Knowledge Enhanced Software Repository Search

点击翻译按钮获取中文摘要

Software stakeholders need to frequently find references from artifacts stored in the project repository to concretize their ideas and to back up their decisions during tasks such as impact analysis, feature location, source code refactoring, and software reuse. Project repository search serves as one key mechanism for stakeholders to find relevant information during this process. In the past decade, researchers have adopted keyword-based text retrieval methods to search for relevant artifacts aiming to satisfy stakeholders' information needs during those tasks. However, the current text retrieval methods usually lack a sufficient understanding of the project target domain. This has resulted in a considerable gap between the search queries and the software artifacts. Moreover, the output of those methods is often a flat list of artifacts in which the interconnections among the artifacts are disregarded. Inference on those connections, however, are essential to perform complex software engineering tasks. Due to such limitations, great effort is still needed for the stakeholders to effectively locate and infer from relevant artifacts in the project repositories. The main goal of this program is to investigate knowledge enhanced information retrieval techniques in order to strengthen the analysis and decision support for the project stakeholders. Knowledge about the target domains is indispensable for understanding the artifact semantics and for analyzing their connections. Therefore, leveraging such knowledge during repository searches can potentially overcome the limitations of current keyword-based methods. In particular, we aim to achieve three objectives: 1) design effective artifact processing techniques and tools to extract the domain knowledge from heterogeneous software artifacts; 2) develop a knowledge enhanced searching system that can adequately understand and address human queries during software engineering tasks; and 3) devise interaction techniques and corresponding tools that are embedded in the searching process to refine the extracted knowledge. Given the wide application of repository search during software development, our research results can be applied to support the information needs of various project stakeholders. The underlying algorithms for knowledge extraction and artifact comparison can be used to enhance many analytic tools that require drawing inferences from artifact semantics. The explicit knowledge representation and interaction techniques also offer advantages to support knowledge sharing for large projects. This research program will train 10 HQP (2 Ph.D., 3 MSc, 5 BSc) through hands-on research practices in large-scale software analyses. All HQP will be well-equipped with extensive knowledge of software engineering and expertise on natural language processing and knowledge representation techniques. The HQP working on objective 2 and 3 will also develop considerable experience in human-centered design.