Predicting Tastes from Friend Relationships

In the last few years, online social networks have become an important part of people’s lives. They have made data describing the connections between people more accessible than it has ever been in the past. In light of this new technology, we can now explore the predictive power of social data. In this paper, we present a method to predict target variables that describe individuals based upon their relationships to others and knowledge of the target variables for some of those other individuals. In the next section, we do a survey of the previous research on using social graphs as predictive instruments. We then discuss in Section 3 how we apply machine learning to this problem by constructing an appropriate support vector machine. In Section 4, we describe how we obtained test data from the Facebook API[4], as well as the features and target variables that we used. In Section 5, we analyze how our algorithm performs when trained with the Facebook data using ten-fold cross validation.