Gay, GregoryHeimdahl, Mats2020-09-022020-09-022013-04-15https://hdl.handle.net/11299/215918There are a class of problems - such as deciding on a series of development practices - that challenge AI approaches for two reasons: (1) well-defined data is necessary, but it is not clear which factors are important and (2) multiple recommendations must be made, and dependence must be considered. Recommender systems offer inspiration for addressing these problems. First, they can make use of data models that broadly pair a series of recommendations with generalized information like project descriptions and design documents. More importantly, they offer feedback mechanisms to refine the calculated recommendations. Detailed feedback could be factored back into the data model to, over time, build evidence and context for recommendations. Existing algorithms and data models would be amenable to the addition of feedback mechanisms, and the use of these dynamic models could lower start-up costs and generate more accurate results as the model grows. We believe that feedback-driven dynamic prediction models will become an exciting research topic in the field of AI-based software engineering.en-USReport