For algorithms, spotting trolling or harassment online largely relies on monitoring keywords -- a blunt instrument that doesn’t account for context or the evolution of conversations. But a new machine learning algorithm might change that.
A new study out of Caltech, "Finding Social Media Trolls: Dynamic Keyword Selection Methods for Rapidly-Evolving Online Debates” created a proof of concept project that was able to use machine learning to identify harassment online, using “word embedding models” that are better at understanding context.
We take a closer look at this new method and its promise.
Guests:
Anima Anandkumar, professor of computing and mathematical sciences at Caltech; one of the principal investigators on the study
Michael Alvarez, professor of political science at Caltech; he is one of the principal investigators on the study