Basic Guide to Creating a Markov Chain-Driven Twitter Bot
Over the past few weeks Ive been playing around with different ways of making Twitter bots (automated Twitter accounts). I dont have much to show for it yet most of my experimentation has been with @schlinkbot, which probably doesnt look too impressive at this point. But Ive learned some foundational stuff.
Anyway, in this post Im going to be going over how to use a Ruby gem called Twitter Ebooks by GitHub user mispy to make a Twitter bot. There are of course other ways of doing this, but Twitter Ebooks has some nice features. Note: Im still new to this gem/framework, so excuse any problems. But I did get the bot up and tweeting and replying.
You can read more about Twitter Ebooks at You must login or register to view this content.
Wait, What is This?
Right, so Twitter Ebooks is basically a framework written in Ruby that allows users to create their own Twitter bots. You can think of it like a Twitter bot factory. We want to use it to make one Twitter bot.
Our bot, using Twitter Ebooks, will connect to Twitter using the You must login or register to view this content. This will allow us to send tweets and listen for mentions of our bot. Actually running the bot involves us: (1) creating a twitter account for the bot, (2) registering a Twitter app for the bot, (3) creating a bot with Twitter Ebooks, then (4) running the Ruby code that runs the bot.
Twitter Ebooks, and the bots created with it, are written in a programming language called Ruby. Its whats called a Ruby gem, which well need to download in a special way in order to use.
Where Will My Bot Get Its Content From?
Twitter Ebooks uses a pseudo-Markov generator to generate text for the bot to tweet. Basically its going to mash-up text from a real Twitter account (like your personal account perhaps), and spit it back out all mashed up. Hopefully that sounds cool to you. It can be pretty poetic!
What Youll Need to Set Up a Twitter Bot This Way
Youll need a basic Ruby coding environment. I think the bare minimum would be a fresh install of XCode or Command Line Tools, which you can download from Apple, and RVM. Youll also need a code editor like Sublime Text 2 or 3. If you can install Ruby gems youre ready to roll.
Setup Step 1: Configuration
First step is to install the Twitter Ebooks Ruby gem. To do this open your Terminal and enter
. Hopefully it installs successfully. If it doesnt you probably dont have XCode or Command Line Tools set up correctly, or you dont have RVM (or an equivalent Ruby version manager) installed correctly (see above).
gem install twitter_ebooks
Now youll want to use the Unix commands
to make a folder for your bot project. When youre in the proper directory, run
(or whatever you want to call your bot). Now open the
ebooks new my_test_bot
file in your code editor. This is where you tell your bot how to act and thus will be where well do most of our coding.
This would be a good time to go create the Twitter account that will be your bot. For example, I went and made @schlinkbot. Write down the password somewhere.
Now we need to create a Twitter app. While logged in as your bot account, head over to [link]https://apps.twitter.com/[/link]. Hit the create new app button in the top right and fill out the form. It doesnt really matter what you put here, just be sure to leave Callback URL blank and agree to terms. If it asks, choose the highest level of permissions (read, write, direct messages). (Note: Since writing this blog post, I think Twitter has changed their API rules such that for an account to create a new account, it needs to have a phone number attached to it. If, like me, your one real phone number is attached to your real, non-bot account, You must login or register to view this content.may be useful here.)
Lets head over to the Permissions tab and make sure our app asks for Read, Write and Access direct messages. With the checked, hit Update Settings. Now go to the Keys and Access Tokens tab. Scroll down and click the Generate My Access Token and Token Secret button.
Cool. We now have 4 long token/secrets on this page. These 4 codes will allow our app to read and post to the bots Twitter account. The 4 codes are: consumer key (API key), consumer secret (API secret), access token, and access token secret.
Now, while keeping your browser open to that page, open
in your text editor.
By default youre given a bare-bones example, but Id suggest replacing that with my slightly more built-up example below, which I based off of the bot example in You must login or register to view this content.
NOTE: Here is a You must login or register to view this content. of the above code if thats easier for you to read/copy-and-paste.
First off, lets fill in those 4 codes. The first two go up top where it says
. The second two go down at the bottom at
. In all 4 cases, put the codes between the quotation marks.
Next, change out
to the handle of your Twitterbot. And change
to whatever real Twitter account you want your bot to imitate. This real account with populate something that Twitter Ebooks calls the model.
OK, lets save that
file and go back to the command line. Now run the following lines from within your bots directory:
ebooks archive sts10 corpus/sts10.json
ebooks consume corpus/sts10.json
This went and grabbed your real Twitter accounts tweets so that your bot can mimic it. If you have trouble, this is You must login or register to view this content. where I found this series of commands.
Now, to actually run the bot, enter
. Since we only entered the bare minimum into the
file, your bot literally does nothing at this point, but if you saw something like:
@schlinkbot: Loading model model/sts10.model
@schlinkbot: starting tweet stream
Were in good shape. To stop your bot, hit Control + c in the terminal.
Step 2: Give Your Bot Instructions on How to Act
In my example above, I give you two important lines of code:
scheduler.every '57m' do
statement = model.make_statement(140)
Using a scheduler, this code tweets a Markov-generated statement that is 140 characters long every 57 minutes. Obviously you can make it tweet more or less often by changing that
Now knowing how the model generates Markov text, lets look at how wed instruct the bot to reply to users with a Markov statement in addition to sending statements out every 57 minutes.
# Reply to a mention
statement = model.make_statment(120)
So now instead of just tweeting out the 140 characters, we are replying to whatever tweet mentioned our bot.
@ryanvailbrown The last-ditch missile plan failed, sig up their Nobel Peace Prizes.
Schlink Bot (@schlinkbot) December 20, 2014
Restarting Your Bot
As I said above, to stop your bot hit Control + c in Terminal. To start it again, navigate to the correct directory and run
Running Your Bot NOT On Your Local Machine (i.e. Semi-Permanently)
As youve probably surmised, following this guide means youll only be able to run your bot on your local machine. As in, your bot will only tweet/reply to tweets while youre running the Ruby program. To have your bot run forever, youll need to push your finished code to a cloud hosting service like [url=As youve probably surmised, following this guide means youll only be able to run your bot on your local machine. As in, your bot will only tweet/reply to tweets while youre running the Ruby program. To have your bot run forever, youll need to push your finished code to a cloud hosting service like You must login or register to view this content., as the You must login or register to view this content.suggests. I havent done this yet, mostly because I forget how Heroku works and Im still frequently playing with my
There are plenty more event handlers in the
file to play with
for example. For more ideas, check out You must login or register to view this content.