As part of a new study, an artificial intelligence system was developed, which locates texts on social networks that are suspected to have been written by anorexic girls. At the end of the study, the system achieved close to 91% success.
The death of the model Karin Bowman from anorexia has raised the need for early detection of women and men suffering from eating disorders, in order to provide them with the necessary treatment and support as soon as possible.
This is the goal that was the basis of the research conducted by Rabbi Professor Yaakov HaCohen-Kerner, Dean of the Faculty of Engineering and Computer Science at the Lev Academic Center.
Prof. HaCohen-Kerner turned to Prof. Eitan Bacher from the Department of Psychology at the Hebrew University, the former chief psychologist of the two Hadassah hospitals in Jerusalem and a world-renowned expert in the field of anorexia, to accompany the research.
As part of the research, 200 posts were collected, 100 of which were collected from groups and subnets known as platforms used by girls in their teens and early 20s, who suffer from the disorder. For the purpose of criticism, 100 other posts (of a similar length to the first group of posts) were selected, including challenging posts, such as posts by girls who deal intensively with issues of weight, nutrition and sports, but are not anorexic.
Based on the use of Prof. Bacher's set of words, the system was able to achieve about 76% success, and from there they continued to define and integrate additional feature families.
At the end of the process, the system achieved close to 91% success using 10 characteristic families, while Prof. Bacher defined a target of 80% as success. "The system actually determines whether there is a risk that the author of the post suffers from anorexia, and warns of this in a way that can allow professionals to continue to determine whether this is indeed true from a clinical point of view, and help her," concludes Prof. HaCohen-Kerner.