Knowledge-Based Systems最新文献

{"title":"Heterophily-aware dynamic hypergraph for semi-supervised classification","authors":"Shaojun Liang ,&nbsp;Ying Zheng ,&nbsp;Housheng Su ,&nbsp;Lei Zhang ,&nbsp;Yi Yang","doi":"10.1016/j.knosys.2025.114435","DOIUrl":"10.1016/j.knosys.2025.114435","url":null,"abstract":"<div><div>Hypergraph neural networks, as high-order graph neural networks, excel in handling intricate relationships within non-Euclidean infinite-dimensional spaces. However, conventional homophily assumption-based hypergraph methods exhibit limited effectiveness in semi-supervised classification scenarios involving heterophily problem, where neighboring nodes often belong to dissimilar categories. To address this challenge, this paper proposes a Heterophily-Aware Dynamic Hypergraph (HADHG) framework grounded in heterophily assumption through label domain analysis. The framework comprises three key components: a hypergraph-oriented label propagation method for deriving class-specific label features, a label tensor construction approach characterizing node-level heterophily intensity via 2D tensors, and a center attention mechanism that dynamically optimizes hypergraph structures. By enabling nodes to dynamically reconfigure the local graph structure based on microscopic heterophily intensity, HADHG effectively mitigates heterophily interference. Comprehensive experiments using real-flight data from Unmanned Aerial Vehicles and the public Gear dataset highlight the framework’s superiority over state-of-the-art methods. The codes and datasets are openly available at <span><span>https://github.com/DL-LEO/HADHG</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114435"},"PeriodicalIF":7.6,"publicationDate":"2025-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145120251","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}

{"title":"Multi-domain weakly decoupled domain generalization network for fault diagnosis under unknown operating conditions","authors":"Yawei Sun ,&nbsp;Hongfeng Tao ,&nbsp;Vladimir Stojanovic","doi":"10.1016/j.knosys.2025.114452","DOIUrl":"10.1016/j.knosys.2025.114452","url":null,"abstract":"<div><div>The utilization of transfer learning strategies to solve cross-domain fault diagnosis problems has achieved significant results. However, most existing multi-source domain generalization fault diagnosis methods use a single classifier or introduce auxiliary classifiers, focusing on learning domain-invariant features or global feature distribution matching. Furthermore, since the data distributions of different source domains may be significantly different, this may lose the data distribution information specific to each source domain. In addition, how to reduce the variation in risk between samples within the same domain training is also a challenging issue. Finally, it is also crucial to balance the predictive outputs of multiple classifiers to adapt them to the data distribution of the target domain. Based on the above challenges, this paper proposes a multi-domain weakly decoupled domain generalization network for fault diagnosis under unknown operating conditions. Feature weakly decoupled mechanism is achieved by employing multiple classifiers and incorporating the variance of samples within the same sample domain as a penalty term. This reduces the model’s sensitivity to changes in the extreme distribution of samples within the domain. Classifier weakly decoupled mechanism, on the other hand, reduces the inter-domain risk variance by minimizing the loss of variance in the predicted output of the source domain classifiers. This improves the robustness of the model to inter-domain distributional changes and covariate changes. Experimental results on three datasets validate the effectiveness and general applicability of the proposed approach.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114452"},"PeriodicalIF":7.6,"publicationDate":"2025-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145108882","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}

{"title":"A unified multi-subgraph pre-training framework for spatio-temporal graph","authors":"Mingze Zhong ,&nbsp;Zexuan Long ,&nbsp;Xinglei Wang ,&nbsp;Tao Cheng ,&nbsp;Meng Fang ,&nbsp;Ling Chen","doi":"10.1016/j.knosys.2025.114428","DOIUrl":"10.1016/j.knosys.2025.114428","url":null,"abstract":"<div><div>Spatio-temporal graph (STG) learning has shown great potential in capturing complex spatio-temporal dependencies and has achieved significant success in various fields such as traffic flow prediction, climate forecasting, and epidemiological spread research. By learning general features from spatio-temporal graphs, pre-trained graph models can capture hidden semantic information in the data, thereby enhancing the learning effect of downstream tasks and improving overall model performance. However, most existing spatio-temporal graph learning methods use the entire graph for training, which may not fully capture local structure and feature information. In addition, existing methods usually adopt sequence modeling techniques without fully considering the time decay effect, i.e., the need to apply decaying attention to distant time steps. To address these issues, this paper proposes a <u>u</u>nified dual-phase <u>m</u>ulti-<u>s</u>ubgraph pre-training <u>s</u>patio-<u>t</u>emporal graph framework (UMSST). Specifically, in the first phase, the framework learns the global representation of the spatio-temporal graph and locates key graph nodes, while learning the “unit representations” of these key nodes. In the second phase, multiple spatio-temporal subgraphs are constructed based on these “unit representations” to further capture the implicit encoding information of more general features around the corresponding subgraphs, thereby helping the model make full use of general features. Experimental results on real datasets show that the proposed pre-trained spatio-temporal graph framework significantly improves the performance of downstream tasks and demonstrates its effectiveness in comparison with recent strong baseline models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114428"},"PeriodicalIF":7.6,"publicationDate":"2025-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145108887","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}

