Computer Science Seminar - Dr. S. Pandey, Chalmers University Gothenburg

Title:

Revolutionizing Software Engineering: AI-based Design Pattern Recognition and Application Insights

Abstract:

The talk delves into the cutting-edge advancements in Design Pattern Recognition (DPR) within Software Engineering, leveraging AI techniques. We introduce a novel DPR approach utilizing a cross-lingual Programming Language Model (PLM) called TransDPR. Based on a Facebook pre-trained model, this method achieves an impressive 90% accuracy and an F1-score of 0.88 on open-source projects, showcasing its potential in identifying design patterns with high precision.

Moving forward, the talk will discuss the comparison between Word-Based and AST-Based Models. We showcase the promise of modern language models like Code2Vec and Word2Vec in DPR, with Code2Vec and Word2Vec producing average F1 scores of 0.781 and 0.690 on open-source Java programs. The results highlight the effectiveness of these models in discerning specific design patterns and their potential in practical software engineering contexts. The exploration advances further with more advanced large language models and employs state-of-the-art language models, to extract semantic information from Java programs. By applying the k-nearest neighbors approach to predict implemented design patterns, the study showcases the effectiveness of different language models in capturing semantic information relevant to design patterns, providing insights into automated design pattern recognition.

Finally, based on the interview study, the talk will focus on the practical application of design patterns in the automotive industry. Through an interview-based study, we uncover how industry professionals view and utilize design patterns, revealing a fine perspective that goes beyond academic theories.