Using Approximate String Matching for Verifying Cosmetic Product Names in the Cosmetic Notification Request Form

Abstract

Objectives: To develop a program for checking cosmetic product names in the cosmetic notification request form using approximate string matching by Levenshtein edit distance and to evaluate the program's performance by comparing its verifying result with that from experts. Method: The program development framework was based on the system development life cycle theory. The problems on verifying cosmetic product names were analyzed with two problems identified including being time consuming and inaccuracy in the verification of product names. A web-based application for verifying cosmetic product names was developed. The main algorithm in the application was an approximate string matching by Levenshtein edit distance. The performance of the application was assessed using F1. Results: The n-gram technique was applied to create a set of appropriate features for increasing the performance in string matching. The appropriate n-gram in this study was trigram with thresholds at 80, 85, and 90. The threshold at 80 meant that two strings with a text similarity greater than or equal to 80 percent were defined as the same string. In comparing the efficiency of the program using the 8 engines of Thai word segmentation, the best engines at thresholds set to 85 and 90, were the attacut and tltk, respectively. There were no significant differences in the results on verification of cosmetic product names of the program and those from experts. Conclusion: This program was developed for verifying the names of cosmetic products in the notification request form using approximate string matching. The results of verification on cosmetic product names are accurate and consistent to those of experts. It can be used to support authorities in preliminary verification of information in the cosmetic notification request form.