Automated Software Engineering最新文献

{"title":"VulFileHunter: leveraging description augmentation with large language models for vulnerability-relevant file uncovery","authors":"Wang Tao,&nbsp;Wang Zhen,&nbsp;Zhang Wenhao,&nbsp;Xu Daocheng,&nbsp;Ren Xiaoxue","doi":"10.1007/s10515-026-00678-x","DOIUrl":"10.1007/s10515-026-00678-x","url":null,"abstract":"<div><p>Open-source software (OSS) enhances modern software development by providing public access to its code, fostering innovation, and reducing costs. However, it also introduces vulnerabilities in the software supply chain, necessitating vigilant management and maintenance. The management of OSS vulnerabilities often relies on public advisories such as the National Vulnerability Database (NVD), which documents vulnerabilities through Common Vulnerabilities and Exposures (CVE) entries. While these entries are crucial, they often suffer from inconsistent quality and lack integration with the broader software supply chain, limiting their effectiveness. To address these challenges, we propose <span>VulFileHunter</span>, a general framework designed to automatically augment CVE entries and locate vulnerability-relevant files. Our approach uniquely integrates a three-phase step, including CVE description augmentation, candidate vulnerability-relevant file retrieval, and vulnerability-relevant file localization. Experimental results show that <span>VulFileHunter</span> achieves an F1 score of 0.72, representing more than 60% improvements over baselines. An ablation study reveals that each phase designed in <span>VulFileHunter</span> is essential for achieving superior performance for locating vulnerability-relevant files. These findings highlight that <span>VulFileHunter</span> serves as a powerful tool for locating vulnerability-relevant files, laying the foundation for finer-grained software supply chain protection. The replication package is available at https://github.com/XSig-Opensource/VulFileHunter.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-10-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132218","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":"Correction: Intelligent automated code review and software quality evaluation: a survey of AutoML and large language model techniques","authors":"Rohan Appasaheb Borgalli,&nbsp;Uma Raj,&nbsp;Shwetha Shetty","doi":"10.1007/s10515-026-00682-1","DOIUrl":"10.1007/s10515-026-00682-1","url":null,"abstract":"","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149132084","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":"The fools are certain; the wise are doubtful: Exploring LLM confidence in code completion","authors":"Zoe Kotti,&nbsp;Konstantina Dritsa,&nbsp;Diomidis Spinellis,&nbsp;Panos Louridas","doi":"10.1007/s10515-026-00683-0","DOIUrl":"10.1007/s10515-026-00683-0","url":null,"abstract":"<div><p>Code completion entails the task of providing missing tokens given a surrounding context. It can boost developer productivity and serve as a code discovery tool. Code completion has recently been approached with Large Language Models (LLMs) fine-tuned on code (code LLMs). The performance of code LLMs can be assessed with downstream and intrinsic metrics. Downstream metrics are usually employed to evaluate the practical utility of a model, but can be unreliable and require complex calculations and domain-specific knowledge. In contrast, intrinsic metrics such as perplexity, entropy, and mutual information, which measure model confidence or uncertainty, are simple, versatile, and universal across LLMs and tasks, and have been proposed as proxies for functional correctness and hallucination risk in LLM-generated code. Motivated by this, we evaluate the confidence of LLMs when generating code by measuring code perplexity across programming languages, models, and datasets using various LLMs, and a sample of 2254 files from 881 GitHub projects. We find that strongly-typed languages exhibit lower perplexity than dynamically typed languages. Scripting languages also demonstrate higher perplexity. Shell appears universally high in perplexity, whereas Java appears low. Code perplexity depends on the employed LLM; under a fixed model, relative language-level rankings are largely stable across evaluation corpora. Although code comments modestly increase perplexity, the language ranking based on perplexity is barely affected by their presence. LLM researchers, developers, and users can use our findings to assess the suitability of LLM-based code completion in specific software projects based on how language, model choice, and code characteristics impact model confidence.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10515-026-00683-0.