Accurate Context Extraction from Unstructured Text Based on Deep Learning

Information today is the pillar of the development and the economic growth. A reliable information comes from a flawless exploitation of the appropriate data. The semantic analysis of data is one of the key factors that have a direct impact on the quality of the information that is extracted from it. The subject matter of a given document is closely related to the context of the information conveyed by that document. The reliable identification of the context of a document is, in our opinion, an important added value for an efficient semantic exploitation of the data. In this paper we propose a new approach for the accurate identification of the context associated with a document, given a precise definition of the context. The system has been evaluated on corpora from the "New York Times" and on a context dataset, called "WikiContext", that we have built ourselves for this purpose. Our experiments have shown that our system outperforms the most advanced context extraction systems.