Knowledge-Based Systems最新文献_第2页

Semi-supervised pairwise transfer learning based on multi-source domain adaptation: A case study on EEG-based emotion recognition 基于多源领域适应的半监督成对迁移学习：基于脑电图的情绪识别案例研究

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-28 DOI: 10.1016/j.knosys.2024.112669

Chao Ren , Jinbo Chen , Rui Li , Weihao Zheng , Yijiang Chen , Yikun Yang , Xiaowei Zhang , Bin Hu

Negative transfer mitigation in transfer learning and universal-model establishment are crucial in electroencephalography (EEG)-based emotion recognition research. This study proposed a multi-source domain adaptation pairwise transfer learning method (named PLMSDANet) for EEG-based emotion recognition. PLMSDANet reduced the impact of negative transfer using a semi-supervised strategy that introduced a limited set of target-labeled data and selected the most compatible source domains for further training. In addition, a two-stage feature extractor was employed. Initially, we used a general feature extractor to capture the common spatial–spectral features of all the domains. Subsequently, we created independent branches for each pair of source and target domains to learn specific features from each source domain, incorporating discrepancy loss to harmonize the classification results of the different source domains. Furthermore, pairwise learning was used to solve the problem of the intra-domain distribution of sample classes. Finally, a cross-subject strategy was used to validate the public datasets SEED and SEED-IV extensively, achieving average emotion recognition accuracies of 90.09% and 73.08%, respectively. In summary, PLMSDANet combines multi-source domain transfer learning with semi-supervised paired learning methods, effectively transferring knowledge from multiple source domains to the target domain while enhancing the distinguishability between classes. The experimental results show that the PLMSDANet method effectively mitigates the negative-transfer issue and demonstrates excellent recognition performance, surpassing that of state-of-the-art methods.

在基于脑电图（EEG）的情感识别研究中，迁移学习中的负迁移缓解和通用模型的建立至关重要。本研究为基于脑电图的情感识别提出了一种多源领域适应性成对迁移学习方法（命名为 PLMSDANet）。PLMSDANet 采用半监督策略，引入有限的目标标记数据集，并选择最兼容的源域进行进一步训练，从而减少了负迁移的影响。此外，我们还采用了两阶段特征提取器。首先，我们使用通用特征提取器来捕捉所有域的共同空间-光谱特征。随后，我们为每对源域和目标域创建了独立分支，以学习每个源域的特定特征，并结合差异损失来协调不同源域的分类结果。此外，我们还利用成对学习来解决样本类别在域内分布的问题。最后，采用跨主体策略对公共数据集 SEED 和 SEED-IV 进行了广泛验证，平均情感识别准确率分别达到 90.09% 和 73.08%。总之，PLMSDANet 将多源领域迁移学习与半监督配对学习方法相结合，有效地将多个源领域的知识迁移到目标领域，同时增强了类之间的可区分性。实验结果表明，PLMSDANet 方法有效地缓解了负迁移问题，并表现出优异的识别性能，超过了最先进的方法。

{"title":"Semi-supervised pairwise transfer learning based on multi-source domain adaptation: A case study on EEG-based emotion recognition","authors":"Chao Ren , Jinbo Chen , Rui Li , Weihao Zheng , Yijiang Chen , Yikun Yang , Xiaowei Zhang , Bin Hu","doi":"10.1016/j.knosys.2024.112669","DOIUrl":"10.1016/j.knosys.2024.112669","url":null,"abstract":"<div><div>Negative transfer mitigation in transfer learning and universal-model establishment are crucial in electroencephalography (EEG)-based emotion recognition research. This study proposed a multi-source domain adaptation pairwise transfer learning method (named PLMSDANet) for EEG-based emotion recognition. PLMSDANet reduced the impact of negative transfer using a semi-supervised strategy that introduced a limited set of target-labeled data and selected the most compatible source domains for further training. In addition, a two-stage feature extractor was employed. Initially, we used a general feature extractor to capture the common spatial–spectral features of all the domains. Subsequently, we created independent branches for each pair of source and target domains to learn specific features from each source domain, incorporating discrepancy loss to harmonize the classification results of the different source domains. Furthermore, pairwise learning was used to solve the problem of the intra-domain distribution of sample classes. Finally, a cross-subject strategy was used to validate the public datasets SEED and SEED-IV extensively, achieving average emotion recognition accuracies of 90.09% and 73.08%, respectively. In summary, PLMSDANet combines multi-source domain transfer learning with semi-supervised paired learning methods, effectively transferring knowledge from multiple source domains to the target domain while enhancing the distinguishability between classes. The experimental results show that the PLMSDANet method effectively mitigates the negative-transfer issue and demonstrates excellent recognition performance, surpassing that of state-of-the-art methods.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112669"},"PeriodicalIF":7.2,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142578534","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

Enhancing collaboration in multi-agent reinforcement learning with correlated trajectories 利用相关轨迹加强多代理强化学习中的协作

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-28 DOI: 10.1016/j.knosys.2024.112665

Siying Wang , Hongfei Du , Yang Zhou , Zhitong Zhao , Ruoning Zhang , Wenyu Chen

Collaborative behaviors in human social activities can be modeled with multi-agent reinforcement learning and used to train the collaborative policies of agents to achieve efficient cooperation. In general, agents with similar behaviors have a certain behavioral common cognition and are more likely to understand the intentions of both parties then to form cooperative policies. Traditional approaches focus on the collaborative allocation process between agents, ignoring the effects of similar behaviors and common cognition characteristics in collaborative interactions. In order to better establish collaborative relationships between agents, we propose a novel multi-agent reinforcement learning collaborative algorithm based on the similarity of agents’ behavioral features. In this model, the interactions of agents are established as a graph neural network. Specifically, the Pearson correlation coefficient is proposed to compute the similarity of the history trajectories of the agents as a means of determining their behavioral common cognition, which is used to establish the weights of the edges in the modeled graph neural network. In addition, we design a transformer-encoder structured state information complementation module to enhance the decision representation of the agents. The experimental results on Predator–Prey and StarCraft II show that the proposed method can effectively enhance the collaborative behaviors between agents and improve the training efficiency of collaborative models.

