Knowledge-Based Systems最新文献_第7页

Dual-graph regularized sparse robust adaptive concept factorization

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-17 DOI: 10.1016/j.knosys.2025.113299

Weizhi Xiong , Yanrong Ma , Jun Ma

{"title":"Dual-graph regularized sparse robust adaptive concept factorization","authors":"Weizhi Xiong , Yanrong Ma , Jun Ma","doi":"10.1016/j.knosys.2025.113299","DOIUrl":"10.1016/j.knosys.2025.113299","url":null,"abstract":"<div><div>Due to the inability of traditional concept factorization methods to fully capture the intricate local and global manifold structures within the raw data space, they are unable to obtain detailed structural information effectively. To address this limitation, we put forward a concept factorization approach named sparse dual-graph regularized concept factorization with stable adaptive spectral clustering (SDCFSAS). Primarily, SDCFSAS leverages Dot-Product Weighting and stable adaptive spectral clustering to construct a similarity matrix that learns intrinsic features of the data, especially nonlinear or non-convex structures. Besides, by utilizing a robust estimator to filter the side effects of outlier points, it ensures that normal samples play a pivotal function in the construction of the model, enhancing the robustness and reliability of the model. Furthermore, the introduction of <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn><mo>,</mo><mi>r</mi></mrow></msub></math></span>-norm (<span><math><mrow><mn>1</mn><mo>≤</mo><mi>r</mi><mo>≤</mo><mn>2</mn></mrow></math></span>) taking for a measure on the deviation term further strengthens the robustness. Additionally, the computable sparse <span><math><msub><mrow><mi>L</mi></mrow><mrow><mn>2</mn><mo>,</mo><mi>p</mi></mrow></msub></math></span>-norm (<span><math><mrow><mn>0</mn><mo><</mo><mi>p</mi><mo>≤</mo><mn>1</mn></mrow></math></span>) regularization terms are employed to establish a sparse model, improving the model’s generalization capability, computational efficiency, and noise reduction. Finally, The performance of algorithm used to solve SDCFSAS is studied in detail, especially its convergence and computational complexity. To demonstrate the clustering performance and recognition ability of our SDCFSAS, we proceed comparative experiments on eight real-world datasets against other similar state-of-the-art algorithms. Moreover, statistical analysis is employed to validate the results, which showcase the significant advantages of our approach.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113299"},"PeriodicalIF":7.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682617","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

Knowledge-based natural answer generation via effective graph learning

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-17 DOI: 10.1016/j.knosys.2025.113288

Zedong Liu , Jianxin Li , Yongle Huang , Ningning Cui , Lili Pei

{"title":"Knowledge-based natural answer generation via effective graph learning","authors":"Zedong Liu , Jianxin Li , Yongle Huang , Ningning Cui , Lili Pei","doi":"10.1016/j.knosys.2025.113288","DOIUrl":"10.1016/j.knosys.2025.113288","url":null,"abstract":"<div><h3>Objectives:</h3><div>Natural Answer Generation (NAG) aims to generate natural and fluent answers to user questions. Existing NAG methods typically employ fixed-hop retrieval to construct knowledge graphs and utilize attention-based networks for answer generation. However, these approaches lack interpretability, struggle to filter out redundant information in the graph, and are computationally intensive.</div></div><div><h3>Methods:</h3><div>To address these issues, this paper introduces an innovative approach <span><math><mi>AdaptQA</mi></math></span> model. Initially, AdaptQA constructs a knowledge graph from the knowledge base (KB) using an adaptive multi-hop retrieval algorithm. Subsequently, it generates answers through the Graph-based Mamba module (GBM), effectively filtering out redundant information. Finally, the answers are optimized using a pre-trained large language model to enhance their fluency and accuracy.</div></div><div><h3>Novelty:</h3><div>The proposed AdaptQA model introduces a new approach to NAG by improving the completeness of the knowledge graph and optimizing question answers. This method overcomes the limitations of existing NAG techniques by reducing the complexity of model inference.</div></div><div><h3>Findings:</h3><div>Through extensive experiments on two benchmark datasets, HotpotQA and WikiHop, AdaptQA demonstrates superior performance, significantly outperforming existing NAG methods. Specifically, AdaptQA achieves an accuracy of 94.47% on the HotpotQA dataset and 91.38% on the WikiHop dataset.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113288"},"PeriodicalIF":7.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682632","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

Cross-city transfer learning for traffic forecasting via incremental distribution rectification

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-17 DOI: 10.1016/j.knosys.2025.113336

