This study explores the application of deep learning transformer models, including LSTM, Bi-LSTM, GPT2-LSTM, BERT, and XLNet, integrated with Explainable AI (XAI), to detect hedonic emotional states by analyzing eudaimonic behaviors of online users. Hedonic emotional states, characterized by pleasure-seeking and immediate gratification, are inferred from eudaimonic behaviors, which reflect purpose-driven and meaningful online interactions. The proposed methodology leverages the sequential processing capabilities of LSTM and Bi-LSTM for temporal pattern recognition, while GPT2-LSTM enhances contextual understanding of user-generated text. BERT and XLNet, with their bidirectional and permutation-based architectures, capture nuanced semantic relationships in online content. XAI techniques, such as SHAP and LIME, are employed to provide interpretability, ensuring transparent model decisions. The models are trained on diverse datasets from social media platforms, evaluating their performance in classifying emotional states. Results demonstrate high accuracy and interpretability, offering insights into user well-being and behavior. This approach advances emotion detection for mental health monitoring and personalized digital interventions.

Keywords: Deep Learning, Transformers, LSTM, Bi-LSTM, GPT2-LSTM, BERT, XLNet, XAI, Hedonic Emotions, Eudaimonic Behavior, Emotion Detection, Social Media Analysis, Explainable AI, Mental Health Monitoring.