Say hello to @LimerickTweets.  It’s a bot that retweets some of the latest tweets in the form of limericks.

I'm not a fan of facial hair

oh okay cool see if I care

— Sophie (@__sophs) March 1, 2015

I only live twice

— JesusChrist (@FakeShemetov) February 27, 2015

I'm craving fried rice.

— ϻ.thor (@trvpmt) February 27, 2015

I respect his intentions there.

— Nino (@NinoBeige) March 1, 2015

I’ve been interested in generating limericks for a couple years now.  When I was new to programming and NLP concepts, it was one of the first things I thought about doing.  After all, limericks are pretty easy to think about computationally: you have a just need to identify three 8/9-syllable phrases that rhyme and a 5-syllable rhyming couplet — with thousands of lines of data, these wouldn’t be all that hard to find.  It also seemed like the results would be interesting!  Short tweets, the kind that would be only 5 or 8 syllables, are often ambiguous, emotive and poetic.  it seemed like stringing several together might result in a piece of poetry that was able to suggest real, genuine sentiment.

Naturally, it seemed like a limerick bot (particularly a twitter limerick bot)  would be an obvious and overdone project. I checked, however, and while there are a couple of GitHub accounts with “work in progress” bots, I couldn’t actually find one that had made it to the web.  It seemed like the perfect opportunity to build one myself and see what happened.

Implementation

I built my bot in Python, using the Twitter Temboo API to scrape recent tweets.  It quickly accumulated several thousand and updates regularly.  Using NLTK and the CMU Pronouncing Dictionary (of course), I assigned each one a syllable count and looked up its pronunciation so that I could find rhymes.  Finding the best set of three words that rhyme with one another is actually kind of an interesting problem.  How do you figure out which three words in a dataset of one thousand rhyme best?  My solution was to hash tweets into buckets based on syllable-count and the pronunciation of the last word in the tweet, starting with the first vowel of the word.  For instance, “cat” translated to “K AE T” which hashed into the “AE T” bucket — just like “sat” (“S AE T”) or “drat” (“D R AE T”).  It wasn’t going to match longer words to rhyme with the ends of sentences, like “go in” and “cohen”, but I felt really dirty with every bulky data structure I thought of.  There are a lot of tweets, the words used in them aren’t big, and so I wasn’t particularly worried about not matching every possible rhyme.  To post, the program would choose one 5-syllable bucket with at least two tweets that end in distinct words (we group all posts that end in the same word together so we don’t accidentally try to rhyme a word with itself), and, similarly, one 8-syllable bucket containing at least 3 tweets.  These tweets are then posted in limerick order and deleted from the word list.

Some other interesting problems included working with decimal numbers and unicode.  I wanted “#YOLO #420” to both have the correct number of syllables and rhyme with “twenty” (“scenty”?), but the CMU rhyming dictionary doesn’t translate straight numbers.  My solution was to replace every instance of a decimal number with a handcrafted verbal representation when finding rhymes.  It’s a little hacky, but it does the job.

Posting Tweets

LimerickTweets posts its limericks by retweeting the original author of the tweet in the order of the limerick.  This is done both to add some legitimacy to the tweets — it’s proof I’m not making them up — and to make the bot engage with the world around it.  By retweeting the original poster, it creates a dialogue with them and their friends, inviting the user to see what their words have become a part of.  In order to comply with Twitter logistics, each tweet is used only once. The program scrapes for new tweets just before it posts.

Since it is frequently checking for new tweets that match a certain pattern, you might call this a “watcher” bot using Michael Cook’s nomenclature.  It is only generative insofar as it’s given content.

Results

The bot has been active for about a week now, and has tweeted about 100 retweet limericks.  It maintains about six followers (usually from among those whose posts were retweeted), some of whom engage with the limericks by retweeting their favorites.  One woman began replying with names for the limericks, and sending words of comfort to posters who were complaining about their lives.  I’ll be interested in seeing how the world reacts with time and visibility.

The results themselves are spectacular, in my humble opinion.  They rhyme well, the poetry they create is culturally rich and meaningful, and it even has managed to capture some of the trending topics (as it turns out, a lot of things rhyme with “dress”, “blue”, “black” and “gold”).  While not the most novel thing in the entire world, I think LimerickTweets contributes a stylized spotlight to posters and uses their sentiments to craft novel, reactive art.

Some Additional Thoughts

A week or so back, I mentioned this assignment to one of my directing professors, who is largely unfamiliar with tech and computation.  He was exceptionally confused by the notion of data scraping, and was left without a lot of words when I asked what he might do with a bot and a large repo of tweets to generate culture.  He thought about it for a while before saying “the thing you have to remember is that, what makes drama unique is its intimacy.  Many art forms like to generalize, to give some abstract overview of the human condition.  Drama tries to take one small instance of the human condition, and bring its audience in very intimate contact with it.  If you want your work to be dramatic, you should keep this in mind.”

It’s valid for a bot or an app to take in thousands of lines of data and condense it into something a person can visualize and understand.  I feel like this is generally what we think of when we think of this sort of data processing.  But perhaps an app influenced by the realm of theatre isn’t going to do this.  Instead, it’s the job of the theatre to highlight single people and their plights in a stylized way that allows us to understand them differently, if not better.  Perhaps it is more “dramatic” for an app to showcase individual tweets in an artistic framework than to try and make a statement about the thousands of tweets it searched through to find them.  I’m interested in further exploring this notion of maintaining intimacy alongside large datasets.

The Other Bot

Since I felt like perhaps this bot wasn’t the most interesting and original thing on earth, I decided to make another one.  However, I’ve yet to set up a twitter account for this one.  Each account needs a separate phone number in order to access the API?  How is that supposed to scale?

