Information Processing & Management最新文献_第10页

Unifying RGB, thermal infrared, infrared, and text for person re-identification: A multi-modal dataset and vision-language transformer 统一RGB、热红外、红外和文本用于人物再识别：一个多模态数据集和视觉语言转换器

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2026-01-07 DOI: 10.1016/j.ipm.2025.104589

Muhammad Umair, Zhou Jun, Muhammad Hammad Musaddiq, Ahmad Muhammad

Person re-identification (ReID) under real-world multi-modal settings remains constrained by the lack of unified, diverse datasets and modality-aware learning strategies. To bridge this gap, we propose Multi-modal ReID (MM-ReID), a large-scale dataset encompassing 8537 unique IDs, 0.85 million images with three aligned image modalities RGB, Infrared (IR), Thermal Infrared (TI), and 0.83 million natural language descriptions. MM-ReID captures diverse scenarios, including indoor/outdoor, cross-camera views, day and night, and cloth changes, offering a comprehensive foundation for multi-modal ReID research. To build one unified multi-modal person Re-ID, we introduce Cross-Modal Semantic Anchoring (CMSA): CMSA injects fixed vision-language embeddings as parameter-free semantic anchors that steer a ViT towards a modality-agnostic, language-aware space, enabling rich semantic transfer through text-vision alignment. Our training incorporates two synergistic loss functions: Caption-Adaptive Triplet loss dynamically adjusts the triplet margin according to caption similarity, forcing harder negatives when textual descriptions overlap and yielding stronger discrimination. Caption-Aware CIM-T loss (Cross-Identity Inter-modal Margin with Text) simultaneously enlarges inter-identity gaps and contracts intra-identity distances across RGB-IR-TI views, guided by caption context to resolve ambiguous appearances. Our method attains 79.4 mAP and 97.5 R-5 on the Market1501-MM dataset, representing improvements of +1.4 mAP and +0.7 R-5 over prior SOTA approaches. Extensive experiments on MM-ReID demonstrate superior generalization and adaptability across unseen modalities and domains. Our approach establishes a new paradigm for modality-extensible and interpretable multi-modal ReID research.

现实世界多模态环境下的人再识别（ReID）仍然受到缺乏统一的、多样化的数据集和模态感知学习策略的限制。为了弥补这一差距，我们提出了多模态ReID (MM-ReID)，这是一个包含8537个唯一id的大型数据集，85万张图像，具有三种排列的图像模式RGB，红外（IR），热红外（TI）和83万自然语言描述。MM-ReID捕获了多种场景，包括室内/室外、跨镜头视图、白天和夜晚以及布料变化，为多模态ReID研究提供了全面的基础。为了构建统一的多模态人Re-ID，我们引入了跨模态语义锚定（CMSA）： CMSA注入固定的视觉语言嵌入作为无参数语义锚定，将ViT引导到模态不可知的语言感知空间，通过文本视觉对齐实现丰富的语义传递。我们的训练包含两个协同损失函数：标题自适应三重损失根据标题相似度动态调整三重边界，当文本描述重叠时强制使用更硬的否定，并产生更强的辨别。字幕感知的CIM-T损失（文本跨同一性多模态边距）在RGB-IR-TI视图中同时扩大了同一性之间的差距，并缩小了同一性之间的距离，在标题上下文的指导下解决了模糊的外观。我们的方法在Market1501-MM数据集上获得了79.4 mAP和97.5 R-5，比之前的SOTA方法提高了+1.4 mAP和+0.7 R-5。大量的实验表明，MM-ReID在未知的模式和领域具有卓越的泛化和适应性。我们的方法为模态可扩展和可解释的多模态ReID研究建立了一个新的范式。

{"title":"Unifying RGB, thermal infrared, infrared, and text for person re-identification: A multi-modal dataset and vision-language transformer","authors":"Muhammad Umair, Zhou Jun, Muhammad Hammad Musaddiq, Ahmad Muhammad","doi":"10.1016/j.ipm.2025.104589","DOIUrl":"10.1016/j.ipm.2025.104589","url":null,"abstract":"<div><div>Person re-identification (ReID) under real-world multi-modal settings remains constrained by the lack of unified, diverse datasets and modality-aware learning strategies. To bridge this gap, we propose Multi-modal ReID (MM-ReID), a large-scale dataset encompassing 8537 unique IDs, 0.85 million images with three aligned image modalities RGB, Infrared (IR), Thermal Infrared (TI), and 0.83 million natural language descriptions. MM-ReID captures diverse scenarios, including indoor/outdoor, cross-camera views, day and night, and cloth changes, offering a comprehensive foundation for multi-modal ReID research. To build one unified multi-modal person Re-ID, we introduce Cross-Modal Semantic Anchoring (CMSA): CMSA injects fixed vision-language embeddings as parameter-free semantic anchors that steer a ViT towards a modality-agnostic, language-aware space, enabling rich semantic transfer through text-vision alignment. Our training incorporates two synergistic loss functions: Caption-Adaptive Triplet loss dynamically adjusts the triplet margin according to caption similarity, forcing harder negatives when textual descriptions overlap and yielding stronger discrimination. Caption-Aware CIM-T loss (Cross-Identity Inter-modal Margin with Text) simultaneously enlarges inter-identity gaps and contracts intra-identity distances across RGB-IR-TI views, guided by caption context to resolve ambiguous appearances. Our method attains 79.4 mAP and 97.5 R-5 on the Market1501-MM dataset, representing improvements of +1.4 mAP and +0.7 R-5 over prior SOTA approaches. Extensive experiments on MM-ReID demonstrate superior generalization and adaptability across unseen modalities and domains. Our approach establishes a new paradigm for modality-extensible and interpretable multi-modal ReID research.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 4","pages":"Article 104589"},"PeriodicalIF":6.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Modality augmentation and task-aware dual-modal LoRAs for multi-task multimodal federated learning 面向多任务多模态联邦学习的模态增强和任务感知双模态lora

