Information Processing & Management最新文献_第7页

Deep learning-based hypernetwork dismantling for effectively hindering structural recovery 基于深度学习的超网络拆解，有效阻碍结构恢复

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

Information Processing & Management

Pub Date : 2025-12-17 DOI: 10.1016/j.ipm.2025.104551

Bin Tang , Hanqiang Deng , Yuxian Duan , Heng Zhang , Jian Huang , Jiarui Zhang

Network dismantling aims to disrupt network structure and function by removing the smallest set of nodes. It has been extensively adopted in many real-world applications such as preventing virus propagation and disrupting terrorist communications. Traditional approaches, however, are almost exclusively built on simple graph that capture only pairwise interactions, thereby overlooking the higher-order, group-wise dependencies that are naturally encoded by hypernetworks. In this work, we propose a hypernetwork dismantling framework based on deep learning, which can be trained purely on small synthetic hypernetworks and then applied for various real-world hypernetworks. In this framework, we also design a novel inductive hypergraph attention neural network with two-level aggregated hypergraph attention neural network layers to ensure the generalization and effectiveness of the framework. Besides, we design a novel node labeling strategy explicitly incorporating network resilience, ensuring the learned optimal hypernetwork dismantling strategy inflicts enduring structural damage, hindering recovery. Extensive experiments on two types of large synthetic and 9 real-world hypernetworks demonstrate that our framework significantly outperforms the state-of-the-art methods. Specifically, our framework achieves an overall performance improvement of 17 % on synthetic hypernetworks and 20 % on real-world hypernetworks. In summary, it achieves superior disruption with fewer node removals and delivers more persistent damage, fundamentally impairing system resilience.

网络拆除的目的是通过移除最小的节点集来破坏网络的结构和功能。它已广泛应用于许多现实世界的应用，如防止病毒传播和破坏恐怖主义通信。然而，传统的方法几乎完全建立在简单的图上，只捕获成对的交互，从而忽略了由超网络自然编码的高阶、组智能依赖关系。在这项工作中，我们提出了一个基于深度学习的超网络拆解框架，该框架可以纯粹在小型合成超网络上进行训练，然后应用于各种现实世界的超网络。在此框架中，我们还设计了一种新型的归纳超图注意神经网络，该网络具有两层聚合的超图注意神经网络层，以保证框架的泛化和有效性。此外，我们设计了一种明确结合网络弹性的新颖节点标记策略，确保学习到的最优超网络拆除策略造成持久的结构破坏，阻碍恢复。在两种类型的大型合成网络和9个真实世界的超网络上进行的大量实验表明，我们的框架明显优于最先进的方法。具体来说，我们的框架在合成超网络上实现了17%的整体性能提升，在真实世界的超网络上实现了20%的整体性能提升。总而言之，它以更少的节点移除实现了更好的中断，并带来了更持久的破坏，从根本上削弱了系统的弹性。

{"title":"Deep learning-based hypernetwork dismantling for effectively hindering structural recovery","authors":"Bin Tang , Hanqiang Deng , Yuxian Duan , Heng Zhang , Jian Huang , Jiarui Zhang","doi":"10.1016/j.ipm.2025.104551","DOIUrl":"10.1016/j.ipm.2025.104551","url":null,"abstract":"<div><div>Network dismantling aims to disrupt network structure and function by removing the smallest set of nodes. It has been extensively adopted in many real-world applications such as preventing virus propagation and disrupting terrorist communications. Traditional approaches, however, are almost exclusively built on simple graph that capture only pairwise interactions, thereby overlooking the higher-order, group-wise dependencies that are naturally encoded by hypernetworks. In this work, we propose a hypernetwork dismantling framework based on deep learning, which can be trained purely on small synthetic hypernetworks and then applied for various real-world hypernetworks. In this framework, we also design a novel inductive hypergraph attention neural network with two-level aggregated hypergraph attention neural network layers to ensure the generalization and effectiveness of the framework. Besides, we design a novel node labeling strategy explicitly incorporating network resilience, ensuring the learned optimal hypernetwork dismantling strategy inflicts enduring structural damage, hindering recovery. Extensive experiments on two types of large synthetic and 9 real-world hypernetworks demonstrate that our framework significantly outperforms the state-of-the-art methods. Specifically, our framework achieves an overall performance improvement of 17 % on synthetic hypernetworks and 20 % on real-world hypernetworks. In summary, it achieves superior disruption with fewer node removals and delivers more persistent damage, fundamentally impairing system resilience.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104551"},"PeriodicalIF":6.9,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790405","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

KGNS: Knowledge graph-driven neighbor selection for long-tail recommendations KGNS：知识图驱动的长尾推荐邻居选择

