Matteo Esposito
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Matteo Esposito has a total of 77 co-authors. Here are the top co-authors by number of shared publications:
- Valentina Lenarduzzi
University of Oulu
30 shared publications - Davide Taibi
University of Southern Denmark
26 shared publications - Alexander Bakhtin
University of Oulu
12 shared publications - Mikel Robredo
University of Oulu, Oulu, Finland
12 shared publications - Noman Ahmad
University of Oulu Finland
8 shared publications - Davide Falessi
University of Rome Tor Vergata, Rome, Italy
6 shared publications
Latest Publications
- 9 October 2026
- Published by Springer Nature
- p. 174-182
Analyzing the Evolution of Structural Communities Within Microservice Architecture
- 9 October 2026
- Published by Springer Nature
- p. 174-182
Analyzing the Evolution of Structural Communities Within Microservice Architecture
- Open Access
- 3 September 2026
- Published by Springer Nature in Software Quality Journal
- Vol. 34 (3) , 50
On the Cohesion of Hub-like microservices
Microservice architecture has gained traction as a powerful way to build distributed, scalable systems. However, analyzing its quality requires considering aspects unique to this architectural style. In particular, cohesion of microservices is considered important for evaluating microservice maintainability and quality, while few such metrics are available. Hub-like microservices are defined as an architectural anti-pattern, assumed to emerge when services centralize too many dependencies or responsibilities, thereby exhibiting low cohesion. However, this assumption has not been empirically validated. This study investigates whether Hub-like microservices indeed display lower internal cohesion than other services. We analyzed 123 microservices from a dataset of 25 microservice systems by leveraging an existing labeling of Hub-like microservices and computing eight cohesion metrics. We examined correlations between cohesion and several Hub-like detection techniques, including centrality and clustering-based metrics, the Minimum Description Length principle, and the Arcan approach. Our findings reveal weak to moderate negative correlations between service interface data cohesion and network centrality. Conversely, class relation cohesion is positively correlated with clustering-based Hub-like indicators, as well as shows a relation to all categorical Hub-like indicators with a positive effect size, due to the opposite interpretation of this metric. Thus, the common assumption that Hub-like services suffer from poor internal cohesion is not supported by our findings, with only a specific combination of a Hub-like detection method and a cohesion metric behaving as conjectured.
- 3 September 2026
- Published by Springer Nature in Software Quality Journal
- Vol. 34 (3) , 50
Open AccessOn the Cohesion of Hub-like microservices
Microservice architecture has gained traction as a powerful way to build distributed, scalable systems. However, analyzing its quality requires considering aspects unique to this architectural style. In particular, cohesion of microservices is considered important for evaluating microservice maintainability and quality, while few such metrics are available. Hub-like microservices are defined as an architectural anti-pattern, assumed to emerge when services centralize too many dependencies or responsibilities, thereby exhibiting low cohesion. However, this assumption has not been empirically validated. This study investigates whether Hub-like microservices indeed display lower internal cohesion than other services. We analyzed 123 microservices from a dataset of 25 microservice systems by leveraging an existing labeling of Hub-like microservices and computing eight cohesion metrics. We examined correlations between cohesion and several Hub-like detection techniques, including centrality and clustering-based metrics, the Minimum Description Length principle, and the Arcan approach. Our findings reveal weak to moderate negative correlations between service interface data cohesion and network centrality. Conversely, class relation cohesion is positively correlated with clustering-based Hub-like indicators, as well as shows a relation to all categorical Hub-like indicators with a positive effect size, due to the opposite interpretation of this metric. Thus, the common assumption that Hub-like services suffer from poor internal cohesion is not supported by our findings, with only a specific combination of a Hub-like detection method and a cohesion metric behaving as conjectured.