{"title":"Feature reduction causal network (FRCN): A novel approach for analyzing coupling relationships in radar system","authors":"Chenfeng Wang ,&nbsp;Xiaoguang Gao ,&nbsp;Zidong Wang ,&nbsp;Bo Li ,&nbsp;Kaifang Wan ,&nbsp;Xinyu Li ,&nbsp;Chuchao He","doi":"10.1016/j.knosys.2025.114484","DOIUrl":"10.1016/j.knosys.2025.114484","url":null,"abstract":"<div><div>To evaluate radar performance in complex electromagnetic environments, a compact and efficient causal model is required to model such a complex, nonlinear high-stakes problem. Hence, in this paper, we propose a feature reduction causal network (FRCN). Firstly, to determine the number of hidden layer features in the FRCN, a feature extraction strategy is designed using the intrinsic dimension (ID) of raw data as key prior knowledge, thereby reducing modeling complexity and improving computational efficiency. Then, to further reveal the causal relationships between features and the final objective, a Bayesian network (BN) is constructed in the task layer, intuitively showing the coupling relationships through a directed graph and providing interpretability for decisions on high-stakes problems. Moreover, we extend the layer-wise relevance propagation to the BN in the FRCN, enabling bidirectional reasoning throughout the entire process, which is beneficial to understand the model and its behavior in a human-understandable way. In experiments, it is proved that ID plays a significance role in feature number selection. Next, we design a new interpretable evaluation indicator, called decision-specific average edge relevance, to quantify interpretability. Compared to eight representative models, FRCN not only achieves higher accuracy but also provides stronger interpretability in terms of relevance, informativeness, and trustworthiness. A detailed analysis of a radar system enhances the understanding of coupling relationships among various factors, thereby validating the effectiveness of FRCN in feature reduction, interpretability, and trustworthiness for high-dimensional, complex, and nonlinear data.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114484"},"PeriodicalIF":7.6,"publicationDate":"2025-09-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145159235","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}

{"title":"Provide explainable clues: A generative traceable method for knowledge graph completion","authors":"Ziqi Ma ,&nbsp;Jinpeng Li ,&nbsp;Hang Yu","doi":"10.1016/j.knosys.2025.114426","DOIUrl":"10.1016/j.knosys.2025.114426","url":null,"abstract":"<div><div>Improving the quality of Knowledge Graph Completion (KGC) results is an essential topic in the field of knowledge graphs. Recently, generative models (GMs) have gained widespread attention for addressing the generalization issues of traditional approaches. However, the black-box nature of generative models often leads to hallucinations, which reduce the model’s performance. Most methods attempt to mitigate this issue through retrieval enhancement and decoding constraints. However, they overlook one major cause of hallucinations–poor explainability. Based on this concept, we propose a <strong>G</strong>enerative <strong>T</strong>raceable <strong>M</strong>ethod, namely GTM, which aims to improve the KGC capability of GMs by exploring the inhibitory effect of explainability on hallucinations. In GTM, a clue tracker is used to find contextual evidence for explainability. In addition, to measure explainability clues, we propose a context-aware analyzer, which enhances the understanding of context through group analogy. In the reasoning phase, we ensure the validity of the generated results by integrating the interpretive capability of clues. Extensive experiments have demonstrated that GTM can adapt to various KGC tasks and significantly enhance the performance of KGC models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114426"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145108946","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}