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

Information Processing & Management

Pub Date : 2025-12-17 DOI: 10.1016/j.ipm.2025.104561

Zhipeng Zhang , Yao Zhang , Yonggong Ren

Long-tail recommendations face persistent challenges due to popularity bias in user interaction data and sparse interactions for long-tail items. Existing knowledge graph (KG)-based approaches often amplify these issues through indiscriminate neighborhood selection, leading to biased user representations and noisy item embeddings, which ultimately result in a trade-off between accuracy and diversity. To address these limitations, we propose KGNS, a KG-driven neighbor selection approach that strategically reconstructs neighborhoods for both users and items by leveraging rich KG semantics. On the user side, KGNS employs a long-tail neighbor selector to identify semantically relevant long-tail items, reconstructing user neighborhoods to mitigate popularity bias and better capture genuine long-tail interests. On the item side, a co-occurrence neighbor selector enhances long-tail item embeddings by introducing high-quality, semantically correlated neighbors without introducing noise. Through multi-task training, KGNS optimizes the model to recommend top N items that balance both mainstream and long-tail preferences. Extensive experiments on three real-world datasets demonstrate that KGNS not only enhances long-tail recommendation performance but also maintains high overall accuracy, achieving an average improvement of 6.75 % in accuracy and 3.85 % in diversity over state-of-the-art baselines. The code is available at: https://github.com/ZZP-RS/KGNS.

由于用户交互数据中的流行度偏差和长尾项目的稀疏交互，长尾推荐面临持续的挑战。现有的基于知识图（KG）的方法往往通过不加区分的邻域选择放大这些问题，导致有偏见的用户表示和有噪声的项目嵌入，最终导致准确性和多样性之间的权衡。为了解决这些限制，我们提出了KGNS，这是一种KG驱动的邻居选择方法，通过利用丰富的KG语义战略性地重建用户和物品的邻居。在用户端，KGNS使用长尾邻居选择器来识别语义相关的长尾条目，重构用户社区以减轻流行偏差，更好地捕获真正的长尾兴趣。在项目方面，共现邻居选择器通过引入高质量、语义相关的邻居而不引入噪声来增强长尾项目嵌入。通过多任务训练，KGNS对模型进行优化，推荐出平衡主流偏好和长尾偏好的top N项。在三个真实数据集上进行的大量实验表明，KGNS不仅增强了长尾推荐性能，而且保持了较高的整体准确性，在最先进的基线上实现了6.75%的准确率和3.85%的多样性平均提高。代码可从https://github.com/ZZP-RS/KGNS获得。

{"title":"KGNS: Knowledge graph-driven neighbor selection for long-tail recommendations","authors":"Zhipeng Zhang , Yao Zhang , Yonggong Ren","doi":"10.1016/j.ipm.2025.104561","DOIUrl":"10.1016/j.ipm.2025.104561","url":null,"abstract":"<div><div>Long-tail recommendations face persistent challenges due to popularity bias in user interaction data and sparse interactions for long-tail items. Existing knowledge graph (KG)-based approaches often amplify these issues through indiscriminate neighborhood selection, leading to biased user representations and noisy item embeddings, which ultimately result in a trade-off between accuracy and diversity. To address these limitations, we propose KGNS, a KG-driven neighbor selection approach that strategically reconstructs neighborhoods for both users and items by leveraging rich KG semantics. On the user side, KGNS employs a long-tail neighbor selector to identify semantically relevant long-tail items, reconstructing user neighborhoods to mitigate popularity bias and better capture genuine long-tail interests. On the item side, a co-occurrence neighbor selector enhances long-tail item embeddings by introducing high-quality, semantically correlated neighbors without introducing noise. Through multi-task training, KGNS optimizes the model to recommend top <em>N</em> items that balance both mainstream and long-tail preferences. Extensive experiments on three real-world datasets demonstrate that KGNS not only enhances long-tail recommendation performance but also maintains high overall accuracy, achieving an average improvement of 6.75 % in accuracy and 3.85 % in diversity over state-of-the-art baselines. The code is available at: <span><span>https://github.com/ZZP-RS/KGNS</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104561"},"PeriodicalIF":6.9,"publicationDate":"2025-12-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790406","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

Improving human-machine collaborative event detection in chinese texts by pursuing high recall 追求高查全率，改进中文文本人机协同事件检测

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

Information Processing & Management

Pub Date : 2025-12-16 DOI: 10.1016/j.ipm.2025.104541

Jiashun Duan , Yan Pan , Wei Wu , Fangfang Li , Xiang Zhao , Xin Zhang

Existing event detection models cannot meet the demands of applications that require high-quality results, where human-machine collaboration is often needed to ensure quality, but this approach comes with low efficiency. To address this, we introduce the event trigger recommendation task, aiming to ensure high recall so that humans rarely need to re-read and check the text, thereby improving efficiency. And then, we propose SpETR, a Span-based Event Trigger Recommendation model that recommends candidate triggers by calculating the two-stage joint confidence of trigger identification and classification. At last, in terms of model training, a semantic-based boundary smoothing method is proposed, along with improved focal loss, to enhance the model’s recall rate. The experimental results show that SpETR achieves high recall rates, reaching 97.76 % and 98.69 % on the DuEE and Self-built datasets, respectively, when recommending 3 candidates, with improvements of 4.30 % and 5.21 % compared to the optimal baseline. Additionally, employing the SpETR for human-machine collaborative event detection achieves up to 30.6 % time saving, which significantly improves efficiency.