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CodeCritic: towards effective code quality evaluation with large language models","authors":"Yanlin Wang,&nbsp;Yanxian Huang,&nbsp;Sida Xu,&nbsp;Ensheng Shi,&nbsp;Jiachi Chen,&nbsp;Wanjun Zhong,&nbsp;Zibin Zheng","doi":"10.1007/s10515-026-00632-x","DOIUrl":"10.1007/s10515-026-00632-x","url":null,"abstract":"<div><p>Code quality evaluation aims to assess code quality, providing critical insights for a deeper comprehension of source code. Previous studies typically focus on a single dimension of code quality, lacking a comprehensive perspective. Recent advancements in large language models (LLMs) have shown promise in evaluating the quality of natural language across multiple dimensions, suggesting their potential for a more comprehensive evaluation of code. In this paper, we explore the potential of LLMs for comprehensively evaluating generated code by (1) creating a diverse code quality evaluation dataset <b>CodeCriticData</b> based on eight code evaluation criteria, covering three aspects of code quality: correctness, efficiency, and readability.; (2) conducting an extensive empirical study to assess the code quality evaluation ability of four mainstream LLMs using CodeCriticData; and (3) introducing <b>CodeCritic</b>, an LLM-based method designed to enhance code quality evaluation. Our empirical results show that while GPT-4 outperforms other LLMs in code evaluation, its proprietary nature and cost limit its accessibility. To address this, we present CodeCritic, an open-source LLM fine-tuned on CodeCriticData, which achieves state-of-the-art results. CodeCritic demonstrates an average improvement of 7.0% across the eight evaluation dimensions, with some dimensions showing improvements up to 19.4%. In pairwise comparisons with baselines, CodeCritic achieves a victory rate of 58% to 88%. Additionally, CodeCritic proves to be versatile, showing strong performance across five different programming languages, and highlighting its cross-language generalizability. Our source code and dataset are available at: https://github.com/DeepSoftwareAnalytics/CodeCritic</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133396","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 layered formal verification of ChaCha20-Poly1305 in TLS implementations","authors":"Xiangdong Li,&nbsp;Hui Li,&nbsp;Jingjing Guan,&nbsp;Xiaolei Wang,&nbsp;Yunan Zhou","doi":"10.1007/s10515-026-00674-1","DOIUrl":"10.1007/s10515-026-00674-1","url":null,"abstract":"<div><p>ChaCha20-Poly1305 is a widely used authenticated encryption scheme deployed in protocols such as TLS 1.3 and QUIC. Although its security has been extensively analyzed at the standard level, the translation from natural language specifications to low-level implementations still depends on human expertise and remains error-prone. In this paper, we instantiate a layered verification workflow that connects an RFC-level High Spec, implementation-oriented Low Specs, and concrete C implementations. By separating RFC-level and implementation-oriented Cryptol models, the workflow helps manage the semantic gap between abstract algorithm descriptions and implementation-level details. To make the abstraction gap explicit, we prove auxiliary model-consistency lemmas between the RFC-level and implementation-oriented Cryptol models. We demonstrate the effectiveness of this workflow on the baseline C implementations of ChaCha20-Poly1305 in two independently developed libraries, OpenSSL and openHiTLS. The result is a case study of applying Cryptol, SMT solving, and SAW-based equivalence checking to existing ChaCha20-Poly1305 implementations in mainstream cryptographic libraries.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149133139","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":"TLSS_ICFS: A novel two-stage cross-project defect prediction approach","authors":"Hang Fu,&nbsp;Ran Mo,&nbsp;Xinya Mu,&nbsp;Wei Ding,&nbsp;Haopeng Song,&nbsp;Chaochao Wu,&nbsp;ZengYang Li,&nbsp;Yutao Ma","doi":"10.1007/s10515-026-00686-x","DOIUrl":"10.1007/s10515-026-00686-x","url":null,"abstract":"<div><p>Cross-project defect prediction (CPDP) leverages external source projects to predict defects in projects with limited historical data. Traditional CPDP methods have several limitations: 1) Source selection ignores the domain adaptation potential of candidate projects; 2) Transfer learning, applied separately after source selection, leads to unstable performance; 3) For Filter-based feature selection methods, MIC/Pearson focus on univariate feature-class correlation but ignore inter-feature redundancy, while CFS often under-selects features and fails to balance relevance and generalization in CPDP scenarios. To address these issues, we propose TLSS_ICFS, a two-stage CPDP framework that integrates source selection, transfer learning, and feature selection for collaborative optimization. TLSS embeds Kernel Mean Matching (KMM) into source selection to evaluate post-transfer performance, which can address the decoupling issue between source selection and transfer learning. ICFS improves CFS with backtracking-free greedy forward selection, incremental contribution tracking, dynamic truncation, and random restart to avoid local optima and feature under-selection. Experiments on PROMISE (41 datasets) and ADPDS (33 datasets) show: Using the stable Logistic Regression (LR) classifier, TLSS outperforms baseline methods by 9.60%–43.58% in F1-score and 2.09%–29.52% in AUC on the PROMISE dataset; on the ADPDS dataset, TLSS is slightly inferior to CFPS/Bellwether (with no statistically significant difference) but outperforms all other baselines. Although feature selection alone yields marginal gains, ICFS stably enhances TLSS performance. TLSS_ICFS outperforms five state-of-the-art (SOTA) CPDP methods, with median F1-score improvements of 14.02%–56.39% and median AUC improvements of 1.40%–27.94% on the PROMISE dataset, and 13.50%–39.31% (F1-score) and 0.27%–21.55% (AUC) on the ADPDS dataset. TLSS_ICFS mitigates source-target distribution divergence, overcomes the limitations of traditional CPDP methods, and provides a stable, high-performance solution for cross-project defect prediction in data-scarce scenarios.