人类社会活动中的合作行为可以用多代理强化学习来建模，并用于训练代理的合作策略，以实现高效合作。一般来说，行为相似的代理具有一定的行为共同认知，更容易理解双方的意图，进而形成合作策略。传统方法侧重于代理之间的协作分配过程，忽视了相似行为和共同认知特征在协作互动中的影响。为了更好地建立代理之间的合作关系，我们提出了一种基于代理行为特征相似性的新型多代理强化学习合作算法。在这个模型中，代理之间的互动关系被建立为一个图神经网络。具体来说，我们提出用皮尔逊相关系数来计算各代理历史轨迹的相似性，以此来确定它们的行为共同认知，并以此来建立模型图神经网络中各条边的权重。此外，我们还设计了一个变换器-编码器结构的状态信息互补模块，以增强代理的决策表征。捕食者-猎物》和《星际争霸 II》的实验结果表明，所提出的方法能有效增强代理之间的协作行为，提高协作模型的训练效率。

{"title":"Enhancing collaboration in multi-agent reinforcement learning with correlated trajectories","authors":"Siying Wang , Hongfei Du , Yang Zhou , Zhitong Zhao , Ruoning Zhang , Wenyu Chen","doi":"10.1016/j.knosys.2024.112665","DOIUrl":"10.1016/j.knosys.2024.112665","url":null,"abstract":"<div><div>Collaborative behaviors in human social activities can be modeled with multi-agent reinforcement learning and used to train the collaborative policies of agents to achieve efficient cooperation. In general, agents with similar behaviors have a certain behavioral common cognition and are more likely to understand the intentions of both parties then to form cooperative policies. Traditional approaches focus on the collaborative allocation process between agents, ignoring the effects of similar behaviors and common cognition characteristics in collaborative interactions. In order to better establish collaborative relationships between agents, we propose a novel multi-agent reinforcement learning collaborative algorithm based on the similarity of agents’ behavioral features. In this model, the interactions of agents are established as a graph neural network. Specifically, the Pearson correlation coefficient is proposed to compute the similarity of the history trajectories of the agents as a means of determining their behavioral common cognition, which is used to establish the weights of the edges in the modeled graph neural network. In addition, we design a transformer-encoder structured state information complementation module to enhance the decision representation of the agents. The experimental results on Predator–Prey and StarCraft II show that the proposed method can effectively enhance the collaborative behaviors between agents and improve the training efficiency of collaborative models.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112665"},"PeriodicalIF":7.2,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142553655","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

CSP-Net: Common spatial pattern empowered neural networks for EEG-based motor imagery classification CSP-Net：基于脑电图的运动图像分类的通用空间模式增强型神经网络

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-28 DOI: 10.1016/j.knosys.2024.112668

Xue Jiang, Lubin Meng, Xinru Chen, Yifan Xu, Dongrui Wu

Electroencephalogram-based motor imagery (MI) classification is an important paradigm of non-invasive brain–computer interfaces. Common spatial pattern (CSP), which exploits different energy distributions on the scalp while performing different MI tasks, is very popular in MI classification. Convolutional neural networks (CNNs) have also achieved great success, due to their powerful learning capabilities. This paper proposes two CSP-empowered neural networks (CSP-Nets), which integrate knowledge-driven CSP filters with data-driven CNNs to enhance the performance in MI classification. CSP-Net-1 directly adds a CSP layer before a CNN to improve the input discriminability. CSP-Net-2 replaces a convolutional layer in CNN with a CSP layer. The CSP layer parameters in both CSP-Nets are initialized with CSP filters designed from the training data. During training, they can either be kept fixed or optimized using gradient descent. Experiments on four public MI datasets demonstrated that the two CSP-Nets consistently improved over their CNN backbones, in both within-subject and cross-subject classifications. They are particularly useful when the number of training samples is very small. Our work demonstrates the advantage of integrating knowledge-driven traditional machine learning with data-driven deep learning in EEG-based brain–computer interfaces.

基于脑电图的运动图像（MI）分类是非侵入式脑机接口的一个重要范例。共同空间模式（CSP）利用了执行不同运动图像任务时头皮上的不同能量分布，在运动图像分类中非常流行。卷积神经网络（CNN）凭借其强大的学习能力也取得了巨大成功。本文提出了两种 CSP 赋能神经网络（CSP-Nets），它们将知识驱动的 CSP 过滤器与数据驱动的 CNN 相集成，以提高 MI 分类的性能。CSP-Net-1 直接在 CNN 之前添加一个 CSP 层，以提高输入的可辨别性。CSP-Net-2 用 CSP 层取代了 CNN 中的卷积层。两个 CSP 网络中的 CSP 层参数都是用根据训练数据设计的 CSP 过滤器初始化的。在训练过程中，它们既可以保持固定，也可以使用梯度下降法进行优化。在四个公共 MI 数据集上进行的实验表明，这两个 CSP 网络在主体内和跨主体分类方面都比其 CNN 骨干网络有持续的改进。在训练样本数量很少的情况下，它们尤其有用。我们的工作证明了在基于脑电图的脑机接口中将知识驱动的传统机器学习与数据驱动的深度学习相结合的优势。

