Co-Bi-LSTM: Opinion Mining on Twitter Data Using Convolutional Neural Network with Optimized Bidirectional LSTM Model

Abstract

The proliferation of social networking platforms has generated a substantial volume of user-generated content, posing significant challenges for text classification due to its diverse nature. Sentiment analysis or opinion mining, is crucial for extracting insights from user opinions and emotions regarding various entities and events. This research classifies tweets into positive and negative sentiments using Twitter data for predictive analysis in domains such as consumer behaviour and election outcomes. Two Kaggle datasets like Sentiment140 and News Headlines Dataset for Sarcasm Detection are used. The pre-processing phase includes cleaning, tokenization, and padding. Word embedding like skip-gram can capture semantic relationships and are used in neural network architectures for word2vec conversion. This paper proposes a hybrid model called Convolutional Optimized Bidirectional LSTM (CO-Bi-LSTM), combining Convolutional Neural Networks (CNN) with an Optimized Bidirectional Long Short-Term Memory (O-Bi-LSTM) network, enhanced by the Hybrid Hippopotamus based Zebra Optimization Algorithm (HH-ZOA). The model's performance is evaluated using metrics such as accuracy, F-measure, and precision, demonstrating its efficacy in sentiment analysis of Twitter data.