“The Street Preacher” - A Hyper-Local Twitter Bot

I walk through Yonge & Dundas Square in Toronto every day.

So. Many. People.

That intersection, which some call Toronto’s equivalent of Times Square, has a large number of street preachers. Loud, startling, obnoxious people that yell warnings of doom or urge repentance. Silly people.

I decided to use Twitter’s real-time streaming API to make an extremely specific location-based Twitter bot. The purpose? To respond to you if you tweet near the street preachers at Yonge & Dundas, with similar messages. Call it art, or a statement about society, or making fun of those preachers, whatever - I call it a fun technical and social experiment.

Using an excellent ArsTechnica article as a guide, I created a quick Python script that watches the Twitter stream for a given area, and replies to tweets in a very specific location. (±10 meters or so, by my guess.) If you’re one of the lucky few to tweet within those bounds, you’ll get a reply from @yonge_dundas:


A day later, I decided to clean up the script (rewrite it in Ruby, too) and open-source it. Well, here it is, in a quick Github gist:

# "The Street Preacher"
# hyper-local twitter bot
# by Peter Sobot (psobot.com)
# November 29, 2011
# ---------------------------
# Instructions:
#   Place phrases in phrases.txt (one per line)
#   Set a latitude and longitude (@from_lat, @from_lng)
#   Set a radius (currently in degrees lat/long)
#   Set your twitter account's username (to prevent feedback)
#   Enter your Twitter application keys and OAuth credentials
#     (get them from https://dev.twitter.com/)
#   Run:
#     `ruby preacher.rb start`
#   ???
#   Profit! (Not really.)
# ---------------------------
# Defaults are set to Yonge & Dundas Square, Toronto.
# @yonge_dundas is a twitter clone of the notorious street preachers
# that live at that intersection.
# ---------------------------

require 'rubygems'
require 'tweetstream'
require 'twitter'
require 'logger'

PHRASES = Dir.pwd + '/phrases.txt'
EXCLUDED_USERS = Dir.pwd + '/excluded_users.txt'
USER_TWEET_TIME_LIMIT = 10800 #in seconds, how often is too often? 3 hours.
ERROR_TIMEOUT = 300           #in seconds, how long do we wait if Twitter barks at us?

# Set this to your account's username, so it doesn't feedback loop.
USERNAME = 'yonge_dundas'

@logger = Logger.new STDERR

# Twitter phrases:
@phrases = IO.readlines(PHRASES).collect{|p|p.chomp}.compact.reject{|n|n.empty?}
def random_phrase
  @phrases.sort_by{ rand }.first

def excluded_users
  IO.readlines(EXCLUDED_USERS).collect{|p|p.chomp}.compact.reject{|n|n.empty?} + [USERNAME]

# Let's store a list of people and times we've tweeted at them, to avoid spam
@user_cache = {}

def hit_recently user_id
  !@user_cache[user_id].nil? && (Time.now - @user_cache[user_id]) < USER_TWEET_TIME_LIMIT

# Tweet from:
@from_lat = 43.65641564830964
@from_lng = -79.38105940818787
radius = 0.001 # catch area in degrees lat/lng

consumer_key = "your_twitter_consumer_key_here"
consumer_secret = "your_twitter_consumer_secret_here"
oauth_token = "your_oauth_token_here"
oauth_token_secret = "your_oauth_token_secret"

# Confiruationses
Twitter.configure do |config|
  config.consumer_key = consumer_key
  config.consumer_secret = consumer_secret
  config.oauth_token = oauth_token
  config.oauth_token_secret = oauth_token_secret

TweetStream.configure do |config|
  config.consumer_key = consumer_key
  config.consumer_secret = consumer_secret
  config.oauth_token = oauth_token
  config.oauth_token_secret = oauth_token_secret
  config.auth_method = :oauth
  config.parser = :yajl

# Let's make us a bounding box to give Twitter's streaming API
N = @from_lat + radius
S = @from_lat - radius
E = @from_lng + radius
W = @from_lng - radius

def parse_tweet status
  return if excluded_users.include? status[:user][:screen_name] 

  if status[:coordinates] and status[:coordinates][:type] == 'Point'
    lng, lat = status[:coordinates][:coordinates]

    if    lng < [E, W].max \
      and lng > [E, W].min \
      and lat < [N, S].max \
      and lat > [N, S].min

      @logger.info "Got one! Replying to @#{status[:user][:screen_name]}:"
      @logger.info "\t#{status[:id]}: \"#{status[:text]}\""

      if not status[:in_reply_to_user_id] \
        and not status[:retweeted] \
        and status[:entities][:user_mentions].empty? \
        and not hit_recently(status[:user][:id])

        tweet = Twitter.update(
          "@#{status[:user][:screen_name]} #{random_phrase}",
          :in_reply_to_status_id => status[:id],
          :lat => @from_lat,
          :long => @from_lng,
          :display_coordinates => true

        @user_cache[status[:user][:id]] = Time.now

        @logger.info "\t#{tweet[:id]}: \"#{tweet[:text]}\""
        @logger.info "Didn't reply - tweet was mention, retweet, reply, or spammy."
        @logger.info "In reply to: " + status[:in_reply_to_user_id].inspect
        @logger.info "Retweeted? " + status[:retweeted].inspect
        @logger.info "Mentioned: " + status[:entities][:user_mentions].inspect
        @logger.info "User last hit at: " + @user_cache[status[:user][:id]].inspect

      km_away = Math.sqrt(((lat - @from_lat) * 111)**2 + ((lng - @from_lng) * 79)**2)
      @logger.info "Tweet not within bounding box:\t#{km_away} km away."
rescue Exception => ex
  @logger.error ex.message
  @logger.error ex.backtrace.join "\n"

client = TweetStream::Daemon.new('preacher', :log_output => true)
client.on_error { |message| @logger.error message }
client.on_reconnect { |timeout, retries| @logger.error "Reconnect: timeout = #{timeout}, retries = #{retries}" }

# Start filtering based on location
  @logger.info "Starting up the Street Preacher..."
  client.locations("#{W},#{S},#{E},#{N}") { |status| parse_tweet status }
rescue HTTP::Parser::Error => ex
  # Although TweetStream should recover from
  # disconnections, it fails to do so properly.
  @logger.error "HTTP Parser error encountered - let's sleep for #{ERROR_TIMEOUT}s."
  @logger.error ex.message
  @logger.error ex.backtrace.join "\n"

Feel free to fork it, repurpose it, and do whatever! (Just keep my name at the top, if you please.)


Now read this

A Better Music Workflow?

I produce a lot of music. I don’t necessarily release a lot of it (or finish all of it) but I have at least 100 songs I consider developed enough to listen to, and about 300 other song files that are just riffs, beats and vocal ideas... Continue →