JMIR Publications

ABSTRACT

Background:

Online social media provides common people with a platform to express their emotions conveniently and anonymously. There have been nearly two million messages in a particular Chinese social media data source with several thousand more each day, which becomes impossible to analyze manually. However, it has been identified as an important data source for suicide prevention related to depression disorder.

Objective:

We propose in this paper a distant supervision approach to achieve a system that can automatically identify textual comments at high suicide risk.

Methods:

To avoid expensive manual data annotations, we first explore a knowledge graph method to produce approximate annotations for distant supervision, based on which a deep learning architecture is built and refined by the interactions with psychology experts. There are three annotation levels: free annotations (zero cost), easy annotations (by psychology students), and hard annotations (by psychology experts).

Results:

Our system is evaluated accordingly and shown that the performance at each level is promising. Combined with several significant psychology features from user blogs, we obtained 80.75% Precision, 75.41% Recall and 77.98% F1 score for the hardest test data.

Conclusions:

In this paper, we proposed a distant supervision approach to achieve an automatic system that can classify high/low suicide risk of social media comments. The model can therefore provide volunteers with early warnings to prevent social media users from committing suicide.