Automated Parsing of Geographical Addresses: A Multilayer Feedforward Neural Network Based Approach

Delivering mails and parcels is an age-old problem. With the advent of E-Commerce, one can now get almost everything delivered. The core piece of this delivery puzzle is the delivery address. Though there has been a lot of technological advancement in supply chain management yet, delivery address verification remains manually intensive or at the best, dependent on some hardcoded rule-based process. Addresses are found in a large number of patterns and variations. In order to understand any address, we need to extract essential components of the address. They are majorly influenced by the syntax and semantics of an address. Traditionally, component extraction is done using hard-coded rules based on the expert knowledge of how addresses are written in a particular geography. This paper proposes a machine learning based approach for address parsing using a neural network. The proposed approach is compared to the existing methods and its advantages are depicted in terms of accuracy, scalability, and support for variations in addresses.