The People Who Dance
A dancer has several spellings, a few partners, a festival circuit, and sometimes a world title. None of it is written anywhere a computer can read. This is how TangoTube builds a list of 10,947 people from the titles strangers typed, without putting anyone’s name on a dance they were not in, and then hands each person the keys to their own page.
- dancers
- 10,947dancers
- “this person is in this video”
- 278,110“this person is in this video”
- real partnerships
- 5,559real partnerships
- world champions
- 91world champions
- dancers on the map of the scene
- 90dancers on the map of the scene
Part 01
One person, many strings
A recording cannot be libelled; a person can. Putting someone’s name on a performance they were not in is a mistake with a face attached. So the people list is built by rules that refuse when unsure, and a lone nickname, the riskiest string of all, is written down as a clue but never asserted as a credit.
Part 02
One door for every new name
Any similarity threshold is wrong most often right next to itself. The band where the doorkeeper refuses turns those cases into questions for a person instead of coin tosses, and a name that becomes a new person is kept off the site until someone has looked at it.
Part 03
Partners are counted, not typed
The map is a by-product of the videos. Remove a video and the line it drew is gone the next night. Of 14,525 pairs who have shared a video, 5,559 are partnerships, and only those appear as “frequent partners” or as lines on the map in the next scene.
Part 04
Draw the scene once a night, then let people walk it
No behavioural data and no model are involved in these recommendations. They are the lines the night job drew, read one hop out.
Part 05
Ninety-one certain names, and the keys to your own page
Tour dates an owner posts appear on their page, on each festival’s page, and in the weekly edition of everyone who follows them. What an owner cannot do is make a performance disappear, because the notebook belongs to everyone.
Take-aways, without the tango
- 01Model people as points with many strings attached, and grade the string that matched.
- 02One door for new entities, with a “too close to call” exit that asks a person.
- 03Count relationships from evidence on a schedule. Never hand-enter a graph you can derive.
- 04Certain lists are also tests. Link them by hand and fail loudly.
- 05Let the subject own the facts about themselves, and keep the evidence about events out of everyone’s reach.