The first machine learning project I ever worked on was trying to predict political ideology based on someones twitter account. You can find my writeup here.
It uses directed follower information and basic linear regression to predict political leaning. As a training set is used the follower information of members of the United States Congress and their dw_nominate scores to learn weights for twitter users.
It was the final project for CS 221: Artificial Intelligence, Principals and Techniques at Stanford University.
The main point of this project was my implementation of stochastic gradient descent, which I did from scratch.