AIBRD: Automated Identification of Bug Report Description

ABSTRACT Bug report descriptions usually contain three types of key information: observed behavior, expected behavior, and steps to reproduce. These elements help developers identify, reproduce, and fix software bugs. Manually identifying them in large bug reports is time‐consuming, so automation is needed. However, existing methods mainly rely on features extracted from individual sentences and neglect contextual information, which leads to poor performance in bug report sentence classification. To address this problem, we construct a semantic knowledge base to capture contextual information and then employ a multidimensional fusion strategy to integrate features from multiple sources for bug report sentence classification. The semantic knowledge base consists of three components: a keyword co‐occurrence matrix, synonym sets, and causality pair sets, which together capture contextual information. We then fuse the semantic knowledge base features with BERT word embeddings through a cross‐attention mechanism, followed by the concatenation of discourse pattern features to improve classification accuracy. We evaluated our method on a context‐dependent dataset, achieving an accuracy of 73.4% and a Hamming loss of 0.167. Experimental results indicate that AIBRD outperforms existing methods. Compared with the state‐of‐the‐art baseline DEMIBuD‐H, AIBRD improves macro‐average accuracy by 3.3% and macro‐average F1 score by 7.0%.