Approximate proper name matching

Abstract

Approximate proper-name matching uses concepts of approximate string matching and applies them to special case of finding ‘close’ or ‘similar’ names, to an input name, from a large database of names. Such Proper-Name-Approximate matching finds applications in situations where a user is unsure of how a person’s name is spelled, such as in a telephone directory search system or a library search system where a user wishes to search books on an author’s name.In this report we examine this problem in two main aspects: How to organize data efficiently, so as to obtain relevant results quickly, and how to develop suitable search techniques which would rank results suitably. We suggest four new data organization techniques to replace the current standard technique, Soundex, and we suggest refinements to the currently available search techniques. We then assess the performance of the developed techniques and compare them against the currently available ones. We also show that the developed techniques provide us with better a result faster that is they take lesser time per query than the current methods. In the course of evaluation we also suggest a new assessment technique (weighted recall) which gives a better measurement of system performance than the standard assessment techniques.