dystopedia
A Markov chain Twitter bot based on the titles of deleted Wikipedia articles.
The data
The data in titles.txt is drawn from the first version of deletionpedia, which was a project to archive articles that had been deleted from Wikipedia or might be deleted soon. It closed in 2011, but the database archive of deletionpedia v1 is accessible at the Internet Archive here. In the compressed archive deletionpediadbatleycom_w-20110712-wikidump.7z, there is a file with the titles of all articles archived on deletionpedia called deletionpediadbatleycom_w-20110712-titles.txt. Every article title is listed twice, once with just the title, then another time with the title and date of deletion. I cleaned this up by sorting it for unique items, then took every other line from this file:
sort -u titles.txt
awk 'NR%2==0' titles.txt
Getting a full list of deleted Wikipedia articles
You can get titles of deleted Wikipedia articles from Quarry, the live SQL query service built by @yuvipanda. However, this has a lot more profanity, gibberish, hate speech, personal attacks, and other nasty things that aren't really funny in a bot like this. Getting the entire log in the English-language Wikipedia is a big task that takes more time than Quarry allows for queries, but you can get a subset sample here. You would also have to replace all underscores with spaces for markovify to work.