Knowledge-Based Systems最新文献

BDGKT: Bidirectional dynamic graph knowledge tracing BDGKT：双向动态图知识跟踪

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115532

Xinjia Ou , Tao Huang , Shengze Hu , Huali Yang , Zhuoran Xu , Junjie Hu , Jing Geng

Knowledge tracing (KT) aims to model the evolution of students’ knowledge states by analyzing their historical learning trajectories and predicting future performance. However, current KT methods primarily focus on unidirectional relationship modeling, overlooking the bidirectional dynamic interaction mechanisms between learners and questions. Student knowledge states shape question adaptability through group patterns (e.g., difficulty calibration), whereas dynamic transformation of question features provides progressive guidance signals for knowledge advancement across learning stages. In this study, we propose a novel bidirectional dynamic graph KT (BDGKT) method for modeling the information flow between students and questions while capturing knowledge state evolution and question characteristic transformation. Specifically, we first introduce a dynamic graph construction based on homogeneous student groups that uses a spatiotemporal constraint strategy to reduce computational costs while improving information propagation quality. Subsequently, we design a bidirectional message propagation mechanism to capture time-evolving bidirectional dynamic signals. To update question nodes (from students to questions), we introduce a state-aware attention mechanism that aggregates student nodes and responses, revealing group-level question commonalities. By contrast, to update student nodes (from questions to students), we propose an evolution mechanism that aggregates question nodes and responses based on timestamps, allowing us to track the evolution of student knowledge states. Extensive experiments on four real-world datasets validate the effectiveness and compatibility of our method. Furthermore, BDGKT improves interpretability by exploring question absolute information (group-agnostic) and relative information (group-dependent).

知识追踪（Knowledge tracing， KT）旨在通过分析学生的历史学习轨迹和预测学生的未来表现，来模拟学生知识状态的演变。然而，目前的KT方法主要侧重于单向关系建模，忽略了学习者与问题之间的双向动态交互机制。学生的知识状态通过群体模式（如难度校准）塑造问题的适应性，而问题特征的动态转换为跨学习阶段的知识进步提供了渐进式的指导信号。在本研究中，我们提出了一种新的双向动态图KT （BDGKT）方法来建模学生与问题之间的信息流，同时捕捉知识状态演变和问题特征转换。具体来说，我们首先引入了一种基于同质学生群体的动态图构建方法，该方法使用时空约束策略来降低计算成本，同时提高信息传播质量。随后，我们设计了一种双向消息传播机制来捕获随时间变化的双向动态信号。为了更新问题节点（从学生到问题），我们引入了一个状态感知的关注机制，该机制聚合了学生节点和响应，揭示了组级问题的共性。相比之下，为了更新学生节点（从问题到学生），我们提出了一种基于时间戳聚合问题节点和响应的进化机制，使我们能够跟踪学生知识状态的演变。在四个真实数据集上的大量实验验证了我们方法的有效性和兼容性。此外，BDGKT通过探索问题绝对信息（群体不可知）和相对信息（群体依赖）来提高可解释性。

{"title":"BDGKT: Bidirectional dynamic graph knowledge tracing","authors":"Xinjia Ou , Tao Huang , Shengze Hu , Huali Yang , Zhuoran Xu , Junjie Hu , Jing Geng","doi":"10.1016/j.knosys.2026.115532","DOIUrl":"10.1016/j.knosys.2026.115532","url":null,"abstract":"<div><div>Knowledge tracing (KT) aims to model the evolution of students’ knowledge states by analyzing their historical learning trajectories and predicting future performance. However, current KT methods primarily focus on unidirectional relationship modeling, overlooking the bidirectional dynamic interaction mechanisms between learners and questions. Student knowledge states shape question adaptability through group patterns (e.g., difficulty calibration), whereas dynamic transformation of question features provides progressive guidance signals for knowledge advancement across learning stages. In this study, we propose a novel bidirectional dynamic graph KT (BDGKT) method for modeling the information flow between students and questions while capturing knowledge state evolution and question characteristic transformation. Specifically, we first introduce a dynamic graph construction based on homogeneous student groups that uses a spatiotemporal constraint strategy to reduce computational costs while improving information propagation quality. Subsequently, we design a bidirectional message propagation mechanism to capture time-evolving bidirectional dynamic signals. To update question nodes (from students to questions), we introduce a state-aware attention mechanism that aggregates student nodes and responses, revealing group-level question commonalities. By contrast, to update student nodes (from questions to students), we propose an evolution mechanism that aggregates question nodes and responses based on timestamps, allowing us to track the evolution of student knowledge states. Extensive experiments on four real-world datasets validate the effectiveness and compatibility of our method. Furthermore, BDGKT improves interpretability by exploring question absolute information (group-agnostic) and relative information (group-dependent).</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115532"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175181","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

