Dependence of Binary Associations on Co-occurrence Granularity in News Documents

We describe and formalize an approach to correlate binary associations (such as between entities and events, between persons and events, etc.) implied by News documents on the co-occurrence granularity (such as document-level, paragraph-level, sentence-level, etc.) of the corresponding text phrases in the documents. Specifically, we present both qualitative and quantitative characterization of searching News documents: former in terms of the nature of the content and the queries, and latter in terms of a metric obtained by adapting the notions of precision and recall. Specifically, the approach tries to reduce the manual effort required to analyze the News documents to compare the three alternatives for granularity of co-occurrence. Furthermore, the analysis suggests ways to improve retrieval performance as illustrated by applying our findings to News documents for the year 2005.