Detecting Early Signatures of Persuasion in Information Cascades
INDIANA UNIV AT BLOOMINGTON
Pagination or Media Count:
Indiana University team worked on two different core functions to refine the persuasion detection system that we are building i the feature selection system for organic and promoted content classification ii user influence detection. University of Michigan team worked on i a formal definition of rumor and ii with the help of two human annotators to label some rumors from our Boston Marathon Explosion dataset and refine the codebook of rumor. The inter rater reliability is at first 0.46 and then improved to 0.6 after several rounds of modification of codebook. We have also been working on improving the performance of our rumor detection system. With human annotators, we had some very preliminar evaluation of our system.
- Information Science