HiSURF: Hierarchical semantic-guided unified radiance field for generalizing across unseen scenes HiSURF：分层语义引导的统一辐射场，用于在未见场景中进行泛化

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115530

Qiang Liu , Teng Wang , Zhiguo Zhang , Jun Nie , Xiao Lu , Chunyang Sheng , Shibin Song , Qiaoqiao Sun , Haixia Wang

Recent advancements in neural field representations have significantly improved novel view synthesis for seen scenes. However, generalizing seen representations to unseen scenes remains challenging. Addressing this problem, we propose the Hierarchical Semantic-guided Unified Radiance Field (HiSURF) to leverage hierarchical semantic attributes from seen scenes as prior knowledge. The synthesis of scene representations for unseen environments can be enabled by establishing an interpretable mapping between semantic attributes and visual features. Specifically, HiSURF consists of a local semantic embedding module, a global semantic mapping module, and a composite rendering module. For a scene with multiple objects, the local module disentangles attributes of objects to generate fine object-level triplanes, which preserve structural and surface details for objects. At the same time, the global module utilizes attributes of the holistic scene to construct a coarse scene-level triplane, which ensures layout consistency and contextual coherence for the scene. Then, the composite rendering module integrates features from both object-level and scene-level triplanes for high-quality novel view synthesis. Experimental results on the ClevrTex and Kubric datasets demonstrate that our HiSURF not only outperforms existing approaches in novel view synthesis but also exhibits superior generalization capability to unseen scenes.

神经场表征的最新进展显著改善了对已见场景的新视图合成。然而，将看到的表示推广到看不见的场景仍然具有挑战性。为了解决这个问题，我们提出了分层语义引导的统一辐射场（HiSURF），以利用来自所见场景的分层语义属性作为先验知识。通过在语义属性和视觉特征之间建立可解释的映射，可以实现对不可见环境的场景表示的综合。HiSURF由局部语义嵌入模块、全局语义映射模块和复合呈现模块组成。对于具有多个物体的场景，局部模块分解物体的属性，生成精细的物体级三平面，保留物体的结构和表面细节。同时，全局模块利用整体场景的属性构建粗场景级三平面，保证场景布局的一致性和上下文的连贯性。然后，合成渲染模块集成了对象级和场景级三平面的特征，以实现高质量的新视图合成。在ClevrTex和Kubric数据集上的实验结果表明，我们的HiSURF不仅在新视图合成方面优于现有方法，而且对未见过的场景也表现出优越的泛化能力。

{"title":"HiSURF: Hierarchical semantic-guided unified radiance field for generalizing across unseen scenes","authors":"Qiang Liu , Teng Wang , Zhiguo Zhang , Jun Nie , Xiao Lu , Chunyang Sheng , Shibin Song , Qiaoqiao Sun , Haixia Wang","doi":"10.1016/j.knosys.2026.115530","DOIUrl":"10.1016/j.knosys.2026.115530","url":null,"abstract":"<div><div>Recent advancements in neural field representations have significantly improved novel view synthesis for seen scenes. However, generalizing seen representations to unseen scenes remains challenging. Addressing this problem, we propose the Hierarchical Semantic-guided Unified Radiance Field (HiSURF) to leverage hierarchical semantic attributes from seen scenes as prior knowledge. The synthesis of scene representations for unseen environments can be enabled by establishing an interpretable mapping between semantic attributes and visual features. Specifically, HiSURF consists of a local semantic embedding module, a global semantic mapping module, and a composite rendering module. For a scene with multiple objects, the local module disentangles attributes of objects to generate fine object-level triplanes, which preserve structural and surface details for objects. At the same time, the global module utilizes attributes of the holistic scene to construct a coarse scene-level triplane, which ensures layout consistency and contextual coherence for the scene. Then, the composite rendering module integrates features from both object-level and scene-level triplanes for high-quality novel view synthesis. Experimental results on the ClevrTex and Kubric datasets demonstrate that our HiSURF not only outperforms existing approaches in novel view synthesis but also exhibits superior generalization capability to unseen scenes.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115530"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175149","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

FasterGCN: Accelerating and enhancing graph convolutional network for recommendation FasterGCN：加速和增强用于推荐的图卷积网络

