IEEE Access (Jan 2025)
Using Deep Learning Transformers for Detection of Hedonic Emotional States by Analyzing Eudaimonic Behavior of Online Users
Abstract
Hedonic emotions represent a significant concept in the study of linguistics, encapsulates the patterns of positivity, pleasures, activities, and enjoyment. These emotions play a critical role in shaping eudaimonic behavior in that users reflect the personal growth of individual, personality, social connections, and values, leading to positive psychology. This behavior refers to taking actions and practices that contribute to a deep sense of purpose, meaning, and personal growth in life. The significance of detecting these emotions extends beyond psychological well-being, impacting fields such as marketing, mental health, and education. The area of hedonic emotion remains underexplored. In this study, for classification of hedonic emotions, the main aim to carry out an extensive empirical analysis by applying computational models of shallow machine learning, ensemble learning, and advanced Deep Learning (DL) algorithms. Moreover, diverse feature extraction techniques including textual features such as Term Frequency-Inverse Document Frequency (TF-IDF), Part-of-Speech (PoS) and word embeddings such as word2vec, GloVe will be explored. In addition, the state-of-the-art sentence embeddings have also been computed. The experimentation has been carried out on a standard dataset by analyzing the patterns of positive emotions and activities to understand the user’s behavior in the digital world. This empirical analysis of results measured with standard performance measures reveals the highest accuracy of 97% with transformer-based model RoBERTa using advanced embedding approach, to effectively detect and identify the hedonic emotion patterns from online textual content. By bridging the existing research gap, our research contributes to a more comprehensive understanding of emotional dynamics in digital interactions.
Keywords
