Surveillance of sentiment and affect in open source text

New topics of discussion emerge continuously as political and cultural environments shift. These topics can convey a range of ideas, stories, values, and shared experiences. In open source data, discourse is often infused with emotionally-charged prose expressed by large and diverse pools of authors. Manually identifying and analyzing the emotional content of combined discussions from millions of individuals is an impossible task for any single analyst or decision maker. Technologies that can automatically identify emerging topics of discussion and align them with public opinions can provide early warning to acts of violence or other threats [1]. We present an approach to detecting basic emotions associated with topics by combining work in topic modeling with affect analysis. We apply the correspondence LDA topic model (CorrLDA2) [2] to correlate emotional states with topics. Using this technique, topics have associated emotions and emotion categories are correlated over a corpus of text.