Leveraging citation intent for publication relatedness

This paper proposes a novel Intent-aware citation-based (IC) method for measuring publication relatedness using citation intents. Solely relying on citation counts and citation co-occurence patterns, traditional citation-based methods treat all citations as equally informative and ignore citation intents, which reflects the rhetorical functions behind the citation contexts, leading to many tenuous publication links and obscuring the differential impact of citation intents. To address these limitations, our method discovers the citation intents for each citation context and retains meaningful inter-paper links only when they are derived from citation contexts sharing same intent. Our method is applied across three citation-based relations including Direct citation ( $$\text {IC}_{\text {DC}}$$ ), Bibliographic coupling ( $$\text {IC}_{\text {BC}}$$ ), Co-citation ( $$\text {IC}_{\text {CC}}$$ ) and is evaluated through network clustering algorithms to assess textual coherence. Experimental results demonstrate that IC significantly outperforms traditional approaches while revealing the differential impact of citation intents. Particularly, the group of Comparison intents (Differences and Similarities) excels in most cases, with improvements of 0.0389 for $$\text {IC}_{\text {BC}}$$ (NLP dataset) and 0.0659 for $$\text {IC}_{\text {DC}}$$ (Biomedical dataset). Besides, Extends, Motivation, Uses, and Future Work also show strong potential, with Extends improving $$\text {IC}_{\text {DC}}$$ by 0.0327 on the NLP dataset, and Motivation achieving a 0.0423 gain on the Biomedical dataset. Conversely, Background exhibits the worst performance among all intents, indicating minimal contribution to publication relatedness. Beyond effectiveness in paper–paper relationship modeling, our approach has broader implications for bibliometric analysis by enabling systematic investigations of how citation intent shapes scholarly relationships.