- 29 August 2026
- Published by Springer Nature
- p. 295-311
Towards LLM Accelerated Rapid Reviews for Software Tool Discovery - Case for Log Anomaly Detection
- 29 August 2026
- Published by Springer Nature
- p. 295-311
Towards LLM Accelerated Rapid Reviews for Software Tool Discovery - Case for Log Anomaly Detection
- Open Access
- 8 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 507-518
Descriptor: An Improved Microservice Dataset of Logs and Metrics (LO2v2)
LO2v2 is an updated and improved multi-modal dataset capturing logs, metrics, and traces from a production-ready microservice system implementing the OAuth2.0 authorization framework. The dataset extends a prior version by introducing a redesigned data collection pipeline that enhances reliability, reduces sources of data leakage, and increases the diversity and representativeness of recorded system behaviors. LO2v2 contains 640GB of data with 115 independent runs of a comprehensive test suite composed of 54 correct and error-inducing API scenarios executed over seven microservices and a MySQL database. In total, there are around 2.2 million metric files and 80,000 log files containing almost 2.6 billion log lines. Test execution order, duration, and inter-test idle periods are randomized to better reflect realistic system activity. The dataset includes 1136 container-level and 49 host-level metrics sampled at one-second intervals and extensive Docker-level service logs. However, due to limitations of injecting the Jaeger tracing agent, the maximum number of traces obtained during a test for each service is 20. All data are stored in plaintext formats, with metrics additionally provided as CSV files to facilitate time-series analysis. LO2v2 is designed to support research on anomaly detection, multi-modal observability, fault analysis, and benchmarking for microservice-based systems. The dataset offers a reproducible and extensible foundation for studies aiming to evaluate monitoring models, explore multi-modal fusion approaches, or develop new techniques for analyzing complex distributed architectures.
- 8 June 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 507-518
Open AccessDescriptor: An Improved Microservice Dataset of Logs and Metrics (LO2v2)
LO2v2 is an updated and improved multi-modal dataset capturing logs, metrics, and traces from a production-ready microservice system implementing the OAuth2.0 authorization framework. The dataset extends a prior version by introducing a redesigned data collection pipeline that enhances reliability, reduces sources of data leakage, and increases the diversity and representativeness of recorded system behaviors. LO2v2 contains 640GB of data with 115 independent runs of a comprehensive test suite composed of 54 correct and error-inducing API scenarios executed over seven microservices and a MySQL database. In total, there are around 2.2 million metric files and 80,000 log files containing almost 2.6 billion log lines. Test execution order, duration, and inter-test idle periods are randomized to better reflect realistic system activity. The dataset includes 1136 container-level and 49 host-level metrics sampled at one-second intervals and extensive Docker-level service logs. However, due to limitations of injecting the Jaeger tracing agent, the maximum number of traces obtained during a test for each service is 20. All data are stored in plaintext formats, with metrics additionally provided as CSV files to facilitate time-series analysis. LO2v2 is designed to support research on anomaly detection, multi-modal observability, fault analysis, and benchmarking for microservice-based systems. The dataset offers a reproducible and extensible foundation for studies aiming to evaluate monitoring models, explore multi-modal fusion approaches, or develop new techniques for analyzing complex distributed architectures.
- Open Access
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Software
- Vol. 43 (3) , 106-110
Agents Don’t Just Think, They Communicate: Communication and Coordination Protocols for Agentic AI
This article compares agent communication and coordination protocols for agent–tool and agent–agent interaction, highlights the key tradeoffs in building interoperable, observable, and governable agent ecosystems, and provides guidance on selecting communication protocols based on practical usage scenarios.
- 20 May 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Software
- Vol. 43 (3) , 106-110
Open AccessAgents Don’t Just Think, They Communicate: Communication and Coordination Protocols for Agentic AI
This article compares agent communication and coordination protocols for agent–tool and agent–agent interaction, highlights the key tradeoffs in building interoperable, observable, and governable agent ecosystems, and provides guidance on selecting communication protocols based on practical usage scenarios.
- Open Access
- 7 May 2026
- Published by Springer Nature in Empirical Software Engineering
- Vol. 31 (5) , 134
Analyzing the ripple effects of refactoring
Context: Short development cycles and continuous delivery pressure often push developers toward expedients that lead to poor design and hard-to-maintain systems. A common remedy is code refactoring, which reduces complexity and improves maintainability, though often seen as costly and risky. Objective: We investigate the long-term effects of refactoring to provide recommendations that support strategic development decisions. Method: We assess refactoring impact through change- and defect-proneness analysis, as well as benefit/effort evaluation. Results: Most refactorings have short-lived effects, persisting for fewer than 10 changes. Structural refactorings may last over 190 changes, with significant differences across families. Medium-lived refactorings (9–19 changes) prove the most stable and efficient, while longer-lasting ones become increasingly defect-prone and costly. Conclusions: Refactorings differ in sustainability. Medium-duration refactorings strike the best balance between stability and maintenance cost, while structural ones, though impactful, pose higher long-term risks. These insights guide prioritization of refactoring types to maximize benefit and minimize technical debt.
