Built a Danbooru autocomplete cleaning assistant using this dataset
Hi, I used this dataset to add a new optional feature to my Forge Neo extension, TagComplete Neo Multi-CSV.
Repository:
https://github.com/ukr8b3g-cmyk/TagComplete-Neo-Multi-CSV
The new feature is called Danbooru Tag Cleaning Assist.
It uses the correction data from Grio43/Tag_cleaning to add lightweight hints to Danbooru autocomplete results, such as:
remove-heavy correction warnings
related added-tag hints
category-aware matching
The source Parquet is not queried during normal autocomplete. It is processed once into a small local SQLite index, and only the currently displayed autocomplete candidates are checked.
In my current build, the 1,739,622 source rows are reduced to about 9,041 tag records and 8,266 related-pair records, producing a SQLite database of about 0.9 MB.
The feature is optional and disabled by default. It does not automatically replace tags, and remove is not treated as meaning that a Danbooru tag is deprecated.
Thanks for publishing this dataset. It turned out to be very useful for adding correction-aware autocomplete assistance.