Finding semantic correlation in structured data from unstructured notes

摘要

In financial reports, sometimes, unstructured notes explain the nuances of the financial data described in structured form. The problem of interest here is to parse such unstructured notes and find the semantic correlation that it implies in the structured data. While techniques like Named Entity Recognition have been used for extracting information from unstructured text, it has been observed that in order to discover meaningful semantic correlation in structured data, it is important to parse the unstructured notes in accordance with the terms and relations specified in the ontology of the domain. In this paper we examine this problem and present a mechanism to address the extraction of relations from unstructured notes and determine the semantic correlation of such relations with the structured data. Once such correlations are established, it becomes easier to understand the structured data, or have an automated question-answering system respond to user queries.