Code RED: Reactive Emotion Difference for Stress Detection on Social Media.

EasyChair Preprint 10659

Abstract

The prevalence of stress-related issues has become increasingly evident in recent years. Many of the population use social media platforms as an outlet to talk about life situations and express stress. Furthermore, detecting stress has become a critical matter due to the impact of stress on our daily lives, and the consequences of not detecting stress early on can lead to severe physical and mental health complications. Therefore social media stress detection is an emerging field that leverages machine learning and deep learning techniques to identify stress indicators in social media posts. While most of the works in this field focus on analyzing the posts’ textual contents, many ignore the social support cues that could aid the stress detection process. However, this study proposes a stress detection method by leveraging the emotional content of posts, social support or comments inspired by multidisciplinary (social science) theories. We build a classifier based on the emotional difference between the initial post and the responses or reactions it receives. We utilize a state-of-the-art transformer-based emotion classifier and two publicly available datasets. Our approach achieves a better stress classification by incorporating social support emotions. Our main contributions are the novelty and utility of the new approach to stress detection on social media and the expansion of the datasets to include social support. This study showcases and proves the validity of using social support to detect stress on social media.

Keyphrases: Emotion Analysis, Natural Language Processing, mental health, social media, social support, stress detection

@booklet{EasyChair:10659,
  author    = {Zeyad Alghamdi and Faisal Alatawi and Mansooreh Karami and Tharindu Kumarage and Ahmadreza Mosallanezhad and Huan Liu},
  title     = {Code RED: Reactive Emotion Difference for Stress Detection on Social Media.},
  howpublished = {EasyChair Preprint 10659},
  year      = {EasyChair, 2023}}