Code Complexity and Version History for Enhancing Hybrid Bug Localization

Abstract

Software projects are not void from bugs when they are released, so the developers keep receiving bug reports that describe technical issues. The process of identifying the buggy code files that correspond to the submitted bug reports is called bug localization. Automating the bug localization process can speed up bug fixing and improve the productivity of the developers, especially with a large number of submitted bug reports. Several automatic bug localization approaches were proposed in the literature reviews which are based on the textual and /or semantic similarity among the bug reports and the source code files. Nevertheless, none of the previous approaches made use of the source code complexity despite its importance; as high complexity source code files have higher probabilities to be modified than the low complexity files and are prone to bug occurrences. To improve the accuracy of the automatic bug localization task, this paper proposes a Hybrid Bug Localization approach (HBL) that makes full use of textual and semantic features of source code files, previously fixed bug reports, in addition to the source code complexity and version history properties. The effectiveness of the proposed approach was assessed using three open-source Java projects, ZXing, SWT, and AspectJ, of different sizes. Experimental results showed that the proposed approach outperforms several state-of-the-art approaches in terms of the mean average precision (MAP) and the mean reciprocal rank (MRR) metrics.

Two-Level Information-Retrieval-Based Model for Bug Localization Based on Bug Reports

“…To overcome this issue, a semantic similarity feature is used to increase the effectiveness of the fault localization tasks. There are many approaches used in the literature to achieve semantic similarity by building the vectors that represent words; these vectors are called word-embedding vectors and capture the semantic relations between words [11]. The most utilized approaches for word embedding are Word2Vec and GloVe [11].…”

“…There are many approaches used in the literature to achieve semantic similarity by building the vectors that represent words; these vectors are called word-embedding vectors and capture the semantic relations between words [11]. The most utilized approaches for word embedding are Word2Vec and GloVe [11]. Word2Vec is a family of model optimizations and architectures that uses a large dataset to learn word embeddings [16].…”

“…For the similarity in text between the bug report and source code files, some previous studies used a Vector Space Model (VSM), such as [5,6]. Other studies, such as [4,[11][12][13][14], used a revised Vector Space Model (rVSM), which was proposed by Zhou et al [15]; this approach considers the length of the source code file, as a longer source file tends to have a higher probability of containing bugs [15]. In the proposed model, rVSM is used and it represents the bug report and source code file in the form of vectors of term weights.…”

“…The previously repaired bug reports feature was effective in enhancing the bug localization process and it is widely used in the literature; for example, it has been used in [4][5][6]11,12,14,15]. Bug reports that have been repaired previously could potentially lead to newer, similar bugs because similar bugs usually fix similar files [19].…”

AttentiveBugLocator: A Bug Localization Model using Attention-based SemanticFeatures and Information Retrieval

“…Hence, AttentiveBugLocator reduces the lexical mismatch by compactly learning semantic and contextual information with careful fusion of five IR features: VSM score [37], stack trace score [28], and code complexity score [21] to improve bug localization. Hence, our main contributions are as follows:…”