# Social support detection from social media texts

## TL;DR

This paper introduces a new NLP task to detect social support in online comments, focusing on identifying support expressions, their targets, and specific groups.

## Contribution

The novel contribution is defining and evaluating the Social Support Detection task with three subtasks and a manually annotated dataset.

## Key findings

- Group-oriented support is more prevalent than individual support in online discourse.
- Combining psycholinguistic and affective features with unigrams improves classification performance.
- Best macro F1-scores achieved range from 0.72 to 0.82 across subtasks.

## Abstract

Social support, conveyed through a multitude of interactions and platforms such as social media, plays a pivotal role in fostering a sense of belonging, aiding resilience in the face of challenges, and enhancing overall well-being. This paper introduces Social Support Detection (SSD) as a Natural Language Processing (NLP) task aimed at identifying supportive interactions within online communities. We define SSD through three subtasks: (1) binary classification of whether a comment expresses social support or not social support, (2) binary classification of the intended support target (individual or group), and (3) multiclass classification of the specific group being supported, including Nation, Other, LGBTQ, Black Community, Religion, and Women. We conducted experiments on a manually annotated dataset of 9,998 YouTube comments. Traditional machine learning models were employed using various combinations of linguistic, psycholinguistic, emotional, and sentiment-based features. Additionally, neural network-based models incorporating word embeddings were evaluated to enhance performance across the subtasks. The results indicate a prevalence of group-oriented support in online discourse, highlighting broader societal dynamics. The findings show that integrating psycholinguistic and affective features with unigram representations improves classification performance. The best macro F1-scores achieved across the subtasks range from 0.72 to 0.82.