Detecting propaganda and illicit content

Social media is the main propaganda tool for terrorist groups and Twitter is not immune.

The challenge

The UN’s Office of Information and Communications Technology needed an automated solution for spotting terrorist propaganda on Twitter. With almost half a million tweets appearing every minute, manual monitoring would hardly be possible. The solution had to be automated.

The solution

deepsense.ai designed a solution that analyzed the entire news stream from a particular account. The team analyzed not single tweets, but the overall communication context. This included all of the users’ messages, the number of connections and followers he or she had and the combination of words used. The model proved that those who disseminate propaganda differ from others - they have fewer followers and publish significantly less information. Knowing this allowed us to filter out news vendors and the people commenting on events as experts. The profile of a suspicious account was built using 100 accounts that were proven disseminators of propaganda.

Later, the model spotted the most suspicious accounts and cherry-picked those which matched the patterns of propaganda-spreading ones.