Automated Detection of Personal Psychological Distress from Social Network Activities

Schwarz Baruch (Education), HUJI, Faculty of Humanities, School of Education
Asterhan Christa, HUJI, School of Education


  • Social network technologies (SNT) and online Social Network Sites (SNS) are immensely popular and have become an integral part of people's everyday functioning and social lives worldwide, and especially among adolescents.
  • The negative effects and danger of SNS usage, such as 'cyber bullying' and 'Facebook depression' are tremendous.
  • As of today, no computerized algorithms were developed for automated screening of mental health problems (e.g., depression, social rejection, victimization of bullying).

Our Innovation

  • A novel computerized detection tool for psychological distress and suicide intentions among SNT users based on NLP techniques.
  • Automated, un-intrusive screening of mental health problems (such as depression, anxiety, and suicide ideation).
  • User activity patterns analysis
  • Individual need detection


  • Detectable differences in social network behavior among adolescents who suffer from social rejection and/or depression and those who do not, were found in small scale pilot studies. These differences are often not based on direct, verbal distress references, but on more subtle differences in activity patterns.
  • Further indicators of psychological distress in SNT behavior are extracted by combining human, clinical expertise in mental health, Natural Language Processing (NLP) techniques and vast data sets.
  • A classifier for social network data according to indicators of the authors’ psychological state will be developed and trained to predict the values of variables in the psychological questioner of the participants, given the Deep Neural Networks (DNNs) based representations of their social network data. Using DNNs enhanced the capability of NLP computer algorithms to make deeper interpretations of the meaning of sentences, paragraphs and larger texts.
  • In addition to the text analysis algorithms and the linguistic categories, the use of non-linguistic information derived from the structure of the social network or from non-textual activities of the participants (e.g. uploaded images and emojis) will be considered to improve the algorithms’ predictive strength.


  • Detect SNT users’ distress online in general and SNT young adults’ distress online in particular.
  • Provide psychological and emotional support to social networks users.


Contact for more information:

Anna Pellivert