Bangumi Anime Ranking

Yet another Bangumi anime ranking analysis project.

• Data Last Updated: 2023-04-30
• Analysis Last Updated: 2023-05-01
• Entries: 6565 (= 7673 all ranked animes - 1108 restricted entries)
• As there are too many files, you cannot preview all of them on Github. Clone this repo and view the files locally.
• Enjoys!

PONet Ranking

A side project for better ranking.

• Theory: a simplified ranking system based on partial order network. The ranking is based on how a specific user votes different entries, instead of how a specific entry is voted by different users.
• Dataset: Bangumi15M, contains 8573 valid entries (including restricted ones) and 7,770,854 valid votes.
• Data Last Updated: 2023-05-09
• Analysis Last Updated: 2023-06-01
• See ponet/README.md for more details.
• The result has been published on Bangumi.

Usage

Requirements

Python 3.6+ is required. You can install the dependencies by running the following command.

pip install -r requirements.txt

Run

Since crawling data can be time consuming, I have already crawled the data and put it in data/.

To crawl data by yourself, you can run the following commands, or simply execute run.bat (Windows only).

Task	Script	Output	Function
Get ranking list	spider.py	data/id/
Get entry details	subject.py	data/sub/	api_main()
Pick up available entries	subject.py	data/id/available.txt	ava_main()
Generate CSV containing all entries	subject.py	data/sub.csv	csv_main()
Perform General Statistics	stat.py	data/stat/
Perform Ranking Algorithm Analysis	rank.py	data/rank/

Notes

• All data are stored in UTF-8 encoding. Your spreadsheet software may not recognize it. You should open the CSV file with encoding UTF-8 or UTF-8 with BOM.
• Please configure subject.py Line 31 before running. If you need a fresh start, change to w; but if you want to resume after some blocks, change to a.
• Estimated time to get entry list: around 2 minutes.
• Estimated time to get entry details: 1 hour.
• Refer to Bangumi API for everything about API.
• Although fake user-agent works, it is recommended to configure your own UA.
• The data are crawled and processed in blocks. By default, 10 pages (240 entries) per block. As of 2023-04-30, there are 320 pages in total, which are split into 32 blocks, numbered with their starting page number (1, 11, 21, ..., 311). It is very likely that pages and blocks will increase over time, but there should be no issue if you keep the default block setting.
• Restricted entries can be found in data/id/restricted.txt. They appear in the ranking list but cannot be accessed by API. They are excluded from the sub.csv file.
• The rank field in the CSV file is the rank within available entries. The original rank is not included since it cannot provide any statistical information.
• Because of the restricted entries, it is common to see many entries missing with logs in the console. You can safely ignore them. However, if there are too many missing entries, you may want to check:
  • if the API is down;
  • if you have configured UA correctly;
  • if your IP is blocked by Bangumi.

Results

Field	Value
Total Entries	6565
Total Votes	7,573,022
Average Votes per Entry	1153.3
Median Votes per Entry	395
Average of Average Score	6.6637
Median of Average Score	6.7071
Average of User Votes	7.1854

View data/stat/stat.csv for more details.

View data/stat/ for plots of all fields.

To be continued...

In addition to more plots, there will also be analysis on ranking algorithm, voting behavior, and more.

License

MIT License. See LICENSE for details.

Acknowledgement

Bangumi for providing the data.