Knowledge Graphs: Venturing Out into the Wild

摘要

While we now have vast collections of knowledge at our disposal, it appears that our systems often need further kinds of knowledge that are still missing in most knowledge graphs. This paper argues that we need keep moving further beyond simple collections of encyclopedic facts. Three key directions are (1) aiming at more tightly integrated knowledge, (2) distilling knowledge from text and other unstructured data, and (3) moving towards cognitive and neural approaches to better exploit the available knowledge in intelligent applications.