Classification of business addresses via deep learning

PTA develops sequence-to-sequence language models (Seq2Seq) using the Python framework PyTorch and validates them using previously generated test data in a benchmark. Seq2Seq models essentially consist of two components: an encoder and a decoder. While the encoder analyzes and semantically captures the input text, the decoder decides whether the respective delivery address is a private or business address. Since the address data is transmitted to the customer by the clients via various input channels and interfaces, it is not uncommon for individual address components (e.g., title, first name, last name, company name, street, house number, house number suffix, postal code, city, or district). The model must therefore be able to tolerate such inconsistencies in order to achieve a sufficiently high classification accuracy.