KEAHT: A Knowledge-Enriched Attention-Based Hybrid Transformer Model for Social Sentiment Analysis

Area of Science:

• Natural Language Processing (NLP)
• Artificial Intelligence (AI)
• Computational Social Science

Background:

• Social media platforms are crucial for public opinion on global events.
• Sentiment analysis of unstructured social media data presents challenges.
• Existing Deep Neural Network (DNN) models have limitations in feature handling and accuracy.

Purpose of the Study:

• To propose an advanced AI model for sentiment analysis of complex social issues.
• To address limitations in existing sentiment analysis models, including sequential training and feature importance.
• To provide a robust tool for understanding public opinion during crises like the COVID-19 pandemic and farmer protests.

Main Methods:

• Developed a Knowledge-Enriched Attention-based Hybrid Transformer (KEAHT) model.
• Integrated Latent Dirichlet Allocation (LDA) topic modeling and lexicalized domain ontology for knowledge enrichment.
• Utilized a pre-trained Bidirectional Encoder Representation from Transformer (BERT) for efficient training and attention mechanisms.

Main Results:

• The KEAHT model demonstrated superior performance compared to existing baseline and hybrid models.
• The model accurately handles complex text problems and improves polarity scoring and topic modeling.
• Comparative studies affirmed the credibility and effectiveness of the KEAHT model in NLP applications.

Conclusions:

• The proposed KEAHT model offers a significant advancement in AI-driven sentiment analysis.
• This AI approach is vital for navigating public discourse during global pandemics and political disputes.
• New benchmark datasets for COVID-19 and farmer protest tweets were created to support future research.