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2026-01-09 DOI: 10.1016/j.ipm.2025.104601

Yushi Zeng , Haopeng Ren , Yi Cai , Yingjian Li , Jing Qin , Yaowei Wang

Multimodal Federated Learning (MFL) is a decentralized machine learning paradigm designed to integrate knowledge from clients with diverse modalities into a global model without compromising privacy. Existing MFL methods suffer from two critical limitations: modality bias and task incompatibility. These two limitations stem from modality inconsistency and task heterogeneity among clients, which lead to degraded performance of the global server model on client-specific tasks. To tackle these problems, we introduce a multi-task compatible framework with a modality augmentation(MA) and a task-aware selective local feature aggregation (TA-SLFA). The designed MA and TA-SLFA modules respectively aim to tackle the modality bias and alleviate the task heterogeneity in MFL. Moreover, the task-aware dual-modal Low-Rank Adaptations (LoRAs) are integrated into a vision-language model, enhancing its ability to integrate task-specific features and improve multi-task learning ability. Extensive experiments and ablation analysis are conducted on four common public datasets and the experimental results demonstrate that our proposed model achieves significant improvements in multitask multimodal federated learning.

多模式联邦学习（Multimodal Federated Learning， MFL）是一种分散的机器学习范式，旨在将来自不同模式的客户的知识集成到一个全球模型中，而不会损害隐私。现有的MFL方法存在模态偏差和任务不兼容两大缺陷。这两个限制源于客户机之间的模态不一致和任务异构性，这会导致全局服务器模型在特定于客户机的任务上的性能下降。为了解决这些问题，我们引入了一个具有模态增强（MA）和任务感知选择性局部特征聚合（TA-SLFA）的多任务兼容框架。设计的MA和TA-SLFA模块分别旨在解决多语言教学中的模态偏差和减轻任务异质性。此外，将任务感知双模低秩自适应（LoRAs）集成到视觉语言模型中，增强了其集成特定任务特征的能力，提高了多任务学习能力。在四个常见的公共数据集上进行了大量的实验和分析，实验结果表明，我们提出的模型在多任务多模态联邦学习方面取得了显著的进步。

{"title":"Modality augmentation and task-aware dual-modal LoRAs for multi-task multimodal federated learning","authors":"Yushi Zeng , Haopeng Ren , Yi Cai , Yingjian Li , Jing Qin , Yaowei Wang","doi":"10.1016/j.ipm.2025.104601","DOIUrl":"10.1016/j.ipm.2025.104601","url":null,"abstract":"<div><div>Multimodal Federated Learning (MFL) is a decentralized machine learning paradigm designed to integrate knowledge from clients with diverse modalities into a global model without compromising privacy. Existing MFL methods suffer from two critical limitations: modality bias and task incompatibility. These two limitations stem from modality inconsistency and task heterogeneity among clients, which lead to degraded performance of the global server model on client-specific tasks. To tackle these problems, we introduce a multi-task compatible framework with a modality augmentation(MA) and a task-aware selective local feature aggregation (TA-SLFA). The designed MA and TA-SLFA modules respectively aim to tackle the modality bias and alleviate the task heterogeneity in MFL. Moreover, the task-aware dual-modal Low-Rank Adaptations (LoRAs) are integrated into a vision-language model, enhancing its ability to integrate task-specific features and improve multi-task learning ability. Extensive experiments and ablation analysis are conducted on four common public datasets and the experimental results demonstrate that our proposed model achieves significant improvements in multitask multimodal federated learning.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 4","pages":"Article 104601"},"PeriodicalIF":6.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145927681","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Volatility-aware sample re-weighting framework for short-term photovoltaic power forecasting 短期光伏发电预测的波动感知样本重加权框架

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2026-01-12 DOI: 10.1016/j.ipm.2026.104612

Wentao Wang , Haiyan Lu , Ayesha Ubaid , Fanyi Yang , Yan Zhao , Jianzhou Wang

Recent short-term photovoltaic (PV) power forecasting methods have primarily focused on improving model architectures to enhance forecasting accuracy, they often overlook the issue of weather-type imbalance in PV power datasets. To this end, we first introduce a new metric, Mean Accumulated Volatility (MAV), which quantifies the volatility of each sample. By translating unquantified weather-type imbalance into a measurable form of volatility imbalance, we observe that high-MAV samples account for most of the training loss, thereby harming the model’s forecasting accuracy. Then, we further propose ReMAV, a volatility-aware Re-weighting framework that down-weights the losses of high-MAV samples and up-weights those of low-MAV samples based on the MAV-based density. Extensive experiments on eleven baseline forecasting models across three real-world PV power datasets demonstrate that our proposed ReMAV framework effectively handles PV power with weather-type imbalance and consistently outperforms existing baseline models in forecasting accuracy. For example, on the Alice Springs dataset, ReMAV reduces average MAE by 8.53% over baselines, while on the PVOD dataset, MAE drops by 5.46% on average.

