5 ms·

She uses the phrase "spam robot". Is that semantically equivalent to automated? If so, for the love of god, can someone tell me how I can build something that a

by eshvk 13y ago

She uses the phrase "spam robot". Is that semantically equivalent to automated? If so, for the love of god, can someone tell me how I can build something that auto-generates beautiful existential poetry like that? (I am only half joking here)

tessierashpool 13y ago
I only have half an answer: Markov chains. It's probably going through text and using Markov chains to create "similar" text. Then question then becomes what text to go through, how long the Markov chains are, how you vary them, etc. Not long ago I took a workshop on algorithmic music where the lead professor had published a book of haiku. The vast majority of the haiku were algorithmically generated via Markov chains, while a very slim minority were actual haiku from 17th century Japan (iirc) translated into English. Guessing which haiku were "real" and which algorithmic was very hard to do.
- veidr 13y ago
  Do you happen to have more information on that book?
stdbrouw 13y ago
Some ideas from the guy who created @nytimes_ebooks: http://harrisj.tumblr.com/post/23737140672/nytimes-ebooks http://harrisj.tumblr.com/post/23737140672/nytimes-ebooks
simonw 13y ago
Have you played around with markov chains? They're really easy to build and produce very entertaining results - a few years ago I tried feeding all of the SXSW Interactive session titles in to one and got results like "Participatory budgeting crowdsourcing for real time marketing growing a digital culture".
- d23 13y ago
  Can you suggest some good resources for newbies/laymen on the topic? Edit - For anyone else who might be looking, this page gives a nice pythonic example: http://agiliq.com/blog/2009/06/generating-pseudo-random-text-with-markov-chains-u/ http://agiliq.com/blog/2009/06/generating-pseudo-random-text...
  - Groxx 13y ago
    Well, as a very rough overview, all they are is: * read in words * keep track of the frequency of "wordA wordB", "wordB wordC", etc. * use that frequency, e.g.: { "wordA" : {"wordB" : 0.5, "wordC" : 0.25, ...}, "wordB" : {"wordC" : 0.75, ...} } combined with a word, say "wordA", and pick a random word from the frequencies, weighted by frequency. * your word is now the word you just picked. repeat that last step until you get bored. You can go 'up' and 'down' in how many layers of frequency you want to collect, like the frequency of individual word pairs, vs word triples {"wordA wordB" : {"wordC" : 0.25}, etc}, vs sentences, anything. The more you collect, the more 'real' your generation will tend to be, but it takes more data to train it effectively since you want lots of possibilities for each 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' is a single phrase, so it keeps selecting it and it doesn't sound like something anyone would actually say.
comex 13y ago
Its original purpose is affiliate marketing spam - every so often it tweets links to buy crappy ebooks, and the random text was presumably originally an attempt to evade spam detection. The fact that the amusement it provides is entirely accidental in origin (we don't even know for sure whether the operator knows about it now) only adds to the brilliance. (This is briefly mentioned in the article, but since you asked...)
_ZeD_ 13y ago
half joking... have a look at http://www.polygen.org http://www.polygen.org. AFAIK it's only in Italian, but it can generate pretty much anything with a little grammar... -- EDIT -- on http://www.polygen.org/it/grammatiche/tecnologie/eng/manager.grm http://www.polygen.org/it/grammatiche/tecnologie/eng/manager... you can find an online example in English. I found ff cache the page on refresh instead of reloading it, so if you want other generated phrases you need to fool the browser appending some gibberish in get (à la ?foo=1234)
teraflop 13y ago
As the name vaguely alludes to, a lot of the text is extracted from ebooks, including public domain ones from Project Gutenberg. That said, I find it hard to believe that just picking random sentence fragments would give such a high signal-to-noise ratio, so there's probably something else going on. If I were the one pulling the strings of Horse_ebooks, I would probably focus on building a statistical model that uses the number of favorites per tweet as its training data, and looks for linguistic features that are highly correlated with getting lots of favorites.
- eshvk 13y ago
  > If I were the one pulling the strings of Horse_ebooks, I would probably focus on building a statistical model that uses the number of favorites per tweet as its training data, and looks for linguistic features that are highly correlated with getting lots of favorites. Yeah, but the problem in that approach is that you will always hit the cold start problem. What happens at the beginning when there is nothing? And no one likes you? If this was a robot, I can't imagine significant time was spent initially on custom designing these tweets.

