Team Members: Yuxi Pan (yuxpan@cisco.com), Doug Sibley (dosibley@cisco.com), Sean Baird (sebaird@cisco.com)
In the below directories, you can find code used by Team SOLAT IN THE SWEN to perform stance detection on a number of news headlines and article text. Our model is based on an weighted average between gradient-boosted decision trees and a deep convolutional neural network.
Both tree_model and deep_learning_model contain their own README.md files detailing their model and providing instructions for running and installation. The model averaging process is described in tree_model.
To learn more about how stance detection can help to detect fake news and disinformation campaigns, please visit our blog post on blog.talosintelligence.com/2017/06/talos-fake-news-challenge.html.
For those interested, tree_model/README.md has detailed information on how to run our models to duplicate our results.
See Fake News Challenge Official Website for more information. Thank you to Wendy, Melissa, and the entire Talos art team for the graphics, and to Joel and Luci for helping to open source our solution. Big thank you to our leadership team, as well, for allowing us the time to work on this important problem.
Interested in learning how Talos forces the bad guys to innovate? Visit talosintelligence.com