Using BERT for emotion analysis – A novel, resource constrained emotions classification approach with test on a Hungarian media corpus

Datasets and codes for the paper mentioned in the title

bert.py processes texts from etl_*.tsv

gridsearch_*.py fits 100 optimized models and saves results in easily parseable JSON-files

Files included in the repo:

1. etl_*.tsv files are prepared corpora for extracting BERT-embeddings
2. featuresfinal_*.npy are numpy files with BERT-embeddings, labels_*.npy are corresponding labels
3. v1[023]_corpus.tsv files are conventionally preprocessed, tf-idf weighted document-term matrices without the "document" column, v1[023]_labels.tsv files are corresponding labels