Wikipedia receives a huge number of text additions per day, each falling in one of these categories:

  1. Good-faith addition: in 1789
  2. Subtle misinformation: in 1790
  3. Spam: in any CostMart store
  4. Immature vandalism: in YOUR MOM lol
  5. Random keyboard input vandalism: wqefeqesfdqwedqw

This question is about case 5 only.
A detection algorithm considers:

  • The behavior of human fingers
  • The keyboard layout depending on the device (desktop/mobile/etc) and the culture (QWERTY/AZERTY/ЙЦУКЕН/InScript/Arabic/etc)

To train/test the algorithm, I need large quantities of random keyboard input.

  • At least 5000 strings
  • Bonus if for each string the device and culture is known, but it is not strictly required


kdakllkdkldslkdskds, QWERTY
あけさにめはよたかひや, FSKARENテンキー
  • Can you use the random char string generator? – Y.N May 8 '15 at 5:45
  • @Y.N: I need the data to train/test an algorithm. Generated data would not work for this, real-world data is needed. – Nicolas Raoul May 8 '15 at 8:56
  • Could you train it on WP data? eg/ look at reverted edits and pull out the ones where the diff is, say, more than 10 characters and does not contain any dictionary words? This would predominantly be case 5 examples. – Andrew May 8 '15 at 10:11
  • @Andrew: Yeah that would be a great source of data! But it is not trivial to extract, so if someone has done the work I would gladly re-use their data :-) Case 5 is actually a rather rare occurrence, so I guess 95% of the words found by the technique you described would be 1) Words that are not in the dictionary, because any Wikipedia article has 5%-10% of words that are not in my dictionary, like proper nouns, jargon, abbreviations 2) Technical wikicode such as refimprove or pp-move-indef. Cheers! – Nicolas Raoul May 8 '15 at 11:06
  • Not free, but you can ask Amazon MTurk's workers. – Anton Tarasenko May 9 '15 at 16:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.