How to Classify Toxic Comments in Russian Using RuBERT-Toxic

May 24, 2021

In the digital age, safeguarding online communities from toxic remarks is crucial. With the emergence of AI models, we can accurately detect and classify these harmful comments. One such model is RuBERT-Toxic, which is specifically fine-tuned for the Russian language. This blog will guide you through the process of using RuBERT-Toxic for toxic comments classification.

What is RuBERT-Toxic?

RuBERT-Toxic is a modification of the original RuBERT model, tailored to classify toxic comments found in Russian discourse. It has been trained on the Kaggle Russian Language Toxic Comments Dataset, ensuring it understands the nuances of the language.

Understanding the Dataset

The dataset comprises 14,412 Russian-language comments sourced from 2ch. and Pikabu. Out of these, 4,826 comments are labeled as toxic, while 9,586 are considered non-toxic. With an average length of 175 characters, it contains a variety of comment forms.

Performance of RuBERT-Toxic

The classification performance can be evaluated using metrics like Precision (P), Recall (R), and F1 Score (F1), as represented below:


System                     P        R        F1
----------------------------  -------  -------  -------
MNB-Toxic                   87.01%   81.22%   83.21%
M-BERT-Toxic                91.19%   91.10%   91.15%
RuBERT-Toxic                91.91%   92.51%   92.20%
M-USE-CNN-Toxic             89.69%   90.14%   89.91%
M-USE-Trans-Toxic           90.85%   91.92%   91.35%

The RuBERT-Toxic model achieved an impressive F1 score of 92.20%, making it a reliable choice for detecting toxic comments.

How to Use RuBERT-Toxic

To employ the RuBERT-Toxic model for toxic comments classification, follow these simplified steps:

Clone the Repository from GitHub: GitHub Repository.
Install the necessary dependencies listed in the repository.
Load the pre-trained RuBERT-Toxic model using a machine learning library of your choice.
Prepare your text data for input, ensuring they are in the correct format.
Run the model on your data and analyze the results.

Troubleshooting

While working with RuBERT-Toxic, you might encounter some common issues. Here are some troubleshooting tips:

Installation Issues: Ensure all necessary libraries are installed. This often includes TensorFlow or PyTorch.
Data Formatting Errors: Input text must be correctly formatted as specified in the model’s documentation. Check length and character encoding.
Performance Variability: If results vary dramatically, consider adjusting the model configurations or training parameters.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

Why This Matters

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.

By implementing RuBERT-Toxic, you’ll contribute to creating a more positive online environment, helping diminish the spread of toxic language in Russian discussions. Happy coding!