目前的短期光伏发电功率预测方法主要侧重于改进模型架构以提高预测精度，但往往忽略了光伏发电数据集的天气类型不平衡问题。为此，我们首先引入了一个新的度量，平均累积波动率（MAV），它量化了每个样本的波动率。通过将不可量化的天气类型失衡转化为可测量的波动率失衡，我们观察到高mav样本占了大部分训练损失，从而损害了模型的预测精度。然后，我们进一步提出了ReMAV，这是一个波动感知的重新加权框架，它根据基于mavv的密度降低高mav样本的损失权重，提高低mav样本的损失权重。在三个真实光伏发电数据集上的11个基线预测模型上进行的大量实验表明，我们提出的ReMAV框架有效地处理了天气类型不平衡的光伏发电，并且在预测精度上始终优于现有的基线模型。例如，在Alice Springs数据集上，ReMAV使平均MAE比基线降低了8.53%，而在PVOD数据集上，MAE平均下降了5.46%。

{"title":"Volatility-aware sample re-weighting framework for short-term photovoltaic power forecasting","authors":"Wentao Wang , Haiyan Lu , Ayesha Ubaid , Fanyi Yang , Yan Zhao , Jianzhou Wang","doi":"10.1016/j.ipm.2026.104612","DOIUrl":"10.1016/j.ipm.2026.104612","url":null,"abstract":"<div><div>Recent short-term photovoltaic (PV) power forecasting methods have primarily focused on improving model architectures to enhance forecasting accuracy, they often overlook the issue of weather-type imbalance in PV power datasets. To this end, we first introduce a new metric, <strong>Mean Accumulated Volatility (MAV)</strong>, which quantifies the volatility of each sample. By translating unquantified weather-type imbalance into a measurable form of volatility imbalance, we observe that high-MAV samples account for most of the training loss, thereby harming the model’s forecasting accuracy. Then, we further propose <strong>ReMAV</strong>, a volatility-aware <strong>Re</strong>-weighting framework that down-weights the losses of high-MAV samples and up-weights those of low-MAV samples based on the <strong>MAV</strong>-based density. Extensive experiments on eleven baseline forecasting models across three real-world PV power datasets demonstrate that our proposed ReMAV framework effectively handles PV power with weather-type imbalance and consistently outperforms existing baseline models in forecasting accuracy. For example, on the Alice Springs dataset, ReMAV reduces average MAE by 8.53% over baselines, while on the PVOD dataset, MAE drops by 5.46% on average.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 4","pages":"Article 104612"},"PeriodicalIF":6.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978420","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multi-view dynamic perception framework for Chinese harmful meme detection 中文有害模因检测的多视角动态感知框架

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2026-01-15 DOI: 10.1016/j.ipm.2025.104602

Jiapei Hu , Yifan Lyu , Yifan Chen , Kuntao Li , Yun Xue , Jinghua Liang

Chinese harmful memes convey toxicity on social media through varied semantic complexity and diverse modality combinations. However, existing detection methods typically adopt static architectures with fixed interaction patterns, which lack the flexibility to accurately identify harmful cues embedded in heterogeneous semantic and modal content across different memes, hindering a comprehensive understanding of toxic intent. To address this limitation, we propose the Multi-View Dynamic Perception (MDP) framework, a dynamic interaction paradigm specifically designed for Chinese harmful meme detection. Specifically, we develop five types of semantic perception nodes to synchronously extract features from diverse views. These nodes are densely stacked to form two perception branches, respectively guided by textual and visual features, to effectively capture modality-specific cues. To enhance adaptability, each node is equipped with an independent soft router that dynamically regulates information flow and enables flexible interaction patterns tailored to different memes. Furthermore, we introduce a Hierarchical Mutual Learning module to promote complementary representation learning between the two branches via mutual information maximization. Extensive experiments on the publicly available dataset TOXICN MM, comprising 12,000 samples, demonstrate the effectiveness of the proposed framework, with F1 score improvements of 1.06% in harmful meme detection and 2.77% in harmful type identification over the previous state-of-the-art method. We further evaluate the generalization of the MDP framework on a Chinese multimodal sarcasm detection dataset, where the proposed method also achieves competitive results.

