Information and Software Technology最新文献

{"title":"Enhancing building environments: A digital twin approach for informed decision-making and better campus experiences","authors":"Gianni Tumedei,&nbsp;Chiara Ceccarini,&nbsp;Luca Giulianini,&nbsp;Giovanni Delnevo,&nbsp;Catia Prandi","doi":"10.1016/j.infsof.2026.108044","DOIUrl":"10.1016/j.infsof.2026.108044","url":null,"abstract":"<div><h3>Context:</h3><div>Smart buildings are foundational elements of smart cities, integrating technologies such as Building Information Modeling (BIM), Building Management Systems (BMS), and Digital Twins (DTs) to support design, construction, and operations. However, their potential to directly enhance occupant experience and inform individual decision-making remains relatively underexplored, particularly for non-technical users.</div></div><div><h3>Objective:</h3><div>This study investigates the effects of providing accessible, room-level environmental information to all building occupants using data from BIM/BMS systems. The goal is to support building occupants in their daily lives, increasing awareness of environmental conditions and offering added value for their decision-making processes.</div></div><div><h3>Methods:</h3><div>We designed and developed a system that combines real-time sensor readings with short-term environmental forecasts and a comfort computation algorithm. These data are visualized through an interactive wayfinding interface deployed on public touch displays across a university campus. We evaluated the system using a mixed-methods approach, gathering qualitative and quantitative feedback on information quality, design quality, usability, and perceived value.</div></div><div><h3>Results:</h3><div>The evaluation revealed strong interest from non-experts in accessing environmental information. Participants found the system informative, easy to use, and valuable in supporting their everyday activities on campus. The combination of real-time and forecasted data, comfort indicators, and wayfinding contributed to a positive user experience.</div></div><div><h3>Conclusion:</h3><div>Our findings demonstrate the feasibility and value of integrating environmental data into public, user-friendly smart buildings interfaces, highlighting the importance of accessible data in improving sustainability, efficiency, and occupant decision-making in smart campus digital twin environments.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108044"},"PeriodicalIF":4.3,"publicationDate":"2026-02-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174428","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}

{"title":"COTVD: A function-level vulnerability detection framework using chain-of-thought reasoning with large language models","authors":"Yinan Chen ,&nbsp;Xiangping Chen ,&nbsp;Yuan Huang ,&nbsp;Changlin Yang ,&nbsp;Lei Yun","doi":"10.1016/j.infsof.2026.108043","DOIUrl":"10.1016/j.infsof.2026.108043","url":null,"abstract":"<div><h3>Context:</h3><div>With the growth of open-source vulnerability databases, deep learning methods have been widely applied for vulnerability detection. However, most existing approaches focus on coarse-grained detection, predicting only whether a sample contains a vulnerability without providing detailed explanations.</div></div><div><h3>Objective:</h3><div>This study aims to propose a large language model (LLM)-based method for function-level vulnerability detection and analysis, providing detailed insights into detected vulnerabilities.</div></div><div><h3>Methods:</h3><div>We introduce CoTVD, which leverages the reasoning capabilities of LLMs through the Chain of Thought (CoT) approach. An instructive prompt strategy was designed to guide the model in analyzing data and control dependencies related to library function calls, enhancing detection flexibility and accuracy. CoTVD was evaluated on multiple LLMs, including GPT-4o-128K, GPT-3.5-Turbo, Gemini-1.5-Pro, Claude-3.5-Sonnet, Llama-3.1-405B, Qwen2-72B-Instruct-T, and DeepSeek-67B-T.</div></div><div><h3>Results:</h3><div>Experimental results show that CoTVD based on GPT-4o-128K outperforms other models, achieving a recall rate of 94.77%. CoTVD, however, generates a relatively high number of false positives due to its sensitivity to potential security risks, which include not only traditional vulnerabilities but also risks such as information leakage. The study further refines existing dataset labels into finer granularity. In a human evaluation with 10 experts on 200 samples, each participant identified, on average, 1.7 additional vulnerability samples and 1.1 lines of risky code when assisted by CoTVD, demonstrating its effectiveness in real-world scenarios.</div></div><div><h3>Conclusion:</h3><div>CoTVD enables function-level vulnerability detection with detailed analysis and proves effective in identifying vulnerabilities and code risks in practice, offering a novel approach for vulnerability detection and security analysis.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108043"},"PeriodicalIF":4.3,"publicationDate":"2026-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174456","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}

