Padjective Tag Hierarchy

Leaderboard

The Elo-inspired model favours tags that consistently appear earlier in product titles when paired with others. Here are the current top contenders.

Synset	ID	Products	Share
No synsets processed yet.

Top tags bar chart — Top 20 tags by inferred depth.

Biggest losers

Tags that our model predicts are most likely to be pushed to the end of product titles.

tag	component	score
SURFING	0	-551.685868
HARAJUKU	0	-524.147777
SEAT	0	-501.552704
WAITE	0	-497.947028
COVER	0	-491.899984
DIVING	0	-468.987234
凉席	0	-458.307177
XXX	0	-448.434145
CLOTHES	0	-441.018583
DJIBOUTI	0	-437.760267
ORNAMENT	0	-433.813277
SKIRT	0	-428.662407
HOODIE	0	-425.937372
TOY	0	-423.079868
HEADBAND	0	-422.347725
T-SHIRT	0	-419.986086
CIVIC	0	-408.064480
STICKER	0	-407.901515
ROBE	0	-405.289716
VSM 70	714	-404.413936

WordNet synset tagging

We ask GPT models to map each product to the closest WordNet synset.

849 Products classified

49,297 Products remaining

849 No WordNet match

1.7% Coverage so far

Processing rate: 7,358.2 products/day. Remaining work estimated completion: 6 days 16 hours. Projected completion around 2025-10-05T21:25 Last processed at 2025-09-29T04:38:32

Most common synsets

Synset	ID	Products	Share
No synsets processed yet.

Explore all processed synsets →

How the rankings work

Battle generation — tagbattle.py scans each product title, comparing the order of every pair of tags.
Elo-style scoring — ranking.py treats each ordering as a battle, rewarding tags that appear closer to the start.
Visualisation — display.py turns the rankings into shareable tables and charts.

Tags are grouped by connected component so isolated tag families get their own podium.