Result Aggregation for Knowledge-Intensive Multicultural Name Matching

In this paper, we describe a metasearch tool resulting from experiments in aggregating the results of different name matching algorithms on a knowledge-intensive multicultural name matching task. Three retrieval engines that match romanized names were tested on a noisy and predominantly Arabic dataset. One is based on a generic string matching algorithm; another is designed specifically for Arabic names; and the third makes use of culturally-specific matching strategies for multiple cultures. We show that even a relatively naive method for aggregating results significantly increased effectiveness over each of the individual algorithms, resulting in nearly tripling the F-score of the worst-performing algorithm included in the aggregate, and in a 6-point improvement in F-score over the single best-performing algorithm included.