MIT License. Copyright (c) 2016 Jeffrey Alan Meyers. See LICENSE.md

This project contains a command line utility to log tweets from the Twitter Streaming API and load them into Amazon S3 Buckets. You can also download Geofabrik OpenStreetMap extracts, filter out Point-of-Interest nodes and load them data S3 Buckets. The OSM and Twitter log files in S3 are used as input for Elastic MapReduce jobs.

Example output after processing 2GB of tweets and the entire USA OSM extract with the poi_nearby_osm.py MapReduce job. (metro area, number of tweets nearby, point-of-interest)

Dependencies

apt-get update
apt-get install python-dev libgeos-dev libspatialindex-dev \
        build-essential protobuf-compiler libprotobuf-dev
# if you don't already have pip
curl https://bootstrap.pypa.io/get-pip.py | python

Install geotweet command line utility

pip install geotweet

Installing this package will provide you with a python executable named geotweet.

geotweet stream|load|osm [options]
geotweet stream --help                  # store Twitter Streaming API output to log files
geotweet load --help                    # load log files to S3 bucket
geotweet osm --help                     # download osm extracts from geofabrik
                                        # extract POI nodes and load into S3 bucket

Store geograhpic tweets from Twitter Streaming API into --log-dir

usage: geotweet stream [-h] [--log-dir LOG_DIR] [--log-interval LOG_INTERVAL]
                       [--bbox BBOX]

optional arguments:
  -h, --help            show this help message and exit
  --log-dir LOG_DIR     Path to log file directory
  --log-interval LOG_INTERVAL
                        Minutes in each log file
  --bbox BBOX           Bounding Box as 'SW,NE' using 'Lon,Lat' for each
                        point.

Listen for archived files in --log-dir and upload to S3 bucket

usage: geotweet load [-h] [--log-dir LOG_DIR] [--bucket BUCKET]
                     [--region REGION]

optional arguments:
  -h, --help         show this help message and exit
  --log-dir LOG_DIR  Path to log file directory
  --bucket BUCKET    AWS S3 Bucket name
  --region REGION    AWS S3 Region such as 'us-west-2'

Download OSM extracts from GeoFabrik, extract POI nodes and load to S3 Bucket

usage: geotweet osm [-h] [--output OUTPUT] [--states STATES] [--bucket BUCKET]
                    [--region REGION]

optional arguments:
  -h, --help       show this help message and exit
  --output OUTPUT  Location of output files (default=/tmp)
  --states STATES  File containing list of states to download and load
  --bucket BUCKET  AWS S3 Bucket name
  --region REGION  AWS S3 Region such as 'us-west-2'

For geotweet stream the following environment variables must be set. See example_conf/stream-envvars.sh for all options.

• TWITTER_CONSUMER_KEY
• TWITTER_CONSUMER_SECRET
• TWITTER_ACCESS_TOKEN_KEY
• TWITTER_ACCESS_TOKEN_SECRET

For geotweet load|osm See example_conf/load-envvars.sh for all options.

• AWS_ACCESS_KEY_ID
• AWS_SECRET_ACCESS_KEY
• AWS_BUCKET (if not provided as cli param)
• AWS_DEFAULT_REGION (if not provided as cli param)

pip install geotweet

# Twitter
export TWITTER_CONSUMER_KEY="..."
export TWITTER_CONSUMER_SECRET="..."
export TWITTER_ACCESS_TOKEN_KEY="..."
export TWITTER_ACCESS_TOKEN_SECRET="..."

# AWS
export AWS_ACCESS_KEY_ID="..."
export AWS_SECRET_ACCESS_KEY="..."

# start streaming to log files rotate log file every 5 minutes
geotweet stream --log-dir /tmp/geotweet --log-interval 5

# Open and new terminal

# start loading archived log rotations to S3
geotweet load --log-dir /tmp/geotweet --bucket already.created.bucket --region us-west-2

# To download and load OSM data to S3
echo -e "Oregon\nWashington" > states.txt
geotweet osm --bucket already.created.bucket --region us-west-2 --states states.txt

To run as daemon on Ubuntu with Upstart copy example_conf/geotweet-stream.conf and example_conf/geotweet-load.conf to /etc/init and set the environment variables in those files then run:

sudo service geotweet-stream start
sudo service geotweet-load start

Run geotweet stream

Python script running on a cheap VPS (DigitalOcean) will connect to the Twitter Streaming API and filter for tweets inside Continental US.

