Evaluating Identifier Readability Using CodeBERT Embeddings and Self-Attention Bi-LSTM with Explainable Modeling
Engineering, Technology & Applied Science Research · 06.06.2026 · 16 (3) · 36731-36737
DOI: 10.48084/etasr.17996 ↗Kurz gefasst
Fachartikel in der Zeitschrift Engineering, Technology & Applied Science Research, online erschienen am 06.06.2026. Verfasst von Bharat Babaso Mane und 1 weiteren Autor:innen. Die Arbeit nennt 18 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Identifier names are natural language representations in source code that play a significant role in program understanding. Studies on the quality of identifier names generally focus on their role in program knowledge. Identifier names and naming conventions are important for programming understanding. Previous studies have shown a connection between software quality and the quality of identifier names. Recently, Deep Learning (DL) has been used to develop highly efficient models for identifying intelligibility challenges. DL models, such as transformer-based frameworks and attention mechanisms, can acquire contextual and sequential relations between identified tokens. This study proposes an Identifier Readability Analysis Framework using an Explainable Attention-Based Deep Learning (IRAF-XADL) approach, with the primary intention of assessing the quality of identifier names in Python and C++ source code. In the initial stage, a syntax-aware identifier preprocessing pipeline based on language-specific abstract syntax tree parsing is applied to extract identifiers and perform lexical normalization and semantic cleaning. From the normalized identifiers, this study computes ten linguistically and cognitively grounded readability parameters. Semantic and contextual representations are attained using CodeBERT embeddings, which are then processed by a self-attention-based bidirectional long short-term memory model to learn sequential and contextual dependencies. Furthermore, the model is optimized using the AdamW optimizer, enabling improved convergence and overall performance. In the last stage, SHAP-based explainability is integrated for interpreting the contribution of identifier tokens and features to readability predictions. The IRAF-XADL method was experimentally examined on the benchmark Code Snippets: Insights and Readability Dataset, and the results prove the improved performance over the existing approaches in terms of diverse metrics.
Zitieren
APA
Mane, B. B., & Achary, R. (2026). Evaluating Identifier Readability Using CodeBERT Embeddings and Self-Attention Bi-LSTM with Explainable Modeling. Engineering, Technology & Applied Science Research, 16(3), 36731-36737. https://doi.org/10.48084/etasr.17996
Vancouver
Mane BB, Achary R. Evaluating Identifier Readability Using CodeBERT Embeddings and Self-Attention Bi-LSTM with Explainable Modeling. Engineering, Technology & Applied Science Research. 2026;16(3):36731-36737. doi:10.48084/etasr.17996
BibTeX
@article{Mane2026,
author = {Mane, Bharat Babaso and Achary, Rathnakar},
title = {Evaluating Identifier Readability Using CodeBERT Embeddings and Self-Attention Bi-LSTM with Explainable Modeling},
journal = {Engineering, Technology & Applied Science Research},
year = {2026},
volume = {16},
number = {3},
pages = {36731--36737},
doi = {10.48084/etasr.17996}
}
EUVIMED