Relation Discovery from Web Data for Competency Management

This paper describes a technique for automatically discovering associations between people and expertise from an analysis of very large data sources (including web pages, blogs and emails), using a family of algorithms that perform accurate named-entity recognition, assign different weights to terms according to an analysis of document structure, and access distances between terms in a document. My contribution is to add a social networking approach called BuddyFinder which relies on associations within a large enterprise-wide "buddy list" to help delimit the search space and also to provide a form of 'social triangulation' whereby the system can discover documents from your colleagues that contain pertinent information about you. This work has been influential in the information retrieval community generally, as it is the basis of a landmark system that achieved overall first place in every category in the Enterprise Search Track of TREC2006