Doctoral thesis

Holistic recommender systems for software engineering


206 p

Thèse de doctorat: Università della Svizzera italiana, 2017

English The knowledge possessed by developers is often not sufficient to overcome a programming problem. Short of talking to teammates, when available, developers often gather additional knowledge from development artifacts (e.g., project documentation), as well as online resources. The web has become an essential component in the modern developer’s daily life, providing a plethora of information from sources like forums, tutorials, Q&A websites, API documentation, and even video tutorials. Recommender Systems for Software Engineering (RSSE) provide developers with assistance to navigate the information space, automatically suggest useful items, and reduce the time required to locate the needed information. Current RSSEs consider development artifacts as containers of homogeneous information in form of pure text. However, text is a means to represent heterogeneous information provided by, for example, natural language, source code, interchange formats (e.g., XML, JSON), and stack traces. Interpreting the information from a pure textual point of view misses the intrinsic heterogeneity of the artifacts, thus leading to a reductionist approach. We propose the concept of Holistic Recommender Systems for Software Engineering (H-RSSE), i.e., RSSEs that go beyond the textual interpretation of the information contained in development artifacts. Our thesis is that modeling and aggregating information in a holistic fashion enables novel and advanced analyses of development artifacts. To validate our thesis we developed a framework to extract, model and analyze information contained in development artifacts in a reusable meta- information model. We show how RSSEs benefit from a meta-information model, since it enables customized and novel analyses built on top of our framework. The information can be thus reinterpreted from an holistic point of view, preserving its multi-dimensionality, and opening the path towards the concept of holistic recommender systems for software engineering.
  • English
Computer science and technology
License undefined
Persistent URL

Document views: 77 File downloads:
  • 2017INFO005.pdf: 109