| Popis: |
Recent works on handling unstructured text employ multilevel filtering techniques for identifying the key terms in documents and then apply mining techniques to extract necessary information. Though these techniques are more efficient in information retrieval, they cannot be applied directly for information extraction, for documents that are more critical in context and also accuracy cannot be expected. Further, loss of hidden and significant information cannot be tolerated in data critical applications emerging based on unstructured documents. Hence, a novel idea of re-organizing the unstructured textual model into feature enriched structured graphical model by adding spatial, logical, lexical, syntactical and semantic features is proposed. The generated graph depicts relationships across the document at all levels from its micro level token to macro level document. Moreover, a structural pattern identification algorithm for generating an XML schema from the generated graph is also recommended. The experimental outcome for a real-time dataset is presented. |
