{"title":"bjEnet:自然语言语义空间中快速准确的软件缺陷定位方法","authors":"Jiaxuan Han, Cheng Huang, Jiayong Liu","doi":"10.1007/s11219-024-09693-1","DOIUrl":null,"url":null,"abstract":"<p>Automated software bug localization is a significant technology to improve the efficiency of software repair and ensure software quality while promoting the software ecosystem’s stable development. The main objective is to address the semantic matching problem between bug reports and source codes. The appearance of the Transformer structure provides us with a new idea to solve this problem. Transformer-based deep learning models can provide accurate semantic matching results but with a considerable cost (e.g., time). In this paper, we propose a fast and accurate bug localization method named bjEnet based on natural language semantic matching. bjEnet utilizes a pre-trained code language model to transform source codes into code summaries. Then, a code filtering mechanism is employed to exclude source codes unrelated to bug reports, thereby reducing the number of source codes that need to be combined with bug reports for correlation evaluation. Finally, bjEnet uses a BERT-based cross-encoder to localize bugs in the natural language semantic space. The experimental results show that bjEnet is superior to state-of-the-art methods, with an average time to localize a bug report of less than 1 second.</p>","PeriodicalId":21827,"journal":{"name":"Software Quality Journal","volume":"18 1","pages":""},"PeriodicalIF":1.7000,"publicationDate":"2024-07-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"bjEnet: a fast and accurate software bug localization method in natural language semantic space\",\"authors\":\"Jiaxuan Han, Cheng Huang, Jiayong Liu\",\"doi\":\"10.1007/s11219-024-09693-1\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p>Automated software bug localization is a significant technology to improve the efficiency of software repair and ensure software quality while promoting the software ecosystem’s stable development. The main objective is to address the semantic matching problem between bug reports and source codes. The appearance of the Transformer structure provides us with a new idea to solve this problem. Transformer-based deep learning models can provide accurate semantic matching results but with a considerable cost (e.g., time). In this paper, we propose a fast and accurate bug localization method named bjEnet based on natural language semantic matching. bjEnet utilizes a pre-trained code language model to transform source codes into code summaries. Then, a code filtering mechanism is employed to exclude source codes unrelated to bug reports, thereby reducing the number of source codes that need to be combined with bug reports for correlation evaluation. Finally, bjEnet uses a BERT-based cross-encoder to localize bugs in the natural language semantic space. The experimental results show that bjEnet is superior to state-of-the-art methods, with an average time to localize a bug report of less than 1 second.</p>\",\"PeriodicalId\":21827,\"journal\":{\"name\":\"Software Quality Journal\",\"volume\":\"18 1\",\"pages\":\"\"},\"PeriodicalIF\":1.7000,\"publicationDate\":\"2024-07-22\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Software Quality Journal\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://doi.org/10.1007/s11219-024-09693-1\",\"RegionNum\":3,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"COMPUTER SCIENCE, SOFTWARE ENGINEERING\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Software Quality Journal","FirstCategoryId":"94","ListUrlMain":"https://doi.org/10.1007/s11219-024-09693-1","RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"COMPUTER SCIENCE, SOFTWARE ENGINEERING","Score":null,"Total":0}
bjEnet: a fast and accurate software bug localization method in natural language semantic space
Automated software bug localization is a significant technology to improve the efficiency of software repair and ensure software quality while promoting the software ecosystem’s stable development. The main objective is to address the semantic matching problem between bug reports and source codes. The appearance of the Transformer structure provides us with a new idea to solve this problem. Transformer-based deep learning models can provide accurate semantic matching results but with a considerable cost (e.g., time). In this paper, we propose a fast and accurate bug localization method named bjEnet based on natural language semantic matching. bjEnet utilizes a pre-trained code language model to transform source codes into code summaries. Then, a code filtering mechanism is employed to exclude source codes unrelated to bug reports, thereby reducing the number of source codes that need to be combined with bug reports for correlation evaluation. Finally, bjEnet uses a BERT-based cross-encoder to localize bugs in the natural language semantic space. The experimental results show that bjEnet is superior to state-of-the-art methods, with an average time to localize a bug report of less than 1 second.
期刊介绍:
The aims of the Software Quality Journal are:
(1) To promote awareness of the crucial role of quality management in the effective construction of the software systems developed, used, and/or maintained by organizations in pursuit of their business objectives.
(2) To provide a forum of the exchange of experiences and information on software quality management and the methods, tools and products used to measure and achieve it.
(3) To provide a vehicle for the publication of academic papers related to all aspects of software quality.
The Journal addresses all aspects of software quality from both a practical and an academic viewpoint. It invites contributions from practitioners and academics, as well as national and international policy and standard making bodies, and sets out to be the definitive international reference source for such information.
The Journal will accept research, technique, case study, survey and tutorial submissions that address quality-related issues including, but not limited to: internal and external quality standards, management of quality within organizations, technical aspects of quality, quality aspects for product vendors, software measurement and metrics, software testing and other quality assurance techniques, total quality management and cultural aspects. Other technical issues with regard to software quality, including: data management, formal methods, safety critical applications, and CASE.