Optimizing Error Detection in Generated Code Using Metaheuristic Optimized Natural Language Processing

Optimizing Error Detection in Generated Code Using Metaheuristic Optimized Natural Language Processing

연구 분야: Artificial Intelligence

학회: International Conference on Soft Computing and its Engineering Applications

This paper tackles the issue of software defect identification by employing advanced classification methods, leveraging Term Frequency-Inverse Document Frequency encoding to process software source code akin to natural language processing. A semi-synthetic dataset is constructed using the mostly basic Python problems dataset. A code generator is trained to produce potential solutions based on specific prompts. Incorrect answers that do not meet the established test criteria are retained as defective code samples. The study evaluates the efficacy of the AdaBoost classifier in detecting such invalid code. Recognizing the critical influence of hyperparameter settings on classifier performance, several contemporary optimization techniques are employed to fine-tune these settings. Additionally, a modified version of the particle swarm optimization (PSO) algorithm is introduced, which demonstrates the highest performance among the tested methods, achieving an accuracy exceeding 0.983208. This approach underscores the potential of integrating advanced classification techniques with optimized hyperparameter tuning for effective software defect detection.