Authors:

Bryan Zhang

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

The increasing complexity of software systems has led to a growing need for effective analysis and prediction of software failures. Understanding the relationships between different software failures is crucial for preventing similar incidents in the future and improving overall system reliability. The FAIL (Failure Analysis Investigation using LLM) database provides a rich source of information, but its vast size poses challenges in identifying related failures. This research aims to automate the identification and retrieval of related failures from the FAIL database using advanced techniques. It begins with semantic similarity analysis using BERT embeddings to capture nuanced similarities among failure events. Subsequently, a knowledge graph is constructed using DGL-KE, integrating models like TransE, TransR, RESCAL, DistMult, ComplEx, and RotatE to model complex relationships between failure events. Finally, relationship prediction for new failure incidents leverages BERT embeddings in conjunction with the most effective graph embedding model, facilitating accurate and efficient retrieval of relevant failure data. Our experiments reveal that the RotatE model performs best in capturing nuanced relationships between failure events, enabling quick and accurate retrieval of similar past incidents. This combined BERT and knowledge graph approach demonstrates effectiveness in automating software failure analysis, offering a scalable solution for identifying patterns and relationships in failure data. By enhancing the understanding of failure mechanisms and supporting proactive risk management, this research contributes to the development of more robust software systems and improved software reliability. Keywords: Failure Analysis; Semantic Similarity; Knowledge Graph; Software Reliability

Source:

Purdue University / 2024

Topics:

No topics listed

Co-authors:

Bryan Zhang