JMIR Publications

ABSTRACT

Text generation approaches in healthcare communication have evolved along two major paths. The first involves Generative Adversarial Networks (GANs), progressing from basic architectures to specialized variants like TT-GAN and TFGAN, which address challenges in discrete text generation through techniques such as Gumbel-Softmax and Reinforcement Learning. The second path emerges from transformer-based architectures, particularly GPT-2, which uses extensive pre-training and self-attention mechanisms to generate contextually appropriate text. GPT-2's transformer architecture enhances persuasive health communication by generating personalized messages using various strategies like task support, dialogue support, and social support for effective health interventions. The objective of this experiment is to use GPT2 as a generative method to construct the persuasive text in the dataset and compare the performance of Semantic Analysis and Emotion Detection Analysis. We combine sentiment analysis tools (VADER and TextBlob) with emotion detection methods (Text2Emotion and NRCLex) to analyze health coaching messages across different persuasive types: reminders, rewards, suggestions, and praise. TextBlob achieved perfect scores (1.0) in accuracy, precision, recall, and F1-score, outperforming VADER's moderate performance (0.6 accuracy). RoBERTa-sentiment leads with 88% accuracy and 0.8264 F1 score. While transformers excel in accuracy, lexicon-based models like VADER offer better performance-efficiency balance for real-time health communication systems. For emotion detection, anger showed nearly perfect performance across all metrics, while anticipation scored lowest with metrics below 0.5. The emotion detection analysis revealed varying success rates across different emotions, with some categories, like anger, showing near-perfect detection while others, like anticipation, proved more challenging. This research contributes to understanding the emotional dynamics of persuasive health communication and highlights both the capabilities and limitations of current natural language processing tools in analyzing health-related persuasive messaging.