Skip to content

Latest commit

 

History

History
6 lines (5 loc) · 370 Bytes

README.md

File metadata and controls

6 lines (5 loc) · 370 Bytes

Toxic detection (ruBert)

Toxic comments are a serious problem on many social media platforms. One of the topics of the April 14 hackathon from Tinkoff was dedicated to this problem.

We proposed to use the pretrained Rubert model with two fully connected layers for classification. Macro Average score f1 showed 0.76. This is a good result for such a heavy dataset.