Yuanfang Cai
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Yuanfang Cai has a total of 86 co-authors. Here are the top co-authors by number of shared publications:
- Rick Kazman
University of Hawaii
14 shared publications - Ernst Pisch
Department of Computer Science at Drexel University, Philadelphia, PA, USA
4 shared publications - Jason Lefever
Drexel University, Philadelphia, USA
4 shared publications - Lu Xiao
Stevens Institute of Technology, Hoboken, USA
4 shared publications Qiong Feng
School of Computer Science and Engineering, Nanjing University of Science and Technology, Nanjing, Jiangsu, China
4 shared publications- Humberto Cervantes
Universidad Autónoma Metropolitana Iztapalapa, CDMX, Mexico
3 shared publications
Latest Publications
- Open Access
- 9 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
Faster Code, Deeper Debt? A Multivocal Literature Review on Technical Debt and Its Early Signs in LLM-Assisted Software Development
With the rapid adoption of LLM-assisted coding, the need to manage the technical debt these systems introduce has become urgent. In this paper, we conduct a multivocal literature review of 104 sources (31 formal, 73 grey) to examine how LLM-assisted development contributes to technical debt and what strategies, metrics, and benchmarks exist to mitigate it. We find that LLMs often amplify traditional forms of technical debt, particularly code, design, and documentation debts, while also introducing new LLM-specific debts. Notably, we identify fast-integration debt, where rapidly generated code prioritizes speed over quality, triggering a domino effect that leads to governance debt and increased long-term maintenance costs. Additional emerging categories include prompt, ethical, data, and provenance debt, reflecting new challenges unique to LLM adoption. To address these, strategies suggested in the literature include human-in-the-loop frameworks, prompt engineering, and data quality alignment. In practice, tools such as SonarQube are commonly used to detect technical debt indicators, while research prototypes such as CodeSmellEval are emerging to assess how LLMs contribute to debts. However, no standardized benchmarks or LLM-specific metrics yet exist, leaving an important gap. Based on findings, we outline insights and future directions to ensure reliable integration of LLMs into software engineering workflows.
- 9 June 2026
- Published by Association for Computing Machinery (ACM) in ACM Transactions on Software Engineering and Methodology
Open AccessFaster Code, Deeper Debt? A Multivocal Literature Review on Technical Debt and Its Early Signs in LLM-Assisted Software Development
With the rapid adoption of LLM-assisted coding, the need to manage the technical debt these systems introduce has become urgent. In this paper, we conduct a multivocal literature review of 104 sources (31 formal, 73 grey) to examine how LLM-assisted development contributes to technical debt and what strategies, metrics, and benchmarks exist to mitigate it. We find that LLMs often amplify traditional forms of technical debt, particularly code, design, and documentation debts, while also introducing new LLM-specific debts. Notably, we identify fast-integration debt, where rapidly generated code prioritizes speed over quality, triggering a domino effect that leads to governance debt and increased long-term maintenance costs. Additional emerging categories include prompt, ethical, data, and provenance debt, reflecting new challenges unique to LLM adoption. To address these, strategies suggested in the literature include human-in-the-loop frameworks, prompt engineering, and data quality alignment. In practice, tools such as SonarQube are commonly used to detect technical debt indicators, while research prototypes such as CodeSmellEval are emerging to assess how LLMs contribute to debts. However, no standardized benchmarks or LLM-specific metrics yet exist, leaving an important gap. Based on findings, we outline insights and future directions to ensure reliable integration of LLMs into software engineering workflows.
- Open Access
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 23-30
Improving LLM-assisted code generation through the use of architectural documents and implementation plans.
- 12 April 2026
- Published by Association for Computing Machinery (ACM)
- p. 23-30
Open AccessImproving LLM-assisted code generation through the use of architectural documents and implementation plans.
- 8 April 2026
- Published by Drexel University Libraries
Concern-based management of design debt
Separation of concerns is a longstanding software design principle, and its violation is among the costliest forms of design-level technical debt, elevating defect rates and slowing development as systems grow. Many software engineering tasks are centered around identifying or rectifying this design debt, from refactoring recommenders and architecture recovery algorithms to quality metrics like coupling and cohesion. The common prerequisite across all of these tasks is the ability to identify the discrete concerns of a complex codebase, i.e., cohesive groups of program entities that fulfill a shared responsibility. This thesis proposes that modeling concerns directly is both feasible and valuable for design debt management, grounded in the insight that concerns can be inferred using contextual cues such as class membership. This thesis presents five contributions: (1) an empirical study revealing that existing technical debt tools disagree substantially and are often no more informative than the heuristic that "big files are bad''; (2) NeoDepends, a multi-language extraction tool for program entities, dependencies, and co-change relations; (3) Deicide, a class decomposition algorithm achieving nearly double the co-change alignment of existing approaches; (4) ConcernBERT, an embedding model trained on class membership that outperforms seven baselines by 31–55%; and (5) Concern Deviation, a cohesion metric that outperforms twelve existing metrics against expert judgments. By establishing a stronger foundation for automated concern identification, these contributions advance not only the specific tasks evaluated here but the broader class of software engineering problems for which concern identification is an essential first step.