For each tweet (if Lat-Lon coordinates are provided), extract fields, marshal as JSON and append to a log file. The log files are rotated every 60 minutes.

Example of log entry (1 line with pretty print)

{
   "source" : "<a href=\"http://www.tweet3po.org\" rel=\"nofollow\">Tweet3po</a>",
   "followers_count" : 959,
   "screen_name" : "Tweet3po",
   "tweet_id" : "712897292534087681",
   "friends_count" : 5,
   "location" : "Orlando, FL",
   "timestamp" : "1458802934188",
   "text" : "#HouseBusinessCheck 1750 Mills Ave N 32803 (3/24 02:45) #Orlando #LakeFormosa",
   "created_at" : "Thu Mar 24 07:02:14 +0000 2016",
   "user_id" : 56266341,
   "description" : "Hyper-Local Neighborhood News.",
   "name" : "Tweet3po",
   "lonlat" : [
      -81.36450067,
      28.56774084
   ]
}

Run geotweet load

Listen for log file rotations. Each archived file will be uploaded into an Amazon S3 Bucket.

After log files have been collected for long enough run some MapReduce jobs

git clone https://github.com/meyersj/geotweet.git
cd geotweet
virtualenv env
./env/bin/pip install -r requirements.txt         # plus required system packages

Run MapReduce jobs

cd /path/to/geotweet/bin

MapReduce Word Count from tweets broken down by US, State and County (cached spatial lookup using Shapely/Rtree)

./mrjob_runner state-county-words

MapReduce Word Count of tweets broken down by Metro areas (cached spatial lookup using Shapely/Rtree). Final results will be persisted to MongoDB if GEOTWEET_MONGODB_URI is set to a valid uri.

export GEOTWEET_MONGODB_URI="mongodb://127.0.0.1:27017"
./mrjob_runner metro-words

Output stored in MongoDB db=geotweet as collection=metro_word as documents

{
    metro_area:   "Portland, OR--WA",
    word:         "beautiful",
    count:        142
}

Input is log of geographic tweets and points-of-interest extracted from OSM. For each input record look up metro area and emit data using metro as key

In reduce for each metro area, build index of points-of-interest and do spatial search for nearby POI's for each tweet and emit count for each nearby POI.

Final results will be persisted to MongoDB if GEOTWEET_MONGODB_URI is set to a valid uri.

export GEOTWEET_MONGODB_URI="mongodb://127.0.0.1:27017"
./mrjob_runner poi-nearby

Output stored in MongoDB db=geotweet as collection=metro_osm as documents

{
    metro_area:  "Portland, OR--WA",
    poi:         "Stumptown",
    count:        124
}

Run EMR Job

First build a package that will be used to bootstrap the hadoop nodes

cd /path/to/repo/geotweet

# build and store package in $PWD/dist
python setup.py sdist   

Set all of the required config parameters, set all paths

cp example_conf/mrjob.conf ~/.mrjob.conf
vim ~/.mrjob.conf  

Configure the job with correct input and output buckets

cd /path/to/geotweet/bin
vim emrjob_runner       # make sure you set the `src` and `dst` S3 buckets

./emrjob_runner state-county-words
./emrjob_runner metro-words
./emrjob_runner poi-nearby

Tests available to run after cloning and installing dependencies.

nosetests geotweet/tests/unit/* 

# requires environment variables specified above to be set
# and MongoDB instance running locally
nosetests geotweet/tests/integration/*  

To build a local virtual machine with MongoDB you need virtualbox/vagrant installed and a ubuntu/trusty64 box

git clone https://github.com/meyersj/geotweet.git
cd geotweet
vagrant box add ubuntu/trusty64
vagrant up
vagrant ssh
cd /vagrant/bin

# run mapreduce job
./mrjob_runner poi-nearby

# connect to mongo shell and query results
mongo geotweet
db.metro_osm.find({
    metro_area: "Portland, OR--WA"
}).sort({count:-1})