Information Sciences最新文献_第2页

Lyapunov-based emotion-aware switching in hybrid human-artificial intelligence customer service systems 基于lyapunov的混合人-人工智能客户服务系统的情感感知切换

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-29 DOI: 10.1016/j.ins.2026.123172

Zehan Tan, Henghua Shen

In this paper, we novelly apply the classical Lyapunov stability analysis to hybrid human-Artificial Intelligence (AI) customer service systems. The core idea is to use the Lyapunov ellipsoid of a linear autonomous dynamical system (LADS) to assess the customers’ emotional states and automatically determine whether a switch from the AI agent to a human agent is necessary. This involves two innovations: 1) User emotions are modeled as discrete-time LADSs in the Pleasure–Arousal–Dominance (PAD) space, parameterized by MBTI-specific dynamics matrices; 2) A Lyapunov function defines a safe emotional ellipsoid whose boundary, together with a Lyapunov Decay Rate (LDR), forms a dual-trigger switching mechanism to transfer service from the AI agent to a human agent when the user’s real-time emotional state approaches too fast or crosses the ellipsoid boundary, thus proactively preventing emotional destabilization.

To evaluate the proposed framework, we construct a domain-specific, multi-turn customer service dialogue dataset with PAD annotations. We compare our method with three other existing customer service systems, including methods with Fixed Lyapunov Ellipsoid for All (FLEA), Rule-Based Thresholding (RBT) and No-Switching Baseline (NSB). Comparative experiments demonstrate that the proposed switching mechanism significantly improves reduces negative emotional outcomes, enhances system usability and minimizes unnecessary human intervention.

本文新颖地将经典李雅普诺夫稳定性分析应用于人工智能（AI）混合客户服务系统。其核心思想是使用线性自主动力系统（LADS）的Lyapunov椭球来评估客户的情绪状态，并自动确定是否需要从AI代理切换到人类代理。这涉及两个创新：1)用户情绪建模为快乐-觉醒-支配（PAD）空间中的离散时间lads，由mbti特定的动态矩阵参数化；2) Lyapunov函数定义了一个安全的情绪椭球，该椭球边界与Lyapunov衰减率（Lyapunov Decay Rate， LDR）形成双触发切换机制，当用户实时情绪状态接近太快或越过椭球边界时，将服务从AI智能体转移到人类智能体，从而主动防止情绪不稳定。为了评估提出的框架，我们构建了一个具有PAD注释的特定领域的多轮客户服务对话数据集。我们将我们的方法与其他三种现有的客户服务系统进行了比较，包括所有人的固定Lyapunov椭球（FLEA）、基于规则的阈值（RBT）和无切换基线（NSB）方法。对比实验表明，所提出的切换机制显著改善了负面情绪结果，提高了系统可用性，并最大限度地减少了不必要的人为干预。

{"title":"Lyapunov-based emotion-aware switching in hybrid human-artificial intelligence customer service systems","authors":"Zehan Tan, Henghua Shen","doi":"10.1016/j.ins.2026.123172","DOIUrl":"10.1016/j.ins.2026.123172","url":null,"abstract":"<div><div>In this paper, we novelly apply the classical Lyapunov stability analysis to hybrid human-Artificial Intelligence (AI) customer service systems. The core idea is to use the Lyapunov ellipsoid of a linear autonomous dynamical system (LADS) to assess the customers’ emotional states and automatically determine whether a switch from the AI agent to a human agent is necessary. This involves two innovations: 1) User emotions are modeled as discrete-time LADSs in the Pleasure–Arousal–Dominance (PAD) space, parameterized by MBTI-specific dynamics matrices; 2) A Lyapunov function defines a safe emotional ellipsoid whose boundary, together with a Lyapunov Decay Rate (LDR), forms a dual-trigger switching mechanism to transfer service from the AI agent to a human agent when the user’s real-time emotional state approaches too fast or crosses the ellipsoid boundary, thus proactively preventing emotional destabilization.</div><div>To evaluate the proposed framework, we construct a domain-specific, multi-turn customer service dialogue dataset with PAD annotations. We compare our method with three other existing customer service systems, including methods with Fixed Lyapunov Ellipsoid for All (FLEA), Rule-Based Thresholding (RBT) and No-Switching Baseline (NSB). Comparative experiments demonstrate that the proposed switching mechanism significantly improves reduces negative emotional outcomes, enhances system usability and minimizes unnecessary human intervention.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123172"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190943","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

Beyond static cues: Detecting fine-grained forgeries via temporal inconsistencies in facial dynamics 超越静态线索：通过面部动态的时间不一致性检测细粒度伪造

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-30 DOI: 10.1016/j.ins.2026.123165

Peixu Zhang , Mohan Zhang , Tongyu Wang , Xinyu Yang

Fine-grained facial attribute editing represents a new frontier in deepfake technology, creating hyper-realistic forgeries that evade traditional detection methods by preserving identity and motion consistency. This subtlety poses a dual challenge: existing detectors, tuned for coarse artifacts, are rendered ineffective, and research is hampered by the absence of dedicated benchmark datasets. This paper argues that the key to unmasking these fine-grained forgeries lies not in static appearance, but in the temporal inconsistencies of underlying facial dynamics. To catalyze research in this area, we introduce EditForge, the first large-scale video dataset focused specifically on fine-grained facial attribute editing. Our analysis confirms that dynamic signals provide a powerful forensic trace that is consistently disrupted during the fine-grained forgery process. Building on this insight, we propose Fine-grained Forgery Mamba (F²-Mamba), a novel multimodal detection framework. F²-Mamba synergistically models features from facial dynamics, static appearance, and audio, employing robust alignment mechanisms and a Bi-Mamba architecture to efficiently capture long-range, cross-modal temporal dependencies. Extensive experiments validate that F²-Mamba establishes a new state-of-the-art, achieving area under the ROC curve (AUC) of 99.0% on fine-grained forgeries. This performance signals a paradigm shift towards behavior-based, dynamic analysis, significantly raising the bar for future forgery generation.

