A RoBERTa Based Approach for Address Validation

Yassine Guermazi, Sana Sellami, Omar Boucelma

Abstract: Address verification is becoming more and more mandatory for businesses involved in parcel or mail delivery. Situations where shipments are returned or delivered to the wrong person (or legal entity) are harmful and may incur several costs to the stakeholders. Indeed, addresses often carry incorrect information that need to be corrected prior to any shipment process. In this paper we propose a 2-step address validation approach consisting of (1) standardization and (2) classification, both steps are based on RoBERTa, a pre-trained language model. Experiments have been conducted on real datasets and demonstrate the effectiveness of the approach in comparison to other methods.