Chat Application with AI Harassment Detection

Team: AWTPro SS2019 Group 1 (Jonathan Samuel Will, Sabtain Ahmad, Sukanya Sunder)

Technologies

• Backend:
  • Python / Django framework
  • Nginx web server and Gunicorn on Ubuntu
  • SQLite Database
• Frontend:
  • CSS with Bootstrap, FontAwesome Icons
  • Javascript with AJAX, JQuery
  • GIMP to design some of the visuals

Functioning

• General:
  • Sending messages from the client to the server
  • Messages get classified by server
  • Classification gets added to the message on the client
• Chat application details:
  • Responsive design to work on all devices
  • Smart scrolling: Only scroll down on new messages if not manually scrolled up
  • Everyone gets assigned an identity based on their IP address
  • Participants see their own messages shifted to the right, messages of others shifted to the left
  • Messages get deleted automatically when they are one hour old
  • An identity gets freed up again after the last message from that identity has been deleted
  • If there are no more identities left, every additional participant gets the "Anonymous" default identity
  • Messages can get relabeled based on previous or subsequent messages and their relation to each other
  • Hover with mouse over messages to see additional information regarding the classification and reclassification
• Classifier details:
• Transfer learning with ULMFIT model
• Datasets from Formspring and Twitter
• 3 Classes: Normal, Hatespeech and Offensive