Knowledge-Based Systems最新文献_第10页

STREAM-Net: Spatio-temporal feature fusion network for robust rPPG signal measurement in remote health monitoring STREAM-Net：用于远程健康监测中鲁棒rPPG信号测量的时空特征融合网络

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-11 DOI: 10.1016/j.knosys.2025.114080

Muhammad Usman, Milena Sobotka, Jacek Ruminski

{"title":"STREAM-Net: Spatio-temporal feature fusion network for robust rPPG signal measurement in remote health monitoring","authors":"Muhammad Usman, Milena Sobotka, Jacek Ruminski","doi":"10.1016/j.knosys.2025.114080","DOIUrl":"10.1016/j.knosys.2025.114080","url":null,"abstract":"<div><div>Remote photoplethysmography (rPPG) has become a popular, non-invasive, contactless technique for detecting physiological signals with promising applications. The latest advancements employ deep learning to address the challenges, including motion artifacts, redundancy, and external noise in video-based rPPG signal extraction. To this end, we propose a bilateral spatio-temporal network for estimating blood volume pulse (BVP) signals by analyzing human physiological processes through video frames. The spatio-temporal branches leverage lateral attention and multi-scale feature integration to enhance the extraction of rPPG signals. The spatio-temporal lateral attention module integrates spatial-temporal features at higher and lower resolutions to preserve essential dependency between spatial and temporal data at different scales. Whereas, the multi-scale feature enhancement module encodes high-level features to refine spatial features further with distinct local and global representations. We conducted extensive experiments to validate the effectiveness and superior performance of the proposed method on two benchmark datasets. In cross-dataset validation, STREAM-Net achieved MAE accuracy of 1.151 and RMSE of 2.715 on the UBFC dataset, whereas MAE accuracy of 1.318 and MAPE of 1.384 on the PURE dataset. Cross-dataset testing on both benchmarks demonstrates the efficacy of our proposed approach. Given the importance of repeatability and reliability in clinical measurements, we also quantified predictive uncertainty of our model using Monte Carlo dropout. This analysis exhibited the robust performance and high repeatability of the proposed model, with uncertainty variance ranging from 0.0049 to 0.0113 across different dropout rates. The source code is publicly available at <span><span>https://github.com/usmanraza121/STREAM-Net</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114080"},"PeriodicalIF":7.2,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144654060","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

MultiFeature fusion graph attention network for aspect-based sentiment analysis 面向面向方面情感分析的多特征融合图关注网络

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-11 DOI: 10.1016/j.knosys.2025.114084

Jiaofeng Wang, Hongfang Gong, Xinyu Guo

{"title":"MultiFeature fusion graph attention network for aspect-based sentiment analysis","authors":"Jiaofeng Wang, Hongfang Gong, Xinyu Guo","doi":"10.1016/j.knosys.2025.114084","DOIUrl":"10.1016/j.knosys.2025.114084","url":null,"abstract":"<div><div>Aspect-based sentiment analysis (ABSA) is a fine-grained sentiment analysis task, which aims to predict the sentiment polarity of a given aspect in a sentence. Most of the approaches rely on syntactic and semantic parsing to derive textual insights, often overlooking how aspectual and contextual factors impact model performance. Alternatively, they focus on an in-depth study of the information in the dependency tree, thereby ignoring the importance of the constituent trees. In this work, we introduce a multifeature fusion graph attention network (MFF-GAT) model. The model constructs syntactic, semantic and contextual channels, fusing dependent syntactic information and constituent syntactic information through a gating mechanism. The semantic graph is constructed based on self-attention, and the contextual graph is constructed based on point interaction information. In addition, this study uses the multi-head attention mechanism to interact with aspects and three features and capture aspect-related information. Our MFF-GAT model performs better on the ABSA task than other baseline models, according to experiments conducted on five public datasets.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114084"},"PeriodicalIF":7.2,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144633537","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

HGphormer: Heterophilic Graph Transformer HGphormer：异亲图转换器

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-11 DOI: 10.1016/j.knosys.2025.114031