Banglie Yang , Runze Li , Yijing Wang , Sha Xiang , Shuo Zhu , Cheng Dai , Shengxin Dai , Bing Guo

{"title":"Cross-city transfer learning for traffic forecasting via incremental distribution rectification","authors":"Banglie Yang , Runze Li , Yijing Wang , Sha Xiang , Shuo Zhu , Cheng Dai , Shengxin Dai , Bing Guo","doi":"10.1016/j.knosys.2025.113336","DOIUrl":"10.1016/j.knosys.2025.113336","url":null,"abstract":"<div><div>High-quality traffic forecasting is critical to facilitating urban intelligent transformation. In deep learning enabled urban intelligence era, accurate traffic forecasting requires large-scale data, obtaining which is not always feasible because of limitations posed by scarce equipment resources in cross-city scenarios. To address this problem, existing methods propose to learn transferable meta-knowledge from rich source city data so as to guide the forecasting on the target city. However, they mainly focus on obtaining a source distribution-centric or source and target shared knowledge, rather than target distribution-centric transferable knowledge, which results in unavoidable disturbances due to migration noise. In this case, we propose a cross-city transfer learning method based on Incremental Distribution Rectification from the perspectives of distribution discrepancy quantification and calibration, called Cross-IDR. Specifically, we leverage the Koopman enabled Optimal Transport method to measure the transfer process between distributions, thus modifying the source distribution-centric meta-knowledge to be target distribution-centric. In addition, a Spatio-Temporal Interaction alignment method is proposed to enhance the mining of cross-city interactions between spatial and temporal information. We verify the effectiveness of Cross-IDR on real-world traffic datasets, and the results demonstrate that it outperforms state-of-the-art models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113336"},"PeriodicalIF":7.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143681297","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

User disambiguation learning for precise shared-account marketing: A hierarchical self-attentive sequential recommendation method

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-17 DOI: 10.1016/j.knosys.2025.113328

Weiyi Duan, Decui Liang

{"title":"User disambiguation learning for precise shared-account marketing: A hierarchical self-attentive sequential recommendation method","authors":"Weiyi Duan, Decui Liang","doi":"10.1016/j.knosys.2025.113328","DOIUrl":"10.1016/j.knosys.2025.113328","url":null,"abstract":"<div><div>Precision marketing recommendations face significant challenges due to entanglement of sharing behavior in shared-account marketing. To disentangle shared user behaviors within strict privacy policies, this study proposes the Hierarchical Self-Attentive Sequential Recommendation (HierSASRec) model. HierSASRec employs user disambiguation learning to dynamically identify individual users within shared accounts, utilizing two-level representations for account-item and user-item interactions. By integrating the time intervals of item interactions into the density-based spatial clustering of applications with noise (DBSCAN) method, namely, the time-aware DBSCAN method, HierSASRec automatically extracts user-level sequences beyond fixed user counts, enhancing the identification of similar preferences and close interactions. Through a self-attention mechanism, HierSASRec combines hierarchical interaction information to optimize marketing precision. Additionally, a random user switch mechanism is devised to mitigate noise from long-term sequences, and focuses on immediate marketing decisions for the current user. Experimental validation within real-world datasets underscores the superiority of HierSASRec over state-of-the-art baselines, affirming its practical efficacy in enhancing marketing precision. The code is available at: <span><span>https://github.com/muyunping123/HierSASRec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"315 ","pages":"Article 113328"},"PeriodicalIF":7.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143642906","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 the transferability of adversarial examples through semantic-mixup inputs

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-17 DOI: 10.1016/j.knosys.2025.113325

Fuquan Gan, Yan Wo

{"title":"Improving the transferability of adversarial examples through semantic-mixup inputs","authors":"Fuquan Gan, Yan Wo","doi":"10.1016/j.knosys.2025.113325","DOIUrl":"10.1016/j.knosys.2025.113325","url":null,"abstract":"<div><div>Deep neural networks are highly vulnerable to perturbations that are imperceptible to humans, especially transferable attacks, which can attack multiple models across different architectures without access to the target model, posing a significant security threat to real-world deployments. Various methods have been proposed to enhance the transferability of adversarial examples from different perspectives, among which input transformation-based attacks have shown particularly notable performance. However, existing methods either apply transformations to a single image or perform global transformations, resulting in insufficient enhancement of critical semantic content, which limit the transferability of adversarial samples. We have identified that the mid-low frequency components of an image embody the majority of the critical semantic information and play a dominant role in the model’s decision-making process. To this end, in this paper, we propose a novel input transformation-based attack called Semantic-Mixup Attack (SMA), which employs a primary–secondary relation, considering the input image as primary and a set of images randomly sampled from other categories as secondary. SMA utilizes the mid-low frequency components of the secondary image to represent corresponding semantic features and mixes them locally with the primary image, thereby increasing the diversity of semantic features in the input image and significantly reducing overfitting to the substitute model. By aggregating these semantic feature from other categories, SMA obtains a reasonable and stable gradient direction, which achieves a more global maximum and exhibits much better transferability. Extensive experiments on the ImageNet compatible dataset demonstrate the outstanding performance of our method compared to existing input transformation-based attacks, both on undefended and advanced defense models. Furthermore, our method can be combined with other attacks to further improve the adversarial transferability.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113325"},"PeriodicalIF":7.2,"publicationDate":"2025-03-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682624","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