细粒度的面部属性编辑代表了深度伪造技术的新前沿，通过保持身份和运动一致性来创建超逼真的伪造物，逃避传统的检测方法。这种微妙之处带来了双重挑战：现有的针对粗糙工件进行调优的检测器变得无效，并且由于缺乏专用基准数据集而阻碍了研究。本文认为，揭露这些细粒度伪造的关键不在于静态外观，而在于潜在的面部动态的时间不一致性。为了促进这一领域的研究，我们引入了EditForge，这是第一个专门针对细粒度面部属性编辑的大规模视频数据集。我们的分析证实，动态信号提供了强大的法医痕迹，在细粒度伪造过程中一直被破坏。基于这一见解，我们提出了细粒度伪造曼巴（f2 -曼巴），一种新的多模态检测框架。F2-Mamba协同建模面部动态、静态外观和音频的特征，采用强大的对齐机制和Bi-Mamba架构来有效地捕获远程、跨模态的时间依赖性。大量的实验证实，F2-Mamba建立了一种新的最先进的技术，在细粒度伪造品上实现了99.0%的ROC曲线下面积（AUC）。这种表现标志着向基于行为的动态分析的范式转变，大大提高了未来伪造的标准。

{"title":"Beyond static cues: Detecting fine-grained forgeries via temporal inconsistencies in facial dynamics","authors":"Peixu Zhang , Mohan Zhang , Tongyu Wang , Xinyu Yang","doi":"10.1016/j.ins.2026.123165","DOIUrl":"10.1016/j.ins.2026.123165","url":null,"abstract":"<div><div>Fine-grained facial attribute editing represents a new frontier in deepfake technology, creating hyper-realistic forgeries that evade traditional detection methods by preserving identity and motion consistency. This subtlety poses a dual challenge: existing detectors, tuned for coarse artifacts, are rendered ineffective, and research is hampered by the absence of dedicated benchmark datasets. This paper argues that the key to unmasking these fine-grained forgeries lies not in static appearance, but in the temporal inconsistencies of underlying facial dynamics. To catalyze research in this area, we introduce EditForge, the first large-scale video dataset focused specifically on fine-grained facial attribute editing. Our analysis confirms that dynamic signals provide a powerful forensic trace that is consistently disrupted during the fine-grained forgery process. Building on this insight, we propose Fine-grained Forgery Mamba (F<sup>2</sup>-Mamba), a novel multimodal detection framework. F<sup>2</sup>-Mamba synergistically models features from facial dynamics, static appearance, and audio, employing robust alignment mechanisms and a Bi-Mamba architecture to efficiently capture long-range, cross-modal temporal dependencies. Extensive experiments validate that F<sup>2</sup>-Mamba establishes a new state-of-the-art, achieving area under the ROC curve (AUC) of 99.0% on fine-grained forgeries. This performance signals a paradigm shift towards behavior-based, dynamic analysis, significantly raising the bar for future forgery generation.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123165"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191068","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-branch Meso-Xception network for hybrid-domain feature of deepfake detection 基于混合域特征的双分支Meso-Xception网络深度伪造检测

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-27 DOI: 10.1016/j.ins.2026.123132

Yanan Song, Xiangyuan Chen, Ronghua Xu

The rapid advancement of deep learning-based generative technologies has led to remarkable achievements in deepfake applications using video and image media, particularly in areas such as face swapping and expression transfer. However, these developments have also triggered significant concerns regarding media authenticity and information security. Deepfake content often exhibits various artifacts: in the spatial domain, it may suffer from over-smoothed textures, loss of edge details, or jagged distortions; in the frequency domain, abnormal peaks in high-frequency spectra or noise-induced distortions may appear; and at the semantic level, misaligned keypoints and poor temporal coherence are frequently observed. To address these limitations, this study proposes a network architecture that first performs hybrid-domain feature extraction on deepfake samples. The Xception backbone, optimized through a knowledge distillation strategy to remove redundant layers, is combined with the lightweight MesoNet4 architecture to form a dual-branch backbone that can capture semantic features at different levels. While preserving semantic representation, the overall model size is compressed to just 8.0M parameters, achieving both high-precision detection of deepfake samples (accuracy

\geq

99%) and real-time inference performance (single-frame latency

\leq

10 ms).