中文有害模因通过不同的语义复杂性和不同的情态组合在社交媒体上传递毒性。然而，现有的检测方法通常采用具有固定交互模式的静态架构，缺乏灵活性，无法准确识别跨不同模因的异构语义和模态内容中嵌入的有害线索，阻碍了对有毒意图的全面理解。为了解决这一限制，我们提出了多视图动态感知（MDP）框架，这是一个专门为中文有害模因检测设计的动态交互范式。具体来说，我们开发了五种类型的语义感知节点来同步提取不同视图的特征。这些节点密集堆叠形成两个感知分支，分别由文本和视觉特征引导，以有效捕获特定于模态的线索。为了增强适应性，每个节点都配备了独立的软路由器，动态调节信息流，实现针对不同模因的灵活交互模式。此外，我们引入了一个分层互学习模块，通过互信息最大化来促进两个分支之间的互补表示学习。在公开数据集TOXICN MM上进行的大量实验，包括12,000个样本，证明了所提出框架的有效性，与之前最先进的方法相比，有害模因检测的F1分数提高了1.06%，有害类型识别的F1分数提高了2.77%。我们进一步评估了MDP框架在中文多模态讽刺检测数据集上的泛化效果，该方法也取得了具有竞争力的结果。

{"title":"Multi-view dynamic perception framework for Chinese harmful meme detection","authors":"Jiapei Hu , Yifan Lyu , Yifan Chen , Kuntao Li , Yun Xue , Jinghua Liang","doi":"10.1016/j.ipm.2025.104602","DOIUrl":"10.1016/j.ipm.2025.104602","url":null,"abstract":"<div><div>Chinese harmful memes convey toxicity on social media through varied semantic complexity and diverse modality combinations. However, existing detection methods typically adopt static architectures with fixed interaction patterns, which lack the flexibility to accurately identify harmful cues embedded in heterogeneous semantic and modal content across different memes, hindering a comprehensive understanding of toxic intent. To address this limitation, we propose the Multi-View Dynamic Perception (MDP) framework, a dynamic interaction paradigm specifically designed for Chinese harmful meme detection. Specifically, we develop five types of semantic perception nodes to synchronously extract features from diverse views. These nodes are densely stacked to form two perception branches, respectively guided by textual and visual features, to effectively capture modality-specific cues. To enhance adaptability, each node is equipped with an independent soft router that dynamically regulates information flow and enables flexible interaction patterns tailored to different memes. Furthermore, we introduce a Hierarchical Mutual Learning module to promote complementary representation learning between the two branches via mutual information maximization. Extensive experiments on the publicly available dataset TOXICN MM, comprising 12,000 samples, demonstrate the effectiveness of the proposed framework, with F1 score improvements of 1.06% in harmful meme detection and 2.77% in harmful type identification over the previous state-of-the-art method. We further evaluate the generalization of the MDP framework on a Chinese multimodal sarcasm detection dataset, where the proposed method also achieves competitive results.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 4","pages":"Article 104602"},"PeriodicalIF":6.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145978435","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

TaNSP: An efficient target pattern mining algorithm based on negative sequential pattern TaNSP：一种基于负序模式的高效目标模式挖掘算法

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2026-01-22 DOI: 10.1016/j.ipm.2026.104643

Xiaowen Cui , Xue Dong , Ping Qiu , Chuanhou Sun , Yuhai Zhao , Wenpeng Lu , Xiangjun Dong

Target pattern mining (TPM) aims to return sets of target patterns related to a user-queried target sequence. However, existing TPM research is confined to positive sequential patterns, overlooking negative sequential patterns, which results in limited decision support capabilities. Moreover, introducing negative sequential patterns faces challenges of low mining efficiency and limited pruning techniques. To address these issues, we propose an efficient target pattern mining algorithm based on negative sequential pattern, called TaNSP, to achieve TPM for negative sequence as the user-queried target sequence and output negative sequential patterns containing the target query sequence, while also supporting positive sequential patterns. Specifically, we propose a pruning strategy based on a triple bitmap to guide pattern generation and improve mining efficiency. Then, we propose a pruning strategy to address the limitations of pruning techniques when the negative sequence is the target query sequence. The experimental results on six datasets demonstrate that, compared to the baseline method, TaNSP can increase operational efficiency by more than twice, demonstrating excellent scalability and practicality.

目标模式挖掘（TPM）旨在返回与用户查询的目标序列相关的目标模式集。然而，现有的TPM研究仅限于积极的顺序模式，忽视了消极的顺序模式，导致决策支持能力有限。此外，引入负序模式还面临着挖掘效率低和修剪技术有限的挑战。为了解决这些问题，我们提出了一种高效的基于负序列模式的目标模式挖掘算法TaNSP，将负序列作为用户查询的目标序列实现TPM，输出包含目标查询序列的负序列模式，同时也支持正序列模式。具体来说，我们提出了一种基于三重位图的剪枝策略来指导模式生成，提高挖掘效率。然后，我们提出了一种剪枝策略，以解决当负序列是目标查询序列时剪枝技术的局限性。在6个数据集上的实验结果表明，与基线方法相比，TaNSP的运算效率提高了2倍以上，具有良好的可扩展性和实用性。

