Enhancing Unsupervised Keyword Extraction in Academic Papers through Integrating Highlights with Abstract

Researchers Yi Xiang and Chengzhi Zhang have published a paper demonstrating a significant improvement in automated keyword extraction for academic papers. Their method, detailed in the arXiv preprint "Enhancing Unsupervised Keyword Extraction in Academic Papers through Integrating Highlights with Abstract," addresses a core task in information retrieval. While previous systems primarily relied on a paper's abstract and references, the team identified the often-overlooked 'highlights' section as a rich source of keyword data. This section, which concisely lists key findings, complements the broader context of the abstract.

The study systematically evaluated three input scenarios: abstract-only, highlights-only, and a combined abstract-highlights approach. Experiments were conducted using four different unsupervised extraction models on datasets from Computer Science (CS) and Library and Information Science (LIS). The results were clear: integrating the abstract with the highlights section led to a marked and consistent improvement in keyword extraction performance across the board. The researchers also analyzed the differences in keyword coverage between the two text sources to understand how their variations influence the final output. The associated code and data have been made publicly available, allowing for immediate validation and application by other developers and researchers in the field.