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034971","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":"Security smell detection in Dockerfiles","authors":"Wenbo Zhang,&nbsp;Inah Omoronyia","doi":"10.1007/s10515-026-00684-z","DOIUrl":"10.1007/s10515-026-00684-z","url":null,"abstract":"<div><p>Dockerfiles are a widely adopted format for specifying Infrastructure-as-Code (IaC). However, insecure or misconfigured Dockerfiles frequently introduce critical security smells, including improper privilege use, unpinned dependencies, and secret leakage. State-of-practice tools such as Hadolint, Checkov, and KICS rely on static rules and regular expressions, leading to high false-negative rates and limited explanations for developers—particularly when security smells are semantically interwoven into configuration logic. We present an LLM-based framework that integrates (i) a fine-tuned multi-label risk detector trained on 4,000 manually annotated Dockerfiles across ten high-impact categories, and (ii) a retrieval-augmented interpretation-and-repair module that outputs both human-readable rationales and corrected code. The system is deployed via a VSCode plugin for real-time support. Experiments on a curated test set show that the detector achieves a macro-averaged precision of 0.97 and recall of 0.94, improving recall by 9–28 percentage points over static baselines. A user study further confirms the usability of our approach, demonstrating significant improvements in risk comprehension, identification accuracy, and remediation effort. Our findings highlight the feasibility of LLM-assisted IaC security and provide practical early mitigation for Dockerfile misconfigurations.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://link.springer.com/content/pdf/10.1007/s10515-026-00684-z.pdf","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034561","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AMCD: an adjacency matrix-based code clone detection method","authors":"Bin Hu,&nbsp;Xiancong Pan,&nbsp;Dongjin Yu,&nbsp;Tianyi Hu,&nbsp;Bin Lin","doi":"10.1007/s10515-026-00685-y","DOIUrl":"10.1007/s10515-026-00685-y","url":null,"abstract":"<div><p>Code clone is a significant code smell that is prevalent in software products and affects both development and maintenance. While code reuse can improve development efficiency, duplicated code may result in bug propagation. Thus, detecting code clones is crucial for ensuring software quality. Although there are already many code clone detection approaches, they often perform poorly on large-scale software, suffering from slow runtime or memory overflow. Moreover, few methods can accurately detect code clones with substantial differences. In this paper, we propose AMCD (Adjacency Matrix-based Code Clone Detection), a code clone detection approach with high runtime performance and accuracy. AMCD first extracts code fragments from the repository based on a configurable granularity and sorts them by line count in ascending order. Then, methods with line count differences below a configured threshold are added to a buffer. Finally, parallel similarity computation compares method pairs. AMCD incorporates three different similarity algorithms, enabling flexible detection of various clone types. Experimental results show that AMCD achieves comparable accuracy and recall in Type-1/2 clone detection while outperforming existing techniques in weak Type-3 clone detection. In terms of runtime efficiency, it achieves competitive performance compared to state-of-the-art methods.</p></div>","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034683","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":"Correction: Leveraging text generation for enhanced user story quality","authors":"Carlos Alberto dos Santos,&nbsp;Kevin Bouchard","doi":"10.1007/s10515-026-00672-3","DOIUrl":"10.1007/s10515-026-00672-3","url":null,"abstract":"","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034812","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":"Correction: CodeGAN-MPC: a multi-pattern conditional GAN with differentiable conformance loss for automated test input generation","authors":"Imad H. Hasan,&nbsp;Moayad Y. Potrus","doi":"10.1007/s10515-026-00673-2","DOIUrl":"10.1007/s10515-026-00673-2","url":null,"abstract":"","PeriodicalId":55414,"journal":{"name":"Automated Software Engineering","volume":"33 4","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149034813","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}