Improving the efficacy of approximate searching by personal-name

Abstract

We discuss the design and evaluation of a method to find the information of a person, using his/her name as a search key, even if it has deformations. We present a similarity function that is an edit distance function with costs based on the probabilities of the edit operations but depending on the involved letters and their position. The distance threshold varies with the length of the searched name. The evaluation of the efficacy of approximate matching methods is usually done by subjective relevance judgements. An objective comparison of five methods, reveals that the proposed function highly improves the efficacy: for a recall of 94%, a fallout of 0.2% is obtained.