{"title":"TaNSP: An efficient target pattern mining algorithm based on negative sequential pattern","authors":"Xiaowen Cui , Xue Dong , Ping Qiu , Chuanhou Sun , Yuhai Zhao , Wenpeng Lu , Xiangjun Dong","doi":"10.1016/j.ipm.2026.104643","DOIUrl":"10.1016/j.ipm.2026.104643","url":null,"abstract":"<div><div>Target pattern mining (TPM) aims to return sets of target patterns related to a user-queried target sequence. However, existing TPM research is confined to positive sequential patterns, overlooking negative sequential patterns, which results in limited decision support capabilities. Moreover, introducing negative sequential patterns faces challenges of low mining efficiency and limited pruning techniques. To address these issues, we propose an efficient <u>ta</u>rget pattern mining algorithm based on <u>n</u>egative <u>s</u>equential <u>p</u>attern, called TaNSP, to achieve TPM for negative sequence as the user-queried target sequence and output negative sequential patterns containing the target query sequence, while also supporting positive sequential patterns. Specifically, we propose a pruning strategy based on a triple bitmap to guide pattern generation and improve mining efficiency. Then, we propose a pruning strategy to address the limitations of pruning techniques when the negative sequence is the target query sequence. The experimental results on six datasets demonstrate that, compared to the baseline method, TaNSP can increase operational efficiency by more than twice, demonstrating excellent scalability and practicality.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 4","pages":"Article 104643"},"PeriodicalIF":6.9,"publicationDate":"2026-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146023309","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Graph attention convolutional networks for interpretable multi-hop knowledge graph reasoning 可解释多跳知识图推理的图注意卷积网络

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-06-01 Epub Date: 2025-12-30 DOI: 10.1016/j.ipm.2025.104581

Hao Liu , Dong Li , Bing Zeng , Haopeng Ren

Effective multi-hop reasoning over knowledge graphs is critical for knowledge completion, yet prior methods often struggle to model relation dependencies and capture neighborhood context interactions, thereby limiting path interpretability and predictive performance. To adequately model the interaction between neighborhood information and context, we introduce a graph attention convolutional (GAC) mechanism that aggregates and updates node information within the local first-order neighborhood. We then employ attention mechanisms to generate entity and relation reasoning contexts and construct GAC-based policy networks to reinforce interaction between these contexts and their corresponding neighborhoods. Extensive experiments on five knowledge graphs demonstrate the effectiveness of our method, which achieves notable improvements on FB15K-237, including a 7.6 % relative improvement in Hits@1, a 14.6 % increase in MRR, and a 6.9 % enhancement in path interpretability.

知识图上有效的多跳推理对于知识完成至关重要，然而先前的方法往往难以建立关系依赖关系模型和捕获邻近上下文交互，从而限制了路径的可解释性和预测性能。为了充分模拟邻域信息和上下文之间的相互作用，我们引入了一种图注意卷积（GAC）机制，该机制聚合和更新局部一阶邻域内的节点信息。然后，我们使用注意机制来生成实体和关系推理上下文，并构建基于gac的策略网络来加强这些上下文与其相应邻域之间的交互。在五个知识图上的大量实验证明了我们的方法的有效性，该方法在FB15K-237上取得了显著的改进，其中Hits@1的相对改进为7.6%，MRR提高了14.6%，路径可解释性提高了6.9%。

引用次数: 0

DeepMEL: A multi-agent collaboration framework for multimodal entity linking DeepMEL：用于多模态实体链接的多代理协作框架

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-04-01 Epub Date: 2025-12-09 DOI: 10.1016/j.ipm.2025.104507

Fang Wang , Tianwei Yan , Zonghao Yang , Minghao Hu , Jun Zhang , Zhunchen Luo , Xiaoying Bai

Multimodal Entity Linking (MEL) aims to associate textual and visual mentions with entities in a multimodal knowledge graph. Despite its importance, current methods face challenges such as incomplete contextual information, coarse cross-modal fusion, and the difficulty of jointly large language models (LLMs) and large visual models (LVMs). To address these issues, we propose DeepMEL, a novel framework based on multi-agent collaborative reasoning, which achieves efficient alignment and disambiguation of textual and visual modalities through a role-specialized division strategy. DeepMEL integrates four specialized agents, namely Modal-Fuser, Candidate-Adapter, Entity-Clozer and Role-Orchestrator, to complete end-to-end cross-modal linking through specialized roles and dynamic coordination. DeepMEL adopts a dual-modal alignment path, and combines the fine-grained text semantics generated by the LLM with the structured image representation extracted by the LVM, significantly narrowing the modal gap. We design an adaptive iteration strategy, combines tool-based retrieval and semantic reasoning capabilities to dynamically optimize the candidate set and balance recall and precision. DeepMEL also unifies MEL tasks into a structured cloze prompt to reduce parsing complexity and enhance semantic comprehension. Extensive experiments on five public benchmark datasets demonstrate that DeepMEL achieves state-of-the-art performance, improving ACC by 1 %-57 %. Ablation studies verify the effectiveness of all modules.

多模态实体链接（MEL）旨在将文本和视觉提及与多模态知识图中的实体关联起来。尽管其重要性，但目前的方法面临着诸如上下文信息不完整、跨模态融合粗糙以及大型语言模型（llm）和大型视觉模型（lvm）难以联合的挑战。为了解决这些问题，我们提出了一种基于多智能体协作推理的新框架DeepMEL，该框架通过角色专门化划分策略实现了文本和视觉模式的有效对齐和消歧。DeepMEL集成了Modal-Fuser、Candidate-Adapter、Entity-Clozer和Role-Orchestrator四个专门的agent，通过专门的角色和动态协调完成端到端的跨模态链接。DeepMEL采用双模态对齐路径，将LLM生成的细粒度文本语义与LVM提取的结构化图像表示相结合，显著缩小了模态差距。我们设计了一种自适应迭代策略，结合基于工具的检索和语义推理能力来动态优化候选集，平衡召回率和准确率。DeepMEL还将MEL任务统一为结构化的完形提示，以降低解析复杂性并增强语义理解。在五个公共基准数据集上进行的大量实验表明，DeepMEL达到了最先进的性能，将ACC提高了1% - 57%。烧蚀研究验证了所有模块的有效性。

