Readily observable communications found on Internet social media sites can play a prominent role in spreading information which, when accompanied by subjective statements, can indicate public sentiment and perception. A key component to understanding public opinion is extraction of the aspect toward which sentiment is directed. As a result of message size limitations, Twitter users often share their opinion on events described in linked news stories that they find interesting. Therefore, a natural language analysis of the linked news stories may provide useful information that connects the Twitter-expressed sentiment to its aspect. Our goal is to monitor sentiment towards political actors by evaluating Twitter messages with linked event code data. We introduce a novel link-following approach to automate this process and correlate sentiment-bearing Twitter messages with aspect found in connected news articles. We compare multiple topic extraction approaches based on the information provided in the event codes, including the Goldstein scale, a simple decision tree model, and spin-glass graph clustering. We find that while Goldstein scale is uncorrelated with public sentiment, graph-based event coding schemes can effectively provide useful and nuanced information about the primary topics in a Twitter dataset.
展开▼
