional LSTM and GRU layers with dropout regularization. Total trainable parameters were 37379651.

• The data is split into training and testing sets using train_test_split.

• Each model is compiled with the Adam optimizer, binary cross-entropy loss function, and accuracy metric.

• Models are trained for a specified number of epochs and batch size.

• Training and validation accuracy and loss are monitored during training.

Working

• The code demonstrates how to use the trained model to predict the sentiment of new movie reviews.

• Text pre-processing steps are applied to the new review text given as input.

• The preprocessed text is converted to a sequence of integers using the fitted tokenizer.

• The sequence is padded to match the maximum review length.

• The model predicts the sentiment probability (positive or negative) for the review.

• Based on the predicted probability, the review is classified as "flop movie", "average movie", "good movie", or "blockbuster movie".

• Overall this this code provides a basic framework for sentiment analysis using RNNs. You can experiment with different RNN architectures, hyperparameters.

Use Cases

• Analyze customer reviews and social media posts to understand customer satisfaction with products, services, and brand interactions.
• Identify spam emails by analyzing the language and sentiment used in the message content.
• Track brand sentiment across social media platforms to identify potential issues or crises early on.
• Analyze employee satisfaction surveys and feedback forms to identify potential morale issues or areas for improvement in the workplace.
• Analyze student feedback on online courses to understand their learning experience and identify areas for improvement in course content and delivery.