This second “bot” instead lives on YouTube, and its purpose is to upload generative montages.  It scrapes random vine videos from Twitter, stitches them together with slow fades, and adds slow motion and a dramatic soundtrack before posting the result to YouTube.  It does this by making heavy, abusive use of the Python subprocess library and the ffmpeg on the machine.

Running out of my monthly Temboo limit, I didn’t end up posting a large number of videos.  But you can see them on the YouTube channel I made for my infovis project:

My bot is called @secretsofgreek and its URL is https://twitter.com/secretsofgreek. Ever wonder the secrets behind Greek Letters? Secrets of Greek Org is a bot that discovers secret meanings to Greek Organizations.

mu xi psi: This organization secretly means…. musicology xanthoma prostration

— Secrets of Greek Org (@secretsofgreek) March 3, 2015

My bot was created using twitter. I was apart of a greek organization while attending undergrad, I always found it fascinating that each organization had their own ritual and secret meaning that had been passed down for over 100 years. Some have been “supposedly” leaked while others are still unknown. My goal was to create a bot the explored the secret meanings to greek organizations as a way to unveil the secrets. Most organizations are written in a different language and dont translate directly to an English word. Some of the times I have been surprised as to the formation of the words, but other times it makes absolutely no sense. If I had programed the bot based on adjectives, verbs and nouns it would allow for more sense to be made of secrets. But I appreciate the randomness as you never know when you will recognize a word or what new “secret’ you will learn. The post above made me laugh as it refers to studying music, while having irregular yellow patches of skin and lying down on the ground. When they sound funny, theyre more comical, but if you look up the definitions and piece it together thats when its even more enjoyable.

pi sigma tau: This organization secretly means…. propane staged tumbleweed

— Secrets of Greek Org (@secretsofgreek) March 3, 2015

I also found this tweet to be pretty funny as it declares pi sigma tau to be propane staged tumbleweed. But this is an example of more insync grammatical context I was referring to above instead of having 3 nouns together or 3 verbs, etc.

As I have recently posted this I have not had anyone follow my bot, but my audience is anyone interested in greek organizations. People in the past have asked me what my organization meant and it was nice to have a secret to keep, but I always enjoyed when people tried to guess. This bot stems from that experience, and other people in greek organizations have related similar experiences to me. If the tweets are comical people will enjoy it. According to  Michael Cook’s taxonomies, my bot would fall under the Mash-Ups of greek letters and dictionary words. 

You can find the code here: https://github.com/amyfried/secretofgreeksbot

As my secondary 3X assignment, my Twitter bot creates new corporate mission statements; its Twitter handle is @OurMissionStmt and its associated URL is https://twitter.com/OurMissionStmt

The bot uses context free grammar text generation to produce company mission statements based on a list (I used a list found at http://www.joe-ks.com/archives_feb2001/ManualMSG.htm as basis and added some others that I’ve heard) of corporate-y adverbs, adjectives, nouns, adjectives, and verbs. Working on the Dilbert data scraping assignment led me to stay with the corporate theme of silly buzz words and often useless jargon that I’ve heard in office environments in which I’ve worked as an engineer. The bot was written in Python, using the contextfree module to create the mission statement and the twython wrapper for the Twitter API. The source words were created in a dictionary JSON file that is read by the Python script.

Some of the ones that the script generated were quite amusing to me, because they are both non-sensical and yet ones that could be imagined hanging as a banner above some office space somewhere. My bot has not been running for very long, and I am the only follower for my bot so far, so its tweets have not yet been responded to or favorited. This bot falls under the “Feeds” taxonomy of Michael Cook’s taxonomies, since it generates text content as a continuous stream. As far as who might be interested in such silliness, I suppose anyone who’s worked in a typical cubicle environment and appreciates the absurdity of corporate double-speak would find this bot mildly interesting/amusing. “We strive to collaboratively coordinate orthogonal meta-services while encouraging personal employee growth” or “To synergistically enhance error-free mindshare to set us apart” are a couple that I found to be quite absurd. A few other screenshots from this bot are included below.

For my twitter bot I decided to both poke fun at consumerism and create an ode to the absurdity of the breadth of the amazon product inventory. My bot tweets two random top amazon products in juxtaposition to each other, urging the viewer to buy one or the other. To do this, I scraped the names of the top 500 selling amazon products and created a python script to select 2 products at random and post them to twitter. @this_or_a_that

I wish there was a way to embed url links into the product name for easy purchasing. I also wish I could have figured out how to run this script automatically every hour using cron.

Both images are google image results for amazon ad.

Here’s some funny letters and numbers:

#Rachel Ciavarella IACD spring 2015 CMU
#twitter bot assignment

from twython import Twython
import csv
import random
from random import randrange

twitter = Twython("secret stuff",
"secret stuff")

#populate verb list from file (fill your cart like it's the void in your soul)
f = open('amazon_data.csv','rU')
u = open('amazon_data.csv','rU')

csv_f = csv.reader(f)
csv_u = csv.reader(u)

product_list = []
url_list = []

for row in csv_f:
product_list.append(row[1])

for row in csv_u:
url_list.append(row[0])

#random product from list (pull some products at random off the shelf, you consumerist)
random_product_1 = randrange(0,len(product_list))
random_product_2 = randrange(0,len(product_list))

#print random product (what's that you got in your cart??)
print product_list[random_product_1]
print url_list[random_product_1]

print product_list[random_product_2]
print url_list[random_product_2]

#post two products to twitter(tell the people what they want to buy)
twitter.update_status(status=product_list[random_product_1]+ ' or ' +product_list[random_product_2])