{"title":"Transformer-aware sequence-to-sequence network for personalized tag recommendation in software information sites","authors":"Shubhi Bansal ,&nbsp;Jahnavi Sunchu ,&nbsp;Shahid Shafi Dar,&nbsp;Nagendra Kumar","doi":"10.1016/j.infsof.2026.108061","DOIUrl":"10.1016/j.infsof.2026.108061","url":null,"abstract":"<div><h3>Context:</h3><div>Automatic tag recommendation is crucial for content understanding and retrieval on software information sites. Existing approaches formulate tag recommendation as either multi-label classification problem or sentence matching. However, multi-label classification by treating tags as independent labels, neglects semantic relationships and dependencies among them, leading to inconsistent recommendations. Sentence matching techniques, which rely on lexical similarity, fail to capture contextual information and broader semantic meaning. Moreover, several works leverage limited content sources such as title, body, and code snippet of software objects, leading to data sparsity issues. Extant research provides generic tag suggestions, overlooking users’ expertise and interests.</div></div><div><h3>Objective:</h3><div>To address these limitations, we propose a novel tr<u><strong>AN</strong></u>sformer-based seque<u><strong>N</strong></u>ce-to-sequence framework for pers<u><strong>O</strong></u>nlaised <u><strong>TA</strong></u>g recommenda<u><strong>TION</strong></u>, dubbed as ANNOTATION.</div></div><div><h3>Methods:</h3><div>This approach enables the model to learn dependencies between tags and generate contextually relevant recommendations. To enhance the representation of software objects and mitigate data sparsity, we incorporate valuable information from associated comments, such as clarifications, usage examples, and bug reports. This additional conversational context provides insights into the problem, alternative solutions, and related concepts discussed by the community, resulting in more informed tag recommendations. Furthermore, we personalize tag suggestions by incorporating user profile descriptions and badges, which reflect users’ expertise and interests within specific domains. This ensures that generated tags align with both the software object and the user’s specific knowledge domain, contributing to a more tailored user experience.</div></div><div><h3>Results:</h3><div>Extensive empirical and qualitative evaluations on datasets from Code Review and Stack Overflow demonstrate that our approach significantly outperforms state-of-the-art methods.</div></div><div><h3>Conclusion:</h3><div>Our findings highlight the importance of considering tag dependencies, contextual information, and user preferences for accurate and personalized tag recommendation in software information sites.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108061"},"PeriodicalIF":4.3,"publicationDate":"2026-01-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174455","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}

{"title":"Template-guided interpretable reasoning with execution feedback for LLM-based program repair","authors":"Sichong Hao ,&nbsp;Xianjun Shi ,&nbsp;Hongwei Liu ,&nbsp;Yuyang Yin ,&nbsp;Xi Chen","doi":"10.1016/j.infsof.2026.108058","DOIUrl":"10.1016/j.infsof.2026.108058","url":null,"abstract":"<div><h3>Context:</h3><div>Automated Program Repair (APR) seeks to fix software defects automatically. Large language models (LLMs) show promise in APR, especially when combined with traditional template-based methods. However, existing approaches suffer from low accuracy, poor interpretability, and incompatibility with mainstream closed-source LLMs, leaving the synergy between traditional methods and LLMs underexplored.</div></div><div><h3>Objectives</h3><div>: We propose TRACE, a template-guided reasoning framework that enhances LLMs’ bug-fixing performance across multiple programming languages, while improving repair interpretability to help developers understand and validate generated patches.</div></div><div><h3>Methods:</h3><div>Our approach is novel in integrating both the successful patterns from repair templates across multiple programming languages and the failure experiences from previous execution results into the LLM-driven repair reasoning process. TRACE comprises three components: (1) a multilingual repair template repository that covers 47 common error types and their corresponding repair actions; (2) a repair reasoning engine that, under template guidance, first analyzes root causes and then formulates concrete repair strategies; and (3) a feedback memory buffer that dynamically records execution data from failed attempts to guide subsequent repair iterations. Moreover, our approach enhances the interpretability of results by providing a reasoning trace that explains each patch.</div></div><div><h3>Results</h3><div>: Extensive experiments on the widely used Defects4J and recent multilingual DebugBench benchmarks demonstrate that TRACE achieves state-of-the-art repair performance. TRACE successfully repairs 204 bugs in Defects4J, outperforming the best baseline APR approach by 14.6%. On DebugBench, TRACE demonstrates its multilingual repair ability by achieving state-of-the-art results in Python, Java, and C++.</div></div><div><h3>Conclusions:</h3><div>Our approach bridges the gap between template-based methods and mainstream LLMs by effectively integrating different types of domain knowledge (i.e., common error types, corresponding repair templates, and execution feedback). This not only enhances repair performance across multiple programming languages but also improves result interpretability. Overall, our research highlights the promising prospects of integrating APR-specific domain knowledge with LLM reasoning in practice.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108058"},"PeriodicalIF":4.3,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174453","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}

