file.pdf (913.78 kB)

A Tale of Two Entity Linking and Discovery Systems

Download (913.78 kB)
journal contribution
posted on 01.11.2014 by Kathryn Mazaitis, Richard C. Wang, Frank Lin, Bhavana Dalvi, Jakob Bauer, William W. Cohen

The long-term research agenda of our group is to evaluate the potential of probabilistic logics for complex, large-scale problems which require data resources naturally encoded as relations. In pursuit of this goal, we compared two systems for performing automated entity discovery and linking in English-language text, as submitted to the 2014 TAC Knowledge Base Population Entity Discovery and Linking (EDL) track. Both systems are based on random-walk strategies for measuring similarity within graphs. The first system is PageReactor, a hand-engineering system originally designed for task of wikification. The second is based on ProPPR, a probabilistic logic programming language