Twindle is a set of scripts for Twitter data analysis. The tools include a streaming API client which can store Twitter status updates into a Postgres database as well as a variety of scripts used for data housekeeping, status categorization and trend extraction.
Some typical use cases for twindle include:
- Tracking a specific topic, such as a hashtag or a set of terms, to explore the frequency of its use over time.
- Following users to mine their activity and impact on Twitter (e.g. mentions, retweets, terms used).
Twindle uses a simple Google Spreadsheet (such as the one here) to let the user define terms and users that are to be recorded from the raw Twitter stream. Users which repeatedly match the tracking criteria are automatically added to twindle's search filter to keep track of their further communications. Note that this inbound filtering is very rough and should be refined during a second-level analysis of the collected data.
Status messages are stored by twindle into a fairly simple database structure which can be queried for aggregate analysis. A set of included Python scripts can support such analysis, e.g. by importing lists of users from Twitter lists or by geo-coding user's location settings.
Some of the functions in twindle are useful only when used with elevated API access. This particularly applies to the auto-follow function, as it easily exceeds the 5000 follows included in normal API access.
Twindle is written both in CoffeeScript/NodeJS (used for stream tracking) and Python (used for offline analysis). To run, twindle will also require an instance of Postgres, and, depending on whether you want to run the application in queued mode, RabbitMQ.
For simplicity, the following instructions will assume a fresh install of Ubuntu 13.04; YMMV. To begin, we'll install the system dependencies, including Postgres and RabbitMQ:
sudo add-apt-repository ppa:chris-lea/node.js sudo apt-get update # optional: sudo apt-get install htop tmux sudo apt-get install postgresql-9.1 postgresql-server-dev-9.1 nodejs python-virtualenv git python-dev rabbitmq-server nginx s3cmd supervisor sudo npm install -g coffee-script
Next, we'll set up a Python virtual environment as the working directory for twindle. We'll just put this into the home directory of the ubuntu user:
cd /home/ubuntu virtualenv twindle cd twindle source bin/activate git clone https://github.com/pudo/twindle.git app cd app npm install pip install -r requirements.txt
After this, you need to create a database for twindle:
createuser -P twindle createdb -O twindle -E utf-8 twindle psql -f /home/ubuntu/twindle/app/schema.sql twindle
If you want to use the simple web interface that comes with twindle, you will also need to set up a reverse proxy for it like this:
sudo cp deploy/nginx.conf /etc/nginx/sites-available/twindle sudo ln -s /etc/nginx/sites-available/twindle /etc/nginx/sites-enabled/twindle sudo service nginx restart
You'll need to visit http://dev.twitter.com to set up a Twitter application for twindle. After setting up a basic application, you will be able to view your OAuth consumer credentials and to generate a pair of access credentials through the web interface.
Twindle is configured entirely via environment variables. Have a look at dotenv.tmpl for the available variables and set up the twitter credentials you have created as well as the database configuration. It makes sense to set these variables in the login .bashrc as well as in the supervisor scripts (see below).
Finally, you'll need to set a Google Spreadsheet key to seed the search filters.
You can do this by cloning the sample sheet, then updting the configuration with the new spreadsheet key and adapting the
search terms to your needs. The sheet recognizes two distinct types of search
filters: following a set of terms (
track in the
type column) or a list of
follow in the
type column). For the list of terms, words are split
on a comma. Searches are not case-sensitive and a word will also match as a
Running the tracker
The streaming API tracker can be run in two modes: as a two-process application with an intermediate queue (i.e. the reader frontend will only shuffle statuses from the API stream onto the queue, insertion into SQL is delayed) or as a single process (reader and backend are combined, queueing happens through the node.js event loop). The two programs are:
app.coffeefor the combined application, and
backend.coffeeare the two components of the queue-based script version. Note that the backend need not be running all the time, statuses which have not been persisted will be kept on the queue until a backend is available.
In either case, it makes sense to use a controlled environment for execution,
supervisor. A sample configuration file is included in
deploy/supervisor.conf.tmpl. As supervisor does not evaluate the user's
environment, you must set the configurations explicitly for the processes
managed by supervisor.
analysis/ folder contains a set of Python scripts which can be used to
further analyse the collected updates and to do some housekeeping on the database.
Makefile is included which highlights the usage for some of these scripts.
The dumpraw script will take a batch of stored tweets from the
(where they are initially stored as JSON encoded in a text field), delete them
from the database and save them to a JSON file. The
Makefile shows how this
can be used in conjunction with the
s3cmd command line utility to create a
secondary data store in an S3 bucket.
lists.py will import a Twitter list passed in as its first argument into the
lists table. Imported lists can be used to create blacklists or to track the
activities of a certain subset of users.
While Twitter does have support for statuses with location information, many
users have not activiated this function, or they are posting from devicded
without the location API. In those cases, some information with regards to
geography can be gathered from their user profiles'
location field. As this
is a plain text field, the script uses MapQuests nominatim server to perform
reverse geocoding against the OpenStreetMap database. The results are not very
precise but can serve as a first indicator as to the distribution of messages.
Incrementally perform regular expression-based filtering on the collected data. This script is not abstracted well at the moment and will require further generalization to be of wider use.
Twindle itself does not have any data export function, we're assuming that you will either use the data directly from withtin the database or export it to another format yourself. To save and repeatedly execute SQL queries against twindle, consider using a Freezefile based on the Python dataset package included in the dependencies. This can be used to store a set of queries and repeatedly execute them, e.g. via a cron job.
Copyright (c) 2013, Friedrich Lindenberg
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.