Knowledge-Based Systems最新文献

{"title":"Disentangled and reassociated deep representation for dynamic survival analysis with competing risks","authors":"Chang Cui ,&nbsp;Yongqiang Tang ,&nbsp;Wensheng Zhang","doi":"10.1016/j.knosys.2025.113295","DOIUrl":"10.1016/j.knosys.2025.113295","url":null,"abstract":"<div><div>Survival analysis has been extensively utilized to analyze when the event of interest occurs. However, most of present studies merely focus on single risk and static data, while incapable of handling the scenario where competing risks and longitudinal observations are involved, which is prevalent in clinical practice, especially in the ICU. Although some impressive progress has been made in recent years, they generally utilize a single encoder to learn patient representations and input identical representations into each cause-specific subnetwork to learn the survival distribution of competing risks, thereby neglecting the specificity and association of each risk factor. In this study, we propose a novel model, namely competing risks disentangled and reassociated deep representation for dynamic survival analysis. On one hand, we propose risks-disentangled autoencoders to learn specific representations for each risk factor with contrastive learning. On the other hand, a risks-reassociated representation fusion module is proposed to explicitly learn the association relationships among competing risk representations with attention mechanism. Through extensive experiments on two popular clinical datasets, i.e., MIMIC-III and eICU, we demonstrate that our proposed model achieves advanced survival prediction performance. Visualization and interpretability analysis experiments are also provided to indicate the superior performance of our model.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113295"},"PeriodicalIF":7.2,"publicationDate":"2025-03-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621456","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":"Transferable and discriminative broad network for unsupervised domain adaptation","authors":"Liujian Zhang ,&nbsp;Zhiwen Yu ,&nbsp;Kaixiang Yang ,&nbsp;Bin Wang ,&nbsp;C.L. Philip Chen","doi":"10.1016/j.knosys.2025.113297","DOIUrl":"10.1016/j.knosys.2025.113297","url":null,"abstract":"<div><div>Unsupervised domain adaptation uses labeled data from a source domain to train a robust classifier for an unlabeled target domain with a distinct distribution. The Broad Learning System (BLS), known for its efficiency and effectiveness, is widely applied in classification and regression problems. This paper introduces a novel method named TD-BLS for unsupervised domain adaptation. TD-BLS combines UDA-BLSAE and discriminative BLS into an iterative network. UDA-BLSAE performs domain adaptation and data reconstruction simultaneously, balancing the preservation of intrinsic structure with the reduction of distribution discrepancy. Additionally, the discriminative BLS used in TD-BLS employs pseudo-labeling and manifold learning in the classifier stage to leverage high-confidence predictions and data geometric information. Finally, experiments on multiple public domain adaptation datasets demonstrate that our approach achieves rapid domain adaptation with higher accuracy compared to existing methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113297"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621455","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":"New generation thermal traffic sensor: A novel dataset and monocular 3D thermal vision framework","authors":"Arnd Pettirsch ,&nbsp;Alvaro Garcia-Hernandez","doi":"10.1016/j.knosys.2025.113334","DOIUrl":"10.1016/j.knosys.2025.113334","url":null,"abstract":"<div><div>Applications like traffic safety analysis require highly accurate trajectory data in world coordinates of traffic participants. While systems like LiDAR or stereo cameras can provide such data, they are costly, sensitive to weather and lighting conditions, and may raise privacy concerns. Thermal roadside cameras offer a robust, privacy-compliant alternative. However, monocular thermal cameras face challenges due to the ambiguous relationship between pixel coordinates and world coordinates. Existing methods for monocular 3D detection from RGB roadside cameras often rely on simplifications or the complex task of depth estimation, which limits their effectiveness. Additionally, no dataset currently exists for monocular 3D detection using thermal roadside imagery. This work introduces a dataset of 9,591 thermal images annotated in 3D world coordinates, including detailed camera calibration and surface models. It proposes a lightweight neural network architecture leveraging a projection-based method to incorporate road surface information. By detecting bottom-center contact points in image space and projecting them into 3D, the presented framework efficiently estimates object's position, dimensions, and orientations in 3D. The presented approach outperforms homography-based methods by 25 percentage points in mean average precision (mAP). It achieves real-time performance with 54 FPS on a GPU server and 17 FPS on an NVIDIA Jetson Xavier NX, making it suitable for edge deployment. Unlike RGB-based systems, our method ensures data privacy