- 8 April 2026
- Published by Drexel University Libraries
Concern-based management of design debt
Separation of concerns is a longstanding software design principle, and its violation is among the costliest forms of design-level technical debt, elevating defect rates and slowing development as systems grow. Many software engineering tasks are centered around identifying or rectifying this design debt, from refactoring recommenders and architecture recovery algorithms to quality metrics like coupling and cohesion. The common prerequisite across all of these tasks is the ability to identify the discrete concerns of a complex codebase, i.e., cohesive groups of program entities that fulfill a shared responsibility. This thesis proposes that modeling concerns directly is both feasible and valuable for design debt management, grounded in the insight that concerns can be inferred using contextual cues such as class membership. This thesis presents five contributions: (1) an empirical study revealing that existing technical debt tools disagree substantially and are often no more informative than the heuristic that "big files are bad''; (2) NeoDepends, a multi-language extraction tool for program entities, dependencies, and co-change relations; (3) Deicide, a class decomposition algorithm achieving nearly double the co-change alignment of existing approaches; (4) ConcernBERT, an embedding model trained on class membership that outperforms seven baselines by 31–55%; and (5) Concern Deviation, a cohesion metric that outperforms twelve existing metrics against expert judgments. By establishing a stronger foundation for automated concern identification, these contributions advance not only the specific tasks evaluated here but the broader class of software engineering problems for which concern identification is an essential first step.
- Open Access
- 21 March 2026
- Published by Elsevier in Journal of Systems and Software
- Vol. 237
Automated assessment of the relationship between microservice architectures and performance
- 21 March 2026
- Published by Elsevier in Journal of Systems and Software
- Vol. 237
Open AccessAutomated assessment of the relationship between microservice architectures and performance
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Software
- Vol. 43 (4) , 87-95
Large Language Models as Assistants in Software Architecture Design
To achieve desired software quality attributes, architects must evaluate and choose from a myriad of design concepts, patterns, and technologies. This paper explores how Large Language Models (LLMs) can assist in this decision-making process. We compare design decisions made by LLMs and human architects across several design problems. Our results show that while LLMs suggest relevant options, human architects often disagree with or distrust those suggestions because the models frequently lack context and are inconsistent. Humans still need to make the final decisions in design, but LLMs are valuable as assistants: for generating design options, serving as a sounding board, acting as a creative catalyst and an education aid. We conclude with a discussion of the implications for architectural practice, education, and future research directions.
- 12 February 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Software
- Vol. 43 (4) , 87-95
Large Language Models as Assistants in Software Architecture Design
To achieve desired software quality attributes, architects must evaluate and choose from a myriad of design concepts, patterns, and technologies. This paper explores how Large Language Models (LLMs) can assist in this decision-making process. We compare design decisions made by LLMs and human architects across several design problems. Our results show that while LLMs suggest relevant options, human architects often disagree with or distrust those suggestions because the models frequently lack context and are inconsistent. Humans still need to make the final decisions in design, but LLMs are valuable as assistants: for generating design options, serving as a sounding board, acting as a creative catalyst and an education aid. We conclude with a discussion of the implications for architectural practice, education, and future research directions.
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (10) , 1-15
ConcernBERT: Learning Responsibilities Using Class Membership
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (10) , 1-15
ConcernBERT: Learning Responsibilities Using Class Membership
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (10) , 1-17
Better Together: An LLM-assisted approach to designing software architectures using Attribute-Driven Design
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 52 (10) , 1-17
Better Together: An LLM-assisted approach to designing software architectures using Attribute-Driven Design
- 13 January 2025
- Published by Elsevier in Journal of Systems and Software
- Vol. 222
Multivocal study on microservice dependencies
- 13 January 2025
- Published by Elsevier in Journal of Systems and Software
- Vol. 222
Multivocal study on microservice dependencies
- 1 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 51 (2) , 449-465
A Holistic Approach to Design Understanding Through Concept Explanation
Complex software systems consist of multiple overlapping design structures, such as abstractions, features, crosscutting concerns, or patterns. This is similar to how a human body has multiple interacting subsystems, such as respiratory, digestive, or circulatory. Unlike in the medical domain, software designers do not have an effective way to distinguish, visualize, comprehend, and analyze these interleaving design structures. As a result, developers often struggle through the maze of source code. In this paper, we present an Automated Concept Explanation (ACE) framework that automatically extracts and categorizes major concepts from source code based on the roles that files play in design structures and their topic frequencies. Based on these categorized concepts, ACE recovers four categories of high-level design models using different algorithms and generates a natural language explanation for each. To assess if and how ACE can help developers better understand design structures, we conducted an empirical study where two groups of graduate students were assigned three design comprehension tasks: identifying feature-related files, identifying dependencies among features, and identifying design patterns used, in an open-source project. The results reveal that the students who used ACE can accomplish these tasks much faster and more accurately, and they acknowledged the usefulness of the categorized concepts and structures, multi-type high-level model visualization, and natural language explanations.
