Deep Sentiment Analysis of Twitter Data Using a Hybrid Ghost Convolution Neural Network Model

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Social Science

Background:

• Sentiment analysis of public opinion on platforms like Twitter and Facebook faces challenges with complex training data.
• Traditional sentiment analysis models often exhibit limitations in accuracy and trustworthiness.
• Deep learning algorithms demonstrate significant potential for enhancing sentiment analysis tasks.

Purpose of the Study:

• To assess the dependability of various hybrid sentiment analysis approaches across diverse datasets.
• To compare the performance of hybrid models against single models in sentiment analysis.
• To evaluate the accuracy and computational efficiency of integrated deep learning and Support Vector Machine (SVM) models.

Main Methods:

• Development of hybrid models by integrating Support Vector Machine (SVM), Long Short-Term Memory (LSTM), and Convolutional Neural Network (CNN).
• Evaluation of model dependability and computation time on multiple text datasets, including tweets and reviews.
• Comparison of hybrid models against individual models, focusing on deep learning and SVM integration.

Main Results:

• Hybrid models, particularly those combining deep learning with SVM, consistently outperformed single models across all tested datasets.
• The proposed deep learning-SVM hybrid model achieved high accuracy, reaching 91.3% and 91.5% on specific datasets.
• While LSTM offered higher results, it required longer processing times; CNN demanded less hyperparameter tuning.

Conclusions:

• Hybrid deep learning-SVM models represent a significant advancement in sentiment analysis accuracy.
• The integration of LSTM, CNN, and SVM offers a robust framework for sophisticated sentiment analysis.
• These hybrid approaches provide a more dependable solution for analyzing public opinion from social media data.