<p>Cross-document relation extraction (RE) aims to identify relations between target entities across multiple long documents. Due to the large scale and inherent noise in multi-document corpora, prior works primarily rely on heuristic methods based on bridge entities or capability-limited selectors for evidence sentence selection. However, these approaches often struggle to capture sufficient informative content and latent semantic relations embedded within documents comprehensively. To address this, we propose MAQD, a novel cross-document RE approach that enhances evidence selection through multi-aspect query diversification. Specifically, we construct a knowledge-aware entity graph from the original text and leverage large language models (LLMs) to accomplish multi-faceted query augmentation tasks based on the graph for query diversification. We also employ a multi-aspect reranking and aggregation mechanism to rerank candidate sentences from diverse perspectives and aggregate them to select the most relevant evidence sentences for final relation prediction. Experimental results on the benchmark dataset validate the superiority of MAQD over existing approaches, highlighting its effectiveness in cross-document RE.</p>