Jianshe Wu, Yaolin Liu, Yuqian Wang, Lingjie Zhang, Jingyi Ding

{"title":"HGphormer: Heterophilic Graph Transformer","authors":"Jianshe Wu, Yaolin Liu, Yuqian Wang, Lingjie Zhang, Jingyi Ding","doi":"10.1016/j.knosys.2025.114031","DOIUrl":"10.1016/j.knosys.2025.114031","url":null,"abstract":"<div><div>Graph neural networks (GNNs) have been widely used in various node-level tasks on graphs due to their powerful representation learning ability. Traditional GNNs rely on the homophily assumption that nodes with the same label in a graph tend to be connected to each other. But there are a large number of heterophily graphs in the real world, where most proximal nodes have different labels. So heterophily GNNs has been proposed, which tried to improve the performance of GNNs on heterophily graph by obtaining information from multi-hop neighbor nodes.</div><div>A promising way for heterophily graphs is Graph Transformers (GTs). Without relying on the homophily assumption, GTs aggregate information of nodes depending on their similarity, thus is suitable for both homophily and heterophily graphs. Since the quadratic time complexity of GTs, most of existing GTs focus on how to reduce the complexity and make it applicable for nodes classification, their performance is still unsatisfied in heterophily graphs.</div><div>To solve the above problem, Heterophilic Graph Transformer (HGphormer) is proposed in this paper. In order to reduce the interference between attribute embedding and structure embedding, a parallel architecture of Transformer is proposed. HGphormer also decouples the aggregated information into homophily and heterophily information and uses them adaptively to further improve the accuracy. A sample technique is proposed to sample neighbors from multiple hops and reduce the time complexity. Experiments show that the proposed HGphormer outperforms the state of the art methods on both homophily graph and heterophily graph datasets.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114031"},"PeriodicalIF":7.2,"publicationDate":"2025-07-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144614886","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

Improving adversarial transferability via adaptive ensemble attack with post-optimization 后优化自适应集成攻击提高对抗可转移性

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.114079

Yun Zhang, Yan Wo

{"title":"Improving adversarial transferability via adaptive ensemble attack with post-optimization","authors":"Yun Zhang, Yan Wo","doi":"10.1016/j.knosys.2025.114079","DOIUrl":"10.1016/j.knosys.2025.114079","url":null,"abstract":"<div><div>The transferability of adversarial examples poses serious security threats to deep neural networks (DNNs), as it enables black-box attacks without access to target model details. Studying this phenomenon is crucial for understanding model vulnerabilities and enhancing the robustness and security of DNNs. Model ensemble adversarial attacks have proven to be an effective approach for improving adversarial transferability by leveraging adversarial information of multiple surrogate models. Although existing ensemble attack methods adopt various strategies to integrate surrogate models, they tend to underutilize the strengths of each model or exhibit insufficient model diversity, thus limiting further improvement in attack transferability. To address this issue, we propose an Adaptive Ensemble Attack with Post-Optimization (AEAPO). This method adopts a strategy analogous to mini-batch processing, performing iterative attacks using randomly selected surrogate subsets to reduce memory consumption and preserve model diversity. In the model ensemble attack process, we introduce an adaptive weight adjustment to fully exploit adversarial information from surrogate models and identify commonly vulnerable directions. To mitigate local overfitting to high-weight models and gradient oscillations during weight adaptation, we propose Lookahead gradient optimization with fast and slow gradients for smoother and more generalizable optimization. Additionally, a test model set is introduced in each attack iteration to evaluate and refine adversarial examples, preventing overfitting to the fixed set of surrogate models. Extensive experiments demonstrate that AEAPO outperforms existing methods in black-box scenarios on both normally trained and defense models, and can be effectively combined with various transfer-based attacks, validating its effectiveness and generality.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114079"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144633531","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

Predictive deep reinforcement learning with multi-agent systems for adaptive time series forecasting 基于多智能体系统的自适应时间序列预测深度强化学习

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.113941

Thanveer Shaik , Xiaohui Tao , Lin Li , Haoran Xie , U.R. Acharya , Raj Gururajan , Xujuan Zhou