{"title":"CSP-Net: Common spatial pattern empowered neural networks for EEG-based motor imagery classification","authors":"Xue Jiang, Lubin Meng, Xinru Chen, Yifan Xu, Dongrui Wu","doi":"10.1016/j.knosys.2024.112668","DOIUrl":"10.1016/j.knosys.2024.112668","url":null,"abstract":"<div><div>Electroencephalogram-based motor imagery (MI) classification is an important paradigm of non-invasive brain–computer interfaces. Common spatial pattern (CSP), which exploits different energy distributions on the scalp while performing different MI tasks, is very popular in MI classification. Convolutional neural networks (CNNs) have also achieved great success, due to their powerful learning capabilities. This paper proposes two CSP-empowered neural networks (CSP-Nets), which integrate knowledge-driven CSP filters with data-driven CNNs to enhance the performance in MI classification. CSP-Net-1 directly adds a CSP layer before a CNN to improve the input discriminability. CSP-Net-2 replaces a convolutional layer in CNN with a CSP layer. The CSP layer parameters in both CSP-Nets are initialized with CSP filters designed from the training data. During training, they can either be kept fixed or optimized using gradient descent. Experiments on four public MI datasets demonstrated that the two CSP-Nets consistently improved over their CNN backbones, in both within-subject and cross-subject classifications. They are particularly useful when the number of training samples is very small. Our work demonstrates the advantage of integrating knowledge-driven traditional machine learning with data-driven deep learning in EEG-based brain–computer interfaces.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112668"},"PeriodicalIF":7.2,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142573326","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

Fusing YOLOv5s-MediaPipe-HRV to classify engagement in E-learning: From the perspective of external observations and internal factors 融合 YOLOv5s-MediaPipe-HRV 对电子学习中的参与度进行分类：从外部观察和内部因素的角度

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-28 DOI: 10.1016/j.knosys.2024.112670

Jie Wang , Shuiping Yuan , Tuantuan Lu , Hao Zhao , Yongxiang Zhao

The rapid advancements in computer vision technology present significant potential for the automatic recognition of learner engagement in E-learning. We conducted a two-stage experiment to assess learner engagement based on behavioural (external observations) and physiological (internal factors) cues. Using computer vision technology and wearable sensors, we extracted three feature sets: action, head posture and heart rate variability (HRV). Subsequently, we integrated our constructed YOLOv5s–MediaPipe behaviour detection model with a physiological detection model based on HRV to comprehensively evaluate learners’ behavioural, affective and cognitive engagement. Additionally, we developed a method and criteria for assessing distraction based on behaviour, ultimately creating a comprehensive, efficient, low-cost and easy-to-use system for the automatic recognition of learner engagement. Experimental results showed that our improved YOLOv5s model achieved a mean average precision of 92.2 %, while halving both the number of parameters and model size. Unlike other deep learning-based methods, using MediaPipe–OpenCV for head posture analysis offers advantages in real-time performance, making it lightweight and easy to deploy. Our proposed long short-term memory classifier, based on sensitive HRV metrics and their normalisation, demonstrated satisfactory performance on the test set, with an accuracy = 80 %, precision = 81 %, recall = 80 % and an F1 score = 80 %.

计算机视觉技术的飞速发展为自动识别电子学习中学习者的参与度提供了巨大的潜力。我们进行了一项两阶段实验，根据行为（外部观察）和生理（内部因素）线索来评估学习者的参与度。利用计算机视觉技术和可穿戴传感器，我们提取了三个特征集：动作、头部姿势和心率变异性（HRV）。随后，我们将构建的 YOLOv5s-MediaPipe 行为检测模型与基于心率变异的生理检测模型进行整合，以全面评估学习者的行为、情感和认知参与度。此外，我们还开发了基于行为的分心评估方法和标准，最终创建了一个全面、高效、低成本且易于使用的学习者参与度自动识别系统。实验结果表明，我们改进后的 YOLOv5s 模型平均精确度达到了 92.2%，同时参数数量和模型大小都减少了一半。与其他基于深度学习的方法不同，使用MediaPipe-OpenCV进行头部姿态分析具有实时性强、轻量级、易于部署等优势。我们提出的长短期记忆分类器基于敏感的心率变异指标及其归一化，在测试集上表现令人满意，准确率 = 80%，精确率 = 81%，召回率 = 80%，F1 分数 = 80%。

{"title":"Fusing YOLOv5s-MediaPipe-HRV to classify engagement in E-learning: From the perspective of external observations and internal factors","authors":"Jie Wang , Shuiping Yuan , Tuantuan Lu , Hao Zhao , Yongxiang Zhao","doi":"10.1016/j.knosys.2024.112670","DOIUrl":"10.1016/j.knosys.2024.112670","url":null,"abstract":"<div><div>The rapid advancements in computer vision technology present significant potential for the automatic recognition of learner engagement in E-learning. We conducted a two-stage experiment to assess learner engagement based on behavioural (external observations) and physiological (internal factors) cues. Using computer vision technology and wearable sensors, we extracted three feature sets: action, head posture and heart rate variability (HRV). Subsequently, we integrated our constructed YOLOv5s–MediaPipe behaviour detection model with a physiological detection model based on HRV to comprehensively evaluate learners’ behavioural, affective and cognitive engagement. Additionally, we developed a method and criteria for assessing distraction based on behaviour, ultimately creating a comprehensive, efficient, low-cost and easy-to-use system for the automatic recognition of learner engagement. Experimental results showed that our improved YOLOv5s model achieved a mean average precision of 92.2 %, while halving both the number of parameters and model size. Unlike other deep learning-based methods, using MediaPipe–OpenCV for head posture analysis offers advantages in real-time performance, making it lightweight and easy to deploy. Our proposed long short-term memory classifier, based on sensitive HRV metrics and their normalisation, demonstrated satisfactory performance on the test set, with an accuracy = 80 %, precision = 81 %, recall = 80 % and an F1 score = 80 %.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112670"},"PeriodicalIF":7.2,"publicationDate":"2024-10-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142587439","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