HGCT: Enhancing temporal knowledge graph reasoning through extrapolated historical fact extraction

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-16 DOI: 10.1016/j.knosys.2025.113358

Hoa Dao , Nguyen Phan , Thanh Le , Ngoc-Trung Nguyen

{"title":"HGCT: Enhancing temporal knowledge graph reasoning through extrapolated historical fact extraction","authors":"Hoa Dao , Nguyen Phan , Thanh Le , Ngoc-Trung Nguyen","doi":"10.1016/j.knosys.2025.113358","DOIUrl":"10.1016/j.knosys.2025.113358","url":null,"abstract":"<div><div>Extrapolation on Temporal Knowledge Graphs (TKGs) poses a critical obstacle, garnering significant attention in the academic sphere due to its far-reaching implications across various domains and areas of study. Predicting upcoming events through the analysis of historical data involves a complex task that requires the integration of structural patterns from historical graph data and temporal dynamics, which has been the focus of various recent research efforts. However, existing methods face significant limitations. Many approaches fail to effectively differentiate the importance of historical knowledge, leading to suboptimal message passing. Others struggle to capture both local and global temporal dependencies simultaneously, resulting in incomplete temporal representations. Moreover, conventional embedding techniques often overlook dynamic positional information, which is crucial for robust forecasting. To mitigate these issues, we introduce a forecasting framework for TKGs, History Graph Convolution Transformer (HGCT), which enhances graph embedding by integrating a time-aware self-attention mechanism and convolution operation. This methodology combines a Fact Graph Transformer to organize historical knowledge into a structured framework coming with a Temporal Convolutional Transformer that leverages an innovative positional encoding strategy to distill temporal patterns from historical snapshots, thereby enriching the representation of time series data. Our work introduces a refined ConvTransE variant, named Query-ConvTransE, optimized for query-based information processing, which is consolidated into the decoding component. Evaluation of this approach across a diverse range of six benchmark datasets reveals a performance boost, marked by a 4.33 % increase in Hit@k metric and by a 3.84 % increase in Mean Reciprocal Rank metric relative to the state-of-the-art.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113358"},"PeriodicalIF":7.2,"publicationDate":"2025-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682623","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

Graph out-of-distribution generalization through contrastive learning paradigm

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-15 DOI: 10.1016/j.knosys.2025.113316

Hongyi Du , Xuewei Li , Minglai Shao

引用次数: 0

An efficient method for mining top-k multi-level high utility itemsets

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-15 DOI: 10.1016/j.knosys.2025.113359

Loan T.T. Nguyen , N.T. Tung , Bay Vo

{"title":"An efficient method for mining top-k multi-level high utility itemsets","authors":"Loan T.T. Nguyen , N.T. Tung , Bay Vo","doi":"10.1016/j.knosys.2025.113359","DOIUrl":"10.1016/j.knosys.2025.113359","url":null,"abstract":"<div><div>High utility itemset mining (HUIM) is employed for analysing user behaviour. It is an extended form of frequent itemset mining. Top-k HUIM provides a solution to the challenge of determining a suitable minimum utility threshold. Although HUIM algorithms have been improved and implemented with high performance, mining hierarchical data has not received considerable attention. The memory and runtime requirements of top-k HUIM algorithms on hierarchical databases remain high. This study proposes a method, called top-k multi-level high utility itemset (TK-MLHUI) mining, to make mining more efficient. The method uses several strategies, including sub-tree and local utility, to reduce the search space by applying stricter upper bounds. Any level of hierarchical data may be used with the upper bounds. It also suggests an approach for reducing the data required for scanning through various strategies, including combining items on lists at mining levels and updating promising items. Furthermore, a strategy to efficiently raise the threshold using the utility list of items is introduced. Experiments using various databases allow for the measurement of the method's performance. Compared with a previous method - mlTKO, TK-MLHUI decreases execution time by more than 3700 times while also using less memory.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113359"},"PeriodicalIF":7.2,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682619","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

