Tenfold cross-validation results from application of the naïve Bayes classifier for the different modes of political protest action.

Each labeled row indicates the number of positive (i.e. identified as one of the four modes of political protest participation) in the training data set (Abundance), and the F1-optimal posterior probability (Threshold) for classification, in addition to its the corresponding values of precision (P), recall (R), and combined F1. Precision refers to the probability that the trained classifier will identify a true positive result when applied to new data. Recall refers to the probability that the classifier will identify a positive result out of a sample of politively classified tweets. The F1 statistic combines precision and recall to provide an overall measure of classifier quality. Out-of-domain evaluations are presented parenthetically, adjacent to their corresponding in-domain values.