Besides our recommendation metadata site, a simple GLIA process leading to a set of data for input into a machine learning model is shown here. Guideline summaries are used as source documents in GEM Cutter to define recommendations. These are then evaluated in GLIA as a way of identifying recommendations that could be difficult to implement from those that are easy. Based on this training and test data, we can create a machine learning model to help automate the implementation of guidelines and present developers with valuable feedback regarding their work.
As in all machine learning tasks, the challenge is to create a gold standard, which in this case is a solid set of recommendation assessments. We also need to identify what elements of the recommendation would prevent it from being implemented. It may be specific to a particular audience or it may be something universal, such as vague or ambiguous terms. Our additional metadata may be a source of input for inclusion in the model.