- 7 May 2026
- Published by Springer Nature in Empirical Software Engineering
- Vol. 31 (5) , 134
Open AccessAnalyzing the ripple effects of refactoring
Context: Short development cycles and continuous delivery pressure often push developers toward expedients that lead to poor design and hard-to-maintain systems. A common remedy is code refactoring, which reduces complexity and improves maintainability, though often seen as costly and risky. Objective: We investigate the long-term effects of refactoring to provide recommendations that support strategic development decisions. Method: We assess refactoring impact through change- and defect-proneness analysis, as well as benefit/effort evaluation. Results: Most refactorings have short-lived effects, persisting for fewer than 10 changes. Structural refactorings may last over 190 changes, with significant differences across families. Medium-lived refactorings (9–19 changes) prove the most stable and efficient, while longer-lasting ones become increasingly defect-prone and costly. Conclusions: Refactorings differ in sustainability. Medium-duration refactorings strike the best balance between stability and maintenance cost, while structural ones, though impactful, pose higher long-term risks. These insights guide prioritization of refactoring types to maximize benefit and minimize technical debt.
- Open Access
- 14 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Too Much AI, Too Little Trust: What the Software Engineering Community Really Thinks
Software engineering (SE) research is increasingly focusing on AI for SE. Yet, the community also raises major concerns about how AI is developed, deployed, and used. Trustworthy AI (TAI) offers a framework to address these concerns through lawful, ethical, and robust AI across the AI lifecycle. In this study, we analyze ICSE-FOSE 2026 survey responses to characterize the SE community’s opinion on AI in SE, and to examine how the expressed concerns align with potential ethical value breaches and TAI lifecycle tasks. Results show a largely negative opinion of AI in SE. Reported issues map to 17 ethical values, with the strongest concentration on problems arising during AI use rather than during model development and deployment. These findings suggest an urgent need for actionable guidance and research on the ethical use of AI in SE practice to address technical and ethical shortcomings and strengthen trust in AI-enabled SE.
- 14 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessToo Much AI, Too Little Trust: What the Software Engineering Community Really Thinks
Software engineering (SE) research is increasingly focusing on AI for SE. Yet, the community also raises major concerns about how AI is developed, deployed, and used. Trustworthy AI (TAI) offers a framework to address these concerns through lawful, ethical, and robust AI across the AI lifecycle. In this study, we analyze ICSE-FOSE 2026 survey responses to characterize the SE community’s opinion on AI in SE, and to examine how the expressed concerns align with potential ethical value breaches and TAI lifecycle tasks. Results show a largely negative opinion of AI in SE. Reported issues map to 17 ethical values, with the strongest concentration on problems arising during AI use rather than during model development and deployment. These findings suggest an urgent need for actionable guidance and research on the ethical use of AI in SE practice to address technical and ethical shortcomings and strengthen trust in AI-enabled SE.
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 615-619
SQuaD: The Software Quality Dataset
Software quality research increasingly relies on large-scale datasets that measure both the product and process aspects of software systems. However, existing resources often focus on limited dimensions, such as code smells, technical debt, or refactoring activity, thereby restricting comprehensive analyses across isolated quality dimensions. To address this gap, we present the Software Quality Dataset (SQuaD), a multi-dimensional, time-aware collection of software quality metrics extracted from 450 mature open-source projects across diverse ecosystems, including Apache, Mozilla, FFmpeg, and the Linux kernel. By integrating nine state-of-the-art static analysis tools, i.e., SonarQube, CodeScene, PMD, Understand, CK, JaSoMe, RefactoringMiner, RefactoringMiner++, and PyRef, our dataset unifies over 700 unique metrics at method, class, file, and project levels. Covering a total of 63,586 analyzed project releases, SQuaD also provides version control and issue-tracking histories, software vulnerability data (CVE/CWE), and process metrics proven to enhance Just-In-Time (JIT) defect prediction. The SQuaD enables empirical research on maintainability, technical debt, software evolution, and quality assessment at unprecedented scale. We also outline emerging research directions, including automated dataset updates and cross-project quality modeling to support the continuous evolution of software analytics. The dataset is publicly available on ZENODO (DOI: 10.5281/zenodo.17566690).