and remains effective in diverse weather and lighting conditions, enabling reliable trajectory analysis and near-miss detection for traffic safety applications. Readers can find the dataset here: <span><span>https://doi.org/10.17632/tw6ghtv624.1</span><svg><path></path></svg></span>. The code used in this work is available here: <span><span>https://github.com/4rnd25/new_generation_thermal_traffic_sensor</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113334"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143637197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A similarity learning network for unsupervised deformable brain MRI registration","authors":"Yuan Chang,&nbsp;Zheng Li,&nbsp;Zhenyu Xu","doi":"10.1016/j.knosys.2025.113291","DOIUrl":"10.1016/j.knosys.2025.113291","url":null,"abstract":"<div><div>Deformable image registration is a fundamental task in medical image analysis, aiming to accurately align medical images from different time points or patients. In recent years, learning-based unsupervised deformable registration has received significant attention due to its fast end-to-end registration capability. With the development of deep learning, deformable registration networks that use various advanced network architectures also shown increasingly better registration performance. However, most recent methods have mainly focused on replacing specific layers in networks with advanced network architectures such as Transformers, without specifically addressing the key issues of feature extraction and matching in the registration task itself. In this paper, we explore the key reasons for improving registration performance using Transformers and propose a novel similarity learning network (SLNet) for unsupervised deformable brain MRI registration. In SLNet, we propose: (i) a dual-stream encoder with saliency feature enhancement (SFE) that independently extracts hierarchical features from each image using a dual-stream structure and identifies salient features by computing similarity matrices within features, and (ii) a progressive decoder with similarity feature matching (SFM) that achieves explicit feature matching by computing similarity matrices between features and progressively estimates the final deformation field in a coarse-to-fine manner. Comprehensive experiments are conducted on four publicly available 3D brain MRI datasets (OASIS, IXI, Mindboggle, and LPBA). The results demonstrate that our SLNet achieves state-of-the-art performance, with a DSC improvement of at least 4.7% and an ASD reduction of at least 0.2 mm compared to the representative VoxelMorph.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113291"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143629538","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":"BEFM: A balanced and efficient fine-tuning model in class-incremental learning","authors":"Lize Liu,&nbsp;Jian Ji,&nbsp;Lei Zhao","doi":"10.1016/j.knosys.2025.113298","DOIUrl":"10.1016/j.knosys.2025.113298","url":null,"abstract":"<div><div>The ultimate objective of class-incremental learning (CIL) is to solve the stability–plasticity dilemma by continuously learning new courses without losing what has been learnt from the previous ones. Typical CIL methods tend to favor sample replay and parameter regularization, but recent studies have shown that exploiting and adapting historical models can effectively improve model performance. We present A Balanced and Efficient Fine-tuning Model (BEFM) to improve memory consumption efficiency and prevent catastrophic forgetting in CIL by utilizing historical models. In order to fully utilize various normalization techniques and preserve the model’s high plasticity and stability throughout training, we first create a multi-normalization module to replace the original single batch normalization procedure. On the other hand, we construct a model expansion method with knowledge distillation and offer a logit fine-tuning strategy to increase memory and computational efficiency. The model expansion strategy effectively solves the memory problem caused by feature aggregation by expanding only the deeper models with greater influence when learning a new task, while the knowledge distillation strategy encourages the model to retain the memory of the old task and improves the stability of the model. As a way to address the issue of class imbalance, the logit fine-tuning technique optimizes the standard softmax cross entropy, improving classifier design without adding computational burden. We test our method on the CIFAR10, CIFAR100, and miniImageNet100 datasets under various conditions. According to experimental results, our strategy outperforms current approaches with notable gains in performance. The code is available at <span><span>https://github.com/Lize-Liu/BEFM-main</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113298"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143627964","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":"Synthesizing global and local perspectives in contrastive learning for graph anomaly detection","authors":"Qiqi Yang,&nbsp;Hang Yu,&nbsp;Zhengyang Liu,&nbsp;Pengbo Li,&nbsp;Xue Chen,&nbsp;Xiangfeng Luo","doi":"10.1016/j.knosys.2025.113289","DOIUrl":"10.1016/j.knosys.2025.113289","url":null,"abstract":"<div><div>Graph