Ontology-based similarity between text documents on manifold

This paper firstly utilizes the ontology such as WordNet to build the semantic structures of text documents, and then enhance the semantic similarity among them. Because the correlations between documents make them lie on or close to a smooth low-dimensional manifold so that documents can be well characterized by a manifold within the space of documents, we calculate the similarity between any two semantically structured documents with respect to the intrinsic global manifold structure. This idea has been validated in the conducted text categorization experiments on patent documents.