{"title":"CONKER: CONtrastive knowledge enhanced retrieval for app review bug classification","authors":"Meysam Roostaee ,&nbsp;Seyed Mostafa Fakhrahmad ,&nbsp;Mohammad Hadi Sadreddini","doi":"10.1016/j.infsof.2026.108056","DOIUrl":"10.1016/j.infsof.2026.108056","url":null,"abstract":"<div><h3>Context</h3><div>Accurate and timely identification of software defects from user-generated app reviews is crucial for efficient software maintenance. However, automated classifiers face challenges with the volume and heterogeneity of review text, often struggling with generalization, decision transparency, and learning from limited labeled examples per bug category.</div></div><div><h3>Objective</h3><div>This paper introduces CONKER (CONtrastive Knowledge Enhanced Retrieval), a novel framework to address these issues by improving accuracy and providing inherent interpretability for multi-class app review bug classification.</div></div><div><h3>Methods</h3><div>CONKER employs a two-stage process. First, a transformer-based sentence encoder is fine-tuned using supervised contrastive learning (Multiple Negatives Ranking Loss) on the MABLE benchmark to create a discriminative embedding space. Second, a classification-by-retrieval mechanism assigns labels by matching new reviews against curated class-specific exemplars. The framework is evaluated using 10-fold cross-validation, an ablation study, and a human-in-the-loop study with professional developers.</div></div><div><h3>Results</h3><div>On the MABLE dataset, CONKER achieves a new state-of-the-art weighted F1-score of 0.8133. Critically, it also attains a Macro F1-score over 13 percentage points higher than the strongest baseline, demonstrating superior performance on minority classes. A new zero-shot experiment provides quantitative evidence of its strong cross-platform generalization, where CONKER outperforms a RoBERTa baseline by 10.1 percentage points on a developer-annotated Apple App Store dataset. This is complemented by a human-in-the-loop study that validated the practical utility of CONKER's exemplar-based explanations. Ablation studies confirm that both the contrastive fine-tuning and the per-class retrieval architecture are fundamental to this multi-faceted success.</div></div><div><h3>Conclusion</h3><div>Contrastive learning-driven retrieval is a scalable, interpretable, and high-fidelity approach for nuanced text categorization in empirical software engineering. The CONKER framework demonstrates that by engineering a task-specific semantic space, it is possible to reconcile high predictive accuracy with the transparency required for practical software maintenance workflows.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108056"},"PeriodicalIF":4.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081167","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}

{"title":"A first and fast symbolic approach for data-aware business process compliance checking","authors":"Silvano Colombo Tosatto ,&nbsp;Hannah Burke ,&nbsp;Nick van Beest ,&nbsp;Heerko Groefsema","doi":"10.1016/j.infsof.2026.108047","DOIUrl":"10.1016/j.infsof.2026.108047","url":null,"abstract":"<div><div>State of the art design-time compliance checking techniques are generally limited to verifying control-flow relations. In this paper, we propose an algorithm for verifying design-time compliance of business process models including data. For a faithful resemblance to real-world cases, we include a degree of uncertainty in the outcomes of activities, such that a single execution is not guaranteed to lead to a unique data state. We represent the compliance requirements using conditional obligations. Our proposed approach uses a symbolic representation to compactly represent the possible data states across an execution, and evaluates them as a constraint satisfaction problem to first prune the state space of paths that require no further analysis and finally to evaluate if an execution violates one of the regulatory requirements. The performance of the approach is evaluated against existing algorithms implemented in NuXMV. The results show that our approach significantly outperforms the existing algorithms across the board.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108047"},"PeriodicalIF":4.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174452","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}