DeSGOA: double exponential smoothing gazelle optimization algorithm-based deep learning model for blind source separation DeSGOA：基于双指数平滑瞪羚优化算法的盲源分离深度学习模型

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-26 DOI: 10.1016/j.knosys.2024.112626

N Muhsina , Beegum J Dhoulath

Blind Source Separation (BSS) pertains to a scenario, wherein the sources, as the method used for mixing are not known; only the mixed signals are accessible for subsequent separation. In several applications, it is preferable to retrieve all sources from the mixed signal or, at least isolate a specific source. The research proposes a novel approach, named Double Exponential Smoothing Gazelle Optimization Algorithm-based Generative Adversarial Network (DeSGOA-based GAN), and designed for BSS. The proposed algorithm, DeSGOA combines the power of Double Exponential Smoothing (DES) with the efficiency of the Gazelle Optimization Algorithm (GOA) to achieve superior results in source separation tasks. The research aims to enhance the accuracy and performance of BSS processes using the presented approach. At first, the mixed input signals attained from the dataset are fed to pre-processing phase. This phase aspires to eradicate noise present in the signal via the application of PCA model. The final objective is to capture a important amount of data information in the reduced dataset. Following this, the BSS process is carried out by utilizing the GAN, which is trained through the innovative DeSGOA algorithm. Experimental outcomes illustrate the efficacy of DeSGOA-based GAN method in achieving high-quality source separation, underscoring its potential as a valuable tool in audio signal processing and related applications. Finally, the experimental evaluation illustrated that the presented strategy gained SDR of 35.05, SIR of 11.94, SAR of 8.247, and ISR of 13.02.

盲源分离（BSS）是指在不知道信号源和混合方法的情况下，只有混合信号才能进行后续分离。在一些应用中，最好能从混合信号中检索出所有信号源，或至少分离出特定信号源。研究提出了一种新方法，名为基于双指数平滑瞪羚优化算法的生成对抗网络（DeSGOA-based GAN），专为 BSS 设计。所提出的 DeSGOA 算法结合了双指数平滑（DES）的强大功能和瞪羚优化算法（GOA）的高效率，在源分离任务中取得了卓越的效果。该研究旨在利用所提出的方法提高 BSS 处理的精度和性能。首先，从数据集中获得的混合输入信号被送入预处理阶段。该阶段旨在通过应用 PCA 模型消除信号中存在的噪音。最终目的是在缩小的数据集中获取大量数据信息。随后，利用通过创新的 DeSGOA 算法训练的 GAN 进行 BSS 处理。实验结果表明，基于 DeSGOA 的 GAN 方法在实现高质量音源分离方面非常有效，凸显了其作为音频信号处理和相关应用的重要工具的潜力。最后，实验评估表明，所提出的策略获得了 35.05 的 SDR、11.94 的 SIR、8.247 的 SAR 和 13.02 的 ISR。

{"title":"DeSGOA: double exponential smoothing gazelle optimization algorithm-based deep learning model for blind source separation","authors":"N Muhsina , Beegum J Dhoulath","doi":"10.1016/j.knosys.2024.112626","DOIUrl":"10.1016/j.knosys.2024.112626","url":null,"abstract":"<div><div>Blind Source Separation (BSS) pertains to a scenario, wherein the sources, as the method used for mixing are not known; only the mixed signals are accessible for subsequent separation. In several applications, it is preferable to retrieve all sources from the mixed signal or, at least isolate a specific source. The research proposes a novel approach, named Double Exponential Smoothing Gazelle Optimization Algorithm-based Generative Adversarial Network (DeSGOA-based GAN), and designed for BSS. The proposed algorithm, DeSGOA combines the power of Double Exponential Smoothing (DES) with the efficiency of the Gazelle Optimization Algorithm (GOA) to achieve superior results in source separation tasks. The research aims to enhance the accuracy and performance of BSS processes using the presented approach. At first, the mixed input signals attained from the dataset are fed to pre-processing phase. This phase aspires to eradicate noise present in the signal via the application of PCA model. The final objective is to capture a important amount of data information in the reduced dataset. Following this, the BSS process is carried out by utilizing the GAN, which is trained through the innovative DeSGOA algorithm. Experimental outcomes illustrate the efficacy of DeSGOA-based GAN method in achieving high-quality source separation, underscoring its potential as a valuable tool in audio signal processing and related applications. Finally, the experimental evaluation illustrated that the presented strategy gained SDR of 35.05, SIR of 11.94, SAR of 8.247, and ISR of 13.02.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112626"},"PeriodicalIF":7.2,"publicationDate":"2024-10-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142529406","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

CGG: Category-aware global graph contrastive learning for session-based recommendation CGG：基于会话推荐的分类感知全局图对比学习

