IlyaGusev/rubertconv_toxic_clf

RuBERTConv Toxic Classifier

Model description

Based on rubert-base-cased-conversational model

Intended uses & limitations

How to use

Colab: link

from transformers import pipeline

model_name = "IlyaGusev/rubertconv_toxic_clf"
pipe = pipeline("text-classification", model=model_name, tokenizer=model_name, framework="pt") 

text = "Ты придурок из интернета"
pipe([text])

Training data

Datasets:

• 2ch
• Odnoklassniki
• Toloka Persona Chat Rus
• Koziev's Conversations with toxic words vocabulary

Augmentations:

• ё -> е
• Remove or add "?" or "!"
• Fix CAPS
• Concatenate toxic and non-toxic texts
• Concatenate two non-toxic texts
• Add toxic words from vocabulary
• Add typos
• Mask toxic words with "*", "@", "$"

Training procedure

TBA