Relation Extraction from Financial Reports

This project mainly focuses on using deep learning methods to extract relations
from the so called 10-K SEC financial reports, and adds them to an ontology for
further use. A 10-K report is a comprehensive report submitted by public compa-
nies each year to publish their financial performance. In the US, the 10-K reports
are required by the U.S Security and Exchange Commission (SEC) to provide the
investors with information on a company on which they can base their decisions to
invest. It is far more detailed than the annual report where it describes the com-
pany’s potential to succeed so it is useful for investors to refer to. In this research,
we mainly focus on the distant supervision method to construct the dataset from
the Financial Industry Business Ontology(FIBO) [2] and evaluate the performance
of two distant supervision relation extraction models. Additionally, we discuss the
potential flaws of distant supervision method on this task and investigate some pos-
sible improvements such as anaphora resolution to enhance the knowledge base,
and point out further research direction for the domain-specific relation extraction
area. In addition, this research provides results to Can Erten, a PhD student at the
University of York, who will use the ontology from the reports in his research.