现有的事件检测模型不能满足需要高质量结果的应用程序的需求，这些应用程序通常需要人机协作来保证质量，但是这种方法的效率很低。为了解决这个问题，我们引入了事件触发推荐任务，旨在确保高召回率，以便人类很少需要重新阅读和检查文本，从而提高效率。然后，我们提出了基于跨度的事件触发器推荐模型SpETR，该模型通过计算触发器识别和分类的两阶段联合置信度来推荐候选触发器。最后，在模型训练方面，提出了一种基于语义的边界平滑方法，并改进了焦点损失，以提高模型的召回率。实验结果表明，在推荐3个候选词时，SpETR在DuEE和Self-built数据集上取得了较高的召回率，分别达到97.76%和98.69%，比最优基线分别提高了4.30%和5.21%。此外，采用SpETR进行人机协同事件检测，可节省高达30.6%的时间，大大提高了效率。

{"title":"Improving human-machine collaborative event detection in chinese texts by pursuing high recall","authors":"Jiashun Duan , Yan Pan , Wei Wu , Fangfang Li , Xiang Zhao , Xin Zhang","doi":"10.1016/j.ipm.2025.104541","DOIUrl":"10.1016/j.ipm.2025.104541","url":null,"abstract":"<div><div>Existing event detection models cannot meet the demands of applications that require high-quality results, where human-machine collaboration is often needed to ensure quality, but this approach comes with low efficiency. To address this, we introduce the event trigger recommendation task, aiming to ensure high recall so that humans rarely need to re-read and check the text, thereby improving efficiency. And then, we propose SpETR, a <u>Sp</u>an-based <u>E</u>vent <u>T</u>rigger <u>R</u>ecommendation model that recommends candidate triggers by calculating the two-stage joint confidence of trigger identification and classification. At last, in terms of model training, a semantic-based boundary smoothing method is proposed, along with improved focal loss, to enhance the model’s recall rate. The experimental results show that SpETR achieves high recall rates, reaching 97.76 % and 98.69 % on the DuEE and Self-built datasets, respectively, when recommending 3 candidates, with improvements of 4.30 % and 5.21 % compared to the optimal baseline. Additionally, employing the SpETR for human-machine collaborative event detection achieves up to 30.6 % time saving, which significantly improves efficiency.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104541"},"PeriodicalIF":6.9,"publicationDate":"2025-12-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790407","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

Configurational patterns for forecasting customer satisfaction enhancement based on online reviews: A multi-attribute attitude perspective 基于在线评论预测客户满意度增强的配置模式：多属性态度视角

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

Information Processing & Management

Pub Date : 2025-12-15 DOI: 10.1016/j.ipm.2025.104545

Yong Qin , Chaoguang Luo , Yuyan Luo , Eric W.T. Ngai

In the digital age, automatically identifying and analyzing the causal logic in online reviews to enhance customer satisfaction (CS) is crucial for accurate business decision-making. Despite extensive research grounded in multi-attribute attitude theory (MAAT), existing studies often overlook the nonlinear interrelationships among product or service attributes and their joint impact on CS, limiting a comprehensive understanding of satisfaction formation. To address this gap, this study integrates MAAT with complexity theory to propose an online review-driven framework that models configurational patterns for forecasting CS enhancement. Specifically, we utilize the Top2Vec model to identify product or service attributes and employ a recursive neural tensor network algorithm to calculate the affective distribution vectors of reviews. Then, a bagged neural network model based on affective distribution computing is used to assess the effects of each attribute on CS, and determinant attributes are identified. Based on these core attributes, fuzzy-set qualitative comparative analysis (fsQCA) is applied to identify attribute configuration patterns leading to high CS, and interpretive structural modeling (ISM) and cross-impact matrix multiplication applied to classification (MICMAC) analysis are combined to establish the hierarchical structure and action paths of attributes within the configurations. An empirical study using TripAdvisor reviews of sustainable tourism destinations validates the methodology. By integrating these methods, we can understand how each attribute influences CS, both individually and in combination, and uncover the complex pathways driving high CS. Additionally, practical guidance is provided for businesses to formulate precise customer-oriented management strategies.

在数字时代，自动识别和分析在线评论中的因果逻辑以提高客户满意度（CS）对于准确的业务决策至关重要。尽管在多属性态度理论（MAAT）的基础上进行了广泛的研究，但现有的研究往往忽略了产品或服务属性之间的非线性相互关系及其对CS的共同影响，从而限制了对满意度形成的全面理解。为了解决这一差距，本研究将MAAT与复杂性理论相结合，提出了一个在线评论驱动的框架，该框架为预测CS增强建模配置模式。具体来说，我们利用Top2Vec模型来识别产品或服务属性，并采用递归神经张量网络算法来计算评论的情感分布向量。然后，利用基于情感分布计算的袋装神经网络模型评估各属性对CS的影响，识别出决定属性；基于这些核心属性，应用模糊集定性比较分析（fsQCA）识别导致高CS的属性配置模式，并结合解释结构建模（ISM）和应用于分类的交叉影响矩阵乘法（MICMAC）分析，建立属性在配置中的层次结构和作用路径。一项利用TripAdvisor对可持续旅游目的地的评论进行的实证研究验证了该方法。通过整合这些方法，我们可以了解每个属性是如何影响CS的，无论是单独的还是组合的，并揭示驱动高CS的复杂途径。为企业制定精准的客户导向管理策略提供了实践指导。

