.png){kind=icon}
.png){kind=icon}
Return
ArchSense: Characterizing Component-Internal Implementation-Semantic Evolution via Code Representation
.png){kind=icon}
OA
AI
DOI：10.3390/electronics15132858.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Monitoring software architecture evolution is essential for maintaining system quality, but many practical change-detection pipelines still rely on syntactic differences. Such pipelines can produce large review streams in which minor edits, refactorings, and deeper implementation rewrites are mixed together. In this paper, we introduce ArchSense, a diagnostic framework for characterizing component-internal implementation-semantic evolution. ArchSense combines hierarchical structural alignment with method-level semantic displacement measurement. Instead of treating all code updates as equal, it uses vector representations from GraphCodeBERT to measure implementation-level semantic change intensity and then summarizes this signal within architectural components. We evaluate ArchSense on six large-scale open-source Java systems. The results show that semantic distance is positively associated with a three-annotator construct assessment of implementation change intensity. Additional disagreement analysis links high and low model decisions to concrete source code features, including control flow, call-set, return, exception, synchronization, and test-observable changes. A behavioral proxy check on 23,389 test method updates shows that the default threshold behaves as a conservative high-precision, low-recall review signal for implementation-semantic change. The validated scope of ArchSense is therefore component-internal diagnostic characterization for architecture-oriented review; connector topology, runtime communication, behavioral equivalence, and defect causality require complementary evidence.
Keywords:
software architecture
implementation evolution
semantic code analysis
change characterization