Entity recognition and resolution in semi-structured data

Potentially usable business information exists in unstructured form. This information, although machine readable, resides in unstructured human language texts that are difficult to process by computers. Within this information are references to real world entities, which are the focus of this paper. More specifically, we address the recognition of references to entities and their resolution, in the context of semi-structured data. This kind of data is structured according to a model which defines only generic semantics for its data elements, or includes data elements that contain natural language text. Semi-structured data presents new challenges and opportunities. In this kind of data, grammatical evidence is very often insufficient for entity recognition, since short sentences and simple expressions are predominant. However, the contextual information given by the structure of the data opens new possibilities for innovative techniques. We propose an approach to integrating the support for structured data throughout the complete process. It will be evaluated by studying three entity types in two scenarios with different semi-structured data formats, each one with distinct characteristics.