{"title":"Configurational patterns for forecasting customer satisfaction enhancement based on online reviews: A multi-attribute attitude perspective","authors":"Yong Qin , Chaoguang Luo , Yuyan Luo , Eric W.T. Ngai","doi":"10.1016/j.ipm.2025.104545","DOIUrl":"10.1016/j.ipm.2025.104545","url":null,"abstract":"<div><div>In the digital age, automatically identifying and analyzing the causal logic in online reviews to enhance customer satisfaction (CS) is crucial for accurate business decision-making. Despite extensive research grounded in multi-attribute attitude theory (MAAT), existing studies often overlook the nonlinear interrelationships among product or service attributes and their joint impact on CS, limiting a comprehensive understanding of satisfaction formation. To address this gap, this study integrates MAAT with complexity theory to propose an online review-driven framework that models configurational patterns for forecasting CS enhancement. Specifically, we utilize the Top2Vec model to identify product or service attributes and employ a recursive neural tensor network algorithm to calculate the affective distribution vectors of reviews. Then, a bagged neural network model based on affective distribution computing is used to assess the effects of each attribute on CS, and determinant attributes are identified. Based on these core attributes, fuzzy-set qualitative comparative analysis (fsQCA) is applied to identify attribute configuration patterns leading to high CS, and interpretive structural modeling (ISM) and cross-impact matrix multiplication applied to classification (MICMAC) analysis are combined to establish the hierarchical structure and action paths of attributes within the configurations. An empirical study using TripAdvisor reviews of sustainable tourism destinations validates the methodology. By integrating these methods, we can understand how each attribute influences CS, both individually and in combination, and uncover the complex pathways driving high CS. Additionally, practical guidance is provided for businesses to formulate precise customer-oriented management strategies.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104545"},"PeriodicalIF":6.9,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790313","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

Privacy protection in RAG: A novel method and evaluation framework RAG中的隐私保护：一种新的方法和评估框架

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

Information Processing & Management

Pub Date : 2025-12-15 DOI: 10.1016/j.ipm.2025.104505

Yuan Zhang , Jionghan Wu , Rui Li , Tong Zhang , Yujie Song , Chuanyi Li , Shangqi Wang , Hao Shen , Jiao Yin , Jidong Ge , Bin Luo

Retrieval-augmented generation (RAG) systems, which enhance large language models with domain-specific data, are increasingly deployed in sensitive domains but remain vulnerable to privacy leakage. Existing privacy safeguards often operate at the document or paragraph level, resulting in a coarse-grained trade-off between privacy and utility that may remove essential context or inadequately protect sensitive information. In this work, we propose a novel privacy-preserving RAG method grounded in knowledge graph representations. Our method introduces a privacy-aware processor that operates at the entity and relation granularity, incorporating coordinated modules for re-ranking, filtering, fine-grained synthetic data generation, and content compression. To address computational challenges, we design an efficient two-stage retrieval scheme that combines document-level filtering with graph-level refinement. We further propose two context-aware privacy evaluation metrics, i.e., Attack Extraction and Personal Identification, for measuring inferential privacy risks beyond traditional lexical overlap. Extensive empirical evaluation on medical and legal benchmark datasets demonstrates that our method achieves significant improvements in privacy protection while preserving response utility comparable to non-privacy-preserving RAG baselines. These results highlight the promise of knowledge graph-based, fine-grained privacy interventions for secure deployment of RAG systems in high-stakes applications.

检索增强生成（RAG）系统使用特定于领域的数据增强大型语言模型，越来越多地部署在敏感领域，但仍然容易受到隐私泄露的影响。现有的隐私保护通常在文档或段落级别操作，导致隐私和实用程序之间的粗粒度权衡，可能会删除基本上下文或不能充分保护敏感信息。在这项工作中，我们提出了一种新的基于知识图表示的隐私保护RAG方法。我们的方法引入了一个隐私感知处理器，它在实体和关系粒度上运行，结合了用于重新排序、过滤、细粒度合成数据生成和内容压缩的协调模块。为了解决计算方面的挑战，我们设计了一种高效的两阶段检索方案，该方案结合了文档级过滤和图级细化。我们进一步提出了两个上下文感知的隐私评估指标，即攻击提取和个人识别，用于测量超越传统词汇重叠的推断隐私风险。对医疗和法律基准数据集的广泛实证评估表明，我们的方法在保持响应效用的同时，在隐私保护方面取得了显着改进，与非隐私保护的RAG基线相当。这些结果强调了在高风险应用程序中安全部署RAG系统的基于知识图的细粒度隐私干预的前景。

