This document summarizes a research project on modeling user interests on Twitter using a graph-based approach. The project aims to predict a user's interest profile based on the interests of the users they follow on Twitter. Various graph features are explored as weighting schemes to calculate the influence of followers on a user's interests. Experimental results show that features based on retweets and mentions perform best at predicting interests, with F1 scores around 0.6. A composite model is also proposed that combines predictions from different weighting schemes using learned quality scores. Additionally, a machine learning model is trained to predict interests directly from graph features.