- 1 January 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 51 (2) , 449-465
A Holistic Approach to Design Understanding Through Concept Explanation
Complex software systems consist of multiple overlapping design structures, such as abstractions, features, crosscutting concerns, or patterns. This is similar to how a human body has multiple interacting subsystems, such as respiratory, digestive, or circulatory. Unlike in the medical domain, software designers do not have an effective way to distinguish, visualize, comprehend, and analyze these interleaving design structures. As a result, developers often struggle through the maze of source code. In this paper, we present an Automated Concept Explanation (ACE) framework that automatically extracts and categorizes major concepts from source code based on the roles that files play in design structures and their topic frequencies. Based on these categorized concepts, ACE recovers four categories of high-level design models using different algorithms and generates a natural language explanation for each. To assess if and how ACE can help developers better understand design structures, we conducted an empirical study where two groups of graduate students were assigned three design comprehension tasks: identifying feature-related files, identifying dependencies among features, and identifying design patterns used, in an open-source project. The results reveal that the students who used ACE can accomplish these tasks much faster and more accurately, and they acknowledged the usefulness of the categorized concepts and structures, multi-type high-level model visualization, and natural language explanations.
- 31 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 51 (2) , 484-502
Refactoring Microservices to Microservices in Support of Evolutionary Design
Evolutionary design is a widely accepted practice for defining microservice boundaries. It is performed through a sequence of incremental refactoring tasks (we call it “ microservice refactoring ”), each restructuring only part of a microservice system ( a.k.a., refactoring part ) into well-defined services for improving the architecture in a controlled manner. Despite its popularity in practice, microservice refactoring suffers from insufficient methodological support. While there are numerous studies addressing similar software design tasks, i.e. , software remodularization and microservitization, their approaches prove inadequate when applied to microservice refactoring. Our analysis reveals that their approaches may even degrade the entire architecture in microservice refactoring, as they only optimize the refactoring part in such applications, but neglect the relationships between the refactoring part and the remaining system. As the first response to the need, Micro2Micro is proposed to re-partition the refactoring part while optimizing three quality objectives including the interdependence between the refactoring and non-refactoring parts. In addition, it allows architects to intervene in the decision-making process by interactively incorporating their knowledge into the iterative search for optimal refactoring solutions. An empirical study on 13 open-source projects of different sizes shows that the solutions from Micro2Micro perform well and exhibit quality improvement with an average up to 45% to the original architecture. Users of Micro2Micro found the suggested solutions highly satisfactory. They acknowledge the advantages in terms of infusing human intelligence into decisions, providing immediate quality feedback, and quick exploration capability.
- 31 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Software Engineering
- Vol. 51 (2) , 484-502
Refactoring Microservices to Microservices in Support of Evolutionary Design
Evolutionary design is a widely accepted practice for defining microservice boundaries. It is performed through a sequence of incremental refactoring tasks (we call it “ microservice refactoring ”), each restructuring only part of a microservice system ( a.k.a., refactoring part ) into well-defined services for improving the architecture in a controlled manner. Despite its popularity in practice, microservice refactoring suffers from insufficient methodological support. While there are numerous studies addressing similar software design tasks, i.e. , software remodularization and microservitization, their approaches prove inadequate when applied to microservice refactoring. Our analysis reveals that their approaches may even degrade the entire architecture in microservice refactoring, as they only optimize the refactoring part in such applications, but neglect the relationships between the refactoring part and the remaining system. As the first response to the need, Micro2Micro is proposed to re-partition the refactoring part while optimizing three quality objectives including the interdependence between the refactoring and non-refactoring parts. In addition, it allows architects to intervene in the decision-making process by interactively incorporating their knowledge into the iterative search for optimal refactoring solutions. An empirical study on 13 open-source projects of different sizes shows that the solutions from Micro2Micro perform well and exhibit quality improvement with an average up to 45% to the original architecture. Users of Micro2Micro found the suggested solutions highly satisfactory. They acknowledge the advantages in terms of infusing human intelligence into decisions, providing immediate quality feedback, and quick exploration capability.