{"title":"Privacy protection in RAG: A novel method and evaluation framework","authors":"Yuan Zhang , Jionghan Wu , Rui Li , Tong Zhang , Yujie Song , Chuanyi Li , Shangqi Wang , Hao Shen , Jiao Yin , Jidong Ge , Bin Luo","doi":"10.1016/j.ipm.2025.104505","DOIUrl":"10.1016/j.ipm.2025.104505","url":null,"abstract":"<div><div>Retrieval-augmented generation (RAG) systems, which enhance large language models with domain-specific data, are increasingly deployed in sensitive domains but remain vulnerable to privacy leakage. Existing privacy safeguards often operate at the document or paragraph level, resulting in a coarse-grained trade-off between privacy and utility that may remove essential context or inadequately protect sensitive information. In this work, we propose a novel privacy-preserving RAG method grounded in knowledge graph representations. Our method introduces a privacy-aware processor that operates at the entity and relation granularity, incorporating coordinated modules for re-ranking, filtering, fine-grained synthetic data generation, and content compression. To address computational challenges, we design an efficient two-stage retrieval scheme that combines document-level filtering with graph-level refinement. We further propose two context-aware privacy evaluation metrics, i.e., Attack Extraction and Personal Identification, for measuring inferential privacy risks beyond traditional lexical overlap. Extensive empirical evaluation on medical and legal benchmark datasets demonstrates that our method achieves significant improvements in privacy protection while preserving response utility comparable to non-privacy-preserving RAG baselines. These results highlight the promise of knowledge graph-based, fine-grained privacy interventions for secure deployment of RAG systems in high-stakes applications.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104505"},"PeriodicalIF":6.9,"publicationDate":"2025-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790309","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

Bridging ERP localization barriers: How knowledge sharing drives integration between Chinese firms and global clientele 弥合ERP本地化障碍：知识共享如何推动中国企业与全球客户之间的整合

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

Information Processing & Management

Pub Date : 2025-12-13 DOI: 10.1016/j.ipm.2025.104559

Sajjad Alam, Yingying Zhu, Jin Liu

This study examines the challenges of localizing Enterprise Resource Planning (ERP) systems in China, with a particular focus on managing foreign stakeholder transaction data to ensure seamless implementation and regulatory compliance. It emphasizes the role of knowledge processes in enabling Chinese firms to collaborate effectively with international partners while addressing complex issues related to accounting standards, tax regulations, data privacy, and information security. To explore these relationships, the study applies Partial Least Squares Structural Equation Modeling (PLS-SEM) to survey data collected from 258 firms in the Shanghai Delta region. The results reveal that knowledge-process-driven data security (β = 0.38, p < 0.001) and tax compliance (β = 0.40, p < 0.001) have a significant positive impact on ERP performance. These findings underscore the critical role of knowledge processes in bridging localization gaps, particularly in aligning ERP modules with China's evolving regulatory frameworks. By demonstrating the interplay between knowledge strategies and compliance requirements, this research offers actionable insights for firms navigating ERP localization and for policymakers engaged in cross-border data governance.

本研究探讨了企业资源规划（ERP）系统在中国本地化所面临的挑战，特别关注管理外国利益相关者交易数据，以确保无缝实施和合规。它强调了知识流程的作用，使中国公司能够与国际合作伙伴有效合作，同时解决与会计准则、税收法规、数据隐私和信息安全相关的复杂问题。为了探讨这些关系，本研究运用偏最小二乘结构方程模型（PLS-SEM）对上海三角洲地区258家企业的数据进行了调查。结果表明，知识过程驱动的数据安全性（β = 0.38, p < 0.001）和税收合规性（β = 0.40, p < 0.001）对ERP绩效有显著的正向影响。这些发现强调了知识流程在弥合本地化差距方面的关键作用，特别是在使ERP模块与中国不断发展的监管框架保持一致方面。通过展示知识战略和合规要求之间的相互作用，本研究为企业导航ERP本地化和从事跨境数据治理的政策制定者提供了可操作的见解。

{"title":"Bridging ERP localization barriers: How knowledge sharing drives integration between Chinese firms and global clientele","authors":"Sajjad Alam, Yingying Zhu, Jin Liu","doi":"10.1016/j.ipm.2025.104559","DOIUrl":"10.1016/j.ipm.2025.104559","url":null,"abstract":"<div><div>This study examines the challenges of localizing Enterprise Resource Planning (ERP) systems in China, with a particular focus on managing foreign stakeholder transaction data to ensure seamless implementation and regulatory compliance. It emphasizes the role of knowledge processes in enabling Chinese firms to collaborate effectively with international partners while addressing complex issues related to accounting standards, tax regulations, data privacy, and information security. To explore these relationships, the study applies Partial Least Squares Structural Equation Modeling (PLS-SEM) to survey data collected from 258 firms in the Shanghai Delta region. The results reveal that knowledge-process-driven data security (β = 0.38, <em>p</em> < 0.001) and tax compliance (β = 0.40, <em>p</em> < 0.001) have a significant positive impact on ERP performance. These findings underscore the critical role of knowledge processes in bridging localization gaps, particularly in aligning ERP modules with China's evolving regulatory frameworks. By demonstrating the interplay between knowledge strategies and compliance requirements, this research offers actionable insights for firms navigating ERP localization and for policymakers engaged in cross-border data governance.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104559"},"PeriodicalIF":6.9,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790310","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