{"title":"DeepMEL: A multi-agent collaboration framework for multimodal entity linking","authors":"Fang Wang , Tianwei Yan , Zonghao Yang , Minghao Hu , Jun Zhang , Zhunchen Luo , Xiaoying Bai","doi":"10.1016/j.ipm.2025.104507","DOIUrl":"10.1016/j.ipm.2025.104507","url":null,"abstract":"<div><div>Multimodal Entity Linking (MEL) aims to associate textual and visual mentions with entities in a multimodal knowledge graph. Despite its importance, current methods face challenges such as incomplete contextual information, coarse cross-modal fusion, and the difficulty of jointly large language models (LLMs) and large visual models (LVMs). To address these issues, we propose DeepMEL, a novel framework based on multi-agent collaborative reasoning, which achieves efficient alignment and disambiguation of textual and visual modalities through a role-specialized division strategy. DeepMEL integrates four specialized agents, namely Modal-Fuser, Candidate-Adapter, Entity-Clozer and Role-Orchestrator, to complete end-to-end cross-modal linking through specialized roles and dynamic coordination. DeepMEL adopts a dual-modal alignment path, and combines the fine-grained text semantics generated by the LLM with the structured image representation extracted by the LVM, significantly narrowing the modal gap. We design an adaptive iteration strategy, combines tool-based retrieval and semantic reasoning capabilities to dynamically optimize the candidate set and balance recall and precision. DeepMEL also unifies MEL tasks into a structured cloze prompt to reduce parsing complexity and enhance semantic comprehension. Extensive experiments on five public benchmark datasets demonstrate that DeepMEL achieves state-of-the-art performance, improving ACC by 1 %-57 %. Ablation studies verify the effectiveness of all modules.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104507"},"PeriodicalIF":6.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737354","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Meta-path Sampling-Enhanced Course Recommendation in Heterogeneous Networks 异构网络中元路径抽样增强的课程推荐

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-04-01 Epub Date: 2025-11-20 DOI: 10.1016/j.ipm.2025.104482

Lei Zhang , Mengxiang Ma , Juntao Zhang , Tao Xu , Daojun Han , Linkun Fan , Yanhua Zhao

With the successful development and application of Massive Open Online Courses (MOOCs), course recommendations have received widespread attention from researchers. However, existing course recommendation methods face three challenges: (1) user-course interaction is sparse; (2) insufficient modeling of the multiple interactive semantics of user preferences; and (3) the lack of constraints for user knowledge blind zone preferences. To address these challenges, we propose a novel method called Meta-path Sampling-Enhanced Course Recommendation in Heterogeneous Networks (MSEC-Rec), which improves the accuracy of course recommendations by integrating the multi-interaction semantic information of users. Specifically, we enhance the interactions between users and courses through meta-paths in heterogeneous information networks (HINs) to alleviate the interaction sparsity problem. Then, we design a meta-path sampling strategy to model the semantics of multiple interactions between users and courses. Next, we introduce meta-path negative sampling information in HINs and capture users’ knowledge blindness via the contrastive loss function to optimize the score differences between positive and negative samples. Finally, we conduct experiments on the MOOCCube and XuetangX datasets and compare MSEC-Rec with multiple baselines. Compared with the SOTA method on the MOOCCube dataset, the evaluation metrics HR@K and NDCG@K (K= 5, 10, 20) of MSEC-Rec increased by 0.04%, 3.35%, 5.17%, 2.61%, 4.69%, and 4.2%, respectively, demonstrating its effectiveness. The source code and data are available on GitHub: https://github.com/mmx124/MSEC-Rec.

随着大规模在线开放课程（MOOCs）的成功开发和应用，课程推荐受到了研究者的广泛关注。然而，现有的课程推荐方法面临三个挑战：(1)用户-课程交互稀疏；(2)对用户偏好的多重交互语义建模不足；(3)缺乏对用户知识盲区偏好的约束。为了解决这些挑战，我们提出了一种新的方法，称为元路径采样增强的异构网络课程推荐（MSEC-Rec），该方法通过集成用户的多交互语义信息来提高课程推荐的准确性。具体而言，我们通过异构信息网络（HINs）中的元路径增强用户与课程之间的交互，以缓解交互稀疏问题。然后，我们设计了一个元路径采样策略，对用户和球场之间的多个交互语义进行建模。接下来，我们在HINs中引入元路径负样本信息，通过对比损失函数捕捉用户的知识盲目性，优化正负样本的得分差。最后，我们在MOOCCube和XuetangX数据集上进行了实验，并将MSEC-Rec与多个基线进行了比较。与MOOCCube数据集上的SOTA方法相比，MSEC-Rec的评价指标HR@K和NDCG@K （K= 5、10、20）分别提高了0.04%、3.35%、5.17%、2.61%、4.69%和4.2%，表明其有效性。源代码和数据可在GitHub: https://github.com/mmx124/MSEC-Rec。