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115533

Jiaxin Wu , Chenglong Pang , Guangxiong Chen , Jie Zhao , Jihong Wan

In recommendation systems, graph convolutional networks (GCNs) are widely used to capture high-order user–item interactions. However, deeper GCNs often suffer from over-smoothing, where node representations become indistinguishable. Conversely, excessive efforts to avoid over-smoothing can lead to under-smoothing, resulting in prolonged training and insufficient aggregation of neighborhood information. To address both issues, we propose FasterGCN, a linear GCN specifically designed for recommendation tasks. By distilling potential interaction information (PII) from high-order connectivity, FasterGCN achieves optimal smoothing rapidly and maintains stable performance even with deeper architectures. Moreover, its concise design eliminates complex parameters, enhancing scalability and extensibility. Extensive experiments on six real-world datasets demonstrate that FasterGCN not only consistently outperforms state-of-the-art GCNs but also improves training efficiency by up to 96%, highlighting its potential as a strong backbone for recommender systems.

在推荐系统中，图卷积网络（GCNs）被广泛用于捕获高阶用户-物品交互。然而，更深层次的GCNs经常遭受过度平滑，节点表示变得无法区分。相反，过度努力避免过度平滑会导致平滑不足，导致训练时间延长和邻域信息聚集不足。为了解决这两个问题，我们提出了FasterGCN，这是一个专门为推荐任务设计的线性GCN。FasterGCN通过从高阶连接中提取潜在交互信息（PII），快速实现最优平滑，即使在更深的架构下也能保持稳定的性能。此外，其简洁的设计消除了复杂的参数，增强了可伸缩性和可扩展性。在六个真实数据集上进行的大量实验表明，FasterGCN不仅始终优于最先进的gcn，而且还将训练效率提高了96%，突出了其作为推荐系统强大骨干的潜力。

引用次数: 0

Multi-level dual contrastive learning for cloud API cold-start recommendation 云API冷启动推荐的多级双对比学习

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115534

Mengmeng Sun , Yueshen Xu , Dianlong You , Zhen Chen

A longstanding challenge in cloud API recommender systems is the cold-start problem associated with newly released cloud APIs, which lack historical interactions. Existing approaches typically 1) integrate the content and collaborative embeddings of a cloud API to generate its representation, or 2) adopt API-level alignment strategies that maximize mutual information between them. However, they often assume that cold cloud APIs have similar content features to warm ones, which doesn’t always hold in practice. Additionally, developers frequently combine multiple cloud APIs to create Mashups, indicating that focusing solely on individual cloud APIs fails to capture Mashup preferences. To this end, we propose a multi-level dual contrastive learning (MDCL) framework that explores Mashup preferences to impose effective embedding constraints on low-similarity cold cloud APIs. Specifically, MDCL generates a Mashup preference representation by aggregating the collaborative embeddings of warm cloud APIs based on the Mashup’s interaction history. It then performs group-level alignment between the content embedding of a cloud API and the Mashup’s preference representation, thereby guiding low-similarity cold cloud APIs toward the collaborative space. Furthermore, MDCL integrates API-level alignment and Mashup-API alignment to improve consistency between a cloud API’s content and collaborative embeddings, and to better model interaction patterns between Mashups and cloud APIs. A hybrid training strategy is employed to jointly optimize three alignment objectives: Mashup-API, API-level, and group-level alignment, achieving a better balance between cold-start and warm-start recommendations. Extensive experiments on real-world datasets demonstrate that MDCL outperforms SOTA methods in cold- and warm-start scenarios. Implementation code is available at https://github.com/MengMeng3399/MDCL.

云API推荐系统中一个长期存在的挑战是与新发布的云API相关的冷启动问题，它们缺乏历史交互。现有的方法通常是1)集成云API的内容和协作嵌入来生成其表示，或者2)采用API级对齐策略来最大化它们之间的相互信息。然而，他们通常认为冷云api与热云api具有相似的内容特征，这在实践中并不总是成立。此外，开发人员经常组合多个云api来创建Mashup，这表明仅关注单个云api无法捕获Mashup首选项。为此，我们提出了一个多层次的双重对比学习（MDCL）框架，该框架探索Mashup偏好，以对低相似度冷云api施加有效的嵌入约束。具体来说，MDCL根据Mashup的交互历史，通过聚合热云api的协作嵌入，生成Mashup首选项表示。然后，它在云API的内容嵌入和Mashup的首选项表示之间执行组级对齐，从而将低相似度的冷云API引导到协作空间。此外，MDCL还集成了API级对齐和Mashup-API对齐，以提高云API内容和协作嵌入之间的一致性，并更好地为mashup和云API之间的交互模式建模。采用混合训练策略联合优化三个对齐目标：Mashup-API、api级和组级对齐，在冷启动和热启动建议之间实现更好的平衡。在实际数据集上进行的大量实验表明，MDCL在冷启动和热启动情况下优于SOTA方法。实现代码可从https://github.com/MengMeng3399/MDCL获得。