Citation importance-aware document representation learning for large-scale science mapping 大规模科学制图中基于引文重要性的文献表示学习

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

Information Processing & Management

Pub Date : 2025-12-13 DOI: 10.1016/j.ipm.2025.104557

Zhentao Liang , Nees Jan van Eck , Xuehua Wu , Jin Mao , Gang Li

Effective science mapping relies on high-quality representations of scientific documents. As an important task in scientometrics and information studies, science mapping is often challenged by the complex and heterogeneous nature of citations. While previous studies have attempted to improve document representations by integrating citation and semantic information, the heterogeneity of citations is often overlooked. To address this problem, this study proposes a citation importance-aware contrastive learning framework that refines the supervisory signal. We first develop a scalable measurement of citation importance based on location, frequency, and self-citation characteristics. Citation importance is then integrated into the contrastive learning process through an importance-aware sampling strategy, which selects low-importance citations as “hard negatives”. This forces the model to learn finer-grained representations that distinguish between important and perfunctory citations. To validate the effectiveness of the proposed framework, we fine-tune a SciBERT model and perform extensive evaluations on SciDocs and PubMed benchmark datasets. Results show consistent improvements in both document representation quality and science mapping accuracy. Furthermore, we apply the trained model to over 33 million documents from Web of Science. The resulting map of science accurately visualizes the global and local intellectual structure of science and reveals interdisciplinary research fronts. By operationalizing citation heterogeneity into a scalable computational framework, this study demonstrates how differentiating citations by their importance can be effectively leveraged to improve document representation and science mapping.

有效的科学制图依赖于科学文献的高质量表示。作为科学计量学和信息研究的一项重要任务，科学制图经常受到引文复杂性和异质性的挑战。虽然以前的研究试图通过整合引文和语义信息来改善文献表示，但引文的异质性往往被忽视。为了解决这个问题，本研究提出了一个引用重要性感知的对比学习框架，该框架可以细化监督信号。我们首先根据地点、频率和自引特征开发了可扩展的引文重要性测量方法。然后通过重要性感知抽样策略将引文重要性整合到对比学习过程中，该策略选择低重要性的引文作为“硬否定”。这迫使模型学习细粒度表示，以区分重要引用和敷衍引用。为了验证所提出框架的有效性，我们对SciBERT模型进行了微调，并对SciDocs和PubMed基准数据集进行了广泛的评估。结果表明，在文档表示质量和科学制图精度方面都有一致的提高。此外，我们将训练好的模型应用于Web of Science的3300多万份文档。由此产生的科学地图准确地可视化了全球和当地的科学知识结构，并揭示了跨学科的研究前沿。通过将引文异质性操作到一个可扩展的计算框架中，本研究展示了如何根据引文的重要性来区分引文，从而有效地改善文献表示和科学映射。

{"title":"Citation importance-aware document representation learning for large-scale science mapping","authors":"Zhentao Liang , Nees Jan van Eck , Xuehua Wu , Jin Mao , Gang Li","doi":"10.1016/j.ipm.2025.104557","DOIUrl":"10.1016/j.ipm.2025.104557","url":null,"abstract":"<div><div>Effective science mapping relies on high-quality representations of scientific documents. As an important task in scientometrics and information studies, science mapping is often challenged by the complex and heterogeneous nature of citations. While previous studies have attempted to improve document representations by integrating citation and semantic information, the heterogeneity of citations is often overlooked. To address this problem, this study proposes a citation importance-aware contrastive learning framework that refines the supervisory signal. We first develop a scalable measurement of citation importance based on location, frequency, and self-citation characteristics. Citation importance is then integrated into the contrastive learning process through an importance-aware sampling strategy, which selects low-importance citations as “hard negatives”. This forces the model to learn finer-grained representations that distinguish between important and perfunctory citations. To validate the effectiveness of the proposed framework, we fine-tune a SciBERT model and perform extensive evaluations on SciDocs and PubMed benchmark datasets. Results show consistent improvements in both document representation quality and science mapping accuracy. Furthermore, we apply the trained model to over 33 million documents from Web of Science. The resulting map of science accurately visualizes the global and local intellectual structure of science and reveals interdisciplinary research fronts. By operationalizing citation heterogeneity into a scalable computational framework, this study demonstrates how differentiating citations by their importance can be effectively leveraged to improve document representation and science mapping.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104557"},"PeriodicalIF":6.9,"publicationDate":"2025-12-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145790314","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

SEAttack: A self-evolving jailbreak attack to induce toxic responses for non-toxic queries in large language models SEAttack：一种自我进化的越狱攻击，在大型语言模型中为无害查询诱导有害响应

