CiteArXiv: A Citation-enriched Dataset and Heterogeneous Graph-based Model for Scientific Articles Summarization
Quoc-An Nguyen, Xuan-Hung Le, Thi-Minh-Thu Vu, Hoang-Quynh Le · 2025
The exponential growth of scientific publications creates challenges for researchers in processing information. To improve summarization quality, we present CiteArXiv, a large-scale citation-enriched extension of the ArXiv dataset, where each paper is paired with its citation sentences. We further propose a heterogeneous graph model that integrates internal content and external citation context, using elementary discourse units (EDUs) as extraction granularity. Experimental results show that combining internal and external knowledge enhances scientific summarization performance.