Daniel Nilsson and Mert Yurdakul

Test Scouts (Sweden)

Spotting Twins - A Large Language Model Approach to Duplicate Defect Detection in Software Development

Artificial Intelligence, Machine Learning and Large Language Models has been main discussion topics in social media for some time now both from a general perspective but also specifically within Software Testing. However, the conversation quickly ends up in a discussion if AI will replace the tester or not, instead of in which situations it creates value. We have instead started from the daily activities that we as testers work with and tried to identify where there are bottlenecks and time wasters that could at the same time be suitable for using an AI-based solution to remove those. In this project we have addressed duplicates bug reports as being a huge time waster in large projects in large organizations. Given that defect reports are composed in natural language, the core of this issue resides within the realm of natural language processing. In recent years, large language models, which are based on a neural network architecture called a transformer, have been showing impressive results in natural language processing tasks, namely, understanding, generating, and interacting with human language. Therefore, our proposed method utilizes several large language models to understand given defect reports and to detect duplicates in a dataset of defect reports.

Comprar Tickets
Daniel is a dedicated test personality from Gothenburg, Sweden with more than 20 years’ experience within software testing of complex system. He has mainly worked with embedded software in the automotive industry, including personal vehicles, trucks and busses. Test automation has followed him throughout his entire career, from automating manual scripts to strategical processes and development of complete test environments and frameworks. Daniel has assisted several large global automotive companies in their work with increasing the quality mindset within the organizations, this includes introducing, teaching and coaching test teams in Japan, China, Germany, India and the US in agile test methodology. Daniel is a regular visitor to global test conferences, constantly updating himself on the latest in methods, techniques, and ways of working. He is also active in the local test community and contributing to the development of the tester profession in the Gothenburg area.

Mert pursued his master’s degree in data science & AI at Chalmers University of Technology located in Gothenburg, Sweden. He conducted his master’s thesis at AstraZeneca on automating molecular structure elucidation using machine learning techniques. He has been working at Test Scouts for about 1 year focusing on testing AI-driven systems and developing an internal AI testing tool. He also serves as the industrial supervisor for two groups of thesis students from the University of Gothenburg. Mert is a first-time speaker at international testing conferences but have within his relatively short working career already experience from customer presentations and speaking at local test communities.”