Knowledge Revision for Document Understanding

A key issue in the area of knowledge revision is the definition of efficient and effective (ideal) refinement operators. The paper presents a closed loop incremental learning system, called INCR/CSL, that implements two ideal operators for upward (generalization) and downward (specialization) refinement. It is employed as the knowledge maintenance engine in an intelligent agent for the automated processing of paper documents. Experimental results in the area of document understanding for semantic indexing of documents in a digital library service show that INCR/CSL is able to cope effectively and efficiently with this real-world learning task.