Style Transformer for Authorship Representations - STAR

This is the repository for the Style Transformer for Authorship Representations (STAR) model. We present the weights of our model here.

Also check out our github repo for STAR for replication.

Feature extraction

tokenizer = AutoTokenizer.from_pretrained('roberta-large')
model = AutoModel.from_pretrained('AIDA-UPM/star')

examples = ['My text 1', 'This is another text']

def extract_embeddings(texts):
  encoded_texts = tokenizer(texts)
  with torch.no_grad():
    style_embeddings = model(encoded_texts.input_ids,
                             attention_mask=encoded_texts.attention_mask).pooler_output
  return style_embeddings

print(extract_embeddings(examples))

Citation

@article{Huertas-Tato2023Oct,
	author = {Huertas-Tato, Javier and Martin, Alejandro and Camacho, David},
	title = {{Understanding writing style in social media with a supervised contrastively pre-trained transformer}},
	journal = {arXiv},
	year = {2023},
	month = oct,
	eprint = {2310.11081},
	doi = {10.48550/arXiv.2310.11081}
}