Cordyceps@LT-EDI: Depression Detection with Reddit and Self-training
This work addresses the problem of identifying undiagnosed depression in social media users, but it is incremental as it applies an existing semi-supervised method to a specific dataset and task.
The paper tackled depression severity detection from social media posts by proposing a semi-supervised learning system that uses self-training to classify Reddit posts into severe, moderate, or low depression levels, achieving 3rd place in the LT-EDI@RANLP 2023 shared task.
Depression is debilitating, and not uncommon. Indeed, studies of excessive social media users show correlations with depression, ADHD, and other mental health concerns. Given that there is a large number of people with excessive social media usage, then there is a significant population of potentially undiagnosed users and posts that they create. In this paper, we propose a depression severity detection system using a semi-supervised learning technique to predict if a post is from a user who is experiencing severe, moderate, or low (non-diagnostic) levels of depression. Namely, we use a trained model to classify a large number of unlabelled social media posts from Reddit, then use these generated labels to train a more powerful classifier. We demonstrate our framework on Detecting Signs of Depression from Social Media Text - LT-EDI@RANLP 2023 shared task, where our framework ranks 3rd overall.