Combining Attention-based Gated Bidirectional LSTM and ODCN for Software Defect Prediction

• Item type: Symposium(1)
• 公開日: 2021-08-30
• タイトル: Combining Attention-based Gated Bidirectional LSTM and ODCN for Software Defect Prediction
• タイトル / 言語: en
• タイトル: Combining Attention-based Gated Bidirectional LSTM and ODCN for Software Defect Prediction
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 予測・信頼性
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: Hiroshima University
• 著者所属: Hiroshima University
• 著者所属: Hiroshima University
• 著者所属(英): en
• 著者所属(英): Hiroshima University
• 著者所属(英): en
• 著者所属(英): Hiroshima University
• 著者所属(英): en
• 著者所属(英): Hiroshima University
• 著者名: Dingbang, Fang; Shaoying, Liu; Ai, Liu
• 著者名(英): Dingbang, Fang; Shaoying, Liu; Ai, Liu
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Software reliability plays an important role in the software lifecycle. Traditional defect prediction adopts static metrics as manual features to predict defects. Although static metrics can measure the complexity of software, they lack semantic and structural information about the code. This paper proposes a novel neural network: GB-ODCN, capable of capturing critical features from the source code to predict defects. GL-ODCN consists of a gated bi-directional long short-term memory network (Bi-LSTM) and a one-dimensional convolutional neural network (ODCN). Bi-LSTM constructs dependency of semantic features from the code. Besides, ODCN captures high-level semantic features for judging whether there are defects in the code. Moreover, attention mechanisms based on both networks enhance the importance of features by assigning weights. Experimental results show that GB-ODCN outperforms current state-of-the-art algorithms on several open-source repositories.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Software reliability plays an important role in the software lifecycle. Traditional defect prediction adopts static metrics as manual features to predict defects. Although static metrics can measure the complexity of software, they lack semantic and structural information about the code. This paper proposes a novel neural network: GB-ODCN, capable of capturing critical features from the source code to predict defects. GL-ODCN consists of a gated bi-directional long short-term memory network (Bi-LSTM) and a one-dimensional convolutional neural network (ODCN). Bi-LSTM constructs dependency of semantic features from the code. Besides, ODCN captures high-level semantic features for judging whether there are defects in the code. Moreover, attention mechanisms based on both networks enhance the importance of features by assigning weights. Experimental results show that GB-ODCN outperforms current state-of-the-art algorithms on several open-source repositories.
• 書誌情報: ソフトウェアエンジニアリングシンポジウム2021論文集; 巻 2021, p. 175-180, 発行日 2021-08-30
• 出版者 / 言語: ja
• 出版者: 情報処理学会