{"title":"Multi-level dual contrastive learning for cloud API cold-start recommendation","authors":"Mengmeng Sun , Yueshen Xu , Dianlong You , Zhen Chen","doi":"10.1016/j.knosys.2026.115534","DOIUrl":"10.1016/j.knosys.2026.115534","url":null,"abstract":"<div><div>A longstanding challenge in cloud API recommender systems is the cold-start problem associated with newly released cloud APIs, which lack historical interactions. Existing approaches typically 1) integrate the content and collaborative embeddings of a cloud API to generate its representation, or 2) adopt API-level alignment strategies that maximize mutual information between them. However, they often assume that cold cloud APIs have similar content features to warm ones, which doesn’t always hold in practice. Additionally, developers frequently combine multiple cloud APIs to create Mashups, indicating that focusing solely on individual cloud APIs fails to capture Mashup preferences. To this end, we propose a multi-level dual contrastive learning (MDCL) framework that explores Mashup preferences to impose effective embedding constraints on low-similarity cold cloud APIs. Specifically, MDCL generates a Mashup preference representation by aggregating the collaborative embeddings of warm cloud APIs based on the Mashup’s interaction history. It then performs group-level alignment between the content embedding of a cloud API and the Mashup’s preference representation, thereby guiding low-similarity cold cloud APIs toward the collaborative space. Furthermore, MDCL integrates API-level alignment and Mashup-API alignment to improve consistency between a cloud API’s content and collaborative embeddings, and to better model interaction patterns between Mashups and cloud APIs. A hybrid training strategy is employed to jointly optimize three alignment objectives: Mashup-API, API-level, and group-level alignment, achieving a better balance between cold-start and warm-start recommendations. Extensive experiments on real-world datasets demonstrate that MDCL outperforms SOTA methods in cold- and warm-start scenarios. Implementation code is available at <span><span>https://github.com/MengMeng3399/MDCL</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115534"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175153","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Visual and textual spaces both matter: Taming CLIP for non-IID federated medical image classification 视觉和文本空间都很重要：驯服CLIP用于非iid联合医学图像分类

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115524

Lulu Feng , Shengchao Chen

Federated Learning-based medical image analysis system can offer significant insights to enhance privacy-preserving computer-aided diagnosis (CAD) by accessing both public and private medical data. Adapting pre-trained Vision-Language Foundation Models like CLIP for federated learning-based medical image analysis offers cross-modal insights that boost decision support compared to unimodal visual models. However, effective cross-domain federated adaptation requires intensive fine-tuning and knowledge sharing, challenging in low-resource medical practice due to the divergence between pretrained natural image knowledge and medical imagery. Moreover, the significant statistical heterogeneity (non-IID) of medical data exacerbates these challenges. To address these issues, this paper introduces a Parallel Multimodal Reinforcement framework (PMRFed) that tames CLIP for non-IID federated medical image classification. PMRFed develops client-specific personalized models by reinforcement and constrain local cross-modal alignment, enabling the models to integrate client-specific and globally common knowledge. This approach not only addresses non-IID challenges but also optimizes the trade-off between performance and efficiency. Extensive experiments on real-world medical image classification datasets demonstrate the effectiveness and superiority of our proposed PMRFed.

基于联邦学习的医学图像分析系统可以通过访问公共和私人医疗数据，为增强保护隐私的计算机辅助诊断（CAD）提供重要见解。将预训练的视觉语言基础模型（如CLIP）用于基于联邦学习的医学图像分析，可以提供跨模态的见解，与单模态视觉模型相比，可以提高决策支持。然而，有效的跨域联合自适应需要密集的微调和知识共享，由于预训练的自然图像知识与医学图像之间的差异，在资源匮乏的医疗实践中具有挑战性。此外，医疗数据的显著统计异质性（非iid）加剧了这些挑战。为了解决这些问题，本文引入了一个并行多模态强化框架（PMRFed），该框架将CLIP命名为非iid联邦医学图像分类。PMRFed通过强化和约束局部跨模态对齐来开发客户特定的个性化模型，使模型能够集成客户特定的和全球通用的知识。这种方法不仅解决了非iid挑战，还优化了性能和效率之间的权衡。在实际医学图像分类数据集上的大量实验证明了我们提出的PMRFed的有效性和优越性。

