Multimodal Software Defect Severity Prediction Based on Sentiment Probability

Software systems generate a large number of software bugs during their life cycle, and timely detection and repair of these bugs is a key issue in improving software quality and maintaining software security. Therefore, this paper proposes a severity prediction on affective probabilistic multimodel software bugs. First, this paper uses RoBERTa as a sentiment analysis model and proposes a model training method for defective knowledge enhancement. We use Stack Overflow to construct a manually annotated sentiment probability dataset. Evaluating consistency between sentiment annotators by calculating Fleiss ' kappa values. Next, the difference in the impact of defects of different severity on users is reflected by the probability of sentiment. Using sentiment traits for the next stage of prediction. Finally, these include robust data processing of heterogeneous bug data, a complementary integrated learning framework that incorporates large linguistic and traditional tabular models, and a powerful model integration strategy based on a novel combination of multi-layer stacking and cyclic k-fold integrated bagging. Our comprehensive empirical assessment shows that SPM is more accurate and reliable than the popular defect severity prediction methods.