Spotting depression signs in social media posts using machine learning

We have built one of the first ever created Spanish text user-level depression classifiers, which was trained with hundreds of Telegram posts and reached high accuracy.

Early depression diagnosis is still one of the most pressing human health challenges. Automated high-throughput analysis of social media posts could be an efficient strategy to address this issue.

Here we describe an innovative system that leverages artificial intelligence and machine learning to spot individuals at risk of suffering depression by analysing hundreds of Telegram posts from thousands of individuals. The work is the result of a collaboration by volunteers from SoGoodData, a Spanish NGO that uses data for social good.

The results were presented in the 2023 edition of IberLEF, a shared evaluation campaign for Natural Language Processing (NLP) systems in Spanish and other Iberian languages, in a paper entitled: TextualTherapists at MentalRiskES-IberLEF2023: Early Detection of Depression using a User-level Feature-based Machine Learning Approach