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 615-619
Open AccessSQuaD: The Software Quality Dataset
Software quality research increasingly relies on large-scale datasets that measure both the product and process aspects of software systems. However, existing resources often focus on limited dimensions, such as code smells, technical debt, or refactoring activity, thereby restricting comprehensive analyses across isolated quality dimensions. To address this gap, we present the Software Quality Dataset (SQuaD), a multi-dimensional, time-aware collection of software quality metrics extracted from 450 mature open-source projects across diverse ecosystems, including Apache, Mozilla, FFmpeg, and the Linux kernel. By integrating nine state-of-the-art static analysis tools, i.e., SonarQube, CodeScene, PMD, Understand, CK, JaSoMe, RefactoringMiner, RefactoringMiner++, and PyRef, our dataset unifies over 700 unique metrics at method, class, file, and project levels. Covering a total of 63,586 analyzed project releases, SQuaD also provides version control and issue-tracking histories, software vulnerability data (CVE/CWE), and process metrics proven to enhance Just-In-Time (JIT) defect prediction. The SQuaD enables empirical research on maintainability, technical debt, software evolution, and quality assessment at unprecedented scale. We also outline emerging research directions, including automated dataset updates and cross-project quality modeling to support the continuous evolution of software analytics. The dataset is publicly available on ZENODO (DOI: 10.5281/zenodo.17566690).
- Open Access
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1036-1037
Running Large Language Models at Scale for Mining Software Repositories: Lessons Learned from HPC-Based Batch Inference
The rapid diffusion of Large Language Models (LLMs) is fundamentally changing how Mining Software Repositories (MSR) research is conducted, particularly for studies that rely on unstructured textual artifacts such as commit messages, issue discussions, pull request reviews, and practitioner-generated content. While recent work has demonstrated the potential of LLMs to support classification, summarization, and qualitative analysis tasks, the majority of existing approaches rely on interactive or API-based executions [2, 3, 6]. Such execution models are poorly suited for large-scale empirical MSR studies, where thousands or hundreds of thousands of artifacts must be processed in a controlled, reproducible, and cost-aware manner. This tutorial addresses a central methodological challenge now faced by the MSR community: how to operationalize LLM-based analyses at repository scale without sacrificing empirical rigor. Drawing from multiple large-scale empirical studies, we report lessons learned from executing LLM-driven analyses on supercomputing infrastructure using batch-oriented workflows. In this context, we focus on two technologies that represent de-facto standards in modern high-performance computing (HPC) environments. First, we discuss SLURM batch scheduling via sbatch [4], the most widely adopted workload manager across academic and industrial HPC centers, which enables deterministic execution, explicit resource allocation, and repeatable experimentation [4]. Second, we present vLLM [8], a high-throughput, memory-efficient inference engine that has rapidly emerged as a reference solution for serving large language models on shared GPU infrastructures, enabling efficient utilization of scarce accelerator resources [5, 7]. Through concrete examples grounded in MSR datasets, the tutorial demonstrates how batch inference fundamentally reshapes experimental design. Rather than treating LLMs as interactive tools, batch execution forces researchers to define explicit inputs, prompts, execution parameters, and output schemas upfront, thereby making LLM usage a first-class methodological component of the empirical pipeline. We highlight common pitfalls observed in ad-hoc LLM usage, including uncontrolled variability, hidden costs, and irreproducible results, and show how HPC-based execution mitigates these issues. Importantly, the tutorial emphasizes that LLM outputs should be treated as judgments rather than ground truth. We discuss validation strategies that combine multiple independent LLMs with targeted human-in-the-loop sampling to balance scalability and reliability. These strategies enable researchers to quantify disagreement, identify systematic failure modes, and maintain human oversight where it is most needed, without compromising the benefits of large-scale automation. This tutorial bridges MSR methodology and widely used HPC execution models, equipping researchers with transferable, infrastructure-agnostic practices to responsibly integrate LLMs into repository-level studies. It highlights new opportunities for rigorous, reproducible, and scalable LLM-driven empirical software engineering research and lays the groundwork for future MSR methodological standards. Acknowledgments. This work has been partially funded by projects Business Finland 6GSoft [1], the Research Council of Finland MuFAno (grants n. 349487 and 349488), and the Ministry of Education and Culture, Finland FAST (Finnish Software Engineering Doctoral Research Network).
