XRAG-CJG: A Trust-Aware and Explainable Retrieval Framework for Reviewer Recommendation

Abstract

Assigning suitable reviewers to academic submissions remains challenging due to the limitations of similarity-based methods and the limited interpretability of modern learning-based approaches. This paper proposes XRAG-CJG, a trust-aware and explainable reviewer recommendation framework that combines semantic retrieval with graph-based reasoning. The framework integrates SPECTER2 embeddings and FAISS retrieval with multi-dimensional feature alignment across methods, tasks, domains, and modalities, and contextual concepts. A Causal Justification Graph (CJG) models relationships between query abstracts, reviewer profiles, and publications to generate evidence-based explanations. The framework was evaluated through a human-centered ranking study involving multiple research abstracts and reviewer candidates. Experimental results demonstrate higher Top-3 overlap and improved ranking consistency compared with a retrieval-augmented baseline, indicating more reliable identification of suitable reviewers. While exact ranking agreement remains challenging because reviewer assignment is inherently subjective, the proposed framework consistently recommends relevant reviewers and provides transparent, evidence-based justifications. These results demonstrate that integrating semantic retrieval with structured reasoning improves both the effectiveness and explainability of reviewer recommendation.