Journal publication · 2026
Liu, J., Cheng, X., Wei, S., & Wang, G. (2026). Designing a computational framework for identifying suspicious citation groups in heterogeneous academic networks: A graph learning approach. Information Systems Frontiers, 28(4), 1651–1673. https://doi.org/10.1007/s10796-026-10770-y
Explore the original contribution
Interactive paper explorationStart
A link or highly cited paper may look legitimate when examined alone.
Reconsider
Combine local connections, global positioning and semantic relevance to expose collective structural deviations.
Implication
Present the flagged network as evidence for expert review, with explicit uncertainty.
Detection approaches often look for extreme individual nodes, high citation counts, or unusually intense pairwise links.
SusCG models suspicious citation groups through the combination of local cohesion, global structural roles, and semantic relevance. The anomaly can reside in the configuration of relationships rather than any individual citation.
Concept explorer
Select a concept to see its role in this paper’s contribution.
The paper's deep graph-learning artifact combines journal and paper representations to identify candidate suspicious citation groups.
When misconduct depends on coordination, frame the anomaly at the group level.
Jointly model local interaction patterns and global structural roles.
Evaluate structural irregularities against textual relevance, so topical cohesion is not automatically treated as misconduct.
SusCG combines network structure learning, journal behavior similarity, paper representation learning, and journal–paper local proximity. Evaluation examines real-world and semi-realistic networks.
PDF pp. 5–11, 13–17; printed pp. 1655–1661, 1663–1667.
Three design principles explain how to recognise coordination-driven anomalies without reducing them to isolated outliers or conflating density with wrongdoing.
PDF pp. 19–20; printed pp. 1669–1670.
Review the actual subgraph, reciprocity, citation flows and semantic relevance as structured evidence in an editorial investigation.
PDF pp. 20–21; printed pp. 1670–1671.
Use SusCG for early warning and decision support. A flagged group is a candidate for investigation, not a verdict of misconduct or grounds for automatic sanctions.
PDF p. 21; printed p. 1671.
Why this matters
Publishers and research-integrity teams need evidence that distinguishes coordinated manipulation from legitimate intellectual communities, especially as citation-based assessment expands.
Discuss this researchThe real-world dataset includes 935 journals but only five confirmed journal cases. The authors describe this evaluation as a stress test under data scarcity, not proof of broad generalisability. Dense interdisciplinary communities can still produce false positives. Adaptation to fraud or other networks is proposed, not validated here.
Four-part architecture, real and semi-realistic data, baseline and ablation comparisons.
Named principles, human oversight and data-scarcity limitations.
This exploration is based on the author-supplied full text. It distinguishes the paper’s own contribution from established concepts it draws on. For formal citation, consult the publication.