- 13 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 1036-1037
Open AccessRunning Large Language Models at Scale for Mining Software Repositories: Lessons Learned from HPC-Based Batch Inference
The rapid diffusion of Large Language Models (LLMs) is fundamentally changing how Mining Software Repositories (MSR) research is conducted, particularly for studies that rely on unstructured textual artifacts such as commit messages, issue discussions, pull request reviews, and practitioner-generated content. While recent work has demonstrated the potential of LLMs to support classification, summarization, and qualitative analysis tasks, the majority of existing approaches rely on interactive or API-based executions [2, 3, 6]. Such execution models are poorly suited for large-scale empirical MSR studies, where thousands or hundreds of thousands of artifacts must be processed in a controlled, reproducible, and cost-aware manner. This tutorial addresses a central methodological challenge now faced by the MSR community: how to operationalize LLM-based analyses at repository scale without sacrificing empirical rigor. Drawing from multiple large-scale empirical studies, we report lessons learned from executing LLM-driven analyses on supercomputing infrastructure using batch-oriented workflows. In this context, we focus on two technologies that represent de-facto standards in modern high-performance computing (HPC) environments. First, we discuss SLURM batch scheduling via sbatch [4], the most widely adopted workload manager across academic and industrial HPC centers, which enables deterministic execution, explicit resource allocation, and repeatable experimentation [4]. Second, we present vLLM [8], a high-throughput, memory-efficient inference engine that has rapidly emerged as a reference solution for serving large language models on shared GPU infrastructures, enabling efficient utilization of scarce accelerator resources [5, 7]. Through concrete examples grounded in MSR datasets, the tutorial demonstrates how batch inference fundamentally reshapes experimental design. Rather than treating LLMs as interactive tools, batch execution forces researchers to define explicit inputs, prompts, execution parameters, and output schemas upfront, thereby making LLM usage a first-class methodological component of the empirical pipeline. We highlight common pitfalls observed in ad-hoc LLM usage, including uncontrolled variability, hidden costs, and irreproducible results, and show how HPC-based execution mitigates these issues. Importantly, the tutorial emphasizes that LLM outputs should be treated as judgments rather than ground truth. We discuss validation strategies that combine multiple independent LLMs with targeted human-in-the-loop sampling to balance scalability and reliability. These strategies enable researchers to quantify disagreement, identify systematic failure modes, and maintain human oversight where it is most needed, without compromising the benefits of large-scale automation. This tutorial bridges MSR methodology and widely used HPC execution models, equipping researchers with transferable, infrastructure-agnostic practices to responsibly integrate LLMs into repository-level studies. It highlights new opportunities for rigorous, reproducible, and scalable LLM-driven empirical software engineering research and lays the groundwork for future MSR methodological standards. Acknowledgments. This work has been partially funded by projects Business Finland 6GSoft [1], the Research Council of Finland MuFAno (grants n. 349487 and 349488), and the Ministry of Education and Culture, Finland FAST (Finnish Software Engineering Doctoral Research Network).
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 501-505
Sovereign-by-Design: A Reference Architecture for AI and Blockchain Enabled Systems
Digital sovereignty has emerged as a central concern for modern software-intensive systems, driven by the dominance of non-sovereign cloud infrastructures, the rapid adoption of Generative AI, and increasingly stringent regulatory requirements. While existing initiatives address governance, compliance, and security in isolation, they provide limited guidance on how sovereignty can be operationalized at the architectural level. In this paper, we argue that sovereignty must be treated as a first-class architectural property rather than a purely regulatory objective. We introduce a Sovereign Reference Architecture that integrates self-sovereign identity, blockchain-based trust and auditability, sovereign data governance, and Generative AI deployed under explicit architectural control. The architecture explicitly captures the dual role of Generative AI as both a source of governance risk and an enabler of compliance, accountability, and continuous assurance when properly constrained. By framing sovereignty as an architectural quality attribute, our work bridges regulatory intent and concrete system design, offering a coherent foundation for building auditable, evolvable, and jurisdiction-aware AI-enabled systems. The proposed reference architecture provides a principled starting point for future research and practice at the intersection of software architecture, Generative AI, and digital sovereignty.
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 501-505
Open AccessSovereign-by-Design: A Reference Architecture for AI and Blockchain Enabled Systems
Digital sovereignty has emerged as a central concern for modern software-intensive systems, driven by the dominance of non-sovereign cloud infrastructures, the rapid adoption of Generative AI, and increasingly stringent regulatory requirements. While existing initiatives address governance, compliance, and security in isolation, they provide limited guidance on how sovereignty can be operationalized at the architectural level. In this paper, we argue that sovereignty must be treated as a first-class architectural property rather than a purely regulatory objective. We introduce a Sovereign Reference Architecture that integrates self-sovereign identity, blockchain-based trust and auditability, sovereign data governance, and Generative AI deployed under explicit architectural control. The architecture explicitly captures the dual role of Generative AI as both a source of governance risk and an enabler of compliance, accountability, and continuous assurance when properly constrained. By framing sovereignty as an architectural quality attribute, our work bridges regulatory intent and concrete system design, offering a coherent foundation for building auditable, evolvable, and jurisdiction-aware AI-enabled systems. The proposed reference architecture provides a principled starting point for future research and practice at the intersection of software architecture, Generative AI, and digital sovereignty.