Improving Sentiment Analysis for Social Media Applications Using an Ensemble Deep Learning Language Model.

Area of Science:

• Natural Language Processing
• Computational Social Science
• Artificial Intelligence

Background:

• The rapid growth of social media data, especially during the coronavirus pandemic, necessitates effective methods for understanding user opinions.
• Traditional feature-based sentiment analysis techniques struggle with the nuances and scale of pandemic-related social media discourse.
• Deep learning models offer enhanced representation capabilities for complex language patterns.

Purpose of the Study:

• To improve sentiment classification performance on pandemic-related social media data.
• To develop a customized deep learning framework incorporating advanced word embedding and a long short-term memory (LSTM) network.
• To create a hybrid ensemble model combining the proposed LSTM classifier with other state-of-the-art sentiment analysis methods.

Main Methods:

• Developed a deep learning model utilizing advanced word embedding for contextual understanding and a long short-term memory (LSTM) network.
• Implemented an ensemble approach combining the baseline LSTM model with established sentiment analysis classifiers.
• Trained and evaluated models on a custom Twitter dataset of coronavirus hashtags and public Amazon/Yelp review datasets.

Main Results:

• The proposed word embedding and LSTM network framework effectively learns contextual word relations and handles rare words by recognizing morphological patterns (suffixes/prefixes).
• The hybrid ensemble model demonstrated superior performance by leveraging the strengths of diverse state-of-the-art sentiment analysis techniques.
• Statistical analysis confirmed that the developed models significantly outperformed existing methods in sentiment classification accuracy.

Conclusions:

• The customized deep learning framework provides a robust approach for sentiment analysis in emerging situations like pandemics.
• Ensemble modeling effectively captures diverse analytical perspectives, leading to enhanced accuracy in sentiment classification.
• The study validates the efficacy of advanced deep learning techniques for analyzing large-scale, dynamic social media data.