1️⃣ The project provides open-source comment data for researchers to conduct in-depth research on the Danmaku corpus, which mainly focuses on the subculture bullet-screen comment corpus (including but not limited to the Guichu, animation and e-sports type).
2️⃣ Corpus is maintained by TinyTalks, a Community for NLP research in Short Text in Chinese.
- Video Corpus
- Video-Channel Network
- Danmaku Comment Corpus
- Danmaku Language Models
- Danmaku Marked Data
- Citation
The video corpus were more than 1 milions plays during the years from 2017 to 2020. The attributes can be as follow.
| Attribute | explanation |
|---|---|
| Bvno | 视频识别号 |
| Tname | 主标签 |
| Tag_name | 标签列表 |
| Owner_mid | 发布者Id |
| Title | 标题 |
| Pubdate | 发布时间 |
| Duration | 持续时间 |
| View | 观看数量 |
| Danmaku | 弹幕数量 |
| Reply | 转发数量 |
| Favorite | 收藏数量 |
| Coin | 投币数量 |
| Share | 分享数量 |
| Like | 喜欢数量 |
Source: https://pan.bnuz.edu.cn/l/J5z6nP password:bnuz
https://pan.bnuz.edu.cn/l/g1ydM2 password:bnuz
| Attribute | explanation |
|---|---|
| videoType | 视频类型 |
| relation | 属于 |
| channel | 频道标签 |
The attributes detail can as follow.
| Attribute | type | explanation | Default |
|---|---|---|---|
| text | (str) | 弹幕文本 | |
| dm_time | (float) | 弹幕在视频中的位置，单位为秒 | 0.0 |
| send_time | (float) | 弹幕发送的时间 | time.time() |
| crc32_id | (str) | 弹幕发送者 UID 经 CRC32 算法取摘要后的值 | None |
| color | (str) | 弹幕十六进制颜色 | "ffffff" |
| weight | (int) | 弹幕在弹幕列表显示的权重 | -1 |
| id_ | (int) | 弹幕 ID | -1 |
| id_str | (str) | 弹幕字符串 ID | "" |
| action | (str) | 暂不清楚 | "" |
| mode | (Mode) | 弹幕模式 | Mode.FLY |
| font_size | (FontSize) | 弹幕字体大小 | FontSize.NORMAL |
| is_sub | (bool) | 是否为字幕弹幕 | False |
| pool | (int) | 暂不清楚 | -1 |
| attr | (int) | 暂不清楚 | -1 |
| Type | Original Source | Text Source | Password |
|---|---|---|---|
| 鬼畜 |  |
|
bnuz |
| 电竞 | |
|
bnuz |
| 动漫 | |
|
bnuz |
| 疫情 | |
|
bnuz |
疫情类 https://pan.bnuz.edu.cn/l/aoMMOM(密码：bnuz)
https://pan.bnuz.edu.cn/l/onFbAO(密码：bnuz)
https://pan.bnuz.edu.cn/l/QJGkNF(密码：bnuz)
| Mdeols | Description | Mask Accuracy | Link |
|---|---|---|---|
| chinese_danmaku_roberta | fine-tuned version of uer/chinese_roberta_L-8_H-512 on an Danmaku Corpus(2000k raw data) dataset. | 0.7780 |  |
数据使用utf-8编码，逗号分隔的csv保存
数据分为数字信息、文本信息、标注分类（未标注数据无标注信息）
数字信息包含：弹幕在视频中出现的时间点、展示模式、字号、字体颜色、发送时间、弹幕池编号、发送用户编号、在弹幕数据库中的编号，信息用逗号分隔
标注类别：0高兴、1难过、2愤怒、3惊、4负样本
https://pan.bnuz.edu.cn/l/snpijm password: bnuz
citation: https://github.com/MelkiorOno/DanmakuMarked-data
If you use this corpus in your research, please cite this repository.
@article{
QBTS202209010,
author = {陈鑫,张以欣,吴俊潮,郭凌宇,余泽汇 & 杨静},
title = {历时弹幕语料库的构建与探析——以青年亚文化弹幕为例},
journal = {情报探索},
volume = {No.299},
number = {77-85},
year = {2022},
issn = {1005-8095},
doi ={10.3969/j.issn.1005－8095.2022.09.010}
}