OXIOM - Posting Invoices using the CatBoost Algorithm and Embedded Representations

The employment of artificial intelligence (AI) opens up possibilities for the automation of many everyday tasks, including accounting. Proper data extraction, as well as accurate and fast posting of invoices, have become crucial in modern business. Manual invoice posting demands time and expertise, while AI automates patterns, freeing accountants for complex tasks. This paper presents classification models based on machine learning algorithms, which efficiently solve this problem. Special attention was paid to the CatBoost algorithm due to its ability to work with categorical data, while embedding models were used to represent some categorical features. The combination of quality data collection and preprocessing with application of appropriate machine learning algorithms and natural language processing (NLP) concepts resulted in a model with high accuracy.