Pretrained models for code have exhibited promising performance across various code-related tasks, such as code summarization, code completion, code translation, and bug detection. However, despite their success, the majority of current models still represent code as a token sequence, which may not adequately capture the essence of the underlying code structure.

Requirements review is the process that reviewers read documents, make suggestions, and help improve the quality of requirements, which is a major factor that contributes to the success or failure of software. However, manually reviewing is a time-consuming and challenging task that requires high domain knowledge and expertise. To address the problem, we developed a requirements review tool, called Rebot, which automates the requirements parsing, quality classification, and suggestions generation. The core of Rebot is a neural network-based quality model which fuses multi-modal information (visual and textual information) of requirements documents to classify their quality levels (high, medium, low). The model is trained and evaluated on a real industrial requirements documents dataset which is collected from ZTE corporation. The experiments show the model achieves 81.3% accuracy in classifying the quality into three levels. To further validate Rebot, we deployed it in a live software development project. We evaluated the correctness, usefulness, and feasibility of Rebot by conducting a questionnaire with the users. Around 76.5% of Rebot's users believe Rebot can support requirements review by providing reliable quality classification results with revision suggestions. Furthermore, Around 88% of the users believe Rebot helps reduce the workload of reviewers and increase the development efficiency.

In the field of requirement document quality assessment, existing methods mainly focused on textual patterns of requirements. Actually, the cognitive process that experts read and qualitatively measure a requirement document is from outward appearance to inner essence. Inspired by this intuition, this paper proposed a Multimodal Requirement Document Quality Analyzer (MRDQA), a neural model which combines the textual content with the visual rendering of requirement documents for quality assessing. MRDQA can capture implicit quality indicators which do not exist in requirement text, such as tables, diagrams, and visual layout. We evaluated MRDQA on the requirement documents collected from ZTE and achieved 81.3% accuracy in classifying their quality into three levels (high, medium, and low). We have successfully applied MRDQA as a pre-filter in ZTE’s requirement review system. It identifies low and medium quality requirements, thereby allows review experts to focus only on high-quality requirements. With this mechanism, the workload can be greatly reduced and the requirement review process can be accelerated.

We present PTracer, a Linux kernel patch trace bot based on an improved PatchNet. PTracer continuously monitors new patches in the git repository of the mainline Linux kernel, filters out unconcerned ones, classifies the rest as bug-fixing or non bug-fixing patches, and reports bug-fixing patches to the kernel experts of commercial operating systems. We use the patches in February 2019 of the mainline Linux kernel to perform the test. As a result, PTracer recommended 151 patches to CGEL kernel experts out of 5,142, and 102 of which were accepted. PTracer has been successfully applied to a commercial operating system and has the advantages of improving software quality and saving labor cost.