Abstract

Citations are scientists’ tools for grounding their innovations and findings in the existing collective knowledge. However, not all citations are semantically identical. Scientists use citations at different parts of their work to convey precise information. Hence, to understand scientific documents best, it is crucial for machines to recognize the intent behind each citation. Current state-of-the-art methods rely on contextual sentences surrounding each citation to classify the intent. In the absence of the actual content, these approaches become unusable. In this work, we propose a text-free citation intent classification method. The proposed method uses a knowledge graph built on top of the SciCite dataset to extract citation information for publications and learn to predict citation intent. We study this problem in both transductive and inductive settings. Our experimental results show that we can achieve a comparable macro F1 score to word embedding contentbased methods by only relying on a knowledge graph. Specifically, we achieve macro F1 scores of 62.16 and 59.81 in the transductive and inductive settings, respectively, on the linklevel SciCite dataset.

Date

March 15, 2026

Authors

Xinwei Du, Kian Ahrabian, Arun Baalaaji Sankar Ananthan, Richard Delwin Myloth, Jay Pujara