{"title":"Visual and textual spaces both matter: Taming CLIP for non-IID federated medical image classification","authors":"Lulu Feng , Shengchao Chen","doi":"10.1016/j.knosys.2026.115524","DOIUrl":"10.1016/j.knosys.2026.115524","url":null,"abstract":"<div><div>Federated Learning-based medical image analysis system can offer significant insights to enhance privacy-preserving computer-aided diagnosis (CAD) by accessing both public and private medical data. Adapting pre-trained Vision-Language Foundation Models like CLIP for federated learning-based medical image analysis offers cross-modal insights that boost decision support compared to unimodal visual models. However, effective cross-domain federated adaptation requires intensive fine-tuning and knowledge sharing, challenging in low-resource medical practice due to the divergence between pretrained natural image knowledge and medical imagery. Moreover, the significant statistical heterogeneity (non-IID) of medical data exacerbates these challenges. To address these issues, this paper introduces a Parallel Multimodal Reinforcement framework (<span><strong>PMRFed</strong></span>) that tames CLIP for non-IID federated medical image classification. <span><strong>PMRFed</strong></span> develops client-specific personalized models by reinforcement and constrain local cross-modal alignment, enabling the models to integrate client-specific and globally common knowledge. This approach not only addresses non-IID challenges but also optimizes the trade-off between performance and efficiency. Extensive experiments on real-world medical image classification datasets demonstrate the effectiveness and superiority of our proposed <span><strong>PMRFed</strong></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115524"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175145","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Improved LSTNet-Driven hyperchaotic sequence optimization and its application in multi-Image encryption 改进lstnet驱动的超混沌序列优化及其在多图像加密中的应用

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115529

Yongzhang Li, Ye Tao

Existing deep-learning-based chaotic image encryption schemes suffer from insufficient performance and are limited to simple chaotic systems, hindering their practical applicability. Therefore, we have developed a new deep learning-based multi-image chaotic encryption framework. First, by integrating optimization test functions, we constructed a 2D hyperchaotic system (ASHM) for generating chaotic sequences. Next, we propose a network architecture, ChaosAutoFormer, which is trained on chaotic sequences and efficiently generates new random sequences. The generated sequences pass standard randomness tests. Subsequently, we applied these sequences to a multi-image encryption system and employed lightweight encryption methods such as circular shifting, deep rearrangement, and adaptive path selection, balancing both encryption security and efficiency. The new sequences generated by deep learning overcome the defects caused by the direct use of chaotic sequences, and the complexity of the deep learning structure makes it resistant to various attacks. Simulation results indicate that the proposed algorithm achieves a key space of 2³¹⁸ values, an information entropy of 7.9993, an NPCR of 99.6145%, and an UACI of 33.6364%, making it suitable for scenarios that require high encryption security.

现有的基于深度学习的混沌图像加密方案存在性能不足，且仅限于简单的混沌系统，阻碍了其实际应用。因此，我们开发了一种新的基于深度学习的多图像混沌加密框架。首先，通过对优化测试函数的积分，构造了用于生成混沌序列的二维超混沌系统（ASHM）。接下来，我们提出了一种网络架构ChaosAutoFormer，它在混沌序列上进行训练，并有效地生成新的随机序列。生成的序列通过标准的随机性测试。随后，我们将这些序列应用于多图像加密系统，并采用了循环移位、深度重排和自适应路径选择等轻量级加密方法，平衡了加密的安全性和效率。深度学习生成的新序列克服了直接使用混沌序列所带来的缺陷，深度学习结构的复杂性使其能够抵抗各种攻击。仿真结果表明，该算法的密钥空间为2318个值，信息熵为7.9993，NPCR为99.6145%，UACI为33.6364%，适用于对加密安全性要求较高的场景。

{"title":"Improved LSTNet-Driven hyperchaotic sequence optimization and its application in multi-Image encryption","authors":"Yongzhang Li, Ye Tao","doi":"10.1016/j.knosys.2026.115529","DOIUrl":"10.1016/j.knosys.2026.115529","url":null,"abstract":"<div><div>Existing deep-learning-based chaotic image encryption schemes suffer from insufficient performance and are limited to simple chaotic systems, hindering their practical applicability. Therefore, we have developed a new deep learning-based multi-image chaotic encryption framework. First, by integrating optimization test functions, we constructed a 2D hyperchaotic system (ASHM) for generating chaotic sequences. Next, we propose a network architecture, ChaosAutoFormer, which is trained on chaotic sequences and efficiently generates new random sequences. The generated sequences pass standard randomness tests. Subsequently, we applied these sequences to a multi-image encryption system and employed lightweight encryption methods such as circular shifting, deep rearrangement, and adaptive path selection, balancing both encryption security and efficiency. The new sequences generated by deep learning overcome the defects caused by the direct use of chaotic sequences, and the complexity of the deep learning structure makes it resistant to various attacks. Simulation results indicate that the proposed algorithm achieves a key space of 2<sup>318</sup> values, an information entropy of 7.9993, an NPCR of 99.6145%, and an UACI of 33.6364%, making it suitable for scenarios that require high encryption security.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115529"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175146","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

