Automated Peer Review and Citation-Grounded Models

Peer review depends on expert judgment, but editors and program committees increasingly face large submission volumes, uneven reviewer expertise, incomplete citation coverage, and reviews whose recommendations are difficult to reconcile. This paper presents Citation-Grounded Review Assistance (CGRA), a 2020-era framework for automated support of scholarly peer review. CGRA does not replace reviewers. It creates structured evidence for editors by combining scientific document representations, citation recommendation, citation-function analysis, reviewer-assignment signals, review-argument mining, and calibrated confidence gates. For each submitted manuscript, the system builds a claim and citation profile, checks whether central claims are supported by relevant prior work, ranks reviewer-paper expertise matches, and analyzes review text for evidence, requests, references, and unsupported assertions. A controlled study over peer-review and scientific-document corpora shows that CGRA improves missing-citation detection F1 from 0.42 to 0.55, improves reviewer-assignment precision at five from 0.48 to 0.61, and raises review-action classification macro-F1 from 0.63 to 0.71 while routing 18% of low-confidence decisions to manual editorial review. The main finding is that automated peer-review support is most useful when citation evidence and model confidence are visible, auditable, and subordinate to human editorial decisions.