russian-discord-chat-logs

Dataset Overview

The data is stored in Parquet format, which provides efficient storage, fast query performance, and maintains precise data types. Each row represents a single unique chat message that has passed through several stages of rigorous cleaning, filtering, and semantic analysis.

Schema

Column	Type	Description
message	string	The cleaned text of the Discord message (containing at least 30% Cyrillic characters).
semantic_score	float32	A normalized value (0.0 - 100.0) representing the message's information and semantic density.

Processing Pipeline

To ensure high data quality and remove noise, the raw logs went through the following automated pipeline:

1. Extraction: Raw message logs were parsed and extracted from multiple nested folders containing messages.json files within the Discord data package.
2. Deduplication: Identical messages were merged, leaving only unique text entries.
3. Length Trimming (Outlier Removal): All unique messages were sorted by length. The top 10% (shortest messages) and bottom 10% (longest messages) were trimmed to eliminate extreme outliers (e.g., single-word spam or massive copy-pastes).
4. Cleaning:
   • URL Removal: All hyperlinks (http, https, ftp, www.) were stripped from the text.
   • Emoji Filter: Messages consisting entirely of emojis and whitespaces were discarded.
5. Language Filtering: To filter out multi-language noise, code snippets, and random characters, the pipeline calculated the ratio of Cyrillic characters. Only messages where at least 30% of the text consists of Russian letters ([а-яА-ЯёЁ]) were retained.

How Semantic Score is Calculated

The semantic_score is not just a measure of length, but a representation of Semantic Density. The calculation involves:

• Vectorization: Each message is converted into a high-dimensional vector (embedding) using the paraphrase-multilingual-MiniLM-L12-v2 Sentence-Transformer model. This model excels at understanding context and semantic relationships between words across multiple languages.
• L2-Norm Calculation: We calculate the Euclidean magnitude (L2-norm) of the embedding vector (the square root of the sum of squared vector components). Complex and contextually rich sentences typically result in higher vector norms.
• Length Weighting: To balance the score, we apply a logarithmic weight based on the character length of the message (log(len + 1)). This prevents long, repetitive sentences from dominating while ensuring that very short phrases (like "Ок") receive lower scores.
• Min-Max Scaling: The final raw values (calculated as L2-Norm * Length Weight) are normalized to a 0 to 100% scale.
  • 100.0: The most semantically dense message in the dataset.
  • 0.0: The message with the least information density (e.g., simple interjections or basic phrases).

The final dataset is fully sorted by semantic_score in descending order, putting the most valuable and meaningful discussions at the top.

Usage

from datasets import load_dataset

dataset = load_dataset("KvaytG/russian-discord-chat-logs", split="train")

License

This dataset is released under the Apache License 2.0.

Citation

@misc{kvaytg_russian_discord_chat_logs,
  author       = {KvaytG},
  title        = {Russian Discord Chat Logs: A Semantically Ranked Dataset},
  year         = {2026},
  publisher    = {Hugging Face},
  journal      = {Hugging Face Datasets},
  url          = {https://huggingface.co/datasets/KvaytG/russian-discord-chat-logs},
  note         = {Processed Discord messages with semantic density scoring using paraphrase-multilingual-MiniLM-L12-v2}
}