data has shown explosive growth, with application scenarios covering social networks, e-commerce networks, financial transaction networks, etc. In this context, graph anomaly detection is particularly important, aiming to prevent various malicious activities. Existing approaches, however, are still limited in that they either ignore global information and focus only on aggregating neighbor information of the target node, or they utilize global context as a supervisory signal while ignoring local information. In certain scenarios, anomalies can only be detected in a single view (global or local). Furthermore, the issue of class imbalance in graph-based anomaly detection is exacerbated by the significant disparity between the number of benign user samples and anomalous samples in real-world scenarios. As a solution to the above challenges, we present a framework for synthesizing Global and Local perspectives in Contrastive Learning (GALCL). GALCL leverages multi-view contrast to integrate both global and local information. By using node-graph and node-subgraph cross-scale contrasts, the framework enhances the prominence of local and global information, thereby capturing anomaly information that might be missed by focusing solely on the global or local level. In addition, a class-wise loss function is adopted to alleviate class imbalances on the graph. Comprehensive experiments conducted on eight real-world datasets demonstrate that our method outperforms the current state-of-the-art methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113289"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621232","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":"Bidirectional alignment text-embeddings with decoupled contrastive for sequential recommendation","authors":"Piao Tong ,&nbsp;Qiao Liu ,&nbsp;Zhipeng Zhang ,&nbsp;Yuke Wang ,&nbsp;Tian Lan","doi":"10.1016/j.knosys.2025.113290","DOIUrl":"10.1016/j.knosys.2025.113290","url":null,"abstract":"<div><div>The key challenge in sequential recommendation is to accurately predict the next item based on historical interaction sequences by learning effective sequence representations. Existing models typically optimize sequence representations using the next ground-truth item as the supervised signal. However, this approach often results in biased interest representations and neglects the benefits of bidirectional supervision, leading to incomplete sequence representations and semantic mismatches. To address these limitations, we propose <strong>ADRec</strong> for bidirectional sequence–item <strong>A</strong>lignment text-embeddings with <strong>D</strong>ecoupled contrastive learning for sequential <strong>Rec</strong>ommendation based only on text data. ADRec combines self-supervised and supervised signals derived from intrinsic correlations in recommendation data, to enhance semantic consistency between sequence and ground-truth item representations, improving recommendation performance. Specifically, we introduce a hybrid learning mechanism that integrates an unsupervised contrastive learning paradigm to decouple sequence and item representations and supervised contrastive learning to achieve bidirectional semantic alignment. Additionally, a dual-momentum queue mechanism is devised to expand the diversity of negative samples with limited resources, optimizing the quality of user interest representations in the text modality. Extensive experiments on six public datasets show that ADRec consistently outperforms state-of-the-art methods by learning superior sequence representations. The code is publicly available at <span><span>https://github.com/pppiao/ADRec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113290"},"PeriodicalIF":7.2,"publicationDate":"2025-03-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621391","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":"Twin Q-learning-driven forest ecosystem optimization for feature selection","authors":"Hongbo Zhang,&nbsp;Jinlong Li,&nbsp;Xiaofeng Yue,&nbsp;Xueliang Gao,&nbsp;Haohuan Nan","doi":"10.1016/j.knosys.2025.113323","DOIUrl":"10.1016/j.knosys.2025.113323","url":null,"abstract":"<div><div>Feature selection (FS) enhances the performance of the classification model by selecting relevant features and discarding unnecessary ones. Due to the efficiency of metaheuristic algorithms in solving FS problems, they have drawn much attention. However, the previous metaheuristic-based FS methods have drawbacks, such as easily falling into local optima and limited utilization of FS characteristics. To address these problems, we propose a novel twin Q-learning-driven forest ecosystem optimization named TQFEO for FS problems. Initially, an ordinal number initialization strategy is developed to guarantee the quality of initial individuals at the initial stage. Specifically, a twin Q-learning-driven forest ecosystem is constructed to ensure the algorithm's adaptive capability. Furthermore, a fitness-variance-evaluation-based status detection strategy is proposed to perceive optimization status. If an abnormality is detected, low-quality individuals are to be processed. Finally, a Manhattan distance guides position update and elite random walk strategy is designed to maintain population diversity and accelerate the convergence rate. Experimental results on 20 benchmark datasets across various domains demonstrate that TQFEO outperforms conventional and recent metaheuristic algorithms.