GSMS: Integrating graph structures and multi-curvature space mapping for entity alignment via generative adversarial training GSMS：通过生成对抗训练集成图结构和多曲率空间映射用于实体对齐

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-10 DOI: 10.1016/j.knosys.2026.115536

Linlin Ding , Mengjunyao Si , Mo Li , Yishan Pan , Xin Wang

Entity Alignment (EA) aims to identify equivalent real-world entities across different knowledge graphs. These graphs exhibit a mixture of structural forms, such as hierarchies, cycles, and chains, which correspond to different geometric behaviors. However, most existing methods learn representations in a single geometric space, implicitly assuming uniform structural regularity, which limits their ability to capture diverse relational semantics and nonlinear dependencies in graphs with mixed or irregular topologies. To address this limitation, we propose a novel GSMS model, which integrates Graph Structural signals with Multi-curvature Space mapping under a generative adversarial training framework. GSMS unifies structural enhancement, multi-curvature geometric mapping, and adversarial training into a cohesive framework that strengthens both the discriminative capacity and robustness of entity representations. Specifically, it first enhances structural representations by leveraging second-order and triangular-ring relations while suppressing noise through stacked adaptive edge-weight updates. Then, it embeds entities into Euclidean, hyperbolic, and spherical spaces and adaptively fuses these complementary geometric features via a geometry-gated fusion module. Subsequently, a generative adversarial scheme aligns structural and geometric embeddings by treating the latter as “real” samples, thereby enforcing geometric consistency and improving robustness. Extensive experiments on multiple benchmark cross-lingual knowledge graph datasets demonstrate that GSMS consistently outperforms state-of-the-art methods, achieving notable improvements across various evaluation metrics, particularly under sparse and structurally heterogeneous settings.

实体对齐（Entity Alignment， EA）的目的是在不同的知识图谱中识别等价的真实世界实体。这些图展示了混合的结构形式，如层次、循环和链，它们对应于不同的几何行为。然而，大多数现有方法在单个几何空间中学习表示，隐含地假设统一的结构规则，这限制了它们在混合或不规则拓扑的图中捕获不同关系语义和非线性依赖的能力。为了解决这一限制，我们提出了一种新的GSMS模型，该模型在生成对抗训练框架下将图结构信号与多曲率空间映射集成在一起。GSMS将结构增强、多曲率几何映射和对抗性训练统一到一个内聚框架中，增强了实体表示的判别能力和鲁棒性。具体来说，它首先通过利用二阶和三角环关系来增强结构表征，同时通过堆叠自适应边权更新来抑制噪声。然后，它将实体嵌入欧几里德空间、双曲空间和球面空间，并通过几何门控融合模块自适应地融合这些互补的几何特征。随后，生成对抗方案通过将后者视为“真实”样本来对齐结构和几何嵌入，从而增强几何一致性并提高鲁棒性。在多个基准跨语言知识图谱数据集上进行的大量实验表明，GSMS始终优于最先进的方法，在各种评估指标上取得了显着改进，特别是在稀疏和结构异构设置下。

