eRsik Depression Severity Level Challenge
• Task 2 – Severity level of Depressions
• Test data 70 subjects, Training data 20 subjects
• We don’t have the ground truth, but we have the subjects BDI questionnaires results.
• Data is in XML format contains (Title, Date, Text, Info)
eRsik Depression Level Estimation Task
• Four evaluation measures were introduced to evaluate the
1. Average Hit Rate (AHR) computes the ratio of cases where the automatic
questionnaire has exactly the same answer as the real questionnaire.
2. Average Closeness Rate (ACR) is a less stringent measure that considers the
distance between each real answer and the answer submitted by the
3. ADODL and DHCR, were oriented to compute how effective the systems are
at estimating the overall depression level of the individual. These two
measures compute the deviation between the total depression score (sum
of all responses in the questionnaire) of the real questionnaire vs the
questionnaire submitted by the participants.
• Is the data validated by a psychiatrist/linguist?
• How/When is the data annotated?
• What is the criteria of a respondents?
• Do we split demographic (male/female, Age range, Status)
• How many respondents do we need for each label?
• How many data (string length) is needed from each respondents?
• How long do we scrap the data from a respondents?