What is the best way to learn text style in an article? By text style I mean special characteristics and patterns inherent to different authors/group's writing style. For-example, author attribution problem is about learning who wrote an article.

I've seen some researchers applying VAEs and GANs to text to learn the style distribution of sentences in an article (Some references: here, here, here). However, VAEs and GANs usually have worse performance in language compared to vision tasks.

My question is how to capture style information using LLMs? Is averaging the last hidden states of sentences/paragraphs based on the "max_length" attribute of the model a good approach? Is there a better way to capture style information with LLMs?



You must log in to answer this question.