{"title":"Predictive deep reinforcement learning with multi-agent systems for adaptive time series forecasting","authors":"Thanveer Shaik , Xiaohui Tao , Lin Li , Haoran Xie , U.R. Acharya , Raj Gururajan , Xujuan Zhou","doi":"10.1016/j.knosys.2025.113941","DOIUrl":"10.1016/j.knosys.2025.113941","url":null,"abstract":"<div><div>Reinforcement learning has been increasingly applied in monitoring applications because of its ability to learn from previous experiences and make adaptive decisions. However, existing machine learning-based health monitoring applications are mostly supervised learning algorithms, trained on labels, and they cannot make adaptive decisions in an uncertain, complex environment. This study proposes a novel and generic system, predictive deep reinforcement learning (PDRL), with multiple RL agents in a time series forecasting environment. The proposed generic framework accommodates virtual Deep Q Network (DQN) agents to monitor predicted future states of a complex environment with a well-defined reward policy so that the agent learns existing knowledge while maximizing their rewards. In the evaluation process of the proposed framework, three DRL agents were deployed to monitor a subject’s future heart rate, respiration, and temperature predicted using a BiLSTM model. With each iteration, the three agents were able to learn the associated patterns, and their cumulative rewards gradually increased. It outperformed the baseline models for all three monitoring agents. The proposed PDRL framework achieves state-of-the-art performance in time series forecasting by effectively integrating reinforcement learning agents with deep learning-based prediction. The proposed DRL agents and deep learning model in the PDRL framework are customized to enable transfer learning in other forecasting applications like traffic and weather, and monitor their states. The PDRL framework is able to learn the future states of the traffic and weather forecasting, and the cumulative rewards are gradually increasing over each episode.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 113941"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144605858","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

Large-scale fine-grained image retrieval via Proxy Mask Pooling and multilateral semantic relations 基于代理掩码池和多边语义关系的大规模细粒度图像检索

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.114018

Xianxian Zeng , Jie Zhou , Dunhao Liu , Jun Yuan , Ming Yin , Weichao Xu , Shun Liu

{"title":"Large-scale fine-grained image retrieval via Proxy Mask Pooling and multilateral semantic relations","authors":"Xianxian Zeng , Jie Zhou , Dunhao Liu , Jun Yuan , Ming Yin , Weichao Xu , Shun Liu","doi":"10.1016/j.knosys.2025.114018","DOIUrl":"10.1016/j.knosys.2025.114018","url":null,"abstract":"<div><div>Fine-grained image retrieval poses significant challenges due to subtle inter-class differences and large intra-class variations. Attention mechanisms have been widely adopted to enhance feature discrimination, yet their serial application remains underexplored due to computational constraints and the risk of over-focusing on redundant information. Meanwhile, recent advances in attribute-aware deep hashing have underscored the interpretability of hash codes, offering a promising avenue for improving retrieval performance. To address these challenges, we propose a novel deep learning framework with two key innovations. First, the <strong>Proxy Mask Pooling Module</strong> mitigates the trade-off between computational efficiency and model performance in serial attention mechanisms while enhancing spatial and positional information. By guiding the model to focus on diverse and semantically rich regions, this module improves the discriminative power of fine-grained features at both local and global levels. Second, inspired by the interpretability of hash codes as attribute descriptors, we introduce the <strong>Multilateral Semantic Tuple Relation Strategy</strong>, which leverages a hypergraph-based structure and a novel loss function to model intricate multi-way relationships among images. This design effectively strengthens semantic representation and retrieval accuracy. Extensive experiments on multiple benchmark fine-grained image retrieval datasets demonstrate the superiority of our method, achieving notable improvements in retrieval precision, computational efficiency, and generalization ability over state-of-the-art models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114018"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144605861","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

Synthetic tabular data generation using a VAE-GAN architecture 使用VAE-GAN架构生成合成表格数据

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.113997

Dmitry Anshelevich, Gilad Katz

引用次数: 0

Tri-level interaction fusion network for graph similarity learning 图相似度学习的三层交互融合网络

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.113969

Yan Kang , Hao Peng , Yue Peng , Jing Guo , Ying Lin

{"title":"Tri-level interaction fusion network for graph similarity learning","authors":"Yan Kang , Hao Peng , Yue Peng , Jing Guo , Ying Lin","doi":"10.1016/j.knosys.2025.113969","DOIUrl":"10.1016/j.knosys.2025.113969","url":null,"abstract":"<div><div>Graph similarity learning is vital in various domains, such as chemical molecular structure comparison and transportation network optimization. Although existing graph similarity learning methods are effective, the following challenges still exist: (i) how to enrich node representation, (ii) how to capture and fuse multi-level graph interaction information, and (iii) how to effectively learn rich graph interaction features. To address these challenges, we propose a novel three-level interaction fusion network (TIFN) by fusing node-node, node-graph, and graph-graph interaction information at different stages to capture the complex interdependencies between graphs. At the enhanced node embedding learning stage, we first propose a skip-connected multi-layer graph isomorphism network to extract high-quality node features and present a novel style-based multi-head attention mechanism to capture long-range dependencies between nodes. At the dual graph interaction learning stage, we model graph dependencies from multi-granularity and multi-perspective views. Specifically, we effectively incorporate a coarse-fine grained aggregation module and a node-graph interaction comparison module to learn rich graph interaction features. At the similarity score stage, global-level graph-graph interactions transform the aggregated features into a final score. Extensive experimental results demonstrate the performance of TIFN by comparing it with 15 baselines on three benchmark datasets. Notably, on the LINUX dataset, TIFN achieves approximately 45.79% improvement in MSE compared to the state-of-the-art method.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 113969"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144614885","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

