Problem: Digital propaganda increasingly uses images to make messages more appealing and to be shared more widely.
NCIA are working with a number of organisations to help identify propaganda and inflammatory information in the public information space (traditional and social media).
Outcome: To find a solution which can identify objects and text within images and identify relationships between key elements of the image and associated hashtags. Images should be scored with a value related to the level of correlation between the image and a reference set of graphical objects and text terms.
(Any solution which could be integrated into the open source KNIME analytics platform (www.knime.org) would be particularly welcome.)
Any hackathon solutions to this challenge which perform well with the test data set and others will be taken forward and used at a NATO exercise soon after the hackathon ends. There is potential for hackathon teams to continue to work with NCIA if both parties wish.
Datasets: Training set of 486 labelled images (representing positive and negative propaganda), to be tested against labelled images prepared in testing set.
Training set - total of 486 images
o 235 represents NATO negative propaganda
o 251 are neutral or represents positive propaganda
· Testing set - total of 50 images
o 25 represents NATO negative propaganda
o 25 are neutral or represents positive propaganda