{"title":"Meta-path Sampling-Enhanced Course Recommendation in Heterogeneous Networks","authors":"Lei Zhang , Mengxiang Ma , Juntao Zhang , Tao Xu , Daojun Han , Linkun Fan , Yanhua Zhao","doi":"10.1016/j.ipm.2025.104482","DOIUrl":"10.1016/j.ipm.2025.104482","url":null,"abstract":"<div><div>With the successful development and application of Massive Open Online Courses (MOOCs), course recommendations have received widespread attention from researchers. However, existing course recommendation methods face three challenges: (1) user-course interaction is sparse; (2) insufficient modeling of the multiple interactive semantics of user preferences; and (3) the lack of constraints for user knowledge blind zone preferences. To address these challenges, we propose a novel method called Meta-path Sampling-Enhanced Course Recommendation in Heterogeneous Networks (MSEC-Rec), which improves the accuracy of course recommendations by integrating the multi-interaction semantic information of users. Specifically, we enhance the interactions between users and courses through meta-paths in heterogeneous information networks (HINs) to alleviate the interaction sparsity problem. Then, we design a meta-path sampling strategy to model the semantics of multiple interactions between users and courses. Next, we introduce meta-path negative sampling information in HINs and capture users’ knowledge blindness via the contrastive loss function to optimize the score differences between positive and negative samples. Finally, we conduct experiments on the MOOCCube and XuetangX datasets and compare MSEC-Rec with multiple baselines. Compared with the SOTA method on the MOOCCube dataset, the evaluation metrics HR@K and NDCG@K (K= 5, 10, 20) of MSEC-Rec increased by 0.04%, 3.35%, 5.17%, 2.61%, 4.69%, and 4.2%, respectively, demonstrating its effectiveness. The source code and data are available on GitHub: <span><span>https://github.com/mmx124/MSEC-Rec</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104482"},"PeriodicalIF":6.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145555231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

MPRG:A unified framework for knowledge graph reasoning via pattern-aware relation graph MPRG：基于模式感知关系图的知识图推理的统一框架

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-04-01 Epub Date: 2025-12-23 DOI: 10.1016/j.ipm.2025.104570

Zhiwen Xie , Meiling Wang , Po Hu

Knowledge graph reasoning (KGR) aims to infer missing links in knowledge graphs (KGs). Traditional methods in transductive settings suffer from poor generalizability to unseen entities and relations. Recently, inductive paradigms have emerged to address this limitation by learning a unified reasoning model transferable across KGs. However, current inductive approaches typically rely on relation co-occurrence interaction, which fails to capture richer relation patterns such as symmetry and composition patterns that have been shown to be critical for effective KGR. In this paper, we propose a Multi-Pattern Relational Graph (MPRG) model, a novel inductive reasoning framework that explicitly models complex relational patterns in KGs. Our method first mines interpretable reasoning rules to extract structural relation patterns, which are then used to construct a pattern-aware relational graph. Build upon this relational graph, a pattern-aware relation representation module is applied to learn expressive relation embeddings through structured message passing. These enriched relation representations are further used to guide entity-level reasoning via a relation-guided entity reasoning module. We evaluate MPRG on 43 KGR benchmark datasets across transductive, inductive, and fully-inductive settings. Experimental results show that MPRG consistently achieves state-of-the-art performance with particularly strong results under the inductive setting. Moreover, MPRG achieves substantial average improvements of 2.1 % in MRR and 2.9 % in Hit@10 across all 43 datasets over existing methods, demonstrating its effectiveness and strong generalization capability across diverse KGR tasks.

知识图推理（Knowledge graph reasoning， KGR）旨在推断知识图中缺失的环节。传统方法在转导设置中存在对不可见实体和关系的泛化能力差的问题。最近出现了归纳范式，通过学习可在KGR之间转移的统一推理模型来解决这一限制。然而，目前的归纳方法通常依赖于关系共现交互作用，无法捕获更丰富的关系模式，如对称性和组合模式，这些模式已被证明对有效的KGR至关重要。本文提出了一个多模式关系图（MPRG）模型，该模型是一种新的归纳推理框架，可以显式地对KGs中的复杂关系模式进行建模。我们的方法首先挖掘可解释的推理规则来提取结构关系模式，然后将其用于构建模式感知的关系图。在此关系图的基础上，应用模式感知关系表示模块，通过结构化消息传递学习表达关系嵌入。这些丰富的关系表示通过关系导向的实体推理模块进一步用于指导实体级推理。我们在43个KGR基准数据集上评估了MPRG，包括传导、感应和全感应设置。实验结果表明，在感应设置下，MPRG始终能够达到最先进的性能，并且效果特别强。此外，与现有方法相比，MPRG在所有43个数据集上的MRR平均提高了2.1%，Hit@10平均提高了2.9%，证明了其在不同KGR任务中的有效性和强大的泛化能力。