基于深度学习的生成技术的快速发展使得使用视频和图像媒体的深度伪造应用取得了显着成就，特别是在面部交换和表情转移等领域。然而，这些发展也引发了对媒体真实性和信息安全的重大关注。深度伪造的内容通常会呈现出各种各样的伪影：在空间域中，它可能会遭受过度平滑的纹理、边缘细节的丢失或锯齿状的扭曲；在频域，高频频谱可能出现异常峰或噪声引起的畸变；在语义层面上，经常观察到关键点不对齐和时间连贯性差。为了解决这些限制，本研究提出了一种网络架构，首先对deepfake样本进行混合域特征提取。异常主干通过知识蒸馏策略进行优化以去除冗余层，并与轻量级MesoNet4架构结合形成双分支主干，可以捕获不同层次的语义特征。在保留语义表示的同时，整体模型大小被压缩到仅8.0M个参数，实现了对deepfake样本的高精度检测（准确率≥99%）和实时推理性能（单帧延迟≤10 ms）。

{"title":"Dual-branch Meso-Xception network for hybrid-domain feature of deepfake detection","authors":"Yanan Song, Xiangyuan Chen, Ronghua Xu","doi":"10.1016/j.ins.2026.123132","DOIUrl":"10.1016/j.ins.2026.123132","url":null,"abstract":"<div><div>The rapid advancement of deep learning-based generative technologies has led to remarkable achievements in deepfake applications using video and image media, particularly in areas such as face swapping and expression transfer. However, these developments have also triggered significant concerns regarding media authenticity and information security. Deepfake content often exhibits various artifacts: in the spatial domain, it may suffer from over-smoothed textures, loss of edge details, or jagged distortions; in the frequency domain, abnormal peaks in high-frequency spectra or noise-induced distortions may appear; and at the semantic level, misaligned keypoints and poor temporal coherence are frequently observed. To address these limitations, this study proposes a network architecture that first performs hybrid-domain feature extraction on deepfake samples. The Xception backbone, optimized through a knowledge distillation strategy to remove redundant layers, is combined with the lightweight MesoNet4 architecture to form a dual-branch backbone that can capture semantic features at different levels. While preserving semantic representation, the overall model size is compressed to just 8.0M parameters, achieving both high-precision detection of deepfake samples (accuracy <span><math><mo>≥</mo></math></span> 99%) and real-time inference performance (single-frame latency <span><math><mo>≤</mo></math></span> 10 ms).</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123132"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081919","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

A novel graph-theoretic and data visualization framework for spatiotemporal bubble analysis in turbulent flows 湍流中时空气泡分析的一种新的图论和数据可视化框架

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-29 DOI: 10.1016/j.ins.2026.123164

Alaa A. Najim , Safa A. Najim

This paper presents a novel methodology for analyzing bubble movement in turbulent flows by framing multi-object tracking as a global optimization problem on a spatiotemporal graph. Bubble detections across frames are modeled as nodes in a directed graph, with tracking addressed through a minimum cost flow algorithm. Unlike sequential trackers, our approach considers the entire temporal context to achieve globally optimal paths, reducing identity switches and fragmentation caused by occlusions and interactions. The pipeline integrates anisotropic Gaussian filtering and Otsu’s thresholding with a cost function enforcing movement and appearance consistency. Experimental validation across flow rates (

Q_{1} = 1.0

,

Q_{2} = 2.0

,

Q_{3} = 3.0

) identifies distinct systems: stable homogeneous, transitional interference, and saturated aggregation-dominated. The method simultaneously measures temporal evolution of bubble count

N (t)

, mean diameter

\overset{―}{d} (t)

, and velocity

\overset{―}{v} (t)

, revealing critical phenomena like saturation effects where increased airflow leads to bubble aggregation rather than an increased count.

本文提出了一种分析湍流中气泡运动的新方法，该方法将多目标跟踪视为一个时空图上的全局优化问题。跨帧的气泡检测被建模为有向图中的节点，通过最小成本流算法进行跟踪。与顺序跟踪器不同，我们的方法考虑了整个时间背景来实现全局最优路径，减少了由闭塞和相互作用引起的身份切换和碎片化。该管道集成了各向异性高斯滤波和Otsu阈值法，并使用成本函数强制运动和外观一致性。不同流量（Q1=1.0, Q2=2.0, Q3=3.0）的实验验证确定了不同的系统：稳定均匀，过渡干扰和饱和聚集主导。该方法同时测量气泡计数N(t)、平均直径d - (t)和速度v - (t)的时间演变，揭示了关键现象，如饱和效应，其中气流增加导致气泡聚集而不是增加计数。

{"title":"A novel graph-theoretic and data visualization framework for spatiotemporal bubble analysis in turbulent flows","authors":"Alaa A. Najim , Safa A. Najim","doi":"10.1016/j.ins.2026.123164","DOIUrl":"10.1016/j.ins.2026.123164","url":null,"abstract":"<div><div>This paper presents a novel methodology for analyzing bubble movement in turbulent flows by framing multi-object tracking as a global optimization problem on a spatiotemporal graph. Bubble detections across frames are modeled as nodes in a directed graph, with tracking addressed through a minimum cost flow algorithm. Unlike sequential trackers, our approach considers the entire temporal context to achieve globally optimal paths, reducing identity switches and fragmentation caused by occlusions and interactions. The pipeline integrates anisotropic Gaussian filtering and Otsu’s thresholding with a cost function enforcing movement and appearance consistency. Experimental validation across flow rates (<span><math><msub><mi>Q</mi><mn>1</mn></msub><mo>=</mo><mn>1.0</mn></math></span>, <span><math><msub><mi>Q</mi><mn>2</mn></msub><mo>=</mo><mn>2.0</mn></math></span>, <span><math><msub><mi>Q</mi><mn>3</mn></msub><mo>=</mo><mn>3.0</mn></math></span>) identifies distinct systems: stable homogeneous, transitional interference, and saturated aggregation-dominated. The method simultaneously measures temporal evolution of bubble count <span><math><mi>N</mi><mo>(</mo><mi>t</mi><mo>)</mo></math></span>, mean diameter <span><math><mover><mi>d</mi><mo>―</mo></mover><mo>(</mo><mi>t</mi><mo>)</mo></math></span>, and velocity <span><math><mover><mi>v</mi><mo>―</mo></mover><mo>(</mo><mi>t</mi><mo>)</mo></math></span>, revealing critical phenomena like saturation effects where increased airflow leads to bubble aggregation rather than an increased count.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123164"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190945","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

