Explainie -explaining Information Extraction

Business Intelligence (BI) over unstructured text is under intense scrutiny both in the industry and research. Recent work in this field includes automatic integrating of unstructured text into business analytics, model recognition, and probabilistic databases to handle uncertainty of Information Extraction (IE). However, still an open issue is how to handle IE quality, which is a part of ETL like process for the BI. Precision of IE is still too low for BI and, according to Sunita Sarawagi in recent survey on IE, we are still far from a comprehensive quality model for IE. Currently the BI user has neither methodology nor tools, which would help him to discover if the result is an unexpected fact or an error in IE. In this work we present preliminary results on developing methodology and tool (ExplainIE), which helps users to debug unexpected results. ExplainIE presents results within BI tool and auxiliary view on low level detail (e.g., entity graph). We consider two kinds of users: BI and IE expert.