by eshvk 13y ago

tessierashpool 13y ago

I only have half an answer: Markov chains. It's probably going through text and using Markov chains to create "similar" text. Then question then becomes what text to go through, how long the Markov chains are, how you vary them, etc. Not long ago I took a workshop on algorithmic music where the lead professor had published a book of haiku. The vast majority of the haiku were algorithmically generated via Markov chains, while a very slim minority were actual haiku from 17th century Japan (iirc) translated into English. Guessing which haiku were "real" and which algorithmic was very hard to do.

veidr 13y ago
Do you happen to have more information on that book?

stdbrouw 13y ago

Some ideas from the guy who created @nytimes_ebooks: http://harrisj.tumblr.com/post/23737140672/nytimes-ebooks http://harrisj.tumblr.com/post/23737140672/nytimes-ebooks

simonw 13y ago

Have you played around with markov chains? They're really easy to build and produce very entertaining results - a few years ago I tried feeding all of the SXSW Interactive session titles in to one and got results like "Participatory budgeting crowdsourcing for real time marketing growing a digital culture".

d23 13y ago
Can you suggest some good resources for newbies/laymen on the topic? Edit - For anyone else who might be looking, this page gives a nice pythonic example: http://agiliq.com/blog/2009/06/generating-pseudo-random-text-with-markov-chains-u/ http://agiliq.com/blog/2009/06/generating-pseudo-random-text...
- Groxx 13y ago
  Well, as a very rough overview, all they are is: * read in words * keep track of the frequency of "wordA wordB", "wordB wordC", etc. * use that frequency, e.g.: { "wordA" : {"wordB" : 0.5, "wordC" : 0.25, ...}, "wordB" : {"wordC" : 0.75, ...} } combined with a word, say "wordA", and pick a random word from the frequencies, weighted by frequency. * your word is now the word you just picked. repeat that last step until you get bored. You can go 'up' and 'down' in how many layers of frequency you want to collect, like the frequency of individual word pairs, vs word triples {"wordA wordB" : {"wordC" : 0.25}, etc}, vs sentences, anything. The more you collect, the more 'real' your generation will tend to be, but it takes more data to train it effectively since you want lots of possibilities for each 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' so it doesn't repeat itself. Otherwise it might think that the only thing that comes after 'key' is a single phrase, so it keeps selecting it and it doesn't sound like something anyone would actually say.

comex 13y ago

Its original purpose is affiliate marketing spam - every so often it tweets links to buy crappy ebooks, and the random text was presumably originally an attempt to evade spam detection. The fact that the amusement it provides is entirely accidental in origin (we don't even know for sure whether the operator knows about it now) only adds to the brilliance. (This is briefly mentioned in the article, but since you asked...)

_ZeD_ 13y ago

half joking... have a look at http://www.polygen.org http://www.polygen.org. AFAIK it's only in Italian, but it can generate pretty much anything with a little grammar... -- EDIT -- on http://www.polygen.org/it/grammatiche/tecnologie/eng/manager.grm http://www.polygen.org/it/grammatiche/tecnologie/eng/manager... you can find an online example in English. I found ff cache the page on refresh instead of reloading it, so if you want other generated phrases you need to fool the browser appending some gibberish in get (à la ?foo=1234)

teraflop 13y ago

As the name vaguely alludes to, a lot of the text is extracted from ebooks, including public domain ones from Project Gutenberg. That said, I find it hard to believe that just picking random sentence fragments would give such a high signal-to-noise ratio, so there's probably something else going on. If I were the one pulling the strings of Horse_ebooks, I would probably focus on building a statistical model that uses the number of favorites per tweet as its training data, and looks for linguistic features that are highly correlated with getting lots of favorites.

eshvk 13y ago
> If I were the one pulling the strings of Horse_ebooks, I would probably focus on building a statistical model that uses the number of favorites per tweet as its training data, and looks for linguistic features that are highly correlated with getting lots of favorites. Yeah, but the problem in that approach is that you will always hit the cold start problem. What happens at the beginning when there is nothing? And no one likes you? If this was a robot, I can't imagine significant time was spent initially on custom designing these tweets.