{"title":"Synergistic enhancement of requirement-to-code traceability: A framework combining large language model based data augmentation and an advanced encoder","authors":"Jianzhang Zhang ,&nbsp;Jialong Zhou ,&nbsp;Nan Niu ,&nbsp;Jinping Hua ,&nbsp;Chuang Liu","doi":"10.1016/j.infsof.2026.108045","DOIUrl":"10.1016/j.infsof.2026.108045","url":null,"abstract":"<div><h3>Context:</h3><div>Automated requirement-to-code traceability link recovery, essential for industrial system quality and safety, is critically hindered by the scarcity of labeled data, especially for methods relying on supervised fine-tuning.</div></div><div><h3>Objective:</h3><div>To address this bottleneck, this paper proposes and validates a synergistic framework that integrates large language model (LLM)-driven data augmentation with an advanced encoder.</div></div><div><h3>Methods:</h3><div>We first optimize data augmentation through a systematic evaluation of bi-directional and zero/few-shot prompting strategies. Building on the augmented data, we enhance an established state-of-the-art pre-trained language model based method by incorporating an encoder distinguished by a broader pre-training corpus and an extended context window. We conduct experiments on four public datasets, including rigorous validity checks and cross-project evaluations, to quantify the distinct contributions of our framework’s components.</div></div><div><h3>Results:</h3><div>We demonstrate that the choice among leading LLMs is not a significant performance factor, whereas the prompting strategy is effective. Data augmentation on its own consistently improves the baseline method, providing substantial performance gains of up to 26.66% in average <span><math><msub><mrow><mi>F</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span> score, while applying it to the advanced encoder unlocks a further maximum 10.00% lift in recall. This synergy culminates in a fully optimized framework that decisively outperforms ten established baselines with maximum improvements of 23.13% on <span><math><msub><mrow><mi>F</mi></mrow><mrow><mn>1</mn></mrow></msub></math></span> score and 21.37% on <span><math><msub><mrow><mi>F</mi></mrow><mrow><mn>2</mn></mrow></msub></math></span> score, and achieves a recall of 50.70% in strict cross-project evaluations, significantly surpassing the 21.82% recall of a state-of-the-art RAG approach.</div></div><div><h3>Conclusion:</h3><div>This work contributes a pragmatic and scalable methodology to overcome the data scarcity bottleneck, paving the way for broader industrial adoption of data-driven requirement-to-code traceability.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108045"},"PeriodicalIF":4.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081166","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}

{"title":"A dynamic software start-up competence model","authors":"Nana Assyne ,&nbsp;Alfred Nyadroh ,&nbsp;Emmanuel Adabor ,&nbsp;Emmanuel Antwi-Boasiako ,&nbsp;Isaac Wiafe","doi":"10.1016/j.infsof.2026.108057","DOIUrl":"10.1016/j.infsof.2026.108057","url":null,"abstract":"<div><h3>Context</h3><div>Software startups are important drivers of innovation, job creation, and digital transformation. However, more than 60% fail within the first five years, often due to gaps in competencies that change as the organization evolves.</div></div><div><h3>Objective</h3><div>This study proposes and validates a dynamic competence model for software startups, aligned with the Crowne lifecycle framework comprising Startup, Stabilization, Growth, and Maturity stages.</div></div><div><h3>Method</h3><div>A mixed methods approach was applied, combining qualitative interviews with quantitative surveys. The study identifies evolving competencies across three domains: business, architecture, and innovation.</div></div><div><h3>Results</h3><div>The findings reveal that competencies shift from being “desired” to “required” as startups progress through lifecycle stages. In the Startup stage, business competencies such as corporate structuring are essential, while project management and customer support become critical during Growth and Maturity. Technical competencies like programming and algorithms are vital early on, with advanced skills such as application security gaining importance during scaling. Innovation competencies, particularly problem solving and persistence, remain consistently critical, while creative and abstract thinking change in emphasis as startups mature.</div></div><div><h3>Conclusion</h3><div>The dynamic competence model addresses gaps in static frameworks by offering actionable insights for practitioners to prioritize skill development and resource allocation. It also provides a foundation for researchers to study the evolution of competencies in entrepreneurial contexts. The study recommends extending the model to other sectors and conducting longitudinal studies to enhance its applicability, ultimately contributing to more sustainable software engineering practices and improved startup success.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108057"},"PeriodicalIF":4.3,"publicationDate":"2026-01-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146174426","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}

