Google to Improve Machine Learning Abuse Filters

Google is improving machine learning abuse filters. It is aimed to reclaim identity labels for better training machine learning abuse detection models.

During one of Google's ongoing work on Perspective engineers had found out that the models identified sentences that use the words 'gay', 'lesbian', or 'transgender' as being abusive.

Identity labels like 'gay' and 'lesbian' seem to be over-represented in abusive and toxic online comments. Seeing this the machine learning models attach negative connotations to the label.

The labels and statements which are collected online and in-person as part of Project Respect will go into an open source dataset so developers will be able to teach their own machine learning models which words are used by people to positively identify themselves.