Researchers from the University of São Paulo (USP) in Brazil are hoping to create artificial intelligence (AI) and Twitter-based anxiety and depression prediction models. The models would be used to detect signs of such illnesses before clinical diagnosis. This research was recently published in the journal Language Resources and Evaluation.
The first step of this research was to create a database for the corpus of 47 million publicly posted Portuguese texts and the network of connection between 3,900 Twitter users. They had previously been diagnosed with or treated for mental health problems before the survey. The researchers then used two different datasets: one with users who reported being diagnosed with a mental health problem and the other, a random selection for control purposes.
The study also collected tweets from friends and followers as those with mental health problems had a tendency to follow certain accounts such as discussion forums, influencers, and celebrities who acknowledged their depression. The research team utilized deep learning (AI) and created four text classifiers and word embeddings using models based on bidirectional encoder representations from transformers (BERT). The training input was a sample of 200 tweets from each user.
The team discovered that the BERT model worked best in terms of predicting depression and anxiety. This algorithm studies contextual and semantic relations between words in a sentence, helping them understand what a person with depression likes to discuss, as it usually revolves around themselves and topics such as crisis, death and psychology.
If further research proves successful, the framework can be used to screen prospective sufferers from mental health problems and help families and friends of young people at risk from depression and anxiety.
Deloitte, a multinational professional services network, conducted a survey which highlighted the poor mental health of employees costing Indian employers $14 billion yearly. Deloitte provides consulting on various topics such as audit, financial advisory, tax, enterprise risk and consulting services.
Ivandre Paraboni, professor at USP and one of the authors of the article, said that the signs of depression that can be detected during a visit to the doctor are not necessarily the same as the ones that appear on social media. Usage of the first-person singular pronouns such as ‘I’ and ‘me’ along with topics such as death, crisis and psychology are some of the classic signs of depression. Similarly, the frequent use of heart emoji by depressive users is also identified as a symbol of affection and love.
The research team is developing the database, refining their methods, and improving the models. This is a great step to awareness about mental wellness as World Teen Mental Wellness Day 2023 approaches. It raises attention for the mental wellness of teenagers and in the near future, mental health models and AI can be utilized to create a potential diagnosis at an early stage.