{"title":"Broadening software value from technical to management perspectives: a stakeholder-driven exploration using a modified-Delphi approach","authors":"Chalani Oruthotaarachchi,&nbsp;Janaka Wijayanayake","doi":"10.1016/j.infsof.2026.108046","DOIUrl":"10.1016/j.infsof.2026.108046","url":null,"abstract":"<div><h3>Context</h3><div>Software value is increasingly recognized as a multi-dimensional construct, encompassing technical performance, business alignment, and user-centric factors. While previous research has explored individual aspects of software value such as functionality, cost, and usability, an integrated and holistic discussion remains lacking.</div></div><div><h3>Objectives</h3><div>This research is conducted to emphasize a holistic view of software value through investigating value perspectives of diverse stakeholder categories, shared determinants enabling value alignment, and the external forces affecting such value perceptions.</div></div><div><h3>Methods</h3><div>A hybrid consensus-building approach that combines Delphi approach with Group Discussions was followed. Software process experts, technical experts, and enterprise system users expressed their opinions.</div></div><div><h3>Results</h3><div>Findings reveal distinct stakeholder priorities, where process experts emphasize strategic alignment, enterprise users focus on efficiency, and technical experts prioritize functional integrity. These priorities are influenced by system scope, technological constraints, and business domain. Synthesizing these insights, the study presents a structured software value model, offering practical guidance for practitioners and organizations to align development priorities with stakeholder expectations.</div></div><div><h3>Conclusion</h3><div>Software value is not a static description but differs depending on the role of stakeholders, their workplace contexts, personal priorities, and external circumstances. Further research is needed to investigate how stakeholders’ value perceptions change over time during the software development life cycle.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108046"},"PeriodicalIF":4.3,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146080611","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}

{"title":"An architectural perspective on MLOps: Structures, processes, tools, and stakeholders","authors":"Faezeh Amou Najafabadi,&nbsp;Justus Bogner,&nbsp;Ilias Gerostathopoulos,&nbsp;Patricia Lago","doi":"10.1016/j.infsof.2026.108029","DOIUrl":"10.1016/j.infsof.2026.108029","url":null,"abstract":"<div><h3>Context:</h3><div>Despite the increasing adoption of Machine Learning Operations (MLOps), teams still encounter challenges in effectively applying this paradigm to their projects. While numerous MLOps tools exist, consolidated knowledge to inform architecture design is still lacking.</div></div><div><h3>Objective:</h3><div>In response, our goal is to provide a comprehensive overview of MLOps architectures from a structural and process perspective, the tools mentioned supporting the implementation of architecture components, and the stakeholders responsible for the MLOps process.</div></div><div><h3>Methods:</h3><div>We conduct a systematic mapping study of 93 primary studies to collect and analyze the state of the art knowledge on MLOps systems using automatic, manual, and snowballing-based search strategies. Subsequently, we use card sorting to synthesize the results.</div></div><div><h3>Results:</h3><div>We contribute: (i) a categorization of 39 MLOps architecture components and a description of several MLOps architecture variants; (ii) a systematic map between the components and the existing MLOps tools; (iii) a description of 56 process steps for MLOps systems creation, deployment, and maintenance; and (iv) a description of MLOps stakeholders, their responsibilities, and a systematic map between the process steps and the responsible stakeholders.</div></div><div><h3>Conclusion:</h3><div>Our results serve as an overview of the state of the art in MLOps architectures from a structural and a process perspective to support researchers and practitioners in the architecture design of their MLOps systems.</div></div>","PeriodicalId":54983,"journal":{"name":"Information and Software Technology","volume":"193 ","pages":"Article 108029"},"PeriodicalIF":4.3,"publicationDate":"2026-01-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146081169","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}