Abstract: A computer parses the document to identify a citation, where the citation serves as a pointer to a source reference. The computer determines a location in the document of a textual assertion associated with the citation. The computer calculates relevancy scores between the textual assertion and a corresponding source reference and between the textual assertion and at least one alternate source reference, where the relevancy scores are determined based at least in part on a machine learning algorithm trained with a plurality of training samples. The computer generates a suggested list of at least one of the source references or at least one alternate source reference based on the relevancy scores calculated by the machine learning algorithm and adds a training sample to the plurality of training samples of the machine learning algorithm in response to an action by a user responsive to the suggested list.