Markov-Chain-Based Japanese Twitter Bot

Since this project is strongly optimized for Japanese, other languages are not supported 🍣

Description

• Generate a tweet based on so-called Markov Chain from particular user's tweet history
• Sample: @yootakuti
• My Japanese article: マルコフ連鎖でTwitter Botをつくりました | takuti.me

Installation

If you want to host this bot directly on your local or server machine, you first need to install Ruby gems:

$ gem install bundle
$ bundle install

Note that this application specially depends on kusari, a gem for Japanese Markov chain.

Next, you should generate a directory ipadic/, the IPA dictionary for Japanese tokenization, as described in the Igo documentation.

Additionally, in order to connect to a Twitter account, the following environment variables need to be appropriately set:

export SCREEN_NAME=yootakuti

export CONSUMER_KEY=foo
export CONSUMER_SECRET=bar
export OAUTH_TOKEN=hoge
export OAUTH_TOKEN_SECRET=piyo

FYI: we can use direnv to flexibly configure project-specific environment variables:

$ brew install direnv
$ touch .envrc # and write the above `export` statements
$ direnv allow

Usage

Before enjoying this bot, you must download your tweet history from the Twitter setting page. The downloaded folder must be placed under /path/to/twitter-bot/data/, and the bot will use text column of data/tweets/tweets.csv. Note that this repository contains sample tweets.csv file.

Post on Twitter

After setting the environment variables, we can generate and post a markov tweet as:

$ ruby lib/post_tweet.rb

If you just want to check if a markov tweet is generated correctly, dry-run option is available.

$ ruby lib/post_tweet.rb dry-run

Hourly post by cron

Set your crontab as:

$ echo "01 * * * * /usr/local/rvm/wrappers/ruby-2.2.3/ruby /path/to/twitter-bot/lib/post_tweet.rb" > cron.txt
$ crontab cron.txt

For more detail of RVM+cron setting: RVM: Ruby Version Manager - Using Cron with RVM

Build API server

This repository implements a tiny Sinatra-based API server.

Run:

$ bundle exec foreman start # PORT=5000 by default

Eventually, http://localhost:5000/ and http://localhost:5000/tweet respectively execute lib/post_tweet.rb dry-run and lib/post_tweet.rb.

In case that you publicly build this API server, scheduling a request to /tweet would be an alternative choice to periodically post Markov-chain-based tweet.

Reply daemon

reply_daemon tracks tweets which contain SCREEN_NAME of your bot and replies to all of them:

$ ruby lib/reply_daemon.rb start

Stop the process:

$ ruby lib/reply_daemon.rb stop

Docker

You can easily setup this application as a Docker image:

$ docker build -t takuti/twitter-bot

Once the image has been created, running the scripts in container is straightforward:

$ docker run -it takuti/twitter-bot /bin/sh -c "ruby lib/post_tweet.rb"
$ docker run -it takuti/twitter-bot /bin/sh -c "ruby lib/post_tweet.rb dry-run"

By default, container automatically launches the API sever on port 5000 via bundle exec foreman start, so you can get access to http://localhost:5000/ once a container started running:

$ docker run -it -d -p 5000:5000 takuti/twitter-bot

Notice that, as long as the required environmental variables are properly set in container, http://localhost:5000/tweet also works as we expected.

Deploy on Heroku

Our Docker image enables us to make the API server public on Heroku:

$ heroku create takuti-twitter-bot
$ heroku container:push web

See https://takuti-twitter-bot.herokuapp.com/, for example.

While https://takuti-twitter-bot.herokuapp.com/tweet currently returns an error, you can make it available by configuration of variables.

Deploy on Dokku

Dokku is a Docker-powered OSS PaaS which enables you to build mini-Heroku-like platform on your own server. Similarly to the deployment on Heroku, once you have set up a server with Dokku, the API server can be easily deployed as a Docker image.

Server:

$ dokku apps:create twitter-bot
$ dokku config:set twitter-bot DOKKU_DOCKERFILE_START_CMD="bundle exec foreman start"

Note that setting a way to start running a container to DOKKU_DOCKERFILE_START_CMD is important, because Dokku directly executes a task defined as the web process in Procfile by default. Consequently, your container launches differently from what CMD defines in Dockerfile.

Local:

$ cd /path/to/takuti/twitter-bot
$ git remote add dokku [email protected]:twitter-bot
$ git push dokku master

Eventually, an "Application deployed" message shows up on your local screen with corresponding URL, and you can get access to the API server.