This work was done during my internship at the MSR Cambridge Machine Learning and Perception Group with Yoram Bachrach.
Below we present the relations between Twitter user interests and predicted online personalities using regression coefficients estimated over 10,000 Twitter user profiles. We show that some interests correlate with one attribute value (shown in red or "+") and some with an opposite value (shown in blue or "-"). For example, fashion interest category strongly correlates with the female attribute but gaming and sports with the male attribute, family correlates with users predicted to have the users predicted to have children, health and travel with the users predicted to have higher annual income.
We also show a dendrogram for attributes (rows) and interests (columns). It groups data based on row and column similarities using a hierarchical clustering algorithm. We observe that life satisfaction and optimism attributes, intelligence and education attributes are the most similar; television and entertainment, gaming and sports interest categories are quite similar too.