Learning for Microblogs with Distant Supervision: Political Forecasting with Twitter

Microblogging websites such as Twitter offer a wealth of insight into a population’s current mood. Automated approaches to identify general sentiment toward a particular topic often perform two steps: Topic Identification and Sentiment Analysis. Topic Identification first identifies tweets that are relevant to a desired topic (e.g., a politician or event), and Sentiment Analysis extracts each tweet’s attitude toward the topic. Many techniques for Topic Identification simply involve selecting tweets using a keyword search. Here, we present an approach that instead uses distant supervision to train a classifier on the tweets returned by the search. We show that distant supervision leads to improved performance in the Topic Identification task as well in the downstream Sentiment Analysis stage. We then use a system that incorporates distant supervision into both stages to analyze the sentiment toward President Obama expressed in a dataset of tweets. Our results better correlate with Gallup’s Presidential Job Approval polls than previous work. Finally, we discover a surprising baseline that outperforms previous work without a Topic Identification stage.