A Novel Developer Portrait Model based on Bert-Capsule Network

In order to ensure code quality, it's necessary to construct portraits for developers, which could analyze their behavior to provide personalized programming suggestions. However, most of the existing developer portrait algorithms only use global features and ignore local features extracted from log texts, which leads to the lack of comprehensive personality analysis. To solve this problem, the proposed method proposes a novel developer portrait model, which could describe developers' programming styles more accurately with both global and local information extracted from texts. The proposed model firstly collects the log data produced in the process of continuous integration development. Afterwards, the proposed method proposes the personality portrait model based on BERT-Capsule network, which successfully combines global semantic features and local emotional features. The experimental results show that the proposed BERT-Capsule model can effectively extract the contextual information and the local emotional information of the text, thus improving classification performance of the developer portrait model.