{"title":"GSMS: Integrating graph structures and multi-curvature space mapping for entity alignment via generative adversarial training","authors":"Linlin Ding , Mengjunyao Si , Mo Li , Yishan Pan , Xin Wang","doi":"10.1016/j.knosys.2026.115536","DOIUrl":"10.1016/j.knosys.2026.115536","url":null,"abstract":"<div><div>Entity Alignment (EA) aims to identify equivalent real-world entities across different knowledge graphs. These graphs exhibit a mixture of structural forms, such as hierarchies, cycles, and chains, which correspond to different geometric behaviors. However, most existing methods learn representations in a single geometric space, implicitly assuming uniform structural regularity, which limits their ability to capture diverse relational semantics and nonlinear dependencies in graphs with mixed or irregular topologies. To address this limitation, we propose a novel GSMS model, which integrates <u>G</u>raph <u>S</u>tructural signals with <u>M</u>ulti-curvature <u>S</u>pace mapping under a generative adversarial training framework. GSMS unifies structural enhancement, multi-curvature geometric mapping, and adversarial training into a cohesive framework that strengthens both the discriminative capacity and robustness of entity representations. Specifically, it first enhances structural representations by leveraging second-order and triangular-ring relations while suppressing noise through stacked adaptive edge-weight updates. Then, it embeds entities into Euclidean, hyperbolic, and spherical spaces and adaptively fuses these complementary geometric features via a geometry-gated fusion module. Subsequently, a generative adversarial scheme aligns structural and geometric embeddings by treating the latter as “real” samples, thereby enforcing geometric consistency and improving robustness. Extensive experiments on multiple benchmark cross-lingual knowledge graph datasets demonstrate that GSMS consistently outperforms state-of-the-art methods, achieving notable improvements across various evaluation metrics, particularly under sparse and structurally heterogeneous settings.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115536"},"PeriodicalIF":7.6,"publicationDate":"2026-02-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175155","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Scene-aware memory discrimination: Deciding which personal knowledge stays 情景感知记忆辨别：决定哪些个人知识保留

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-09 DOI: 10.1016/j.knosys.2026.115496

Yijie Zhong , Mengying Guo , Zewei Wang , Zhongyang Li , Dandan Tu , Haofen Wang

Intelligent devices have become deeply integrated into everyday life, generating vast amounts of user interactions that form valuable personal knowledge. Efficient organization of this knowledge in user memory is essential for enabling personalized applications. However, current research on memory writing, management, and reading using large language models (LLMs) faces challenges in filtering irrelevant information and in dealing with rising computational costs. Inspired by the concept of selective attention in the human brain, we introduce a memory discrimination task. To address large-scale interactions and diverse memory standards in this task, we propose a Scene-Aware Memory Discrimination method (SAMD), which comprises two key components: the Gating Unit Module (GUM) and the Cluster Prompting Module (CPM). GUM enhances processing efficiency by filtering out non-memorable interactions and focusing on the salient content most relevant to application demands. CPM establishes adaptive memory standards, guiding LLMs to discern what information should be remembered or discarded. It also analyzes the relationship between user intents and memory contexts to build effective clustering prompts. Comprehensive direct and indirect evaluations demonstrate the effectiveness and generalization of our approach. We independently assess the performance of memory discrimination, showing that SAMD successfully recalls the majority of memorable data and remains robust in dynamic scenarios. Furthermore, when integrated into personalized applications, SAMD significantly enhances both the efficiency and quality of memory construction, leading to better organization of personal knowledge.

智能设备已经深度融入日常生活，产生了大量的用户交互，形成了宝贵的个人知识。在用户内存中有效地组织这些知识对于启用个性化应用程序至关重要。然而，目前使用大型语言模型（llm）进行内存写入、管理和读取的研究面临着过滤不相关信息和处理不断上升的计算成本的挑战。受人类大脑选择性注意概念的启发，我们引入了一个记忆辨别任务。为了解决该任务中的大规模交互和不同的记忆标准，我们提出了一种场景感知记忆识别方法（SAMD），该方法由两个关键组件组成：门控单元模块（GUM）和集群提示模块（CPM）。GUM通过过滤掉难以记忆的交互并专注于与应用程序需求最相关的突出内容来提高处理效率。CPM建立了自适应记忆标准，指导法学硕士辨别哪些信息应该被记住或丢弃。它还分析用户意图和内存上下文之间的关系，以构建有效的聚类提示。全面的直接和间接评价表明了我们方法的有效性和普遍性。我们独立评估了记忆辨别的性能，表明SAMD成功地回忆了大部分可记忆的数据，并在动态场景中保持鲁棒性。此外，当集成到个性化应用中时，SAMD显著提高了记忆构建的效率和质量，从而更好地组织个人知识。

