Tasks
All tasks consist of the detection of mental disorders in users based on their comments posted on Telegram. Given a history of messages about a user, the goal is to identify whether the user suffers from the disorder or not, and his/her attitude to it: against (intends to leave) or in favour of it (intends to continue).
Task 1. Eating disorders detection
1.a. Binary classification
Detect if the user suffers from anorexia or bulimia. Labels will be 0 for “control” (negative, the user does not suffer from eating disorder) or 1 for “suffer” (positive). Let’s see some examples:
suffer: Eating disorders are recognized by a persistent pattern of unhealthy eating or unhealthy dieting. It is an inappropriate eating behaviour and an obsession with weight control. A user is considered to be suffering from the disorder when he/she expresses everyday situations, desires, or actions related to the suffering of such pathology.
User1 (messages):
"sinceramente te recomiendo que vayas a un nutricionista"
"yo tengo anorexia y no es nada lindo, me gustaría poder comer sin sentir culpa como lo hacía antes"
control: The user does not present evidence of suffering from the disorder.
User2 (messages):
"Ustedes son perfectas tal y como están"
"Si ustedes pueden busquen ayuda y si también quieren"
1.b. Simple regression
Provide a probability for the user to suffer anorexia or bulimia. A value of 0 means 100% negative and a value of 1 would be 100% positive.
Task 2. Depression detection
2.a. Binary classification
Detect if the user suffers from depression. Labels will be 0 for “control” (negative, the user does not suffer from depression) or 1 for “suffer” (positive). Let’s see some examples:
suffer: according to the WHO, depression is characterised by persistent sadness, low mood, and a lack of interest or pleasure in activities that were previously rewarding and pleasurable. A user is considered to be suffering from depression when he/she expresses everyday situations, desires, or actions related to the suffering of such pathology.
User2 (messages):
“Hola , estoy realmente mal , no se que hacer con mi vida“
“Porque las cosas no pueden acabar bien”
control: The user does not present evidence of suffering from the disorder.
User4 (messages):
"pareces más menor q yo"
"yo se bailar bachata mas menos"
2.b. Simple regression
Provide a probability for the user to suffer depression. A value of 0 means 100% negative and a value of 1 would be 100% positive.
2.c. Multiclass classification
Decide one among four different classes (“suffer+against”, “suffer+in favour”, “suffer+other”, “control”). The system must return one of these labels for each case.
suffer+against: A person who suffers from the disorder and seeks/offers help or information to get out of the disorder and overcome it. The person is against the disorder.
User1 (messages):
"Yo tengo depresion y soledad , pero salgo ahí fuera e intento apilar toda las particular de motivación que puede encontrar en el día para formar al menos una bola de tierra que pueda para seguir avanzando."
"Hay que salir , hacer deporte , tirar las pastillas a la basura y producir las sustancias químicas que nuestro cerebro necesita."
suffer+in favour: A person who suffers from the disorder and encourages (seeks/provides information) other users to go deeper into the disorder. They are in favour of the disorder.
User2 (messages):
"Hola , estoy realmente mal , no se que hacer con mi vida“
“Porque las cosas no pueden acabar bien”
suffer+other: A person suffering from the disorder and is not related to the above categories.
User3 (messages):
"yo estoy diagnosticado con depresion mayor"
“hay alguna juntada entre los argentinos del grupo ?”
control: A person is considered not to be suffering from the disorder when he/she does not show symptoms of suffering from it. They may be specialists in the subject who are dedicated to helping, people who have suffered from it in the past and are helping others to overcome it, or people who bother other users or talk about a subject other than the disorder.
User4 (messages):
“pareces más menor q yo”
“yo se bailar bachata mas menos”
2.d. Multi-output regression
For each of the previous classes, the system has to provide a probability of belonging to that class. These values, as in task 1.b., are interpreted as 0 for a 100% confidence of the system to not assign the user to a class, and 1 for a 100% probability of assigning the user to a class. Note that the sum of the four probabilities must be 1.
Task 3. Non-defined disorder detection
(cross-domain evaluation)
This is a binary classification (suffer, control) in which participants are encouraged to use the systems developed for subtasks 1.a, 1.b, 2.a, 2.b to identify a different disorder that is unknown to the participant but is related to the previous ones (eating disorder and depression).
3.a. Binary classification
Detect if the user suffers from an unknown disorder. Labels will be 0 for “control” or 1 for “suffer” (positive).
For this task, participants can use the systems developed for subtasks 1.a. and 2.a.
3.b. Simple regression
Provide a probability for the user to suffer from the unknown disorder. A value of 0 means 100% negative and a value of 1 would be 100% positive.
For this task, participants can use the systems developed for subtasks 1.b. and 2.b.