Dual-granularity multi-instance multi-label learning with variational autoencoder

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-15 DOI: 10.1016/j.knosys.2025.113317

Meixia Wang, Yuhai Zhao, Yejiang Wang, Miaomiao Huang, Xuze Liu, Xingwei Wang

{"title":"Dual-granularity multi-instance multi-label learning with variational autoencoder","authors":"Meixia Wang, Yuhai Zhao, Yejiang Wang, Miaomiao Huang, Xuze Liu, Xingwei Wang","doi":"10.1016/j.knosys.2025.113317","DOIUrl":"10.1016/j.knosys.2025.113317","url":null,"abstract":"<div><div>Multi-instance multi-label learning (MIML) is a weakly supervised approach that models relationships between complex objects and multiple labels, where each object is represented as a bag of instances. A key advantage of MIML is its ability to perform both bag-level and instance-level multi-label predictions, relying solely on bag-level labels. However, a significant performance gap persists between instance-level MIML algorithms and fully supervised learning approaches due to the lack of instance-level labels. Existing MIML algorithms address this challenge by treating bag labels as ambiguous and attempting to reduce supervision imprecision. Moreover, they often assume that instances are independent and identically distributed (i.i.d.) and rely on prior knowledge to learn label correlations, which is impractical in real-world scenarios. To address these challenges, we propose MIMLVAE, a novel dual-granularity MIML algorithm based on a variational autoencoder. MIMLVAE employs a graph attention network to dynamically capture label correlations and instance dependencies, eliminating the i.i.d. assumption and prior knowledge. By treating all instances within a bag equally, it infers effective bag-level and instance-level representations for dual-granularity prediction. At the same time, the label encoder captures label-specific prototype representations, facilitating prototype-based classification at both the bag and instance levels without requiring label disambiguation. Furthermore, MIMLVAE integrates a Gaussian mixture model into the shared latent space of features and labels, mitigating posterior collapse and over-regularization. Experiments on six standard MIML datasets demonstrate that MIMLVAE significantly outperforms state-of-the-art methods in both bag-level and instance-level multi-label classification tasks.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113317"},"PeriodicalIF":7.2,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682622","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

Malicious encrypted traffic detection method based on multi-granularity representation under data imbalance conditions

IF 7.2 1区计算机科学

Knowledge-Based Systems Pub Date : 2025-03-15 DOI: 10.1016/j.knosys.2025.113320

Tao Li , Zhiwei Yang , Wenshan Li , Linfeng Du , Xiaolong Lan , Junjiang He

{"title":"Malicious encrypted traffic detection method based on multi-granularity representation under data imbalance conditions","authors":"Tao Li , Zhiwei Yang , Wenshan Li , Linfeng Du , Xiaolong Lan , Junjiang He","doi":"10.1016/j.knosys.2025.113320","DOIUrl":"10.1016/j.knosys.2025.113320","url":null,"abstract":"<div><div>Traffic encryption technology safeguards the secure transmission of user data but also enables attackers to conceal malicious activities. With the widespread adoption of encrypted communication protocols and the growing volume of encrypted traffic, accurately identifying malicious encrypted traffic has become a critical challenge in network security. Traditional methods for encrypted traffic classification rely on manual feature extraction and are hindered by imbalanced data distribution, leading to low classification accuracy. In this paper, we propose a maliciously encrypted traffic detection method based on multi-granularity representation under data imbalance conditions. Firstly, we propose a feature extraction method for encrypted traffic based on multi-granularity representation to fully extract the behavioral and temporal characteristics of the traffic in depth and improve the feature extraction effect. Secondly, to address the problem of imbalanced data distribution in malicious encrypted traffic, we introduce the prompt learning algorithm, which solves the issue of imbalanced malicious encrypted traffic datasets. This method improves detection effectiveness by constructing prompt samples, comprehensively learning the feature space of data samples, and transforming the multi-classification problem into multiple dichotomous classification problems. Lastly, we conduct experiments using real network traffic datasets to validate our proposed method. The results demonstrate that our approach outperforms the MLM model by 30% in both binary and multi-class classification tasks. Furthermore, when compared to other deep learning models, our method improves performance by 20% to 50%. Overall, our proposed method exhibits strong generalization and usability, along with effective capabilities for detecting and classifying malicious encrypted traffic.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"316 ","pages":"Article 113320"},"PeriodicalIF":7.2,"publicationDate":"2025-03-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143682631","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