---
language: "en"
tags:
- bert
- regression
- pytorch
pipeline:
- text-classification
widget:
- text: "We propose a new approach, based on Transformer-based encoding, to highlight extraction. To the best of our knowledge, this is the first attempt to use transformer architectures to address automatic highlight generation. [SEP] Highlights are short sentences used to annotate scientific papers. They complement the abstract content by conveying the main result findings. To automate the process of paper annotation, highlights extraction aims at extracting from 3 to 5 paper sentences via supervised learning. Existing approaches rely on ad hoc linguistic features, which depend on the analyzed context, and apply recurrent neural networks, which are not effective in learning long-range text dependencies. This paper leverages the attention mechanism adopted in transformer models to improve the accuracy of sentence relevance estimation. Unlike existing approaches, it relies on the end-to-end training of a deep regression model. To attend patterns relevant to highlights content it also enriches sentence encodings with a section-level contextualization. The experimental results, achieved on three different benchmark datasets, show that the designed architecture is able to achieve significant performance improvements compared to the state-of-the-art."
- text: "We design a context-aware sentence-level regressor, in which the semantic similarity between candidate sentences and highlights is estimated by also attending the contextual knowledge provided by the other paper sections. [SEP] Highlights are short sentences used to annotate scientific papers. They complement the abstract content by conveying the main result findings. To automate the process of paper annotation, highlights extraction aims at extracting from 3 to 5 paper sentences via supervised learning. Existing approaches rely on ad hoc linguistic features, which depend on the analyzed context, and apply recurrent neural networks, which are not effective in learning long-range text dependencies. This paper leverages the attention mechanism adopted in transformer models to improve the accuracy of sentence relevance estimation. Unlike existing approaches, it relies on the end-to-end training of a deep regression model. To attend patterns relevant to highlights content it also enriches sentence encodings with a section-level contextualization. The experimental results, achieved on three different benchmark datasets, show that the designed architecture is able to achieve significant performance improvements compared to the state-of-the-art."
- text: "Fig. 2, Fig. 3, Fig. 4 show the effect of varying the number K of selected highlights on the extraction performance. As expected, recall values increase while increasing the number of selected highlights, whereas precision values show an opposite trend. [SEP] Highlights are short sentences used to annotate scientific papers. They complement the abstract content by conveying the main result findings. To automate the process of paper annotation, highlights extraction aims at extracting from 3 to 5 paper sentences via supervised learning. Existing approaches rely on ad hoc linguistic features, which depend on the analyzed context, and apply recurrent neural networks, which are not effective in learning long-range text dependencies. This paper leverages the attention mechanism adopted in transformer models to improve the accuracy of sentence relevance estimation. Unlike existing approaches, it relies on the end-to-end training of a deep regression model. To attend patterns relevant to highlights content it also enriches sentence encodings with a section-level contextualization. The experimental results, achieved on three different benchmark datasets, show that the designed architecture is able to achieve significant performance improvements compared to the state-of-the-art."
---

# General Information

This model is trained on journal publications of belonging to the domain: **Computer Science**.

This is an `allenai/scibert_scivocab_cased` model trained in the scientific domain. The model is trained with regression objective to estimate the relevance of a sentence according to the provided context (e.g., the abstract of the scientific paper).

The model is used in the paper 'Transformer-based highlights extraction from scientific papers' published in Knowledge-Based Systems scientific journal.
The model is able to achieve state-of-the-art performance in the task of highlights extraction from scientific papers.

Access to the full paper: [here](https://doi.org/10.1016/j.knosys.2022.109382).

# Usage:

For detailed usage please use the official repository https://github.com/MorenoLaQuatra/THExt .

# References:

If you find it useful, please cite the following paper:

```bibtex
@article{thext,
  title={Transformer-based highlights extraction from scientific papers},
  author={La Quatra, Moreno and Cagliero, Luca},
  journal={Knowledge-Based Systems},
  pages={109382},
  year={2022},
  publisher={Elsevier}
}
```