Using Geometry in Data Extraction

(Deutscher Text weiter unten.)FLIE, Form Labelling for Information Extraction, is the title of a recent paper resulting from an industrial collaboration in the field of insurance. The paper was presented virtually in November 2020 at the Future Technologies Conference. The project lasted a year and finished in July 2020 by delivering a prototype of a system for data extraction from Swiss insurance policies. The work was led by Professor Thomas Hanne and involved Professor Ela Pustulka and a Master student, Phillip Gachnang, who was the research assistant on the project.The project started by writing software to extract data from insurance policies in pdf format and to anonymise it. In the summer of 2019 we visited several brokers in Switzerland, going as far as Lausanne, and used the software to extract data from over 20’000 policies and related documents. The next step consisted in creating data…