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-25 DOI: 10.1016/j.knosys.2024.112661

Mingxin Gan, Xiongtao Zhang, Yuxin Liang

With the auxiliary role of category information in capturing user interests, employing category information to improve session-based recommendation (SBR) is getting an energetic research point. Recent studies organized the category-aware session as the graph structure and utilized the graph neural network to explore the session interest for SBR. However, existing studies only focused on the category information in the current session and failed to overcome inherent sparsity of session data, which resulted in suboptimal SBR performance. To overcome these deficiencies, we propose a Category-aware Global Graph contrastive learning method, namely CGG, for SBR. To be specific, we firstly construct the category-aware global graph based on global item-item transitions, item-category associations and global category-category transitions, which utilizes more sufficient category information across sessions to learn embeddings of categories and items. Furthermore, we design the hierarchical dual-pattern contrastive learning mechanism to model the information interaction of graphical and sequential patterns of a category-aware session, which overcomes the negative influence of sparse session data by injecting self-supervised signals. Extensive experiments on multiple real-world datasets verify that CGG outperforms seven mainstream SBR methods on different measurements.

由于类别信息在捕捉用户兴趣方面的辅助作用，利用类别信息改进基于会话的推荐（SBR）正成为一个充满活力的研究热点。最近的研究将类别感知会话组织成图结构，并利用图神经网络来探索会话兴趣，从而实现基于会话的推荐（SBR）。然而，现有研究只关注当前会话中的类别信息，未能克服会话数据固有的稀疏性，导致 SBR 性能不理想。为了克服这些不足，我们提出了一种用于 SBR 的类别感知全局图对比学习方法，即 CGG。具体来说，我们首先基于全局项目-项目转换、项目-类别关联和全局类别-类别转换构建了类别感知全局图，从而利用跨会话的更充分的类别信息来学习类别和项目的嵌入。此外，我们还设计了分层双模式对比学习机制来模拟类别感知会话的图形模式和顺序模式的信息交互，通过注入自监督信号来克服稀疏会话数据的负面影响。在多个真实世界数据集上进行的广泛实验验证了 CGG 在不同测量指标上优于七种主流 SBR 方法。

{"title":"CGG: Category-aware global graph contrastive learning for session-based recommendation","authors":"Mingxin Gan, Xiongtao Zhang, Yuxin Liang","doi":"10.1016/j.knosys.2024.112661","DOIUrl":"10.1016/j.knosys.2024.112661","url":null,"abstract":"<div><div>With the auxiliary role of category information in capturing user interests, employing category information to improve session-based recommendation (SBR) is getting an energetic research point. Recent studies organized the category-aware session as the graph structure and utilized the graph neural network to explore the session interest for SBR. However, existing studies only focused on the category information in the current session and failed to overcome inherent sparsity of session data, which resulted in suboptimal SBR performance. To overcome these deficiencies, we propose a <strong>C</strong>ategory-aware <strong>G</strong>lobal <strong>G</strong>raph contrastive learning method, namely CGG, for SBR. To be specific, we firstly construct the category-aware global graph based on global item-item transitions, item-category associations and global category-category transitions, which utilizes more sufficient category information across sessions to learn embeddings of categories and items. Furthermore, we design the hierarchical dual-pattern contrastive learning mechanism to model the information interaction of graphical and sequential patterns of a category-aware session, which overcomes the negative influence of sparse session data by injecting self-supervised signals. Extensive experiments on multiple real-world datasets verify that CGG outperforms seven mainstream SBR methods on different measurements.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112661"},"PeriodicalIF":7.2,"publicationDate":"2024-10-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142573328","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 semi-supervised hypergraph regular multi-view NMF with anchor graph embedding 带锚图嵌入的双半监督超图正则多视图 NMF

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-24 DOI: 10.1016/j.knosys.2024.112662

Jianping Mei , Xiangli Li , Yuanjian Mo

Graph regularized nonnegative matrix factorization (GNMF) has been widely used in multi-view clustering tasks due to its good clustering properties. However, it uses a simple graph to describe the complex data relationship of multiple views and tries to obtain a consistent low-dimensional representation, which undoubtedly brings challenges to its clustering performance. In addition, clustering a large amount of high-dimensional data from multiple views undoubtedly faces a huge computational burden. In order to effectively improve the performance and efficiency of GNMF based multi-view clustering algorithm, this paper proposes a dual semi-supervised hypergraph regular multi-view clustering method with anchor graph embedding (DSSHMNMFAE). Specifically, DSSHMNMFAE develops a new anchor selection method to generate anchors and the anchor bipartite graph is constructed to embed the matrix factorization process. DSSHMNMFAE constructs a hypergraph to effectively learn the high-order relationship between data from multiple views. In order to perform semi-supervised learning more efficiently, DSSHMNMFAE integrates pairwise constraint information and label constraint information into the clustering process as dual label information. In addition, DSSHMNMFAE considers the learning of both consistency information and complementarity information, and adopts adaptive measures to distinguish the contributions of different views. We use the alternating iterative algorithm to optimize the objective function of DSSHMNMFAE. The experimental results on eight real datasets show that the performance of DSSHMNMFAE is comparable to other algorithms.

