Quantifying the Intensity of Online Social Support via Large Language Model-Based Evidence Extraction: Development...

Area of Science:

• Computational Linguistics
• Artificial Intelligence
• Social Computing

Background:

• Online social support is crucial for mental health, yet its intensity is understudied.
• Existing models lack comprehensive feature extraction and explainability.
• Prior research focused on support types rather than strength.

Purpose of the Study:

• To develop a deep learning model for predicting online social support intensity.
• To provide human-readable explanations for the model's predictions.
• To quantify the strength of informational and emotional support.

Main Methods:

• A deep learning model was developed, collaborating with a large language model (LLM).
• The model extracts key sentences, computes sentiment scores, and encodes features for intensity classification.
• Evaluated on two human-annotated datasets.

Main Results:

• The model achieved superior accuracy in predicting informational and emotional support intensity compared to baselines.
• Extracted sentences from LLMs served as evidence, explaining model decisions.
• Demonstrated ability to differentiate between strong (explicit intent) and weak (indirect support) support.

Conclusions:

• Quantifying online social support intensity is feasible using LLMs.
• Extracted textual evidence enhances model transparency and interpretability.
• The model offers a novel approach to understanding online social dynamics and mental health support.