RevGNN: Negative Sampling Enhanced Contrastive Graph Learning for Academic Reviewer Recommendation

TL;DR

RevGNN tackles the sparse, bias-prone nature of academic reviewer recommendation by decoupling behavior and knowledge representations and introducing a Pseudo Neg-Label based contrastive learning stage to mitigate false negative sampling. The two-stage encoder (behavior + knowledge) feeds a final interaction-aware decoder, yielding superior performance over a wide range of baselines on real-world Frontiers datasets and the NIPS dataset, across Recall, Precision, HR, and NDCG metrics. Key contributions include the Pseudo Neg-Label sampling strategy, the integration of OAG-LM for knowledge embedding, and a clustering-based self-supervised signal that improves robustness under extreme sparsity. The work demonstrates practical impact for large-scale, privacy-conscious reviewer assignment and opens directions toward federated learning, continual updates, and potential LLM-assisted enhancements.

Abstract

Acquiring reviewers for academic submissions is a challenging recommendation scenario. Recent graph learning-driven models have made remarkable progress in the field of recommendation, but their performance in the academic reviewer recommendation task may suffer from a significant false negative issue. This arises from the assumption that unobserved edges represent negative samples. In fact, the mechanism of anonymous review results in inadequate exposure of interactions between reviewers and submissions, leading to a higher number of unobserved interactions compared to those caused by reviewers declining to participate. Therefore, investigating how to better comprehend the negative labeling of unobserved interactions in academic reviewer recommendations is a significant challenge. This study aims to tackle the ambiguous nature of unobserved interactions in academic reviewer recommendations. Specifically, we propose an unsupervised Pseudo Neg-Label strategy to enhance graph contrastive learning (GCL) for recommending reviewers for academic submissions, which we call RevGNN. RevGNN utilizes a two-stage encoder structure that encodes both scientific knowledge and behavior using Pseudo Neg-Label to approximate review preference. Extensive experiments on three real-world datasets demonstrate that RevGNN outperforms all baselines across four metrics. Additionally, detailed further analyses confirm the effectiveness of each component in RevGNN.

Figures (6)

• Figure 1: Different visibility of items between (A) general and (B) academic reviewer recommendation.
• Figure 2: The framework of RevGNN.
• Figure 3: The stage-2 encoding process. In this stage, the embedding representation of nodes is adjusted by graph contrastive learning. Particularly, we design a Pseudo Neg-Label Sampling strategy to avoid false negative edges for the observation-insufficient scholar-submission interaction graph.
• Figure 4: The network structure of the decoder. The decoder network predicts the probability of review action by considering the attention between the candidate submission and the scholar's historical review.
• Figure 5: The variation of performance with different numbers of cluster $C$.