Speakers: Oduwa Edo-Osagie and Leo Rickayzen – Data Scientists at Depop.
Talk Title: Real-time Chat Moderation using Deep Transformer Networks.
Talk Summary: Moderation is an important aspect of any social platform in order to keep users safe and maintain their trust in the platform. At Depop, one of the ways we do this is by protecting our users from being scammed on our digital marketplace. In order to do this, we need to be able to quickly detect and react to scammers coaxing users off-site to take advantage of them. This talk will present our approach to solving this problem, diving into the architecture of the deep transformer network we employ, along with its implementation, evaluation and interesting productionisation details.
Bio: Leo is a Data Scientist at Depop, the fashion marketplace where the next generation buy, sell and get inspired. For the last year he’s been working as part of the Trust & Safety team to help keep our community of over 20 million users safe. He’s interested in Trust & Safety, Machine Learning and MLOps.
Bio: Oduwa is a data scientist working in Trust and Safety at Depop. Previously, he completed a PhD in machine learning for social media mining at the University of East Anglia. During that time he carried out research projects with Public Health England in Birmingham and the Alan Turing Institute in London. When not working, Oduwa enjoys live music and trashy films.