Classifying address components of Thai mail by natural language processing

Since the writing format of Thai postal address is not fixed, it is difficult to classify the address components. This paper proposes a method to classify address components by using natural language processing (NLP) in order to absorb the nonfixed writing format and a little misspelling. This method finds the zip code and house number and uses them to extract only the address components from the overall destination address block. Secondly, we find the prefix of province that is the largest area component in the address. The province name following the searched prefix is a key to classify the smaller districts such as district and locality by matching in database. In case of a little misspelling, the most similar district in the matched province domain is selected as candidate, and the thresholding determines the district. In experiments, we utilized 500 address samples. The results show 86% accuracy.