Numerous companies are trying to extract structured information from free text. As document repositories grow, automated processes for extracting this "locked" up knowledge become more pressing. However, the critical flaw in the majority of automated approaches, including machine learning and latent semantic analysis, is there is little emphasis on knowledge base construction.
Words are just "vectors" categorizing documents. Relationships between words and concepts embedded within sentences are completely ignored. Taxonomic relationships between concepts are missing.
|
We represent knowledge at the sentence level, we do not miss important relationships between words. Each unit of knowledge is a simple proposition, arranged in a semantic hierarchy, and is easy for users to understand. We map each sentence in a document to one or more simple propositions. Because all semantically equivalent sentences map to the same propositions queries can be made with high precision.
|
Copyright Logical Semantics, Inc. 2011. All rights reserved.
|
|
