Information Extraction in the Business Intelligence Context

Business Intelligence (BI) relies on corporate data to derive information for supporting strategic decisions. Source data internal to the corporation are used to assess productivity, lucrativeness, quality, etc, whereas external data are used to evaluate market share, expansion opportunities, competitiveness, etc. In a typical BI scenario, structured data, usually gathered from application databases, are integrated in Data Warehouses to support analytical tools. There is, however, a wealth of unstructured data which is harder to integrate but can provide valuable information: formal documents, reports, exchanged messages, and numerous websites on the internet. Information Extraction (IE) is one of the main players in harvesting the unstructured information embedded in digital documents. Researchers have developed a wide range of techniques to surface structured information from virtually all types of documents. Access to structured information enables richer analysis of data, which is the main goal of BI. This document describes research in IE and typical approaches to BI. This document is not intended as a survey on IE theory or techniques – for that we recommend the papers by Sarawagi [46] and Nadeau & Sekine [36]. Nor is the document intended to cover a broad range of BI issues, which are detailed in the books by Imhoff et al. [24] and Jarke et al. [28]. The objective of this document is to present typical approaches to IE and BI, to describe research that integrates the two areas, and to provide some insight on what may improve the applicability of IE techniques in BI scenarios. Section 2 presents IE research, analyzing techniques according to the type of input document, the dimensionality of the target extraction, and the variations on the typical IE workflow. Section 3 describes stablished BI technology and approaches to leverage documents in a BI scenario. It also suggests in which areas IE systems could be adapted to facilitate their use in BI tasks. Section 4 concludes the paper.