Automatically breaking down scientific abstracts into subcategories, with little training data

What is it about?

There has been research into breaking down scientific abstracts into subcategories, with supervised learning. This, like most deep learning approaches, requires a vast amount of training data. Such data is available for biomedical abstracts but not other domains. This work presents a transfer learning-based approach that used the available biomedical abstracts with a small corpus of user labeled computer science abstracts, to generate automatic segmentation for computer science abstracts

Featured Image

Photo by CHUTTERSNAP on Unsplash

Photo by CHUTTERSNAP on Unsplash

Why is it important?

Our approach shows a path beyond the need for large training corpora. Additionally, in the course of our investigation, we discovered that the prescribed approach tends to generalize better over outright training with labeled examples.