Classifying HIV risk tweets using tweets from San Diego county

This project aims to characterize and classify tweets that show users exposing HIV risk behaviour through their tweets on the social networking site Twitter. A labeled dataset obtained from doctors in UCSD’s Anti Viral Research Center (AVRC) was used as the dataset. To get a better understanding of the data collected and to build a good classification model, a series of exploratory data analysis (EDA) experiments were performed on the training dataset. The EDA phase of the project revealed information on relevant and irrelevant features as expected. Then a comparative study is performed on classification using models built using logistic regression and Support Vector Machines. We find that logistic regression trumped over Support Vector Machines when the the domain specific terms collected from domain experts were made part of the feature set.