Many datasets of interest today are best described as a linked collection of interrelated objects or, simply, linked data. In the bibliographic domain, for example, the objects are publications, authors and citations, while, in the biomedical domain, protein networks are an example of linked data. In sales data, the items sold together form a linked graph. Link prediction refers to the problem of discovering links that are expected to occur in future between objects that are currently not directly connected. Examples include discovering potentially new drug applications, potential promotional opportunities for sales, and new relations between users in social media. The current NSF Fundamental Research Project is permitting CVDI to improve the state‐of‐the‐art Link Prediction methods to incorporate link interestingness and, in some cases, the type of link; it also is allowing for testing of applicability in several domains. However, when finished (July), there will still be work required to move the research into robust practice. This includes permitting easy incorporation of new data, generating multiple models (to reduce single points of failure), visualizing the underlying graphs, and understanding the nature of the prediction. This proposal seeks to address these issues.