DE-SHAP: A meta-optimization framework leveraging differential evolution for efficient and scalable kernel SHAP explanations DE-SHAP：一个元优化框架，利用差分进化来实现高效和可扩展的内核SHAP解释

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-22 DOI: 10.1016/j.ins.2026.123128

El Arbi Abdellaoui Alaoui , Hayat Sahlaoui , Amine Sallah , Anand Nayyar

The intractable nature of the Kernel SHAP computation presents a significant barrier to implementing rigorous explainable AI in practical large scale machine learning systems. The standard SHAP calculations are based on iterative evaluations of the model averaged over all possible feature subsets, and they become intractable for large dimensional data. To mitigate this difficulty, this paper presents “DE-SHAP”, an evolutionary optimization approach which speeds up ES-method by automatically tuning its parameters via Differential Evolution (DE). DE systematically optimizes two critical parameters background dataset size and Monte Carlo sampling count to minimize computational cost and maintain theoretical soundness within SHAP’s additive feature attribution framework. The proposed framework employs specialized evolutionary operators to ensure convergence efficiency and stability during optimization.

To test and validate the proposed methodology, extensive experiments were performed on various benchmark datasets and model architectures, and the results showed that DE-SHAP reduces computing cost by 52–97%, while the deviation of SHAP values is less than 5% and accuracy of the model remains within a range of approximately 0.5%. Given the fact that it is usually expensive to obtain explanations compared to predictions, these findings provide evidence that DE-SHAP can offer a similar quality of interpretability for a much lower computational cost. By implementing a computationally efficient theoretically justified improvement to a popular interpretability approach. DE-SHAP enables scalable and practical deployment of high-quality model explanations for complex systems with up to 784 input features. This contribution advances the feasibility of rigorous explainable AI in real-world applications, bridging the gap between interpretability research and operational machine learning.

内核SHAP计算的棘手性质为在实际的大规模机器学习系统中实现严格的可解释人工智能提出了一个重大障碍。标准的SHAP计算基于对所有可能的特征子集平均的模型的迭代评估，并且对于大维度数据变得难以处理。为了减轻这一困难，本文提出了一种“DE- shap”进化优化方法，该方法通过差分进化（DE）自动调整es方法的参数来加快es方法的速度。DE系统地优化了背景数据集大小和蒙特卡罗采样计数两个关键参数，以最大限度地减少计算成本，并在SHAP的加性特征归因框架内保持理论的合理性。该框架采用专门的进化算子来保证优化过程的收敛效率和稳定性。为了测试和验证所提出的方法，在各种基准数据集和模型架构上进行了大量实验，结果表明，DE-SHAP将计算成本降低了52-97%，而SHAP值的偏差小于5%，模型的精度保持在约0.5%的范围内。考虑到与预测相比，获得解释通常是昂贵的，这些发现提供了证据，表明DE-SHAP可以以更低的计算成本提供类似质量的可解释性。通过实现对流行的可解释性方法的计算效率理论上合理的改进。DE-SHAP能够为具有多达784个输入特征的复杂系统提供可扩展和实用的高质量模型解释部署。这一贡献促进了在现实世界应用中严格可解释人工智能的可行性，弥合了可解释性研究和操作机器学习之间的差距。

{"title":"DE-SHAP: A meta-optimization framework leveraging differential evolution for efficient and scalable kernel SHAP explanations","authors":"El Arbi Abdellaoui Alaoui , Hayat Sahlaoui , Amine Sallah , Anand Nayyar","doi":"10.1016/j.ins.2026.123128","DOIUrl":"10.1016/j.ins.2026.123128","url":null,"abstract":"<div><div>The intractable nature of the Kernel SHAP computation presents a significant barrier to implementing rigorous explainable AI in practical large scale machine learning systems. The standard SHAP calculations are based on iterative evaluations of the model averaged over all possible feature subsets, and they become intractable for large dimensional data. To mitigate this difficulty, this paper presents “DE-SHAP”, an evolutionary optimization approach which speeds up ES-method by automatically tuning its parameters via Differential Evolution (DE). DE systematically optimizes two critical parameters background dataset size and Monte Carlo sampling count to minimize computational cost and maintain theoretical soundness within SHAP’s additive feature attribution framework. The proposed framework employs specialized evolutionary operators to ensure convergence efficiency and stability during optimization.</div><div>To test and validate the proposed methodology, extensive experiments were performed on various benchmark datasets and model architectures, and the results showed that DE-SHAP reduces computing cost by 52–97%, while the deviation of SHAP values is less than 5% and accuracy of the model remains within a range of approximately 0.5%. Given the fact that it is usually expensive to obtain explanations compared to predictions, these findings provide evidence that DE-SHAP can offer a similar quality of interpretability for a much lower computational cost. By implementing a computationally efficient theoretically justified improvement to a popular interpretability approach. DE-SHAP enables scalable and practical deployment of high-quality model explanations for complex systems with up to 784 input features. This contribution advances the feasibility of rigorous explainable AI in real-world applications, bridging the gap between interpretability research and operational machine learning.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123128"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146190946","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