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113323"},"PeriodicalIF":7.2,"publicationDate":"2025-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621457","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":"TCrossE: Cross-space interaction of bicomplex and quaternion embeddings for temporal knowledge graph completion","authors":"Thanh Vu ,&nbsp;Thanh Le","doi":"10.1016/j.knosys.2025.113321","DOIUrl":"10.1016/j.knosys.2025.113321","url":null,"abstract":"<div><div>Completing temporal knowledge graphs is essential to ensuring their readiness for real-world applications. Temporal knowledge graph completion addresses this challenge by predicting missing temporal facts and enriching the knowledge base over time. However, existing models face key challenges: translation-based models offer interpretability but underperform, while neural network-based models achieve high accuracy but lack transparency in how they capture structural and temporal dependencies. To address these challenges, we propose TCrossE (Temporal Cross-Space Embedding), a novel model that fuses bicomplex and quaternion spaces to enhance the representation of temporal structures. By leveraging rotations in hypercomplex spaces, TCrossE creates hybrid embeddings that effectively model both structural relationships and temporal dependencies. The fusion of bicomplex and quaternion spaces is mathematically motivated and validated through empirical studies. Unlike prior models, TCrossE balances expressiveness and interpretability, ensuring strong performance without sacrificing model transparency. Additionally, our approach optimizes training efficiency, making it more practical for large-scale TKG applications. We evaluate TCrossE on five benchmark datasets: ICEWS14, ICEWS05–15, GDELT, WIKIDATA12k, and YAGO11k, covering a diverse range of temporal knowledge graph structures. Experimental results show that TCrossE outperforms state-of-the-art models, achieving up to 18 % improvement on GDELT and YAGO11k while maintaining competitive performance on other datasets. Furthermore, TCrossE exhibits lower training times, making it suitable for real-world deployment.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113321"},"PeriodicalIF":7.2,"publicationDate":"2025-03-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143621392","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":"Graph knowledge tracing in cognitive situation: Validation of classic assertions in cognitive psychology","authors":"Qianxi Wu ,&nbsp;Weidong Ji ,&nbsp;Guohui Zhou ,&nbsp;Yingchun Yang","doi":"10.1016/j.knosys.2025.113281","DOIUrl":"10.1016/j.knosys.2025.113281","url":null,"abstract":"<div><div>Knowledge Tracing (KT) is a fundamental and challenging task in intelligent education, aiming to trace learners’ knowledge states and learning processes, providing better support and guidance for teaching and addressing mental factors. Previous KT tasks have focused on considering learners’ exposure to extrinsic environmental factors while ignoring the influence of intrinsic psychological factors. Moreover, previous methods have adopted a single perspective in modeling learners’ knowledge states, ignoring the diversity of states in the learning process. To address these issues, we define the concept of <em>cognitive situation</em> through the guidance of cognitive psychology theory to help to explain the extrinsic influence and intrinsic cognition of learners within complex learning environments. Moreover, we design a Cognitive Situation-based Graph KT (CSGKT) model to quantify learners’ influences in the cognitive process by modeling schemas capturing intrinsic characteristics and extrinsic factors through Hyper-Graph Neural Networks (HGNN). Second, we utilize a Directed Graph Convolutional Neural Network (DGCNN) to capture the correlation information between knowledge concepts and structure the learner’s cognitive activities and knowledge states, adding a detailed representation of multiple states of the learning process. In addition, we use the Erase-add Gate to filter out the knowledge states that do not match the learner’s current cognitive activities to stabilize the learner’s due cognition. In our experiments, we selected nine baseline models from three mainstream approaches, including sequence-based approaches, <em>Transformer</em>-based approaches, and complex structure-based approaches. The experimental results show that our models outperform these baseline models. At the same time, we also verify two classic assertions in cognitive psychology, namely, the “short-term memory forgetting of knowledge concepts is mainly caused by interference rather than memory trace fading” and the “cognitive imagery and perceptual function play an equivalent role in the cognitive process”, which further support the feasibility of the model.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113281"},"PeriodicalIF":7.2,"publicationDate":"2025-03-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143592859","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}