Research Highlight Generation with ELMo Contextual Embeddings

Abstract

With the advent of digital publishing and online databases, the volume of textual data generated by scientific research has increased exponentially. This makes it increasingly difficult for academics to keep up with new breakthroughs and synthesise important information for their own work. Abstracts have long been a standard feature of scientific papers, providing a concise summary of the paper's content and main findings. In recent years, some journals have begun to provide research highlights as an additional summary of the paper. The aim of this article is to create research highlights automatically by using various sections of a research paper as input. We employ a pointer-generator network with a coverage mechanism and pretrained ELMo contextual embeddings to generate the highlights. Our experiments shows that the proposed model outperforms several competitive models in the literature in terms of ROUGE, METEOR, BERTScore, and MoverScore metrics.