图形正则化非负矩阵因式分解（GNMF）因其良好的聚类特性已被广泛应用于多视图聚类任务中。然而，它使用简单的图来描述多视图的复杂数据关系，并试图获得一致的低维表示，这无疑给其聚类性能带来了挑战。此外，对来自多个视图的大量高维数据进行聚类无疑面临着巨大的计算负担。为了有效提高基于 GNMF 的多视图聚类算法的性能和效率，本文提出了一种带锚图嵌入的双半监督超图正则多视图聚类方法（DSSHMNMFAE）。具体来说，DSSHMNMFAE 开发了一种新的锚选择方法来生成锚，并构建了锚双方图来嵌入矩阵因式分解过程。DSSHMNMFAE 构建了一个超图，以有效学习多视图数据之间的高阶关系。为了更有效地进行半监督学习，DSSHMNMFAE 将成对约束信息和标签约束信息作为双标签信息集成到聚类过程中。此外，DSSHMNMFAE 还考虑了一致性信息和互补性信息的学习，并采用自适应措施来区分不同观点的贡献。我们采用交替迭代算法来优化 DSSHMNMFAE 的目标函数。在八个真实数据集上的实验结果表明，DSSHMNMFAE 的性能与其他算法相当。

{"title":"Dual semi-supervised hypergraph regular multi-view NMF with anchor graph embedding","authors":"Jianping Mei , Xiangli Li , Yuanjian Mo","doi":"10.1016/j.knosys.2024.112662","DOIUrl":"10.1016/j.knosys.2024.112662","url":null,"abstract":"<div><div>Graph regularized nonnegative matrix factorization (GNMF) has been widely used in multi-view clustering tasks due to its good clustering properties. However, it uses a simple graph to describe the complex data relationship of multiple views and tries to obtain a consistent low-dimensional representation, which undoubtedly brings challenges to its clustering performance. In addition, clustering a large amount of high-dimensional data from multiple views undoubtedly faces a huge computational burden. In order to effectively improve the performance and efficiency of GNMF based multi-view clustering algorithm, this paper proposes a dual semi-supervised hypergraph regular multi-view clustering method with anchor graph embedding (DSSHMNMFAE). Specifically, DSSHMNMFAE develops a new anchor selection method to generate anchors and the anchor bipartite graph is constructed to embed the matrix factorization process. DSSHMNMFAE constructs a hypergraph to effectively learn the high-order relationship between data from multiple views. In order to perform semi-supervised learning more efficiently, DSSHMNMFAE integrates pairwise constraint information and label constraint information into the clustering process as dual label information. In addition, DSSHMNMFAE considers the learning of both consistency information and complementarity information, and adopts adaptive measures to distinguish the contributions of different views. We use the alternating iterative algorithm to optimize the objective function of DSSHMNMFAE. The experimental results on eight real datasets show that the performance of DSSHMNMFAE is comparable to other algorithms.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112662"},"PeriodicalIF":7.2,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142553726","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

Rethinking self-supervised learning for time series forecasting: A temporal perspective 反思时间序列预测的自我监督学习：时间视角

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-24 DOI: 10.1016/j.knosys.2024.112652

Shubao Zhao , Xinxing Zhou , Ming Jin , Zhaoxiang Hou , Chengyi Yang , Zengxiang Li , Qingsong Wen , Yi Wang , Yanlong Wen , Xiaojie Yuan

Self-supervised learning has garnered significant attention for its ability to learn meaningful representations. Recent advancements have introduced self-supervised methods for time series forecasting. However, these efforts have faced limitations due to two primary drawbacks. Firstly, these approaches often borrow techniques from vision and language domains without adequately addressing the unique temporal dependencies inherent in time series data. Secondly, time series often show that the distribution shifts over time, which makes accurate forecasting challenging. In response to these issues, we propose TempSSL, a self-supervised learning framework designed for time series forecasting. TempSSL divides the time series data into context (history data) and target (future data), employing two pre-training strategies: (1) Temporal Masked Modeling (TMM) designed to capture temporal dependencies by reconstructing future time series based on historical context; (2) Temporal Contrastive Learning (TCL) employs context and target as positive samples to enhance discriminative representations and mitigate distribution shifts within the time series. TempSSL’s innovation lies in two key aspects. Firstly, it underscores the importance of temporal dependencies for time series forecasting by designing specific pre-training tasks. Secondly, it effectively integrates contrastive learning and masked modeling, leveraging their respective strengths to develop time series representation with strong instance discriminability and local perceptibility. Extensive experiments across seven widely used benchmark datasets demonstrate that TempSSL consistently outperforms existing self-supervised and end-to-end forecasting methods, achieving improvements ranging from 1.92%

\sim

78.12%. Additionally, TempSSL’s practical effectiveness is further demonstrated through successful application in natural gas demand forecasting.

自监督学习因其学习有意义表征的能力而备受关注。最近的进步为时间序列预测引入了自监督方法。然而，由于存在两个主要缺点，这些努力面临着局限性。首先，这些方法往往借鉴了视觉和语言领域的技术，却没有充分解决时间序列数据固有的独特时间依赖性问题。其次，时间序列数据的分布经常会随时间发生变化，这给准确预测带来了挑战。针对这些问题，我们提出了专为时间序列预测设计的自监督学习框架 TempSSL。TempSSL 将时间序列数据分为上下文（历史数据）和目标（未来数据），并采用两种预训练策略：（1）时序掩蔽建模（TMM），旨在通过基于历史上下文重构未来时间序列来捕捉时序依赖性；（2）时序对比学习（TCL），将上下文和目标作为正样本，以增强判别表征并减轻时间序列内的分布偏移。TempSSL 的创新在于两个关键方面。首先，它通过设计特定的预训练任务，强调了时间序列预测的时间依赖性的重要性。其次，它有效地整合了对比学习和掩蔽建模，利用它们各自的优势，开发出具有很强的实例可辨别性和局部可感知性的时间序列表示。在七个广泛使用的基准数据集上进行的大量实验表明，TempSSL 的性能始终优于现有的自监督和端到端预测方法，提高了 1.92% ∼ 78.12%。此外，TempSSL 在天然气需求预测中的成功应用进一步证明了它的实用有效性。

