Exploiting Unstructured Data Sources for Novel Predictive Applications

A barrier to applying Predictive Analytics more broadly is that more than 95% of information is unstructured (documents, email, web pages, audio, images, etc.) making it difficult for analytic systems to identify relationships between data elements. However, some structure can be extracted using technologies such as natural language parsing, entity extraction, image matching, context-awareness, and behavior modeling. This session will cover some of the Contextual Intelligence work at the Palo Alto Research Center (PARC) in creating new business opportunities based on predictive analytics applied to unstructured data sources. We will demonstrate the use of such methods in a case study of a consumer service that personalizes recommendations by predicting a mobile users future locations and leisure activity category. The service is in trials in Japan during Spring 2010 under the name Machireko.