该研究通过引入Modality Contribution Network (MCNet) 和 Modality Contribution Score (MCS)，量化了在阿尔茨海默病诊断中结构MRI和淀粉样PET的相对贡献，解决了现有AI系统无法为每个患者提供具体模态贡献的问题。
This study addresses the joint optimization of exam-level malignancy prediction and lesion localization in mammography. The authors propose a unified multi-task DETR-based framework that leverages shared features to simultaneously perform image-level classification and region-level localization. For the first time, they systematically evaluate the performance of various modern vision backbones—including ConvNeXtV2, DINOv3, and MambaVision—on this dual objective. Experimental results demonstrate that advanced backbones substantially outperform conventional ResNet architectures: on the OPTIMAM dataset, ConvNeXtV2 achieves 97.96% AUC and 25.08% mAP@.5, while on SGM1k, DINOv3 attains 90.97% AUC and 27.04% mAP@.5, underscoring the critical influence of backbone architecture on multi-task performance in mammographic analysis.
This work addresses the limitations of existing methods for inferring user intent from online travel reviews, which are often susceptible to cascading errors or neglect the structured relationships inherent in domain knowledge. To overcome these issues, the authors propose DKG-MTI, a novel framework that dynamically constructs a user-specific intent knowledge graph during inference and aligns it semantically with a global hotel knowledge graph through structure-aware mechanisms. Integrating fine-tuned large language models, the framework enables joint multi-task reasoning for aspect-level rating prediction and reverse intent generation. By incorporating a knowledge-enhanced mechanism, the approach supports interpretable and scalable intent inference. Experimental results on the TripAdvisor dataset demonstrate that DKG-MTI significantly outperforms strong baseline models in both classification and intent generation tasks.
To address the challenge of systematically integrating and discovering knowledge from vast volumes of unstructured academic papers in China Studies (CS) within Taiwan, this study proposes a generative AI–driven knowledge graph construction paradigm, replacing traditional ontology engineering. Leveraging 1,367 scholarly papers published between 1996 and 2019, we integrate large language model–based triple extraction, entity normalization and mapping, vector database–enabled semantic retrieval, and D3.js–powered interactive visualization to construct the first dynamic domain-specific knowledge graph for Taiwan’s CS field. The resulting graph supports interactive concept-node querying, semantic relationship tracing, and thematic clustering analysis. It effectively uncovers latent scholarly trajectories, emerging research fronts, and structural knowledge gaps, thereby significantly enhancing cross-disciplinary literature discovery efficiency and analytical insight generation.
该研究通过引入Modality Contribution Network (MCNet) 和 Modality Contribution Score (MCS)，量化了在阿尔茨海默病诊断中结构MRI和淀粉样PET的相对贡献，解决了现有AI系统无法为每个患者提供具体模态贡献的问题。
This study addresses the joint optimization of exam-level malignancy prediction and lesion localization in mammography. The authors propose a unified multi-task DETR-based framework that leverages shared features to simultaneously perform image-level classification and region-level localization. For the first time, they systematically evaluate the performance of various modern vision backbones—including ConvNeXtV2, DINOv3, and MambaVision—on this dual objective. Experimental results demonstrate that advanced backbones substantially outperform conventional ResNet architectures: on the OPTIMAM dataset, ConvNeXtV2 achieves 97.96% AUC and 25.08% mAP@.5, while on SGM1k, DINOv3 attains 90.97% AUC and 27.04% mAP@.5, underscoring the critical influence of backbone architecture on multi-task performance in mammographic analysis.
This work addresses the limitations of existing methods for inferring user intent from online travel reviews, which are often susceptible to cascading errors or neglect the structured relationships inherent in domain knowledge. To overcome these issues, the authors propose DKG-MTI, a novel framework that dynamically constructs a user-specific intent knowledge graph during inference and aligns it semantically with a global hotel knowledge graph through structure-aware mechanisms. Integrating fine-tuned large language models, the framework enables joint multi-task reasoning for aspect-level rating prediction and reverse intent generation. By incorporating a knowledge-enhanced mechanism, the approach supports interpretable and scalable intent inference. Experimental results on the TripAdvisor dataset demonstrate that DKG-MTI significantly outperforms strong baseline models in both classification and intent generation tasks.
To address the challenge of systematically integrating and discovering knowledge from vast volumes of unstructured academic papers in China Studies (CS) within Taiwan, this study proposes a generative AI–driven knowledge graph construction paradigm, replacing traditional ontology engineering. Leveraging 1,367 scholarly papers published between 1996 and 2019, we integrate large language model–based triple extraction, entity normalization and mapping, vector database–enabled semantic retrieval, and D3.js–powered interactive visualization to construct the first dynamic domain-specific knowledge graph for Taiwan’s CS field. The resulting graph supports interactive concept-node querying, semantic relationship tracing, and thematic clustering analysis. It effectively uncovers latent scholarly trajectories, emerging research fronts, and structural knowledge gaps, thereby significantly enhancing cross-disciplinary literature discovery efficiency and analytical insight generation.