Understanding Financial Transaction Documents using Natural Language Processing
Prateek Jain, Kunal Verma, Aniket Gaikwad, Pramod Gadde · 2019
In this paper, we share our experiences creating NLP based AI platform for finance - Appzen (http://www.appzen.com). AppZen's auditing technology is being utilized by over 500 enterprise customers including multiple Fortune 500 companies for auditing employee expenses. AppZen's technology can process, analyze and identify relationships between various kinds of transaction documents such as - receipts, invoices, contracts and purchase orders. Each type of transaction document requires custom processing and analysis due to the diversity in language and structure of the document. Contracts typically require deep understanding of the content such as identifying sentence structures, identifying entities and relationships between them compared to receipts and invoices, which are somewhat semi-structured and require a different kind of processing. We elaborate on the challenges we have experienced and use of NLP in conjunction with a lightweight semantic layer to alleviate these challenges.