Abstract
<title>Abstract</title> <p>Cross-domain papers often connect multiple knowledge subspaces, making reviewer recommendation difficult for semantic-first filtering pipelines. Suitable reviewers may receive moderate semantic scores because their historical terminology differs from the query paper, although topic or citation evidence still supports their expertise. To address this boundary-case failure, we propose RECO-Rank, a compensatory multi-evidence ranking model for candidate-set reviewer re-ranking. RECO-Rank calibrates semantic, topic, and citation evidence, aggregates reviewer histories through a semantics-guided main path and a compensation branch, and controls compensation with reliability-aware fusion and top-\((k)\) ranking learning. Experiments on four reviewer-matching datasets show competitive overall performance and clear improvements on several hard top-\((k)\) metrics, especially on SIGIR and KDD. The results suggest that controlled compensation is useful for recovering evidence-complementary reviewers while maintaining top-ranked precision.</p>