language:
- ru
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
dataset_info:
features:
- name: post_title
dtype: string
- name: parent_comment
dtype: string
- name: parent_author
dtype: string
- name: child_comment
dtype: string
- name: child_author
dtype: string
- name: subsite_name
dtype: string
- name: comment_id_parent
dtype: int64
- name: comment_id_child
dtype: int64
- name: replyTo
dtype: int64
- name: parent_likes
dtype: int64
- name: child_likes
dtype: int64
- name: reply_count
dtype: int64
- name: normalized_reply_count
dtype: float64
splits:
- name: train
num_bytes: 3832490635
num_examples: 785296
download_size: 476384423
dataset_size: 3832490635
This dataset contains data from mid-2016 to the end of 2024 from the website DTF.ru
Structure:
– post_title - body of the post;
– parent_comment - parent comment :);
– child_comment - response (child) comment to parent comment;
– subsite_name - subsite name (like a theme);
– comment_id_parent - id of parent comment on dtf.ru
– comment_id_child - id of child comment on dtf.ru
– replyTo - id of parent comment what child comment responses
– parent_likes - number of likes on parent comment
– child_likes - number of likes on child comment
– reply_count - number of responses to the parent comment;
– normalized_reply_count - normalized value of number of responses to the parent comment (from 0 to 1).
– parent_comment_tox - toxity metric for parent comment
– child_comment_tox - toxity metric for child comment
– parent_author - author of parent comment
– child_author - author of child comment
It contains these symbols: a-zA-Zа-яА-Я0-9.,!?;:
Global filter: comments were allowed only from 30 characters or more and from 5 likes or more
Dataset isn't cleaned!
Due to the specifics of moderation on DTF.ru the dataset contains many duplicate rows. There are some types of them:
– "Этот материал был удален по просьбе автора";
– "Комментарий недоступен";
– "Комментарий удален автором поста";
This is mark of deleted comments, they are valuable only in counting them, but not in using them for fine-tuning or analyzing writing.