Weakly supervised object localization via frequency guidance with consistency awareness 带有一致性意识的频率引导弱监督目标定位

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-31 DOI: 10.1016/j.ins.2026.123186

Bingfeng Li , Erdong Shi , Boxiang Lv , Qingshan Chen , Jiayu Zhang , Haoran Feng , Shuai Wang

Weakly supervised object localization (WSOL) aims to localize object regions using only image-level labels, avoiding costly bounding box annotations. Recent methods employ foreground prediction map (FPM) to distinguish foreground from background and improve localization completeness. However, FPM-based approaches often rely on background modeling, which can introduce semantic ambiguity due to the inherent uncertainty of background regions. To address this, we propose the Frequency-Guided and Consistency-Aware (FGCA) Network, which enhances foreground modeling without relying on background modeling. FGCA first employs a Wavelet-Frequency Attention Module (WFAM) to decompose features into low- and high-frequency components, selectively enhancing semantic and fine-grained structural information. Subsequently, a consistency-aware optimization framework is introduced to enhance the semantic coherence and structural integrity of the predicted foreground by incorporating complementary constraints from both local and global perspectives. Specifically, the Semantic-Spatial Consistency Loss (SSCL) enforces fine-grained consistency by integrating category-specific discrimination and pixel-level structural smoothness. In parallel, the Foreground-Global Kullback–Leibler Alignment Loss (FG-KL) regularizes the global semantic distribution of the foreground, guiding the network to emphasize contextually relevant object regions while suppressing background-induced noise. Experiments on standard WSOL benchmarks show that FGCA outperforms state-of-the-art methods, particularly in foreground completeness, boundary precision, and semantic consistency.

弱监督对象定位（WSOL）旨在仅使用图像级别的标签来定位对象区域，避免代价高昂的边界框注释。最近的方法采用前景预测图（FPM）来区分前景和背景，提高定位的完整性。然而，基于fpm的方法往往依赖于背景建模，由于背景区域固有的不确定性，这可能会引入语义模糊。为了解决这个问题，我们提出了频率引导和一致性感知（FGCA）网络，它在不依赖背景建模的情况下增强了前景建模。FGCA首先采用小波频率注意模块（WFAM）将特征分解为低频和高频分量，选择性地增强语义和细粒度结构信息。随后，引入一致性感知优化框架，通过从局部和全局角度结合互补约束，增强预测前景的语义一致性和结构完整性。具体来说，语义空间一致性损失（SSCL）通过集成特定类别的区分和像素级的结构平滑来加强细粒度的一致性。同时，前景-全局Kullback-Leibler对齐损失（FG-KL）对前景的全局语义分布进行正则化，引导网络强调上下文相关的目标区域，同时抑制背景引起的噪声。在标准WSOL基准测试上的实验表明，FGCA优于最先进的方法，特别是在前景完整性、边界精度和语义一致性方面。

{"title":"Weakly supervised object localization via frequency guidance with consistency awareness","authors":"Bingfeng Li , Erdong Shi , Boxiang Lv , Qingshan Chen , Jiayu Zhang , Haoran Feng , Shuai Wang","doi":"10.1016/j.ins.2026.123186","DOIUrl":"10.1016/j.ins.2026.123186","url":null,"abstract":"<div><div>Weakly supervised object localization (WSOL) aims to localize object regions using only image-level labels, avoiding costly bounding box annotations. Recent methods employ foreground prediction map (FPM) to distinguish foreground from background and improve localization completeness. However, FPM-based approaches often rely on background modeling, which can introduce semantic ambiguity due to the inherent uncertainty of background regions. To address this, we propose the Frequency-Guided and Consistency-Aware (FGCA) Network, which enhances foreground modeling without relying on background modeling. FGCA first employs a Wavelet-Frequency Attention Module (WFAM) to decompose features into low- and high-frequency components, selectively enhancing semantic and fine-grained structural information. Subsequently, a consistency-aware optimization framework is introduced to enhance the semantic coherence and structural integrity of the predicted foreground by incorporating complementary constraints from both local and global perspectives. Specifically, the Semantic-Spatial Consistency Loss (SSCL) enforces fine-grained consistency by integrating category-specific discrimination and pixel-level structural smoothness. In parallel, the Foreground-Global Kullback–Leibler Alignment Loss (FG-KL) regularizes the global semantic distribution of the foreground, guiding the network to emphasize contextually relevant object regions while suppressing background-induced noise. Experiments on standard WSOL benchmarks show that FGCA outperforms state-of-the-art methods, particularly in foreground completeness, boundary precision, and semantic consistency.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123186"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191065","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

PMFF-SRNet: A progressive multi-feature fusion network for hyperspectral image reconstruction PMFF-SRNet：用于高光谱图像重建的渐进式多特征融合网络

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-28 DOI: 10.1016/j.ins.2026.123129

Tianhao Yuan, Xia Wang, Qiyang Sun, Yuyang Li

Hyperspectral imaging (HSI) contains abundant spatial and spectral cues, making it advantageous for a wide range of applications, including Earth observation, medicine, as well as agricultural analysis. However, conventional HSI systems are often constrained by high costs, limited acquisition efficiency, and weak adaptability to dynamic scenes. To mitigate these limitations, we propose a progressive multi-feature fusion model, termed PMFF-SRNet, for RGB-to-hyperspectral reconstruction. The proposed model progressively recovers spectral information through multiple stages, which helps reduce spectral redundancy while improving reconstruction efficiency. Furthermore, a Local–Global Spectral Attention (LGSA) module is employed to model spectral features at different granularities, where grouped self-attention focuses on local band interactions while spectral order information contributes to long-range dependency modeling. In addition, a Discrete Wavelet Attention (DWA) module is incorporated into the skip connections to enhance texture and edge restoration by exploiting the multi-scale characteristics of wavelet transforms. Results obtained on the NTIRE benchmarks indicate that PMFF-SRNet achieves competitive reconstruction performance across multiple evaluation metrics, while maintaining a lightweight and computationally efficient architecture. These findings demonstrate the strong potential of PMFF-SRNet for practical hyperspectral reconstruction tasks.

