Certain text, when used in a filename or description, suggests that the uploaded file is more likely to be deleted
Evaluate whether this is a useful tool to flag uploads
Suggested text to search for
Lego
Penis
Scrotum
Erection
Buttocks
Cock ring
Coitus
Penile
Nude
Vagina
Masturbation
Vulva
Testicle
logo
Selfie
Louvre pyramid
Burj khalifa
Eiffel Tower night
Atomium
| Status | Subtype | Assigned | Task | ||
|---|---|---|---|---|---|
| Resolved | None | T357587 [Research EPIC] Media quality investigation on Commons FY24/25 | |||
| Resolved | Cparle | T347963 [Spike] Evaluate text in title that is likely to result in a deletion request |
I did some prelimary work on this while finishing up T344060. Results below. Note this is only for filenames, it's a bit more work to extract descriptions
379015 unique filenames uploaded in 2022 have been deleted
| Total | Deleted | % deletion rate | % of total deletions | |
| Filename containing "logo" | 45105 | 10033 | 22 | 2.6 |
| Filename containing "lego" | 202 | 50 | 25 | <0.1 |
| Filename containing "selfie" | 279 | 103 | 37 | <0.1 |
| Filename containing "nude" | 747 | 179 | 24 | <0.1 |
| Filename containing sexual words | 1667 | 685 | 41 | <0.1 |
| Filename containing famous buildings above | 2 | 2 | 100 | <0.1 |
The notebook used to gather these numbers is here https://gitlab.wikimedia.org/cparle/notebooks/-/blob/main/T347963.ipynb
The table below is also from the notebook. There are 46106 deletions (out of 379015, so 12%) where the filename contains a word that
... which suggests to me this is worth investigating more deeply
| word | total_filenames_containing_word | deleted_filenames_containing_word | deletion_rate | proportion_of_deletions |
| png | 380054 | 62864 | 0.16540807358954254 | 0.1658615094389404 |
| colorado | 36529 | 11291 | 0.3090968819294259 | 0.029790377689537354 |
| logo | 45105 | 10033 | 0.22243653696929386 | 0.026471247839795257 |
| webp | 14778 | 7803 | 0.5280146163215591 | 0.020587575689616507 |
| report | 26742 | 5901 | 0.2206641238501234 | 0.015569304644934896 |
| annual | 22890 | 4915 | 0.2147225862822193 | 0.012967824492434336 |
| work | 14291 | 3320 | 0.23231404380379259 | 0.008759547775154017 |
| flag | 21988 | 3038 | 0.13816627251227942 | 0.008015513897866838 |
| twemoji13 | 3580 | 3014 | 0.8418994413407821 | 0.007952191865757291 |
| gif | 15361 | 2676 | 0.17420740837185078 | 0.007060406580214503 |
| leaves | 8593 | 2496 | 0.29046898638426627 | 0.0065854913393929 |
| photo | 18750 | 2185 | 0.11653333333333334 | 0.005764943339973352 |
| nr | 10034 | 2123 | 0.21158062587203508 | 0.005601361423690355 |
| альбом | 1984 | 1948 | 0.9818548387096774 | 0.005139638272891574 |
| agricultural | 7675 | 1872 | 0.24390879478827363 | 0.004939118504544675 |
| extension | 6288 | 1740 | 0.2767175572519084 | 0.004590847327942166 |
| церковь | 5015 | 1631 | 0.32522432701894316 | 0.004303259765444639 |
| poster | 3584 | 1602 | 0.44698660714285715 | 0.00422674564331227 |
| agent | 5727 | 1573 | 0.27466387288283567 | 0.0041502315211799 |
| union | 14437 | 1559 | 0.10798642377225186 | 0.004113293669115998 |
| cover | 5708 | 1501 | 0.26296426068675544 | 0.003960265424851259 |
| performance | 5587 | 1480 | 0.26490066225165565 | 0.0039048586467554055 |
| trials | 3779 | 1262 | 0.3339507806297962 | 0.0033296835217603523 |
| dr | 9739 | 1204 | 0.12362665571413903 | 0.0031766552774956137 |
| larimer | 2912 | 1189 | 0.40831043956043955 | 0.0031370790074271467 |
| file | 4470 | 1166 | 0.2608501118568233 | 0.003076395393322164 |
| pm | 5933 | 1162 | 0.19585369964604754 | 0.003065841721303906 |
| reclame | 1163 | 1138 | 0.9785038693035254 | 0.003002519689194359 |
| making | 4303 | 1121 | 0.26051591912619104 | 0.002957666583116763 |
| better | 3468 | 1114 | 0.32122260668973474 | 0.002939197657084812 |
| 200d | 5801 | 1107 | 0.19082916738493363 | 0.0029207287310528606 |
| decisions | 2867 | 1062 | 0.3704220439483781 | 0.00280199992084746 |
| america | 8683 | 1049 | 0.12081077968444086 | 0.0027677004867881216 |
Ok having done quite a bit of futile digging I've discovered that neither template links nor wikitext for deleted revisions is stored in the data lake (or dumps), and if we want to find patterns in descriptions/templates related to deletion requests then we'll need to get the wikitext from external storage
So ... closing this ticket, and will add another to do with getting the wikitext for deleted revisions/pages