{"title":"Rethinking self-supervised learning for time series forecasting: A temporal perspective","authors":"Shubao Zhao , Xinxing Zhou , Ming Jin , Zhaoxiang Hou , Chengyi Yang , Zengxiang Li , Qingsong Wen , Yi Wang , Yanlong Wen , Xiaojie Yuan","doi":"10.1016/j.knosys.2024.112652","DOIUrl":"10.1016/j.knosys.2024.112652","url":null,"abstract":"<div><div>Self-supervised learning has garnered significant attention for its ability to learn meaningful representations. Recent advancements have introduced self-supervised methods for time series forecasting. However, these efforts have faced limitations due to two primary drawbacks. Firstly, these approaches often borrow techniques from vision and language domains without adequately addressing the unique temporal dependencies inherent in time series data. Secondly, time series often show that the distribution shifts over time, which makes accurate forecasting challenging. In response to these issues, we propose TempSSL, a self-supervised learning framework designed for time series forecasting. TempSSL divides the time series data into context (history data) and target (future data), employing two pre-training strategies: (1) Temporal Masked Modeling (TMM) designed to capture temporal dependencies by reconstructing future time series based on historical context; (2) Temporal Contrastive Learning (TCL) employs context and target as positive samples to enhance discriminative representations and mitigate distribution shifts within the time series. TempSSL’s innovation lies in two key aspects. Firstly, it underscores the importance of temporal dependencies for time series forecasting by designing specific pre-training tasks. Secondly, it effectively integrates contrastive learning and masked modeling, leveraging their respective strengths to develop time series representation with strong instance discriminability and local perceptibility. Extensive experiments across seven widely used benchmark datasets demonstrate that TempSSL consistently outperforms existing self-supervised and end-to-end forecasting methods, achieving improvements ranging from 1.92% <span><math><mo>∼</mo></math></span> 78.12%. Additionally, TempSSL’s practical effectiveness is further demonstrated through successful application in natural gas demand forecasting.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112652"},"PeriodicalIF":7.2,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142573468","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

Clustering dynamic networks by discriminating roles of vertices and capturing temporality with subsequent feature projection 通过判别顶点的作用和随后的特征投影捕捉时间性，对动态网络进行聚类

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-24 DOI: 10.1016/j.knosys.2024.112660

Yaxiong Ma , Yue Gao , Zengfa Dou , Guohua Huang , Xiaoke Ma

Clustering dynamic networks has gained popularity due to the need to analyze complex systems that evolve over time, which cannot be fully characterized by traditional static models. It is highly non-trivial in comparison to clustering static network since it requires simultaneously to balance clustering accuracy and clustering drift, where clustering accuracy measures how clustering reflects structure of graph at current time, and clustering drift quantifies how clustering smoothes historical snapshot(s). In this study, we propose an algorithm clustering dynamic network by discriminating roles of vertices and capturing temporality with subsequent feature projection (CDN-DRCT). Specifically, clustering accuracy is achieved by factorizing high-order matrix of slice at current time, and vertices are divided into static and dynamic ones by the reconstruction errors. Finally, the proposed algorithm measures temporality of networks with a projection matrix, which connects subsequent features at the previous and current time, thereby enhancing clustering drift of clusters. In this case, temporality of dynamic networks is characterized from vertex and global level, providing a better way to balance clustering accuracy and clustering drift. Experimental results on 10 typical dynamic networks demonstrate the proposed algorithm is superior to baselines in terms of accuracy as well efficiency.

由于需要分析随时间演变的复杂系统，而传统的静态模型无法完全描述这些系统，因此动态网络聚类变得越来越流行。与静态网络聚类相比，动态网络聚类非常复杂，因为它需要同时兼顾聚类精度和聚类漂移，其中聚类精度衡量聚类如何反映当前时间的图结构，而聚类漂移量化聚类如何平滑历史快照。在本研究中，我们提出了一种通过判别顶点的作用和捕捉后续特征投影的时间性对动态网络进行聚类的算法（CDN-DRCT）。具体来说，通过对当前时间切片的高阶矩阵进行因式分解实现聚类精度，并根据重构误差将顶点分为静态顶点和动态顶点。最后，所提出的算法通过投影矩阵来衡量网络的时间性，将上一时间和当前时间的后续特征连接起来，从而增强聚类的漂移。在这种情况下，动态网络的时间性是从顶点和全局层面来描述的，从而为平衡聚类精度和聚类漂移提供了更好的方法。对 10 个典型动态网络的实验结果表明，所提出的算法在准确性和效率方面都优于基线算法。

