Our codebase for our submission to the AffCon 2019 CL-AFF Shared Task. Still in progress.
csv\ contains modified versions of the dataset files provided for the competition, we will not be reuploading the original dataset.
scripts\ contains utility scripts such as our server quick setup script.
lstm_baseline_2label.py is a mutlilabel classification baseline implementation of a naive lstm taking in the default AllenNLP embeddings.
lstm_elmo.py is a baseline implementation of a naive lstm taking ELMo vectors as input and returning both labels as in
cl_aff_embedders.py contains the custom TextFieldEmbedder that can embed ELMo vectors.
simple_seq2vec.py contains a custom Seq2VecPredictor to predict the social or agency value for a single sentence.
Requires the AllenNLP library and Python 3.6 or above to run.
Unsupervised tasks, continuing characterization of performance, application on other datasets