This is a small project related to a data mining course given by Dr. Alexandros Kalousis at the University of Geneva, Department of Computer Science.

Project goals are detailed in doc/project-goals.pdf.

Briefly, the goal is to collect and analyze data gotten from the Twitter microblogging system and build a model to predict if a tweet will be retweeted or not, i.e. to predict if a tweet can be considered as popular or not insight a community.

This project uses python 2.7 thus you obviously need it. pip and virtualenv are the recommended way to install all the required libraries in order to keep everything clean. To build the required packages, you also need gcc (or another C-compiler), gcc-fortran (or another Fortran compiler) and swig. Once the requirements are met, you can process as follow:

• Set up a virtual environment first: virtualenv -p python2.7 env
• And activate it: source env/bin/activate
• Then, to install the required libraries, issue the following commands: pip install numpy pip install -r requirements.txt

And you should be all set. Note that numpy needs to be installed beforehand otherwise matplotlib and scipy will not build.

Create a twitter app developper account. Then, rename scripts/credentials.py.sample to scripts/credentials.py and modify the credential values for valid ones.

You can adapt search tag, language and output json file from scripts/fetch_data.py and then run the file to fetch a dataset.

Remember that, as of Twitter API v1.1, API is limited and you will only be able to fetch a certain amount of data at once. Run the script several times on a several days period if you want to fetch enough data. Also remember that using free Twitter API, you can only fetch information in from a limited time period and not trough all Twitter history. You gotta pay for it if you want more... :(