This post discusses the use of Pointwise Mutual Information (PMI) to sparsify knowledge graphs in supervised machine learning tasks. It explains the need for graph sparsification and provides a concrete example using medical data. The post also delves into the details of using PMI to assess the relevance of edges in the graph and remove irrelevant edges. It concludes with caveats and potential improvements when using PMI for graph sparsification.
Table of contents
PMI: Pointwise Mutual InformationBack to our taskThree Caveats — and potential improvements35 Impressions