{"title":"CLAV: clustering latent vector aggregation for whole slide image retrieval leveraging foundation models","authors":"Alejandro Golfe ,&nbsp;Pablo Meseguer ,&nbsp;Valery Naranjo ,&nbsp;Adrián Colomer","doi":"10.1016/j.knosys.2025.114423","DOIUrl":"10.1016/j.knosys.2025.114423","url":null,"abstract":"<div><div>Content-Based Image Retrieval (CBIR) is crucial in cancer diagnosis, assisting pathologists by providing similar image data from previous records for analysis, especially when there is uncertainty in diagnosing a case. This process supports decision-making by providing valuable reference points to guide the diagnostic process. Foundation models have become increasingly important in the medical field due to their ability to generalize across various tasks and datasets, offering valuable support to pathologists by enhancing the accuracy and efficiency of diagnostic processes. In this article, a foundation model pre-trained on histopathology data is leveraged as a feature extractor without the need for task-specific training, in contrast to existing models that require extensive training to learn significant data representations. The proposed method, Clustering Latent Vector Aggregation (CLAV), condenses the significant feature vectors into a unique representative vector for the Whole Slide Image (WSI). Using a unique feature vector offers the advantage of reducing the size of the memory bank, thereby making the process of querying and retrieving similar WSIs more efficient. The experimental results presented in this study demonstrate that the proposed method enhances performance in CBIR tasks. This article highlights the potential of foundation models to achieve superior retrieval metrics compared to state-of-the-art methods specifically trained for CBIR.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"329 ","pages":"Article 114423"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145095599","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}

{"title":"A novel TriCore scheme for multiple RGB images in telemedicine environments","authors":"Muhammad Tanveer Hussain ,&nbsp;Imrana Shafique ,&nbsp;Shamsher Ullah","doi":"10.1016/j.knosys.2025.114451","DOIUrl":"10.1016/j.knosys.2025.114451","url":null,"abstract":"<div><div>The scientific community is paying high attention to propose efficient solution to the open problem related to security and privacy of still visual communications. Telemedicine is one of the real world applications, experiencing such security concerns related to virtual consultation. Among the presence of already defined encryption schemes, this paper is dedicated to define TriCore, an efficient encryption scheme for multiple RGB images, inspired by three core components: <span><math><msub><mi>σ</mi><mo>∃</mo></msub></math></span>-permutation, Laplacian matrix and 4D hyper chaotic system. The use of Secure Hash Algorithm-256 (SHA-256) in key generation assures the randomness and make it highly sensitive. Also, this paper presents a novel <span><math><msub><mi>σ</mi><mo>∃</mo></msub></math></span>-permutation, to enhance the randomness effect in the corresponding cipher image. The scheme mainly deals with the matrices corresponding to the three channels of the merged image, undergoing <span><math><msub><mi>σ</mi><mo>∃</mo></msub></math></span>-permutation and XOR operations with pair of channel matrices and key matrices, a single cipher image is produced corresponding to multiple RGB images. It prevents the revelation of actual number of shared images and make the scheme more strong. Obtaining the ideal values, experimental results witness the efficiency of TriCore scheme. The entropy is measured as 7.9998. The high resistance against the differential attacks is measured through the Number of Pixel Change rate (99.6158<span><math><mo>%</mo></math></span>) and Unified Average Changing Intensity (33.4621<span><math><mo>%</mo></math></span>). The execution time for 4 images each of size 256 <span><math><mo>×</mo></math></span> 256 is just 0.173470 s. The experimental results show that this paper efficiently facilitates secure communications in telemedicine providing higher security at lowest computational costs.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"329 ","pages":"Article 114451"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145095593","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}

{"title":"Integrating structural and operational knowledge into multi-state system modeling: Application in urban infrastructures","authors":"Henrique O. Caetano ,&nbsp;Luiz Desuó N ,&nbsp;Marco Aiello ,&nbsp;Carlos D. Maciel","doi":"10.1016/j.knosys.2025.114457","DOIUrl":"10.1016/j.knosys.2025.114457","url":null,"abstract":"<div><div>Modern engineering systems, with their increasing complexity driven by technological advancements and growing interdependencies among components, present a challenge to traditional binary-state models. These models, which classify components as either fully operational or failed, are insufficient for capturing the progressive degradation, redundancy mechanisms, and cascading effects observed in real-world systems. Multi-State System (MSS) modeling, which represents intermediate operability states, is a step forward. However, the current literature overlooks a crucial information source: the system’s internal dynamics. These dynamics, which play a crucial role in shaping the system’s behavior, can be leveraged to enhance the learning process in MSS modeling. This study introduces a novel hybrid MSS modeling methodology that incorporates a system’s internal dynamic - such as network topology, redundancy mechanisms, and operational constraints - within an MSS. The methodology is first applied to a Brazilian power system, demonstrating how internal system characteristics influence the state evolution of individual components over time. This evaluation highlights the ability of the model to capture nuanced operational behavior driven by system-level constraints. The methodology is tested on multiple European transmission systems in a second stage to assess its predictive performance in estimating key reliability metrics. The proposed approach consistently outperforms existing models, achieving significantly lower prediction errors by accounting for internal constraints and the system’s dynamics. This work offers a generalizable solution for critical infrastructure planning across domains, enhancing MSS reliability modeling in various engineering systems.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"329 ","pages":"Article 114457"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145095592","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}

