Breaking the silence: leveraging social interaction data to identify high-risk suicide users online using network...

Area of Science:

• Digital mental health
• Computational social science
• Network science

Background:

• Suicidal thought and behavior (STB) is a significant public health concern, often stigmatized and difficult to detect.
• Online environments, despite censorship, offer unique data for understanding STB risk.
• Digital markers derived from social interactions may improve STB risk detection.

Purpose of the Study:

• To develop and validate a machine learning model for predicting highest risk users (HRUs) within an online suicide forum.
• To identify key egocentric network features indicative of heightened suicide risk.
• To explore the utility of social interaction dynamics as digital markers for STB.

Main Methods:

• Collected network data from 192 individuals on an online pro-choice suicide forum, encompassing over 3.2 million interactions.
• Engineered 17 egocentric network features to quantify social interaction and engagement dynamics.
• Trained, validated, and tested a machine learning classifier to predict HRU status, analyzing feature importance.

Main Results:

• The classification model achieved a test AUC of 0.73, indicating effective prediction of HRU status.
• Key predictive features included transitivity, density, and in-degree centrality.
• Predicted HRUs exhibited distinct network properties, including less frequent engagement and 'small world' network structures.

Conclusions:

• Network-based socio-behavioral patterns in online interactions can serve as indicators of heightened suicide risk.
• This study demonstrates the potential of analyzing social dynamics in uncensored online communities for STB research.
• Findings support the integration of network features into future STB descriptive, predictive, and preventative strategies.