France has one of the highest suicide rates in Europe, with approximately 9,000 deaths per year. In Spain, the available statistics show that almost 3 000 people commit suicide every year. The Covid-19 pandemic, the increase in feelings of loneliness and malaise, unfortunately make us fear that these figures will increase. The STOP project (Suicide prevenTion in sOcial Platforms), led by researchers at the Pompeu Fabra University (UPF) in Barcelona, aims to research and analyse patterns of behaviour leading to suicide.

Twitter: a huge database to analyse

Thanks to the collaboration of the Computer Vision Centre of the Autonomous University of Barcelona (UAB) and the Parc Taulí de Sabadell Hospital, these analyses are carried out on Twitter and use artificial intelligence processes. The algorithms designed by the researchers detect these patterns of suicidal behavior with an accuracy of 85%.

The World Health Organization (WHO) has calculated that each suicide has an emotional impact on at least six people around the victim. The STOP project, led by Ana Freire, a researcher at the UPF’s Department of Information and Communication Technologies, has been working on this subject and has recently published some initial results.

According to the researchers, about 8,000 tweets are published on Twitter every second. They contain very valuable information in various fields, but also to analyze mental health problems that could lead to suicide. The observation is twofold: the fact that suicide is a taboo subject and that there is difficulty in accessing psychological consultations means that they can only very rarely benefit from a real diagnosis without mentioning a possible adequate treatment. The analysis of an individual’s social media posts can therefore be key to detecting problems such as depression or eating disorders that can generate, in some cases, suicidal thoughts.

Algorithms defining “high risk” subjects

To perform these advanced analyses of Twitter posts, the researchers trained algorithms to distinguish different patterns. These patterns describe whether the risk of suicide is high or low. The data is labeled by mental health experts and the researchers specify that it is completely anonymous in order to respect the data protection and privacy of users.

The study revealed various characteristics linked to a “high risk of suicide pattern” and a “no risk pattern”. One of these characteristics highlights the strong tendency to talk about the first group as opposed to the second and to use terms related to feelings. These actions are all associated with the anxiety that an individual may feel.

Very specific characteristics, identified thanks to AI

In order to develop these patterns, the researchers have the algorithms explore the images posted, analyze the interactions between users, check the times of connection and publication of Internet users. Jordi González, a researcher at the UAB’s computer vision centre, announced that “there could be a certain correlation between the content of images shared on social networks and the mental health of the user who shares them”.

Ricardo Baeza-Yates, researcher for UPF, highlighted the importance of AI and the algorithms used in the STOP project. According to him, they make it possible to “find new factors in social networks from the use of digital media that can help in an effective diagnosis and contribute to making suicide a less taboo subject in our society”.

Translated from Projet STOP : comment l’analyse de publications sur Twitter peut prévenir les risques de suicide