Visual document retrieval requires understanding heterogeneous and multi-modal content to satisfy implicit information needs. Recent advances use screenshot-based document encoding with fine-grained late interaction to encode holistic information and capture nuanced alignments, significantly improving retrieval performance. However, retrievers are still trained with coarse global relevance labels, without revealing which regions support the match. As a result, retrievers tend to rely on surface-level cues and struggle to capture implicit semantic connections, hindering their ability to handle non-extractive queries. To improve fine-grained relevance modeling, we propose a Attention-Grounded REtriever Enhancement (AGREE) framework. AGREE leverages cross-modal attention from multimodal large language models (MLLMs) as proxy supervision to guide the retriever in identifying relevant document regions. Specifically, AGREE extracts attention maps from the MLLM that highlight which document regions are attended to based on the query. These attention scores serve as local, region-level relevance signals. During training, AGREE combines local signals with the global document-level relevance label to jointly optimize the retriever. This dual-level supervision enables the model to learn not only whether documents match, but also which content drives relevance. Experiments on the challenging visual document retrieval benchmark, ViDoRe V2, show that AGREE significantly outperforms the global-supervision-only baseline by 12.82% and 5.03% in terms of average nDCG@1 and nDCG@5. Quantitative and qualitative analyses further demonstrate that AGREE promotes deeper alignment between query terms and document regions, moving beyond surface-level matching toward more accurate and interpretable retrieval. Our code is available at: https://github.com/VickiCui/AGREE.

The COVID-19 pandemic has resulted in changes in the working environment which shifted the type of systems thinking skills needed for practitioners. These changes include the utilization of a digitalized work environment . To reliably assess practitioners' systems thinking (ST) skills/abilities in a digitalized environment, such as the case of COVID-19, we propose an evaluation approach based on the hybrid weight determination and extended VIKOR model to assess the systems skills of practitioners concerning 7-dimensions of systems thinking. The proposed methodology consists of three phases: the first phase uses a rough set theory to process the assessment data of candidates' systems thinking skills, the ideal interval references of seven systems thinking skills criteria (7-dimension) of practitioners required by an organization is extracted. The second phase is to build a comprehensive weight solution model based on BWM (best-worst method) and entropy weight method (EWM) and analyze the employer's needs under each systems thinking skills dimension. The third phase is to build a new group utility index based on the weight and digital reference and form an extended Vlsekriterijumska Optimizacija I Kompromisno Resenje (E-VIKOR) model to complete the prioritization of practitioners' systems thinking skillset. A case study containing 108 practitioners is conducted to verify the effectiveness of the proposed decision-making model and carry out sensitivity analysis and methods comparison. The results show that the proposed model provides more reliable and robust results for selecting the most appropriate practitioner for the required digitalized job requirements .

In distributed storage systems, the replication mechanisms are usually used to ensure system reliability and data availability. Random replication is widely used in cloud storage systems to prevent data loss. Copyset Replication (CR) as a replication strategy, makes a nearly optimal trade-off between the number of scattered nodes and the probability of data loss. Compared with random replication, CR greatly reduces the probability of data loss caused by node failure. However, CR's random selection strategy makes it difficult to select the optimal copyset based on data characteristics such as calculation and storage. In response to this problem of CR, the Optimal Copyset Replication (OCR) proposed in this paper can select the optimal copyset according to the specified data characteristics and its corresponding node conditions. Finally, combined with Cyberspace Mimicry Defense (CMD) , we implemented OCR in a distributed object storage system and conducted related experiments. When the calculation type data reaches 300,000, the experimental results prove that compared with CR randomly selecting copyset, OCR reduces the data processing time by nearly 10% through selecting the optimal copyset. By setting relevant parameters, OCR can also ensure that the data distribution of each node is relatively uniform, and avoid data skew.