{"title":"MPRG:A unified framework for knowledge graph reasoning via pattern-aware relation graph","authors":"Zhiwen Xie , Meiling Wang , Po Hu","doi":"10.1016/j.ipm.2025.104570","DOIUrl":"10.1016/j.ipm.2025.104570","url":null,"abstract":"<div><div>Knowledge graph reasoning (KGR) aims to infer missing links in knowledge graphs (KGs). Traditional methods in transductive settings suffer from poor generalizability to unseen entities and relations. Recently, inductive paradigms have emerged to address this limitation by learning a unified reasoning model transferable across KGs. However, current inductive approaches typically rely on relation co-occurrence interaction, which fails to capture richer relation patterns such as symmetry and composition patterns that have been shown to be critical for effective KGR. In this paper, we propose a Multi-Pattern Relational Graph (MPRG) model, a novel inductive reasoning framework that explicitly models complex relational patterns in KGs. Our method first mines interpretable reasoning rules to extract structural relation patterns, which are then used to construct a pattern-aware relational graph. Build upon this relational graph, a pattern-aware relation representation module is applied to learn expressive relation embeddings through structured message passing. These enriched relation representations are further used to guide entity-level reasoning via a relation-guided entity reasoning module. We evaluate MPRG on 43 KGR benchmark datasets across transductive, inductive, and fully-inductive settings. Experimental results show that MPRG consistently achieves state-of-the-art performance with particularly strong results under the inductive setting. Moreover, MPRG achieves substantial average improvements of 2.1 % in MRR and 2.9 % in Hit@10 across all 43 datasets over existing methods, demonstrating its effectiveness and strong generalization capability across diverse KGR tasks.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104570"},"PeriodicalIF":6.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145839838","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Label acceptance based label propagation algorithm for community detection 基于标签接受度的标签传播社区检测算法

IF 6.9 1区管理学 Q1 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Processing & Management

Pub Date : 2026-04-01 Epub Date: 2025-12-22 DOI: 10.1016/j.ipm.2025.104573

Xunlian Wu , Anqi Zhang , Jingqi Hu , Han Zhang , Yining Quan , Qiguang Miao , Peng Gang Sun

Community detection plays a crucial role in network analysis. While the Label Propagation Algorithm (LPA) is known for its efficiency, it suffers from unstable results due to random label updates and the inability to capture higher-order structural information. To address these limitations, we propose LALPA (Label Acceptance-based Label Propagation Algorithm) for community detection. LALPA introduces a node importance measure based on neighbor similarity to guide a stable, ordered label update process. To better capture structural information, we reconstruct the network topology by integrating both low-order (adjacent links) and high-order (motif-based) interactions, modeling node influence acceptance. Label acceptance is then determined by combining node importance and influence acceptance. A novel propagation strategy is designed to aggregate labels not only from current neighbors but also from those sharing the same label. Extensive experiments on 10 real-world and 24 synthetic networks show that LALPA consistently outperforms state-of-the-art methods, especially in networks with unobvious community structures. In particular, on all unattributed graphs, LALPA achieves an average performance gain of 2.69 % over the best baseline.

社区检测在网络分析中起着至关重要的作用。虽然标签传播算法（LPA）以其效率而闻名，但由于随机标签更新和无法捕获高阶结构信息，它的结果不稳定。为了解决这些限制，我们提出了LALPA（基于标签接受的标签传播算法）用于社区检测。LALPA引入了一个基于邻居相似度的节点重要性度量来指导一个稳定有序的标签更新过程。为了更好地捕获结构信息，我们通过整合低阶（相邻链接）和高阶（基于图案的）交互来重建网络拓扑，建模节点影响接受度。然后通过结合节点重要性和影响接受度来确定标签接受度。设计了一种新的传播策略，不仅可以聚合当前邻居的标签，还可以聚合共享同一标签的标签。在10个真实网络和24个合成网络上进行的大量实验表明，LALPA始终优于最先进的方法，特别是在具有不明显社区结构的网络中。特别是，在所有未归属图上，LALPA在最佳基线上实现了2.69%的平均性能增益。

{"title":"Label acceptance based label propagation algorithm for community detection","authors":"Xunlian Wu , Anqi Zhang , Jingqi Hu , Han Zhang , Yining Quan , Qiguang Miao , Peng Gang Sun","doi":"10.1016/j.ipm.2025.104573","DOIUrl":"10.1016/j.ipm.2025.104573","url":null,"abstract":"<div><div>Community detection plays a crucial role in network analysis. While the Label Propagation Algorithm (LPA) is known for its efficiency, it suffers from unstable results due to random label updates and the inability to capture higher-order structural information. To address these limitations, we propose LALPA (Label Acceptance-based Label Propagation Algorithm) for community detection. LALPA introduces a node importance measure based on neighbor similarity to guide a stable, ordered label update process. To better capture structural information, we reconstruct the network topology by integrating both low-order (adjacent links) and high-order (motif-based) interactions, modeling node influence acceptance. Label acceptance is then determined by combining node importance and influence acceptance. A novel propagation strategy is designed to aggregate labels not only from current neighbors but also from those sharing the same label. Extensive experiments on 10 real-world and 24 synthetic networks show that LALPA consistently outperforms state-of-the-art methods, especially in networks with unobvious community structures. In particular, on all unattributed graphs, LALPA achieves an average performance gain of 2.69 % over the best baseline.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104573"},"PeriodicalIF":6.9,"publicationDate":"2026-04-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145840334","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"管理学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0