Local, Domain-independent Heuristics for the FEIII Challenge: Lessons and Observations.

The recently concluded Financial Entity Identification and Information Integration (FEIII) competition is an example of a domain-specific entity linking challenge. Given a variety of datasets describing financial institutions, the goal of the competition was to interlink entities referring to the same underlying entity in a minimally supervised manner. In this paper, we present our solution to the challenge. Using local, domain-independent heuristics, namely thresholded block purging and a simple Jaccard matcher, we devised a solution that has execution times of less than a minute on the alloted tasks. Although the method did not achieve competitive precision, it was recall-friendly, suggesting that it is useful both as an easily implemented baseline, as well as a generic preprocessing step for more expensive, precision-friendly algorithms that are fine-tuned for specific domains.