{"title":"Scene-aware memory discrimination: Deciding which personal knowledge stays","authors":"Yijie Zhong , Mengying Guo , Zewei Wang , Zhongyang Li , Dandan Tu , Haofen Wang","doi":"10.1016/j.knosys.2026.115496","DOIUrl":"10.1016/j.knosys.2026.115496","url":null,"abstract":"<div><div>Intelligent devices have become deeply integrated into everyday life, generating vast amounts of user interactions that form valuable personal knowledge. Efficient organization of this knowledge in user memory is essential for enabling personalized applications. However, current research on memory writing, management, and reading using large language models (LLMs) faces challenges in filtering irrelevant information and in dealing with rising computational costs. Inspired by the concept of selective attention in the human brain, we introduce a memory discrimination task. To address large-scale interactions and diverse memory standards in this task, we propose a Scene-Aware Memory Discrimination method (SAMD), which comprises two key components: the Gating Unit Module (GUM) and the Cluster Prompting Module (CPM). GUM enhances processing efficiency by filtering out non-memorable interactions and focusing on the salient content most relevant to application demands. CPM establishes adaptive memory standards, guiding LLMs to discern what information should be remembered or discarded. It also analyzes the relationship between user intents and memory contexts to build effective clustering prompts. Comprehensive direct and indirect evaluations demonstrate the effectiveness and generalization of our approach. We independently assess the performance of memory discrimination, showing that SAMD successfully recalls the majority of memorable data and remains robust in dynamic scenarios. Furthermore, when integrated into personalized applications, SAMD significantly enhances both the efficiency and quality of memory construction, leading to better organization of personal knowledge.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"338 ","pages":"Article 115496"},"PeriodicalIF":7.6,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175152","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

A survey on anomaly segmentation in urban scene understanding with image data 基于图像数据的城市场景理解异常分割研究

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-09 DOI: 10.1016/j.knosys.2026.115521

Yuxuan Zhang , Shuchang Wang , Zhenbo Shi , Wei Yang

Semantic segmentation has made significant advancements over the past decade. However, it typically relies on a closed-set taxonomy, which limits its ability to generalize to objects of unknown categories. This limitation poses security risks in real-world applications, such as autonomous vehicles. To address this issue, anomaly segmentation in urban scene understanding has gained considerable attention, aiming to identify and segment outliers effectively. Considering the rapid progress in anomaly segmentation in recent years, there is no comprehensive survey of the latest developments in this field. In this paper, we systematically summarize recent advancements and introduce a novel perspective to categorize these approaches based on their underlying motivations. We then analyze the performance of each approach on several public leaderboards, demonstrating that this categorization criteria reflects the development trends of recent progress. Additionally, we identify existing challenges and outlook for potential future research directions.

语义分割在过去十年中取得了重大进展。然而，它通常依赖于闭集分类法，这限制了它泛化到未知类别对象的能力。这一限制在自动驾驶汽车等现实应用中带来了安全风险。为了解决这一问题，城市场景理解中的异常分割得到了广泛的关注，旨在有效地识别和分割异常点。由于近年来异常分割研究进展迅速，目前还没有对该领域的最新进展进行全面的综述。在本文中，我们系统地总结了最近的进展，并介绍了一种新的视角，根据它们的潜在动机对这些方法进行分类。然后，我们分析了每种方法在多个公共排行榜上的表现，证明这种分类标准反映了近期进展的发展趋势。此外，我们还确定了现有的挑战和对未来潜在研究方向的展望。

引用次数: 0

Neural probabilistic logic learning: A method for knowledge graph reasoning 神经概率逻辑学习：一种知识图推理方法

IF 7.6 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2026-02-09 DOI: 10.1016/j.knosys.2026.115513

Fengsong Sun , Xianchao Zhang , Jinyu Wang , Zhiguo Jiang

Knowledge graph (KG) reasoning aims to predict missing facts from known data. While rule-based methods achieve high precision, they suffer from scalability limitations in large-scale KGs. Conversely, embedding-based approaches scale efficiently but often compromise precision. To address this trade-off, we propose Neural Probabilistic Logic Learning (NPLL), a novel hybrid framework that simultaneously enhances accuracy and efficiency. NPLL integrates a scoring module to augment the expressive capacity of embedding networks without sacrificing model simplicity or reasoning performance. Furthermore, interpretability is improved through the integration of a Markov Logic Network (MLN) with variational inference. Extensive evaluations on eleven benchmark datasets demonstrate that NPLL consistently outperforms state-of-the-art methods in both accuracy and computational efficiency, yielding substantial improvements in reasoning quality.

知识图（KG）推理旨在从已知数据中预测缺失的事实。虽然基于规则的方法实现了高精度，但它们在大规模的KGs中受到可扩展性的限制。相反，基于嵌入的方法可以有效地扩展，但往往会损害精度。为了解决这种权衡，我们提出了神经概率逻辑学习（NPLL），这是一种同时提高准确性和效率的新型混合框架。NPLL集成了一个评分模块，在不牺牲模型简单性或推理性能的情况下增强嵌入网络的表达能力。此外，通过将马尔可夫逻辑网络（MLN）与变分推理相结合，提高了可解释性。对11个基准数据集的广泛评估表明，NPLL在准确性和计算效率方面始终优于最先进的方法，在推理质量方面取得了实质性的改进。

引用次数: 0