KnutJaegersberg/interpretable_word_embeddings

These word embeddings were computed using the POLAR technique to reproject 'common' word embeddings into roundabout 700 interpretable dimensions of polar opposites (i.e. good/bad). I just used their scripts here: github.com/Sandipan99/POLAR I applied those on the wikidata5m embeddings, 5 million knowledge graph embeddings (SimplE). graphvite.io/docs/latest/pretrained_model.html As the model became too huge, I further filtered it for overlap with fasttext embedding tokens.… See the full description on the dataset page: huggingface.co/datasets/KnutJaegersberg/interpretable_word_embeddings.