Abstract:
Viewpoints are coherent collections of facts that describe a concept from a particular perspective. They are essential for a wide variety of tasks, such as explanation generation and qualitative modeling. We have identified many types of viewpoints and developed a program, the View Retriever, for extracting them from knowledge bases, either singly or in combinations. The View Retriever provides a general solution to the central problem in extracting viewpoints: determining which facts are relevant to requested viewpoints. Our evaluation indicates that viewpoints extracted by the View Retriever are comparable in coherence to those people construct.