Data and code for significance testing for the Information Sciences article on Online suicide prevention through optimized text classification.
Contains predictions on the cross-validation dataset (n=10,000) for each system after optimisation. Results are available for the two tasks (relevance and severity) and both optimisation objectives (F1 and F2). For each system, the _0 column contains the gold standard, and the _1 column contains the predictions.
IPython notebook with code to do pairwise binomial significance testing, and to generate the significance tables as used in the article.
Run from the terminal using jupyter notebook.