高光谱成像（HSI）包含丰富的空间和光谱线索，使其具有广泛的应用优势，包括地球观测，医学和农业分析。然而，传统的HSI系统往往受到高成本、有限的采集效率和对动态场景适应性弱的限制。为了减轻这些限制，我们提出了一种渐进的多特征融合模型，称为PMFF-SRNet，用于rgb到高光谱的重建。该模型通过多阶段逐步恢复光谱信息，减少了光谱冗余，提高了重建效率。此外，采用局部-全局光谱关注（LGSA）模型对不同粒度的光谱特征进行建模，其中分组自关注侧重于局部波段的相互作用，而光谱顺序信息则有助于远程依赖建模。此外，利用小波变换的多尺度特性，将离散小波注意（DWA）模块集成到跳跳连接中，增强纹理和边缘的恢复。在整个基准测试中获得的结果表明，PMFF-SRNet在保持轻量级和计算效率的架构的同时，在多个评估指标上实现了具有竞争力的重建性能。这些发现证明了PMFF-SRNet在实际高光谱重建任务中的强大潜力。

{"title":"PMFF-SRNet: A progressive multi-feature fusion network for hyperspectral image reconstruction","authors":"Tianhao Yuan, Xia Wang, Qiyang Sun, Yuyang Li","doi":"10.1016/j.ins.2026.123129","DOIUrl":"10.1016/j.ins.2026.123129","url":null,"abstract":"<div><div>Hyperspectral imaging (HSI) contains abundant spatial and spectral cues, making it advantageous for a wide range of applications, including Earth observation, medicine, as well as agricultural analysis. However, conventional HSI systems are often constrained by high costs, limited acquisition efficiency, and weak adaptability to dynamic scenes. To mitigate these limitations, we propose a progressive multi-feature fusion model, termed PMFF-SRNet, for RGB-to-hyperspectral reconstruction. The proposed model progressively recovers spectral information through multiple stages, which helps reduce spectral redundancy while improving reconstruction efficiency. Furthermore, a Local–Global Spectral Attention (LGSA) module is employed to model spectral features at different granularities, where grouped self-attention focuses on local band interactions while spectral order information contributes to long-range dependency modeling. In addition, a Discrete Wavelet Attention (DWA) module is incorporated into the skip connections to enhance texture and edge restoration by exploiting the multi-scale characteristics of wavelet transforms. Results obtained on the NTIRE benchmarks indicate that PMFF-SRNet achieves competitive reconstruction performance across multiple evaluation metrics, while maintaining a lightweight and computationally efficient architecture. These findings demonstrate the strong potential of PMFF-SRNet for practical hyperspectral reconstruction tasks.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123129"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191062","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

Hidden policy conditional attribute-based keyword search in federated learning framework 联邦学习框架中基于隐藏策略条件属性的关键字搜索

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-01-27 DOI: 10.1016/j.ins.2026.123150

Zerui Guo , Sha Ma , Qiong Huang

Nowadays, people are paying increasing attention to the security of personal privacy data in artificial intelligence systems. Federated learning is a solution to address the issue of unified data collection for training in artificial intelligence systems. Among these, the Cross-Sample computation process is a stage where personal privacy data is often leaked in federated learning, and a viable mechanism to ensure data confidentiality during the Cross-Sample computation process in federated learning is provided by attribute-based searchable encryption (ABSE). However, the search process in most existing ABSE systems is inherently sequential, which fundamentally precludes their use in scenarios demanding high throughput and concurrent execution. Meanwhile, it is worth noting that most ABSE schemes fail to achieve both rich attribute expression and hidden policy. In response to these limitations, the present work introduces Hidden Policy Conditional Attribute-Based Keyword Search (HP-CABKS), which supports server-side concurrent evaluation in the search phase. Under the Generic Group Model, our scheme satisfies adaptive security against chosen keyword attacks and keyword secrecy. Experimental results demonstrate that our scheme exhibits low time consumption in the search phase. Our scheme exhibits robust security, high efficiency, and strong practicality, making it well-suited for real-world applications such as Cross-Sample computation in federated learning.

