Abstract
With social media a de facto global communication channel used to disseminate news, entertainment, and one's self-revelations, the latter contains double-talk, peculiar insight, and contextual observation about real-world events. The primary objective is to propose a novel pipeline to classify a tweet as either "useful" or "not useful" by using widely-accepted Natural Language Processing (NLP) techniques, and measure the effect of such method based on the change in performance of a Geographical Information System (GIS) artifact. A 1,000 tweet sample is manually tagged and compared to an innovative social media grammar applied by a rule-based social media NLP pipeline. Evaluation underpins answering, prior to content analysis of a tweet, does a method exist to support identifying a tweet as "useful" for subsequent processing? Indeed, "useful" tweet identification via NLP returned precision of 0.9256, recall of 0.6590, and F-measure of 0.7699; consequently GIS social media processing increased 0.2194 over baseline.