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

Information Processing & Management

Pub Date : 2025-12-12 DOI: 10.1016/j.ipm.2025.104544

Huijun Liu , Shasha Li , Bin Ji , Xiaohu Du , Xiaopeng Li , Jun Ma† , Jie Yu

To prevent Large Language Models (LLMs) from being misused by bad actors, model developers delve into designing safety mechanisms to guarantee LLMs generate helpful, honest, and harmless responses. To disclose flaws in the safety mechanisms, researchers conduct adversarial attacks against LLMs to circumvent the safety mechanisms, which are called “jailbreak attack”. Existing jailbreak attacks solely focus on engineering toxic queries and adversarial prompts to induce LLMs to generate toxic responses, losing sight of the study starting from engineering non-toxic queries and adversarial prompts, which also plays an important role in disclosing flaws in safety mechanisms. To fill the research gap, we propose SEAttack, a self-evolving jailbreak attack method to induce LLMs to generate toxic responses for non-toxic queries. Given a non-toxic query, SEAttack initially generates a response, which is more likely to be non-toxic due to the safety mechanisms. Then, it uses multiple iterations of self-evolving to evolve the non-toxic response to a toxic one. To evaluate SEAttack, we construct JailChat, a dataset containing 3000 non-toxic queries. We drive SEAttack to attack eighteen state-of-the-art LLMs, including five closed-source and thirteen open-source LLMs. Experimental results demonstrate that SEAttack achieves up to 89.07 % attack success rate, revealing non-negligible flaws in LLMs’ safety mechanisms. Moreover, we track the changes in the safety mechanisms of four ChatGPT variants. Extensive analyses and human evaluation further validate the effectiveness and rationality of SEAttack.

为了防止大型语言模型（llm）被不良行为者滥用，模型开发人员深入研究设计安全机制，以保证llm生成有用的、诚实的和无害的响应。为了揭示安全机制的缺陷，研究人员对llm进行对抗性攻击，以绕过安全机制，这被称为“越狱攻击”。现有的越狱攻击仅关注工程毒性查询和对抗性提示来诱导llm产生毒性响应，而忽略了从工程无毒查询和对抗性提示开始的研究，这对揭示安全机制的缺陷也起着重要作用。为了填补研究空白，我们提出了SEAttack，一种自我进化的越狱攻击方法，诱导llm为无毒查询生成有毒响应。给定一个无毒查询，SEAttack最初生成一个响应，由于安全机制，该响应更有可能是无毒的。然后，它使用多次自我进化的迭代来进化无毒反应到有毒反应。为了评估SEAttack，我们构建了一个包含3000个无毒查询的数据集JailChat。我们驱动SEAttack攻击18个最先进的llm，包括5个闭源和13个开源llm。实验结果表明，SEAttack的攻击成功率高达89.07%，揭示了llm安全机制中不可忽视的缺陷。此外，我们还跟踪了四种ChatGPT变体的安全机制的变化。大量的分析和人工评估进一步验证了SEAttack的有效性和合理性。

{"title":"SEAttack: A self-evolving jailbreak attack to induce toxic responses for non-toxic queries in large language models","authors":"Huijun Liu , Shasha Li , Bin Ji , Xiaohu Du , Xiaopeng Li , Jun Ma† , Jie Yu","doi":"10.1016/j.ipm.2025.104544","DOIUrl":"10.1016/j.ipm.2025.104544","url":null,"abstract":"<div><div>To prevent Large Language Models (LLMs) from being misused by bad actors, model developers delve into designing safety mechanisms to guarantee LLMs generate helpful, honest, and harmless responses. To disclose flaws in the safety mechanisms, researchers conduct adversarial attacks against LLMs to circumvent the safety mechanisms, which are called “jailbreak attack”. Existing jailbreak attacks solely focus on engineering toxic queries and adversarial prompts to induce LLMs to generate toxic responses, losing sight of the study starting from engineering non-toxic queries and adversarial prompts, which also plays an important role in disclosing flaws in safety mechanisms. To fill the research gap, we propose SEAttack, a self-evolving jailbreak attack method to induce LLMs to generate toxic responses for non-toxic queries. Given a non-toxic query, SEAttack initially generates a response, which is more likely to be non-toxic due to the safety mechanisms. Then, it uses multiple iterations of self-evolving to evolve the non-toxic response to a toxic one. To evaluate SEAttack, we construct JailChat, a dataset containing 3000 non-toxic queries. We drive SEAttack to attack eighteen state-of-the-art LLMs, including five closed-source and thirteen open-source LLMs. Experimental results demonstrate that SEAttack achieves up to 89.07 % attack success rate, revealing non-negligible flaws in LLMs’ safety mechanisms. Moreover, we track the changes in the safety mechanisms of four ChatGPT variants. Extensive analyses and human evaluation further validate the effectiveness and rationality of SEAttack.</div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104544"},"PeriodicalIF":6.9,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737323","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

RADAR: Relation-assisted dual-graph aligning recognition for grounded multimodal named entity recognition 基于多模态命名实体识别的关系辅助双图对齐识别

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

Information Processing & Management

Pub Date : 2025-12-12 DOI: 10.1016/j.ipm.2025.104552

Zai Zhang , Bin Shi , Kai Sun , Hao Wu , Bo Dong

Grounded Multimodal Named Entity Recognition (GMNER) requires the simultaneous identification of textual entities and their corresponding visual regions within images. However, the inability to model visual contextual semantics and the disorganized processing of cross-modal features often lead existing methods to struggle with both visual entity differentiation and bridging the modality gap. We propose RADAR (Relation-Assisted Dual-graph Aligning Recognition), a novel framework that leverages visual relations derived from scene graphs to encode structured context and enhance visual understanding. To achieve fine-grained cross-modal alignment, we design an object-level alignment self-attention mechanism and introduce a dual-graph strategy. Evaluated on the Twitter-GMNER dataset (13,076 image-text pairs), RADAR achieves 60.91 % F1 score, a +4.5 % improvement over the H-Index baseline. The method also demonstrates consistent gains in subtasks, with +3.54 % improvement in EEG metric, validating its effectiveness in multimodal entity alignment.

基于多模态命名实体识别（GMNER）要求同时识别图像中的文本实体及其对应的视觉区域。然而，由于无法对视觉上下文语义进行建模以及跨模态特征的无组织处理，现有方法往往难以区分视觉实体并弥合模态差距。我们提出了RADAR（关系辅助双图对齐识别），这是一个利用来自场景图的视觉关系来编码结构化上下文并增强视觉理解的新框架。为了实现细粒度的跨模态对齐，我们设计了对象级对齐自关注机制，并引入了双图策略。在Twitter-GMNER数据集（13,076对图像-文本）上进行评估，RADAR达到60.91%的F1得分，比H-Index基线提高了+ 4.5%。该方法在子任务中也显示出一致性的增益，EEG度量提高了+ 3.54%，验证了其在多模态实体对齐中的有效性。

引用次数: 0

Regulation and innovation: Unveiling the quadruple-helix-innovation ecosystem of generative AI 监管与创新：揭示生成式人工智能的四螺旋创新生态系统

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

Information Processing & Management

Pub Date : 2025-12-11 DOI: 10.1016/j.ipm.2025.104549

Yu Peng Zhu , Si Qi Liang , Ya Fang Zou , Han Woo Park

The collaboration between generative artificial intelligence (Gen-AI) innovation and regulation is a new trend. How can different regions and subjects balance knowledge innovation in Gen-AI between regulation and innovation? This study introduced an innovative Quadruple Helix model. By analyzing over 6000 pieces of data from five different fields–policy documents, industry patents, academic papers, media news, and posts from X–we obtained knowledge output and collaborative relationships between different Quadruple Helix subjects. We find that governments, industries, and universities play different but interrelated roles in the innovation and regulation of Gen-AI. The participation of civil society has broken through the limitations of the Triple Helix in knowledge innovation, effectively balancing innovation and regulation of Gen-AI knowledge production. This study seeks to provide a new perspective and method for gaining an in-depth understanding of Gen-AI development. We recommend that the government and corporate sector collaboratively construct an innovation and governance system that harmonizes specialized expertise with public consensus, aiming to achieve democratic governance and technological sustainability.

生成式人工智能（Gen-AI）创新与监管之间的协作是一个新趋势。不同地区和学科如何在监管与创新之间平衡Gen-AI的知识创新？本研究引入了一种创新的四螺旋模型。通过分析政策文件、行业专利、学术论文、媒体新闻、x的帖子等5个不同领域的6000多条数据，我们得到了不同四螺旋学科之间的知识输出和协作关系。我们发现，政府、行业和大学在Gen-AI的创新和监管中扮演着不同但相互关联的角色。公民社会的参与，突破了知识创新“三重螺旋”的局限，有效地平衡了Gen-AI知识生产的创新与监管。本研究旨在为深入了解Gen-AI的发展提供新的视角和方法。建议政府和企业共同构建专业知识与公众共识相协调的创新治理体系，以实现民主治理和技术可持续性。

{"title":"Regulation and innovation: Unveiling the quadruple-helix-innovation ecosystem of generative AI","authors":"Yu Peng Zhu , Si Qi Liang , Ya Fang Zou , Han Woo Park","doi":"10.1016/j.ipm.2025.104549","DOIUrl":"10.1016/j.ipm.2025.104549","url":null,"abstract":"<div><div>The collaboration between generative artificial intelligence (Gen-AI) innovation and regulation is a new trend. How can different regions and subjects balance knowledge innovation in Gen-AI between regulation and innovation? This study introduced an innovative Quadruple Helix model. By analyzing over 6000 pieces of data from five different fields–policy documents, industry patents, academic papers, media news, and posts from X–we obtained knowledge output and collaborative relationships between different Quadruple Helix subjects. We find that governments, industries, and universities play different but interrelated roles in the innovation and regulation of Gen-AI. The participation of civil society has broken through the limitations of the Triple Helix in knowledge innovation, effectively balancing innovation and regulation of Gen-AI knowledge production. This study seeks to provide a new perspective and method for gaining an in-depth understanding of Gen-AI development. We recommend that the government and corporate sector collaboratively construct an innovation and governance system that harmonizes specialized expertise with public consensus, aiming to achieve democratic governance and technological sustainability.<!--> </div></div>","PeriodicalId":50365,"journal":{"name":"Information Processing & Management","volume":"63 3","pages":"Article 104549"},"PeriodicalIF":6.9,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737426","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