{"title":"Clustering dynamic networks by discriminating roles of vertices and capturing temporality with subsequent feature projection","authors":"Yaxiong Ma , Yue Gao , Zengfa Dou , Guohua Huang , Xiaoke Ma","doi":"10.1016/j.knosys.2024.112660","DOIUrl":"10.1016/j.knosys.2024.112660","url":null,"abstract":"<div><div>Clustering dynamic networks has gained popularity due to the need to analyze complex systems that evolve over time, which cannot be fully characterized by traditional static models. It is highly non-trivial in comparison to clustering static network since it requires simultaneously to balance clustering accuracy and clustering drift, where clustering accuracy measures how clustering reflects structure of graph at current time, and clustering drift quantifies how clustering smoothes historical snapshot(s). In this study, we propose an algorithm <u><strong>c</strong></u>lustering <u><strong>d</strong></u>ynamic <u><strong>n</strong></u>etwork by <u><strong>d</strong></u>iscriminating <u><strong>r</strong></u>oles of vertices and <u><strong>c</strong></u>apturing <u><strong>t</strong></u>emporality with subsequent feature projection (<strong>CDN-DRCT</strong>). Specifically, clustering accuracy is achieved by factorizing high-order matrix of slice at current time, and vertices are divided into static and dynamic ones by the reconstruction errors. Finally, the proposed algorithm measures temporality of networks with a projection matrix, which connects subsequent features at the previous and current time, thereby enhancing clustering drift of clusters. In this case, temporality of dynamic networks is characterized from vertex and global level, providing a better way to balance clustering accuracy and clustering drift. Experimental results on 10 typical dynamic networks demonstrate the proposed algorithm is superior to baselines in terms of accuracy as well efficiency.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112660"},"PeriodicalIF":7.2,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142553729","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 De-confounded Causal Intervention method for knowledge graph error detection 用于知识图谱错误检测的双重去混淆因果干预方法

IF 7.2 1区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Knowledge-Based Systems

Pub Date : 2024-10-24 DOI: 10.1016/j.knosys.2024.112644

Yunxiao Yang, Jianting Chen, Xiaoying Gao, Yang Xiang

Due to the Knowledge Graph (KG) construction process, erroneous triples are virtually inevitable to be introduced into real-world KGs. Since these errors hinder the expressiveness and applicability of KGs, the development of knowledge graph error detection (KGED) methods is necessary. Despite the overall effectiveness of current KGED methods, their capacity to identify challenging errors is limited. In this work, we conduct empirical studies and find that previous works introduce structural and semantic bias, impeding the identification of erroneous triples, especially in challenging cases. To address this issue, we design a causal graph for the KGED task and propose a Dual De-confounded Causal Intervention (DuDCI) method for debiasing. Firstly, DuDCI utilizes the neighborhood and textual descriptions of triples to calculate their graph and text embeddings. Next, a Causal De-confounded Module is constructed to mitigate the impact of shortcuts caused by the bias through the front-door adjustment. Furthermore, we introduce Disentanglement Constraints to disentangle the information expressed by each embedding, thereby facilitating further bias mitigation. Experimental results on three widely used KGED datasets validate the effectiveness of DuDCI and demonstrate that DuDCI outperforms current KGED methods, with an improvement of at least 2.2%, especially in more challenging noise scenarios.

由于知识图谱（KG）的构建过程，在现实世界的知识图谱中几乎不可避免地会引入错误的三元组。由于这些错误阻碍了知识图谱的表达能力和适用性，因此有必要开发知识图谱错误检测（KGED）方法。尽管目前的知识图谱错误检测（KGED）方法总体上很有效，但它们识别高难度错误的能力有限。在这项工作中，我们进行了实证研究，发现以前的工作引入了结构和语义偏差，阻碍了错误三元组的识别，尤其是在具有挑战性的情况下。为了解决这个问题，我们为 KGED 任务设计了一个因果图，并提出了一种用于除错的双重去混淆因果干预（DuDCI）方法。首先，DuDCI 利用三元组的邻域和文本描述来计算它们的图嵌入和文本嵌入。然后，构建因果去混淆模块，通过前门调整来减轻偏差造成的捷径影响。此外，我们还引入了 "分解约束"（Disentanglement Constraints）来分解每个嵌入所表达的信息，从而进一步减轻偏差。在三个广泛使用的 KGED 数据集上的实验结果验证了 DuDCI 的有效性，并证明 DuDCI 优于当前的 KGED 方法，至少提高了 2.2%，尤其是在更具挑战性的噪声场景中。

{"title":"Dual De-confounded Causal Intervention method for knowledge graph error detection","authors":"Yunxiao Yang, Jianting Chen, Xiaoying Gao, Yang Xiang","doi":"10.1016/j.knosys.2024.112644","DOIUrl":"10.1016/j.knosys.2024.112644","url":null,"abstract":"<div><div>Due to the Knowledge Graph (KG) construction process, erroneous triples are virtually inevitable to be introduced into real-world KGs. Since these errors hinder the expressiveness and applicability of KGs, the development of knowledge graph error detection (KGED) methods is necessary. Despite the overall effectiveness of current KGED methods, their capacity to identify challenging errors is limited. In this work, we conduct empirical studies and find that previous works introduce structural and semantic bias, impeding the identification of erroneous triples, especially in challenging cases. To address this issue, we design a causal graph for the KGED task and propose a Dual De-confounded Causal Intervention (DuDCI) method for debiasing. Firstly, DuDCI utilizes the neighborhood and textual descriptions of triples to calculate their graph and text embeddings. Next, a Causal De-confounded Module is constructed to mitigate the impact of shortcuts caused by the bias through the front-door adjustment. Furthermore, we introduce Disentanglement Constraints to disentangle the information expressed by each embedding, thereby facilitating further bias mitigation. Experimental results on three widely used KGED datasets validate the effectiveness of DuDCI and demonstrate that DuDCI outperforms current KGED methods, with an improvement of at least 2.2%, especially in more challenging noise scenarios.</div></div>","PeriodicalId":49939,"journal":{"name":"Knowledge-Based Systems","volume":"305 ","pages":"Article 112644"},"PeriodicalIF":7.2,"publicationDate":"2024-10-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142553730","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