New model uses artificial intelligence and social media to predict depression and anxiety

AI News

summary: Using Twitter data and applying natural language processing artificial intelligence algorithms, researchers have created new and accurate predictive models for depression and anxiety.

sauce: USPs

Researchers at the University of São Paulo (USP) in Brazil are using artificial intelligence (AI) and Twitter, one of the world’s largest social media platforms, to create predictive models for anxiety disorders and depression. clinical diagnosis.

The study is reported in an article published in the journal Language resources and assessment.

Building a database called SetembroBR was the first step in the research. The name is a reference to the annual suicide awareness and prevention campaign “Yellow September” and the fact that data collection for the study began one day in September.

The second step is still in progress, but the potential to detect if a person is likely to develop depression based solely on their social media friends and followers, without considering their own posts, etc. , provides some preliminary findings.

The database compiled by this group contains a corpus of text (in Portuguese) and information on a network of connections that includes 3,900 Twitter users who reported having been diagnosed or treated for mental health problems prior to the survey. I’m here. The corpus contains all public tweets (no retweets) posted individually by these users, totaling about 47 million short texts.

“First, we manually collected timelines and analyzed tweets by approximately 19,000 users, representing the population of a village or small town. For users who reported having been diagnosed with a problem, and a random selection for control purposes, the last author of this article and a student at USP’s School of Arts, Sciences and Humanities (EACH). Professor Ivandre Paraboni said:

The study also collected tweets from friends and followers. This is based on observations that people with mental health issues tend to follow specific accounts such as discussion forums, influencers and celebrities who publicly admit to depression.

“These people are attracted to each other. increase.

FAPESP also supported project research through the project “Social Media Language Analysis for Early Detection of Mental Health Disorders” led by Paraboni.

Mental health disorders such as depression and anxiety are a global concern. The World Health Organization (WHO) estimated that her 3.8% of the global population, or about 280 million people, were affected by depression, based on 2021 data.

WHO also estimates that the global prevalence of these mental health problems increased by 25% during the COVID-19 pandemic. Tweets were collected for research during this period.

In a recent survey of 784,000 participants by the Brazilian Ministry of Health, 11.3% said they had been diagnosed with depression. Most were women.

Previous research has shown that mental health issues are often reflected in the language used by patients. This discovery has led to a significant body of research on natural language processing (NLP), with particular focus on depression, anxiety, and bipolar disorder. However, most of these studies analyzed English texts and do not always match most Brazilian profiles.


The researchers preprocessed the corpus to remove hashtags, URLs, emoticons, and non-standard characters while preserving the original text.

It then deploys deep learning, an AI technique that teaches computers to process data in ways inspired by the human brain, using models based on bidirectional encoder representations to generate four text classifiers and word embeddings. (a context-sensitive mathematical representation of the relationships between words). From Transformers (BERT), an NLP machine learning algorithm.

These models support neural networks that learn context and meaning by observing relationships in sequential data, such as words in sentences.

The training input consisted of a sample of 200 randomly selected tweets from each user. The parameters were defined by running her cross-validation on the training data five times and calculating the average result.

This is a picture of a computer and a brain
These models support neural networks that learn context and meaning by observing relationships in sequential data, such as words in sentences.Image is in public domain

In conclusion, BERT was the best predictor of depression and anxiety, with statistically significant differences between BERT and the next best option, LogReg. Because the model analyzed sequences of words and complete sentences, people with depression, for example, tended to write about subjects that were relevant to them, using verbs and phrases in the first person, as well as topics such as death. I was able to observe that Crisis and psychology.

“The signs of depression that can be detected during a doctor’s visit aren’t always the same as those that appear on social media,” Paraboni said.

“For example, the use of the first person singular pronouns I and me is very obvious and in psychology this is considered a classic sign of depression. It was also observed to be a symbol of affection and love, although psychologists may not yet characterize it as such.

All collected texts were anonymized. “Neither the actual tweets nor the names of the users were published. To protect the identities of people, we were careful to ensure that user data could not be accessed by students involved in the project,” he said.

Researchers are now expanding databases, improving computational techniques, and upgrading models to screen prospective patients for mental health problems, as well as the families and families of young people at risk of depression and depression. I’m looking to see if I can create a tool for future use to help a friend.

According to a Comscore study released in early March, Brazil ranks third among the world’s most social media consuming countries, behind India and Indonesia but ahead of the United States, Mexico and Argentina. 131.5 million users are online an average of 46 hours per month. The most widely used platforms are YouTube, Facebook, Instagram, TikTok, Kwai and Twitter. They recently changed their rules and started charging for certain services.

About this AI and psychology research news

author: Eloisa Reinert
sauce: USPs
contact: Heloisa Reinert – USP
image: image is public domain

Original research: closed access.
“SetembroBR: A ​​Social Media Corpus for Prediction of Depression and Anxiety Disorders.” Ivandre Paraboni et al. Language resources and assessment


SetembroBR: A ​​Social Media Corpus for Prediction of Depression and Anxiety Disorders

In the current work, we present a new dataset (here called the SetembroBR corpus) for researching and developing predictive models for depression and anxiety disorders in Portuguese based on pre-diagnostic information.

This corpus consists of text and network-related information related to 3900 Twitter users who self-reported their diagnosis or treatment of a mental disorder. social media data.

The current results are intended as a first step to explore how mental health conditions are represented on social media in the Portuguese-speaking world, in support of a pressing issue of great public concern. paves the way for computational applications aimed at

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *