Information Systems最新文献

VCR: Interpretable and interactive debugging of object detection models with visual concepts 具有可视化概念的对象检测模型的可解释和交互式调试

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-12 DOI: 10.1016/j.is.2025.102652

Jie Jeff Xu , Saahir Dhanani , Jorge Piazentin Ono , Wenbin He , Liu Ren , Kexin Rong

Computer vision models can make systematic errors, performing well on average but substantially worse on particular subsets (or slices) of data. In this work, we introduce Visual Concept Reviewer (VCR), a human-in-the-loop slice discovery framework that enables practitioners to interactively discover and understand systematic errors in object-detection models via novel use of visual concepts–semantically meaningful and frequently recurring image segments representing objects, parts, or abstract properties.

Leveraging recent advances in vision foundation models, VCR automatically generates segment-level visual concepts that serve as interpretable primitives for diagnosing issues in object-detection models, while also supporting lightweight human supervision when needed. VCR combines visual concepts with metadata in a tabular format and adapts frequent itemset mining techniques to identify common absences and presences of concepts associated with poor model performance at interactive speeds. VCR also keeps humans in the loop for interpretation and refinement at each step of the slice discovery process. We demonstrate VCR’s effectiveness and scalability through a new evaluation benchmark with 1713 slice discovery settings across three datasets. A user study with six expert industry machine learning scientists and engineers provides qualitative evidence of VCR’s utility in real-world workflows.

计算机视觉模型可能会产生系统错误，平均表现良好，但在特定的数据子集（或切片）上表现得更差。在这项工作中，我们介绍了视觉概念审查器（VCR），这是一个人在循环中的切片发现框架，使从业者能够通过新颖地使用视觉概念（表示对象、部件或抽象属性的语义上有意义且经常重复出现的图像片段）来交互式地发现和理解对象检测模型中的系统错误。利用视觉基础模型的最新进展，VCR自动生成分段级视觉概念，作为对象检测模型中诊断问题的可解释原语，同时在需要时还支持轻量级的人工监督。VCR将可视化概念与表格格式的元数据结合起来，并采用频繁的项集挖掘技术来识别与交互速度较差的模型性能相关的概念的常见缺失和存在。VCR还使人类在切片发现过程的每一步都能进行解释和改进。我们通过一个新的评估基准，在三个数据集上使用1713个切片发现设置，展示了VCR的有效性和可扩展性。六位行业机器学习专家和工程师的用户研究为VCR在实际工作流程中的实用性提供了定性证据。

{"title":"VCR: Interpretable and interactive debugging of object detection models with visual concepts","authors":"Jie Jeff Xu , Saahir Dhanani , Jorge Piazentin Ono , Wenbin He , Liu Ren , Kexin Rong","doi":"10.1016/j.is.2025.102652","DOIUrl":"10.1016/j.is.2025.102652","url":null,"abstract":"<div><div>Computer vision models can make systematic errors, performing well on average but substantially worse on particular subsets (or slices) of data. In this work, we introduce Visual Concept Reviewer (VCR), a human-in-the-loop slice discovery framework that enables practitioners to interactively discover and understand systematic errors in object-detection models via novel use of visual concepts–semantically meaningful and frequently recurring image segments representing objects, parts, or abstract properties.</div><div>Leveraging recent advances in vision foundation models, <span>VCR</span> automatically generates segment-level visual concepts that serve as interpretable primitives for diagnosing issues in object-detection models, while also supporting lightweight human supervision when needed. <span>VCR</span> combines visual concepts with metadata in a tabular format and adapts frequent itemset mining techniques to identify common absences and presences of concepts associated with poor model performance at interactive speeds. <span>VCR</span> also keeps humans in the loop for interpretation and refinement at each step of the slice discovery process. We demonstrate VCR’s effectiveness and scalability through a new evaluation benchmark with 1713 slice discovery settings across three datasets. A user study with six expert industry machine learning scientists and engineers provides qualitative evidence of VCR’s utility in real-world workflows.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102652"},"PeriodicalIF":3.4,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

HILTS: Human-LLM collaboration for effective data labeling HILTS：人类-法学硕士协作有效的数据标签

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-12 DOI: 10.1016/j.is.2025.102660

Juliana Barbosa, Eduarda Alencar, Grace Fan, Aécio Santos, Juliana Freire

The growing complexity and volume of data highlight the importance of learning-based classifiers across diverse tasks, from medical diagnosis to environmental monitoring. A common and impactful use case is data triage—efficiently identifying rare, relevant instances in large, imbalanced datasets. This is crucial for enabling domain experts to focus on what matters most. However, traditional supervised learning approaches often struggle with scalability due to the high cost and time required for manual labeling.

We introduce HILTS (Human-In-the-loop Learn To Sample), a framework designed to tackle these limitations. HILTS leverages Large Language Models (LLMs) for automated initial labeling and strategically incorporates human expertise through advanced active learning techniques. It selects diverse and representative samples for pseudo-labeling and identifies highly uncertain or likely incorrect LLM labels for targeted human review. This focused use of human effort maximizes the value of domain expertise while minimizing annotation overhead.

Our system reduces human labeling effort by up to 80% while outperforming few-shot foundation models such as GPT-4 by over 5% in F1-score in some scenarios—all at a significantly lower cost. HILTS also shows clear improvements over fully automated pseudo-labeling approaches and proves especially effective in handling class imbalance in real-world datasets. Its adaptability and efficiency make it a practical and scalable solution for high-stakes, domain-specific data triage tasks.

不断增长的复杂性和数据量凸显了基于学习的分类器在各种任务中的重要性，从医疗诊断到环境监测。一个常见且有影响力的用例是数据分类——在大型、不平衡的数据集中有效地识别罕见的、相关的实例。这对于使领域专家专注于最重要的事情是至关重要的。然而，传统的监督学习方法由于人工标注的高成本和时间要求，往往难以实现可扩展性。我们介绍了HILTS (Human-In-the-loop Learn To Sample)，这是一个旨在解决这些限制的框架。HILTS利用大型语言模型（llm）进行自动初始标记，并通过先进的主动学习技术战略性地结合人类专业知识。它选择多样化和代表性的样本进行伪标记，并识别高度不确定或可能不正确的LLM标签，用于有针对性的人工审查。这种对人力的集中使用使领域专业知识的价值最大化，同时使注释开销最小化。我们的系统减少了多达80%的人工标记工作，同时在某些情况下，在f1得分上比GPT-4等少量基础模型高出5%以上，所有这些都大大降低了成本。HILTS还显示出比全自动伪标签方法有明显的改进，并证明在处理真实数据集中的类不平衡方面特别有效。它的适应性和效率使其成为高风险、特定于领域的数据分类任务的实用且可扩展的解决方案。

{"title":"HILTS: Human-LLM collaboration for effective data labeling","authors":"Juliana Barbosa, Eduarda Alencar, Grace Fan, Aécio Santos, Juliana Freire","doi":"10.1016/j.is.2025.102660","DOIUrl":"10.1016/j.is.2025.102660","url":null,"abstract":"<div><div>The growing complexity and volume of data highlight the importance of learning-based classifiers across diverse tasks, from medical diagnosis to environmental monitoring. A common and impactful use case is data triage—efficiently identifying rare, relevant instances in large, imbalanced datasets. This is crucial for enabling domain experts to focus on what matters most. However, traditional supervised learning approaches often struggle with scalability due to the high cost and time required for manual labeling.</div><div>We introduce HILTS (Human-In-the-loop Learn To Sample), a framework designed to tackle these limitations. HILTS leverages Large Language Models (LLMs) for automated initial labeling and strategically incorporates human expertise through advanced active learning techniques. It selects diverse and representative samples for pseudo-labeling and identifies highly uncertain or likely incorrect LLM labels for targeted human review. This focused use of human effort maximizes the value of domain expertise while minimizing annotation overhead.</div><div>Our system reduces human labeling effort by up to 80% while outperforming few-shot foundation models such as GPT-4 by over 5% in F1-score in some scenarios—all at a significantly lower cost. HILTS also shows clear improvements over fully automated pseudo-labeling approaches and proves especially effective in handling class imbalance in real-world datasets. Its adaptability and efficiency make it a practical and scalable solution for high-stakes, domain-specific data triage tasks.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102660"},"PeriodicalIF":3.4,"publicationDate":"2025-12-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791324","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

ACTER: Activity Customization through Timely and Explainable Recommendations ACTER：通过及时和可解释的建议定制活动

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-11 DOI: 10.1016/j.is.2025.102666

Anna Dalla Vecchia, Niccolò Marastoni, Barbara Oliboni, Elisa Quintarelli

The proliferation of sensors, including wearable devices, has significantly increased the volume of generated data, opening up new opportunities for personalized recommendations. This paper presents ACTER (Activity Customization through Timely and Explainable Recommendations), an integrated framework to provide contextual, timely, explainable, and user-specific recommendations. Thanks to the sequential rule mining algorithm ALBA (AgedLookBackApriori), we extract totally ordered sequential rules to uncover hidden insights from temporal data, ultimately improving a predefined target parameter related to the selected application domain. An aging mechanism is applied to ensure that recommendations remain relevant, giving more weight to newer information while still considering older data. In addition, our framework leverages historical data to also infer personalized, contextual information, allowing us to adapt the predefined context—usually set at the design stage—more dynamically and expressly. The experimental results of the ACTER evaluation confirm that integrating ad-hoc contexts mined from historical data into the recommender system yields more accurate suggestions.

包括可穿戴设备在内的传感器的激增大大增加了生成的数据量，为个性化推荐开辟了新的机会。本文介绍了ACTER（通过及时和可解释的建议进行活动定制），这是一个集成框架，用于提供上下文相关的、及时的、可解释的和特定于用户的建议。得益于顺序规则挖掘算法ALBA (AgedLookBackApriori)，我们提取了完全有序的顺序规则，以从时间数据中发现隐藏的见解，最终改进了与所选应用程序领域相关的预定义目标参数。使用老化机制来确保建议保持相关性，在考虑旧数据的同时给予新信息更多权重。此外，我们的框架还利用历史数据来推断个性化的上下文信息，从而允许我们更动态、更明确地调整预定义的上下文（通常在设计阶段设置）。ACTER评估的实验结果证实，将从历史数据中挖掘的临时上下文集成到推荐系统中可以产生更准确的建议。

{"title":"ACTER: Activity Customization through Timely and Explainable Recommendations","authors":"Anna Dalla Vecchia, Niccolò Marastoni, Barbara Oliboni, Elisa Quintarelli","doi":"10.1016/j.is.2025.102666","DOIUrl":"10.1016/j.is.2025.102666","url":null,"abstract":"<div><div>The proliferation of sensors, including wearable devices, has significantly increased the volume of generated data, opening up new opportunities for personalized recommendations. This paper presents ACTER (Activity Customization through Timely and Explainable Recommendations), an integrated framework to provide contextual, timely, explainable, and user-specific recommendations. Thanks to the sequential rule mining algorithm ALBA (AgedLookBackApriori), we extract totally ordered sequential rules to uncover hidden insights from temporal data, ultimately improving a predefined target parameter related to the selected application domain. An aging mechanism is applied to ensure that recommendations remain relevant, giving more weight to newer information while still considering older data. In addition, our framework leverages historical data to also infer personalized, contextual information, allowing us to adapt the predefined context—usually set at the design stage—more dynamically and expressly. The experimental results of the ACTER evaluation confirm that integrating ad-hoc contexts mined from historical data into the recommender system yields more accurate suggestions.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102666"},"PeriodicalIF":3.4,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737870","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

From precision to perception: Human-in-the-loop evaluation of keyword extraction for internet-scale contextual advertising 从精确到感知：互联网规模上下文广告关键字提取的人在循环评估

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-11 DOI: 10.1016/j.is.2025.102665

Jingwen Cai , Sara Leckner , Johanna Björklund

Keyword extraction is a foundational task in natural language processing, underpinning countless real-world applications. One of these is contextual advertising, where keywords help predict the topical congruence between ads and their surrounding media contexts to enhance advertising effectiveness. Recent advances in artificial intelligence have improved keyword extraction capabilities but also introduced concerns about computational cost. Moreover, although the end-user experience is of vital importance, human evaluation of keyword extraction performances remains under-explored. This study provides a comparative evaluation of prevalent keyword extraction algorithms with different levels of complexity represented by TF-IDF, KeyBERT, and Llama 2. To evaluate their effectiveness, a mixed-methods approach is employed, combining quantitative benchmarking with qualitative assessments from 855 participants through four survey-based experiments. The findings demonstrate that KeyBERT achieves an effective balance between user preferences and computational efficiency, compared to the other algorithms. We observe a clear overall preference for gold-standard keywords, but there is a misalignment between algorithmic benchmark performance and user ratings. This reveals a long-overlooked gap between traditional precision-focused metrics and user-perceived algorithm efficiency. The study underscores the importance of human-in-the-loop evaluation methodologies and proposes analytical tools to facilitate their implementation.

关键字提取是自然语言处理的一项基础任务，支撑着无数现实世界的应用。其中之一是上下文广告，其中关键词有助于预测广告与周围媒体上下文之间的主题一致性，以提高广告效果。人工智能的最新进展提高了关键字提取能力，但也引入了对计算成本的担忧。此外，尽管最终用户体验至关重要，但关键字提取性能的人类评估仍未得到充分探索。本研究对以TF-IDF、KeyBERT和Llama 2为代表的不同复杂度的流行关键字提取算法进行了比较评价。为了评估其有效性，采用了一种混合方法，通过四个基于调查的实验，将855名参与者的定量基准与定性评估相结合。研究结果表明，与其他算法相比，KeyBERT在用户偏好和计算效率之间实现了有效的平衡。我们观察到对黄金标准关键字的明显总体偏好，但算法基准性能和用户评级之间存在不一致。这揭示了传统的以精度为中心的指标和用户感知的算法效率之间长期被忽视的差距。该研究强调了人在循环评估方法的重要性，并提出了促进其实施的分析工具。

{"title":"From precision to perception: Human-in-the-loop evaluation of keyword extraction for internet-scale contextual advertising","authors":"Jingwen Cai , Sara Leckner , Johanna Björklund","doi":"10.1016/j.is.2025.102665","DOIUrl":"10.1016/j.is.2025.102665","url":null,"abstract":"<div><div>Keyword extraction is a foundational task in natural language processing, underpinning countless real-world applications. One of these is contextual advertising, where keywords help predict the topical congruence between ads and their surrounding media contexts to enhance advertising effectiveness. Recent advances in artificial intelligence have improved keyword extraction capabilities but also introduced concerns about computational cost. Moreover, although the end-user experience is of vital importance, human evaluation of keyword extraction performances remains under-explored. This study provides a comparative evaluation of prevalent keyword extraction algorithms with different levels of complexity represented by TF-IDF, KeyBERT, and Llama 2. To evaluate their effectiveness, a mixed-methods approach is employed, combining quantitative benchmarking with qualitative assessments from 855 participants through four survey-based experiments. The findings demonstrate that KeyBERT achieves an effective balance between user preferences and computational efficiency, compared to the other algorithms. We observe a clear overall preference for gold-standard keywords, but there is a misalignment between algorithmic benchmark performance and user ratings. This reveals a long-overlooked gap between traditional precision-focused metrics and user-perceived algorithm efficiency. The study underscores the importance of human-in-the-loop evaluation methodologies and proposes analytical tools to facilitate their implementation.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102665"},"PeriodicalIF":3.4,"publicationDate":"2025-12-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737933","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Visualizing repetition in process execution variants from partially ordered event data 从部分有序的事件数据中可视化流程执行变体中的重复

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-05 DOI: 10.1016/j.is.2025.102664

Ariba Siddiqui , Francesca Zerbato , Daniel Schuster

Operational processes often exhibit concurrency, where the execution of activities can overlap in time. Moreover, repetitions of activities, both intentional (e.g., iterative tasks) and unintentional (e.g., rework) often occur. Existing process mining techniques and visualizations largely assume sequential event data, making it difficult to analyze repetitions in partially ordered event data, which better captures real-world process behavior. We address this gap by introducing a novel arc-diagram-based visualization that highlights recurring activity patterns within individual process execution variants. This approach allows analysts to intuitively detect repetitions that are otherwise obscured in raw data or traditional variant views. We validate the usefulness and ease of use of the proposed visualization through a user study with process mining experts and provide an implementation of our contribution in an open-source tool, supporting practical adoption.

操作流程通常表现为并发性，其中活动的执行可以在时间上重叠。此外，活动的重复，有意的（例如，迭代任务）和无意的（例如，返工）经常发生。现有的流程挖掘技术和可视化在很大程度上假设事件数据是顺序的，这使得分析部分有序事件数据中的重复变得困难，而部分有序事件数据能够更好地捕捉真实的流程行为。我们通过引入一种新颖的基于弧线图的可视化来解决这一差距，该可视化突出了单个流程执行变体中重复出现的活动模式。这种方法允许分析人员直观地检测在原始数据或传统变体视图中被掩盖的重复。我们通过与过程挖掘专家的用户研究验证了所建议的可视化的有用性和易用性，并在开源工具中提供了我们的贡献的实现，支持实际采用。

引用次数: 0

Efficient allocation of shared resources across multiple processes 跨多个进程有效地分配共享资源

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-05 DOI: 10.1016/j.is.2025.102663

Kiran Busch, Henrik Leopold

Effective resource allocation is crucial for optimizing business processes. Yet, most existing methods focus solely on single-process optimization, overlooking the interdependencies present in multi-process environments. This limitation results in inefficient resource allocation, and scalability challenges. To address this gap, we propose MuProMAC (Multi-Process Multi-Agent Coordination), a novel reinforcement learning-based method designed to optimize resource allocation across multiple interdependent business processes. Unlike prior methods, MuProMAC is the first online resource allocation method that explicitly models the interdependencies between processes and dynamically balances competing resource demands to minimize global average cycle time. We evaluate our method in five multi-process scenarios with different levels of resource contention, comparing it against state-of-the-art online resource allocation methods and three simple baselines. Our results show that MuProMAC is consistently among the top-performing methods in shared-resource environments. It achieves low cycle times and stable performance across different workload conditions, outperforming existing methods through its strong adaptability to evolving business processes and increasing complexity.

有效的资源分配对于优化业务流程至关重要。然而，大多数现有方法只关注单进程优化，忽略了多进程环境中存在的相互依赖性。这种限制导致资源分配效率低下，并对可伸缩性构成挑战。为了解决这一差距，我们提出了MuProMAC（多进程多代理协调），这是一种新的基于强化学习的方法，旨在优化多个相互依赖的业务流程之间的资源分配。与先前的方法不同，MuProMAC是第一个在线资源分配方法，它显式地建模进程之间的相互依赖关系，并动态平衡竞争资源需求，以最小化全局平均周期时间。我们在五个具有不同资源争用水平的多进程场景中评估了我们的方法，并将其与最先进的在线资源分配方法和三个简单的基线进行了比较。我们的结果表明，在共享资源环境中，MuProMAC始终是性能最好的方法之一。它在不同的工作负载条件下实现了低周期时间和稳定的性能，通过对不断发展的业务流程和不断增加的复杂性的强适应性，优于现有的方法。

{"title":"Efficient allocation of shared resources across multiple processes","authors":"Kiran Busch, Henrik Leopold","doi":"10.1016/j.is.2025.102663","DOIUrl":"10.1016/j.is.2025.102663","url":null,"abstract":"<div><div>Effective resource allocation is crucial for optimizing business processes. Yet, most existing methods focus solely on single-process optimization, overlooking the interdependencies present in multi-process environments. This limitation results in inefficient resource allocation, and scalability challenges. To address this gap, we propose MuProMAC (Multi-Process Multi-Agent Coordination), a novel reinforcement learning-based method designed to optimize resource allocation across multiple interdependent business processes. Unlike prior methods, MuProMAC is the first online resource allocation method that explicitly models the interdependencies between processes and dynamically balances competing resource demands to minimize global average cycle time. We evaluate our method in five multi-process scenarios with different levels of resource contention, comparing it against state-of-the-art online resource allocation methods and three simple baselines. Our results show that MuProMAC is consistently among the top-performing methods in shared-resource environments. It achieves low cycle times and stable performance across different workload conditions, outperforming existing methods through its strong adaptability to evolving business processes and increasing complexity.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102663"},"PeriodicalIF":3.4,"publicationDate":"2025-12-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145737934","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

Reflection on compliance monitoring in business processes: Functionalities, application, and tool-support 对业务流程中的遵从性监视的反思：功能、应用程序和工具支持

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-04 DOI: 10.1016/j.is.2025.102650

Linh Thao Ly , Fabrizio Maria Maggi , Marco Montali , Stefanie Rinderle-Ma , Wil M.P. van der Aalst

Together with Information Systems, we celebrate the journal’s 50th anniversary and the 10th anniversary of our joint work on a systematic framework for compliance monitoring functionalities.

我们与《信息系统》杂志一起庆祝该杂志创刊50周年，以及我们就合规监测功能的系统框架共同开展工作10周年。

引用次数: 0

Reflection on the convergence and interplay of edge, fog, and cloud in the AI-driven Internet of Things (IoT) 人工智能驱动的物联网中边缘、雾、云的融合与互动思考

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-03 DOI: 10.1016/j.is.2025.102662

Farshad Firouzi , Bahar Farahani , Alexander Marinšek

As the Information Systems Journal celebrates its 50th Anniversary, we are honored to reflect on the journey and legacy of our 2022 article, “The convergence and interplay of edge, fog, and cloud in the AI-driven Internet of Things (IoT)”. The paper introduced a unified architectural framework that advanced the integration of computing, intelligence, and connectivity across the edge–fog–cloud continuum, establishing a foundational model for scalable, adaptive, context-aware, and trustworthy AI-enabled systems. This reflection highlights how the work has shaped our research trajectories, influenced developments within the broader scientific community, and guided innovation, education, and industrial practice.

在《信息系统杂志》庆祝创刊50周年之际，我们很荣幸地回顾我们2022年的文章《人工智能驱动的物联网（IoT）中边缘、雾和云的融合和相互作用》的历程和遗产。本文介绍了一个统一的架构框架，该框架推进了跨边缘雾云连续体的计算、智能和连接的集成，为可扩展、自适应、上下文感知和可信赖的人工智能支持系统建立了一个基础模型。这种反思强调了这项工作如何塑造了我们的研究轨迹，影响了更广泛的科学界的发展，并指导了创新、教育和工业实践。

引用次数: 0

Exploring cultural commonsense in multilingual large language models: A survey 探索多语言大语言模型中的文化常识：综述

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-12-01 DOI: 10.1016/j.is.2025.102649

Geleta Negasa Binegde, Huaping Zhang

Large language models (LLMs) have demonstrated impressive proficiency in multilingual natural language processing (NLP), yet they frequently struggle with cultural commonsense—the implicit knowledge shaped by societal norms, traditions, and shared experiences. As these models are deployed in diverse linguistic and cultural settings, their ability to understand and apply cultural commonsense becomes crucial for ensuring fairness, inclusivity, and contextual accuracy. This paper presents a systematic review and a large-scale empirical benchmark for evaluating cultural commonsense in multilingual LLMs. Through a comprehensive evaluation of 15 models on the BLEnD dataset, our analysis reveals a critical performance gap of 64.2% between high-resource and low-resource cultures. The results demonstrate significant disparities across model architectures: encoder-only models show more consistent but lower overall performance compared to decoder-based models. We identify key limitations, including data scarcity, representational bias, and inadequate cross-lingual knowledge transfer. Finally, we propose future research directions, such as culturally diverse dataset curation, hybrid knowledge graph architectures, and fairness-aware fine-tuning. The primary contributions of this work are (1) a systematic review of challenges and mitigation strategies for cultural commonsense; (2) a large-scale empirical benchmark that evaluates 15 multilingual LLMs across 13 languages and 16 countries, revealing significant performance disparities; and (3) concrete findings on the effects of model architecture and the limitations of scale in cultural understanding. This research underscores the urgent need to advance cultural commonsense in multilingual LLMs to ensure the development of fair, inclusive, and contextually accurate AI systems globally.

大型语言模型（llm）在多语言自然语言处理（NLP）方面表现出了令人印象深刻的熟练程度，但它们经常与文化常识（由社会规范、传统和共享经验形成的隐性知识）作斗争。由于这些模型被部署在不同的语言和文化环境中，它们理解和应用文化常识的能力对于确保公平性、包容性和上下文准确性至关重要。本文提出了一个系统的审查和大规模的经验基准评估文化常识在多语言法学硕士。通过对BLEnD数据集上的15个模型进行综合评估，我们的分析显示，高资源文化与低资源文化之间的关键绩效差距为64.2%。结果显示了模型架构之间的显著差异：与基于解码器的模型相比，只有编码器的模型显示出更一致但更低的整体性能。我们确定了关键的限制，包括数据稀缺、代表性偏见和跨语言知识转移不足。最后，我们提出了未来的研究方向，如多元文化数据集管理、混合知识图谱架构和公平感知微调。这项工作的主要贡献是：(1)对文化常识的挑战和缓解策略进行了系统的回顾；(2)对16个国家、13种语言的15名多语种法学硕士进行了大规模的实证基准评估，结果显示出显著的绩效差异；(3)关于模型建筑的作用和尺度在文化理解中的局限性的具体发现。这项研究强调了迫切需要在多语言法学硕士中推进文化常识，以确保在全球范围内开发公平、包容和准确的人工智能系统。

{"title":"Exploring cultural commonsense in multilingual large language models: A survey","authors":"Geleta Negasa Binegde, Huaping Zhang","doi":"10.1016/j.is.2025.102649","DOIUrl":"10.1016/j.is.2025.102649","url":null,"abstract":"<div><div>Large language models (LLMs) have demonstrated impressive proficiency in multilingual natural language processing (NLP), yet they frequently struggle with cultural commonsense—the implicit knowledge shaped by societal norms, traditions, and shared experiences. As these models are deployed in diverse linguistic and cultural settings, their ability to understand and apply cultural commonsense becomes crucial for ensuring fairness, inclusivity, and contextual accuracy. This paper presents a systematic review and a large-scale empirical benchmark for evaluating cultural commonsense in multilingual LLMs. Through a comprehensive evaluation of 15 models on the BLEnD dataset, our analysis reveals a critical performance gap of 64.2% between high-resource and low-resource cultures. The results demonstrate significant disparities across model architectures: encoder-only models show more consistent but lower overall performance compared to decoder-based models. We identify key limitations, including data scarcity, representational bias, and inadequate cross-lingual knowledge transfer. Finally, we propose future research directions, such as culturally diverse dataset curation, hybrid knowledge graph architectures, and fairness-aware fine-tuning. The primary contributions of this work are (1) a systematic review of challenges and mitigation strategies for cultural commonsense; (2) a large-scale empirical benchmark that evaluates 15 multilingual LLMs across 13 languages and 16 countries, revealing significant performance disparities; and (3) concrete findings on the effects of model architecture and the limitations of scale in cultural understanding. This research underscores the urgent need to advance cultural commonsense in multilingual LLMs to ensure the development of fair, inclusive, and contextually accurate AI systems globally.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102649"},"PeriodicalIF":3.4,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145652035","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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

IPDRM: A pyramid-based diffusion and contrastive learning framework for sequential recommendation IPDRM：一个基于金字塔的序列推荐扩散和对比学习框架

IF 3.4 2区计算机科学 Q2 COMPUTER SCIENCE, INFORMATION SYSTEMS

Information Systems

Pub Date : 2025-11-29 DOI: 10.1016/j.is.2025.102651

Ruijia Guo, Zhiyuan Chen

Sequential recommendation faces critical challenges in handling data sparsity, noise interference, and ineffective intent modeling. To address these issues, this paper proposes a novel Intent-aware Pyramid Diffusion Recommendation Model (IPDRM) that integrates hierarchical intent modeling with conditional diffusion-based augmentation. The framework employs a pyramid structure to capture multi-granular user intents (base-level item features, mid-level temporal patterns, and top-level semantic abstractions) and utilizes intent-conditioned diffusion to generate semantically consistent augmented views. Contrastive learning is then applied to align representations of original and augmented sequences. Extensive experiments on Tmall and Fliggy datasets demonstrate that IPDRM significantly outperforms state-of-the-art baselines, achieving improvements of up to 20.0 % in HR@5 and 22.5 % in NDCG@5. The model exhibits strong robustness in sparse and noisy scenarios, validated through comprehensive ablation studies and parameter sensitivity analyses. This work provides a effective solution for intent-aware sequential recommendation with both theoretical and practical contributions. The code for the paper is available at https://github.com/CLTCGUO/IPDRM.

顺序推荐在处理数据稀疏性、噪声干扰和无效的意图建模方面面临着严峻的挑战。为了解决这些问题，本文提出了一种新的意图感知金字塔扩散推荐模型（IPDRM），该模型将分层意图建模与基于条件扩散的增强相结合。该框架采用金字塔结构来捕获多粒度的用户意图（基本级项目特征、中级时间模式和顶级语义抽象），并利用意图条件扩散来生成语义一致的增强视图。然后应用对比学习来对齐原始序列和增强序列的表示。在天猫和Fliggy数据集上进行的大量实验表明，IPDRM显著优于最先进的基线，在HR@5和NDCG@5分别实现了20.0 %和22.5 %的改进。通过综合烧蚀研究和参数敏感性分析，该模型在稀疏和噪声情况下表现出较强的鲁棒性。本研究为意向感知顺序推荐提供了一种有效的解决方案，具有理论和实践意义。该论文的代码可在https://github.com/CLTCGUO/IPDRM上获得。

{"title":"IPDRM: A pyramid-based diffusion and contrastive learning framework for sequential recommendation","authors":"Ruijia Guo, Zhiyuan Chen","doi":"10.1016/j.is.2025.102651","DOIUrl":"10.1016/j.is.2025.102651","url":null,"abstract":"<div><div>Sequential recommendation faces critical challenges in handling data sparsity, noise interference, and ineffective intent modeling. To address these issues, this paper proposes a novel Intent-aware Pyramid Diffusion Recommendation Model (IPDRM) that integrates hierarchical intent modeling with conditional diffusion-based augmentation. The framework employs a pyramid structure to capture multi-granular user intents (base-level item features, mid-level temporal patterns, and top-level semantic abstractions) and utilizes intent-conditioned diffusion to generate semantically consistent augmented views. Contrastive learning is then applied to align representations of original and augmented sequences. Extensive experiments on Tmall and Fliggy datasets demonstrate that IPDRM significantly outperforms state-of-the-art baselines, achieving improvements of up to 20.0 % in HR@5 and 22.5 % in NDCG@5. The model exhibits strong robustness in sparse and noisy scenarios, validated through comprehensive ablation studies and parameter sensitivity analyses. This work provides a effective solution for intent-aware sequential recommendation with both theoretical and practical contributions. The code for the paper is available at <span><span>https://github.com/CLTCGUO/IPDRM</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":50363,"journal":{"name":"Information Systems","volume":"138 ","pages":"Article 102651"},"PeriodicalIF":3.4,"publicationDate":"2025-11-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145685680","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"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