{"title":"Stock conditional drawdown at risk portfolio optimization based on gated bidirectional temporal convolution and discrete cosine graph neural networks on hypervariable graphs","authors":"Chia-Hung Wang ,&nbsp;Chiwang Lin","doi":"10.1016/j.knosys.2025.114456","DOIUrl":"10.1016/j.knosys.2025.114456","url":null,"abstract":"<div><div>With the development of machine learning technology, the application of stock prediction in financial portfolio optimization has become increasingly important. This study proposes an intelligent portfolio optimization method that combines gated bidirectional temporal convolution-discrete cosine graph neural network (TDGNN) with the mean-conditional drawdown at risk (Mean-CDaR) model, aiming to improve the risk-return performance of the portfolio. The method consists of two main stages: first, the data is converted into a hypervariable graph through the TDGNN model, the gated bidirectional temporal convolution layer is used to capture the temporal dynamic characteristics, and the discrete cosine graph neural network is combined to effectively model the complex spatiotemporal relationship in the stock market; second, the Mean-CDaR model is used for portfolio optimization, and the maximum drawdown is used as a measurement indicator to achieve precise risk control. Experimental results show that on the CSI 300, S&amp;P500, and Nikkei 225 data sets, TDGNN and Mean-CDaR models perform significantly better than traditional methods, with <span><math><msup><mrow><mi>R</mi></mrow><mn>2</mn></msup></math></span> of 0.9991, 0.9991, and 0.9983, respectively. Under the assumption of no transaction costs, the cumulative returns are 0.42, 0.62, and 0.93, respectively; considering 0.05 % transaction costs, the cumulative returns are 0.1, 0.25, and 0.49, respectively. The study shows that this method not only effectively captures the spatiotemporal dependency of stock data but also effectively controls risks while improving returns, providing investors with a robust and efficient decision support system.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"329 ","pages":"Article 114456"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145057305","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}

{"title":"Structural chain of thoughts for radiology education","authors":"Akash Awasthi ,&nbsp;Brandon Chung ,&nbsp;Anh Mai Vu ,&nbsp;Saba Khan ,&nbsp;Ngan Le ,&nbsp;Zhigang Deng ,&nbsp;Rishi Agrawal ,&nbsp;Carol C. Wu ,&nbsp;Hien Van Nguyen","doi":"10.1016/j.knosys.2025.114433","DOIUrl":"10.1016/j.knosys.2025.114433","url":null,"abstract":"<div><div>Radiology education requires trainees to develop both perceptual and interpretive expertise. However, refinement of these skills is often impeded by the limited availability of mentorship, a consequence of the demanding schedules of experienced radiologists. This lack of personalized guidance makes it difficult for learners to recognize the mistakes they make, understand why those errors occurred and how to refine their perceptual processes. Many of these errors arise from subtle differences in visual attention, such as failing to fixate on an abnormality, allocating an insufficient fixation time, or overlooking an abnormality despite scanning the correct region. Although Large Language Models (LLMs) and Large Multimodal Models (LMMs) have been explored for radiology tasks, they often struggle to detect such fine-grained multimodal variations, particularly when comparing gaze behavior between experts and trainees. To address these limitations, we introduce Structural Chain of Thoughts (SCoT), a novel framework that enhances LLMs and LMMs sensitivity to nuanced multimodal differences by structuring gaze data and radiology report into a thought graph. By leveraging a structural prior, SCoT systematically identifies key perceptual and interpretive discrepancies, allowing models to provide targeted, context-aware feedback. This structured approach not only highlights missed findings but also explains the reasoning behind perceptual errors, turning them into learning opportunities. Applied within radiology education, SCoT bridges the gap between expert and novice performance, offering a scalable solution for AI-driven diagnostic training. We further contribute a simulated dataset of perceptual errors in chest X-ray (CXR) interpretation, facilitating future research into multimodal reasoning and AI-driven medical education. Unlike conventional Chain-of-Thought approaches, SCoT explicitly integrates gaze and textual information into a structured reasoning process, yielding interpretable, fine-grained, and personalized feedback tailored to the unique needs of radiology training. The code and data will be available here: <span><span>GitHub Repository</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"330 ","pages":"Article 114433"},"PeriodicalIF":7.6,"publicationDate":"2025-09-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145108888","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}