如今，人们越来越关注人工智能系统中个人隐私数据的安全性。联邦学习是解决人工智能系统训练中统一数据收集问题的一种解决方案。其中，跨样本计算过程是联邦学习中个人隐私数据容易泄露的阶段，基于属性的可搜索加密（property -based searchable encryption， ABSE）为联邦学习跨样本计算过程中的数据保密性提供了一种可行的机制。然而，大多数现有ABSE系统中的搜索过程本质上是顺序的，这从根本上阻碍了它们在需要高吞吐量和并发执行的场景中的使用。同时，值得注意的是，大多数ABSE方案不能同时实现富属性表达和隐藏策略。为了应对这些限制，本工作引入了基于隐藏策略条件属性的关键字搜索（HP-CABKS），它支持搜索阶段的服务器端并发评估。在通用群模型下，我们的方案满足自适应安全性和关键字保密。实验结果表明，该方案具有较低的搜索耗时。我们的方案具有强大的安全性、高效率和很强的实用性，非常适合联邦学习中的交叉样本计算等实际应用。

{"title":"Hidden policy conditional attribute-based keyword search in federated learning framework","authors":"Zerui Guo , Sha Ma , Qiong Huang","doi":"10.1016/j.ins.2026.123150","DOIUrl":"10.1016/j.ins.2026.123150","url":null,"abstract":"<div><div>Nowadays, people are paying increasing attention to the security of personal privacy data in artificial intelligence systems. Federated learning is a solution to address the issue of unified data collection for training in artificial intelligence systems. Among these, the Cross-Sample computation process is a stage where personal privacy data is often leaked in federated learning, and a viable mechanism to ensure data confidentiality during the Cross-Sample computation process in federated learning is provided by attribute-based searchable encryption (ABSE). However, the search process in most existing ABSE systems is inherently sequential, which fundamentally precludes their use in scenarios demanding high throughput and concurrent execution. Meanwhile, it is worth noting that most ABSE schemes fail to achieve both rich attribute expression and hidden policy. In response to these limitations, the present work introduces Hidden Policy Conditional Attribute-Based Keyword Search (HP-CABKS), which supports server-side concurrent evaluation in the search phase. Under the Generic Group Model, our scheme satisfies adaptive security against chosen keyword attacks and keyword secrecy. Experimental results demonstrate that our scheme exhibits low time consumption in the search phase. Our scheme exhibits robust security, high efficiency, and strong practicality, making it well-suited for real-world applications such as Cross-Sample computation in federated learning.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123150"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191061","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

Communication-efficient decentralized federated graph learning via knowledge distillation under dual heterogeneity 双异构下基于知识蒸馏的高效通信分散联邦图学习

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-02-02 DOI: 10.1016/j.ins.2026.123192

Guojiang Shen , Xu Guo , Haopeng Yuan , Jiaxin Du , Wenyi Zhang , Y. Neil Qu , Xiangjie Kong

Federated Graph Learning (FGL) integrates the privacy-preserving advantages of federated learning with the structural data modeling capabilities of graph neural networks. To mitigate single points of failure and communication bottlenecks inherent in server-based federated learning, we focus on Decentralized Federated Graph Learning (DFGL), which facilitates efficient distributed training in a peer-to-peer (P2P) manner. However, existing approaches predominantly rely on the direct exchange of gradients or model parameters. Under constrained communication resources and dual heterogeneity (data and model), such approaches often suffer from reduced training efficiency and suboptimal performance. To address these issues, we propose DFedGD, a decentralized federated graph learning framework that integrates graph condensation with knowledge distillation. Specifically, DFedGD extracts representative knowledge by synthesizing condensed subgraphs that approximate the global data distribution within a fully decentralized environment. Clients exchange and learn from condensed subgraphs and logits via a mutual knowledge distillation mechanism, thereby enhancing communication efficiency and privacy preservation while naturally supporting model heterogeneity. Furthermore, an alignment mechanism based on shared anchor samples is incorporated to enforce latent representation consistency, effectively mitigating domain shifts arising from non-IID data distributions. Extensive experiments on three publicly available datasets demonstrate that our framework outperforms competitive baselines.

联邦图学习（FGL）将联邦学习的隐私保护优势与图神经网络的结构数据建模能力相结合。为了缓解基于服务器的联邦学习中固有的单点故障和通信瓶颈，我们专注于分散式联邦图学习（DFGL），它以点对点（P2P）的方式促进了高效的分布式训练。然而，现有的方法主要依赖于梯度或模型参数的直接交换。在通信资源受限和数据和模型双重异构的情况下，这种方法往往存在训练效率降低和性能次优的问题。为了解决这些问题，我们提出了DFedGD，这是一个分散的联邦图学习框架，它集成了图冷凝和知识蒸馏。具体来说，DFedGD通过合成在完全分散的环境中近似全局数据分布的压缩子图来提取代表性知识。客户端通过相互的知识蒸馏机制从压缩的子图和逻辑中交换和学习，从而提高了通信效率和隐私保护，同时自然地支持模型的异构性。此外，采用基于共享锚点样本的对齐机制来增强潜在表示一致性，有效减轻非iid数据分布引起的域偏移。在三个公开可用的数据集上进行的大量实验表明，我们的框架优于竞争性基线。

