End-to-End Table Extraction from Annual Reports using DL and NLP

Abstract

Annual reports contain many important data and information – some of this data and information is included in tables. The extraction of these table data is associated with various challenges, including the unstructured nature of PDF documents and the wide variability of table representations. The aim of this master's thesis is to explore an innovative end-to-end solution that enables a user to interface with tabular data within annual reports in PDF format through natural language inputs. The thesis addresses two main challenges: the automated extraction of table data from unstructured PDF documents, and interfacing this data through user inputs in the form of natural language questioning – for example, allowing the user to ask a question about the table content in the annual report like: "What was the profit in 2023?". This aims to make the process of information retrieval easier and more efficient.