Partial multi-label feature selection via adaptive dual-graph regularization 基于自适应双图正则化的部分多标签特征选择

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.114077

Hao Xie, Ivy Liu, Bing Xue, Mengjie Zhang

{"title":"Partial multi-label feature selection via adaptive dual-graph regularization","authors":"Hao Xie, Ivy Liu, Bing Xue, Mengjie Zhang","doi":"10.1016/j.knosys.2025.114077","DOIUrl":"10.1016/j.knosys.2025.114077","url":null,"abstract":"<div><div>Partial Multi-Label Learning (PML) tackles the challenge of developing accurate models based on candidate label sets that include both ground-truth labels and noisy ones. However, high-dimensional data often limits the performance of existing methods. Furthermore, traditional Multi-Label Feature Selection (MFS) methods face challenges in accurately identifying the optimum feature subset from partial multi-label data. To solve these issues, we propose a novel method, Partial Multi-label Feature Selection via Adaptive Dual-graph Regularization (PMFS-ADG). First, we leverage low-rank constraints and sparse representation to model the global relationships among labels, recovering the ground-truth label distribution from the original label space and distinguishing noisy labels. Then, adaptive dual-graph regularization is introduced to learn the non-linear geometric information of both the ground-truth label space and the feature space, enhancing label disambiguation while improving the discriminative ability of the selected features. The <span><math><msub><mi>L</mi><mrow><mn>2</mn><mo>,</mo><mn>1</mn></mrow></msub></math></span>-norm is utilized to impose sparse constraints on the weight matrix, effectively removing irrelevant features. Furthermore, to ensure convergence, we design an efficient alternating optimization algorithm. Experimental results on both synthetic and genuine partial multi-label datasets demonstrate that the proposed method outperforms existing methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114077"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144653871","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

Textual emotion detection with complementary BERT transformers in a Condorcet’s Jury theorem assembly 基于孔多塞陪审团定理集合的互补BERT变换文本情感检测

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-07-10 DOI: 10.1016/j.knosys.2025.114070

Gerardo Bárcena Ruiz , Richard De Jesús Gil Herrera

{"title":"Textual emotion detection with complementary BERT transformers in a Condorcet’s Jury theorem assembly","authors":"Gerardo Bárcena Ruiz , Richard De Jesús Gil Herrera","doi":"10.1016/j.knosys.2025.114070","DOIUrl":"10.1016/j.knosys.2025.114070","url":null,"abstract":"<div><div>This paper explores a novel approach to textual emotion detection (TED) in Spanish and English, leveraging an ensemble of partially trained BERT transformers within a Condorcet’s Jury Theorem (CJT) framework. Recognizing the challenges of limited training data and the complexities of emotion classification, this research investigates whether a combination of BERT models in the CJT ensemble can enhance performance even when individual models have incomplete training. The study evaluates different BERT modalities (BERT, RoBERTa, DistilBERT) and datasets, including SemEval-2018, XED, and Dair-ai/emotion. The main contribution is the development of a CJT ensemble, specifically the Jury Dynamic (JD), a key contribution of this research. This algorithm is designed for deployment in unsupervised production environments, eliminating the need for labeled data or continuous human supervision, leveraging Reinforcement Learning (RL). The Jury Dynamic (JD) adapts to incoming data, making it suitable for real-time applications. Experiments involve retraining BERT models with varying levels of emotional data reduction to simulate incomplete training. Results demonstrate that the CJT ensemble, particularly the JD, can effectively mitigate the negative impacts of limited training data, achieving comparable performance to fully trained models and outperforming individual models. The study highlights the importance of high-quality datasets for TED, particularly in Spanish, and proposes future research directions, including the evaluation of various classifiers and ensemble configurations.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"326 ","pages":"Article 114070"},"PeriodicalIF":7.2,"publicationDate":"2025-07-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"144633535","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