{"title":"Communication-efficient decentralized federated graph learning via knowledge distillation under dual heterogeneity","authors":"Guojiang Shen , Xu Guo , Haopeng Yuan , Jiaxin Du , Wenyi Zhang , Y. Neil Qu , Xiangjie Kong","doi":"10.1016/j.ins.2026.123192","DOIUrl":"10.1016/j.ins.2026.123192","url":null,"abstract":"<div><div>Federated Graph Learning (FGL) integrates the privacy-preserving advantages of federated learning with the structural data modeling capabilities of graph neural networks. To mitigate single points of failure and communication bottlenecks inherent in server-based federated learning, we focus on Decentralized Federated Graph Learning (DFGL), which facilitates efficient distributed training in a peer-to-peer (P2P) manner. However, existing approaches predominantly rely on the direct exchange of gradients or model parameters. Under constrained communication resources and dual heterogeneity (data and model), such approaches often suffer from reduced training efficiency and suboptimal performance. To address these issues, we propose DFedGD, a decentralized federated graph learning framework that integrates graph condensation with knowledge distillation. Specifically, DFedGD extracts representative knowledge by synthesizing condensed subgraphs that approximate the global data distribution within a fully decentralized environment. Clients exchange and learn from condensed subgraphs and logits via a mutual knowledge distillation mechanism, thereby enhancing communication efficiency and privacy preservation while naturally supporting model heterogeneity. Furthermore, an alignment mechanism based on shared anchor samples is incorporated to enforce latent representation consistency, effectively mitigating domain shifts arising from non-IID data distributions. Extensive experiments on three publicly available datasets demonstrate that our framework outperforms competitive baselines.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123192"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191144","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

SWGCN: Synergy weighted graph convolutional network for multi-behavior recommendation SWGCN：多行为推荐的协同加权图卷积网络

IF 6.8 1区计算机科学 0 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Sciences

Pub Date : 2026-05-25 Epub Date: 2026-02-03 DOI: 10.1016/j.ins.2026.123177

Fangda Chen, Yueyang Wang, Chaoli Lou, Min Gao, Qingyu Xiong

Multi-behavior recommendation paradigms have emerged to capture diverse user activities, forecasting primary conversions (e.g., purchases) by leveraging secondary signals like browsing history. However, current graph-based methods often overlook cross-behavioral synergistic signals and the fine-grained intensity of individual actions. Motivated by the need to overcome these shortcomings, we introduce Synergy Weighted Graph Convolutional Network (SWGCN). SWGCN introduces two novel components: a Target Preference Weigher, which adaptively assigns weights to user-item interactions within each behavior, and a Synergy Alignment Task, which guides its training by leveraging an Auxiliary Preference Valuator. This task prioritizes interactions from synergistic signals that more accurately reflect user preferences. The performance of our model is rigorously evaluated through comprehensive tests on three open-source datasets, specifically Taobao, IJCAI, and Beibei. On the Taobao dataset, SWGCN yields relative gains of 112.49% and 156.36% in terms of Hit Ratio (HR) and Normalized Discounted Cumulative Gain (NDCG), respectively. It also yields consistent gains on IJCAI and Beibei, confirming its robustness and generalizability across various datasets. Our implementation is open-sourced and can be accessed via https://github.com/FangdChen/SWGCN.

多行为推荐模式的出现是为了捕捉不同的用户活动，通过利用浏览历史等次要信号来预测主要转换（例如，购买）。然而，目前基于图的方法往往忽略了跨行为的协同信号和个体行为的细粒度强度。为了克服这些缺点，我们引入了协同加权图卷积网络（SWGCN）。SWGCN引入了两个新组件：一个目标偏好加权器，它自适应地为每个行为中的用户-项目交互分配权重；一个协同对齐任务，它通过利用辅助偏好评估器来指导其训练。该任务优先考虑来自更准确反映用户偏好的协同信号的交互。我们的模型的性能通过三个开源数据集，特别是淘宝，IJCAI和贝贝的综合测试进行了严格的评估。在淘宝数据集上，SWGCN在命中率（HR）和归一化贴现累积增益（NDCG）方面的相对收益分别为112.49%和156.36%。它还在IJCAI和Beibei上产生一致的增益，证实了它在各种数据集上的稳健性和泛化性。我们的实现是开源的，可以通过https://github.com/FangdChen/SWGCN访问。

{"title":"SWGCN: Synergy weighted graph convolutional network for multi-behavior recommendation","authors":"Fangda Chen, Yueyang Wang, Chaoli Lou, Min Gao, Qingyu Xiong","doi":"10.1016/j.ins.2026.123177","DOIUrl":"10.1016/j.ins.2026.123177","url":null,"abstract":"<div><div>Multi-behavior recommendation paradigms have emerged to capture diverse user activities, forecasting primary conversions (e.g., purchases) by leveraging secondary signals like browsing history. However, current graph-based methods often overlook cross-behavioral synergistic signals and the fine-grained intensity of individual actions. Motivated by the need to overcome these shortcomings, we introduce <strong>S</strong>ynergy <strong>W</strong>eighted <strong>G</strong>raph <strong>C</strong>onvolutional <strong>N</strong>etwork (<strong>SWGCN</strong>). SWGCN introduces two novel components: a Target Preference Weigher, which adaptively assigns weights to user-item interactions within each behavior, and a Synergy Alignment Task, which guides its training by leveraging an Auxiliary Preference Valuator. This task prioritizes interactions from synergistic signals that more accurately reflect user preferences. The performance of our model is rigorously evaluated through comprehensive tests on three open-source datasets, specifically Taobao, IJCAI, and Beibei. On the Taobao dataset, SWGCN yields relative gains of 112.49% and 156.36% in terms of Hit Ratio (HR) and Normalized Discounted Cumulative Gain (NDCG), respectively. It also yields consistent gains on IJCAI and Beibei, confirming its robustness and generalizability across various datasets. Our implementation is open-sourced and can be accessed via <span><span>https://github.com/FangdChen/SWGCN</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":51063,"journal":{"name":"Information Sciences","volume":"739 ","pages":"Article 123177"},"PeriodicalIF":6.8,"publicationDate":"2026-05-25","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146191143","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