Deep parallel contextual analysis framework based emotion prediction in community wellness communications on social...

Area of Science:

• Computational Social Science
• Natural Language Processing
• Public Health Informatics

Background:

• Understanding public sentiment on social media is vital for community wellness initiatives and policy.
• Short social media texts present challenges for accurate emotion prediction due to limited length and semantic features.

Purpose of the Study:

• To propose a novel framework, Deep Parallel Contextual Analysis Framework (DPCAF), for enhanced emotion prediction in community wellness social media data.
• To address the limitations of short text length and sparse semantic features in social media analysis.

Main Methods:

• Utilized dual word embedding techniques for comprehensive semantic capture and robust representations.
• Fused word, POS, and locational embeddings within a deep parallel layer (CNNs and BiLSTM).
• Employed an attention mechanism for further semantic feature extraction and integrated representations for final emotion prediction.

Main Results:

• DPCAF demonstrated significant improvements over benchmark models, with increases of 4.81% in Precision, 3.44% in Recall, and 10.81% in F1-score.
• Outperformed state-of-the-art models, achieving minimum improvements of 2.65% in Precision, 3.02% in Recall, and 2.53% in F1-score.
• The framework effectively mines deep contextual semantic information from short social media texts.

Conclusions:

• DPCAF offers a robust and effective approach for emotion prediction in the community wellness domain.
• The proposed framework enhances the analysis of public sentiment on social media, contributing to better health awareness and policy guidance.
• Advanced deep learning techniques, including parallel networks and attention mechanisms, are crucial for improving NLP tasks on social media data.