 To accurately detect the intended meaning in social media messages about wildfires, a workflow was developed that uses latent semantic analysis, cosine similarity calculation, affinity propagation, and support vector machine approach to classify Twitter messages with over 90% accuracy.