An Exploratory Study of Address Geocoding Techniques with a Focus on Solutioning for the Indian Address Problem

Abstract

Accurate geocoding of natural language ad- dresses is essential  to any industry dealing with logistics services, like  the postal  service,  e-commerce,  etc., and there are ongoing research efforts in address geocoding by on-demand   service  providers.   However, currently, existing  methods  assume the existence of a standard addressing   scheme  which can be deconstructed  and identified.To build automated,  scalable and self-learning systems that are capable of analyzing information  con- tained in natural language  addresses the systems need, along with user input, verification,  and updating of learning  models. This does not work well in places which do not have  a standard   addressing  scheme or where addresses can not be determined with the help of these addressing schemes due to the massive increase in popu- lation density and urbanization of countries. A number of techniques  have  been proposed  by various  authors to provide  a solution for address  geocoding  of natural language  addresses that have  varying accuracy.  There have  also  been proposed  solutions  as dynamic  models that grow to accommodate  changing landscapes.We also discuss  relevant  issues  with these  systems  such  as data collection, evaluation metrics and bench-marking  to find accurate location. After going through  these proposed solutions and current implementations, we conclude by studying the applicability of these solutions for Indian ad- dresses, their limitations and provide  possible directions for future work.