Risk Mining: Company-Risk Identification from Unstructured Sources
Timothy S. Nugent, Jochen L. Leidner · 2016
Risk permeates all aspects of doing business. However, support tools capable of systematically identifying the complete spectrum of risks that a company might face are currently lacking. Such a tool would need to reliably identify company-risk relationships from unstructured sources, therefore providing a qualitative assessment of risk exposure. We propose a supervised learning approach that combines a weakly-supervised risk taxonomy, named entity tagging and dependency tree analysis in order to perform company-risk relationship classification. We demonstrate that a support vector machine using a tree kernel, trained on hand-annotated articles from Reuters News Archive, is capable of significantly outperforming a selection of alternative classification algorithms. To our knowledge, this is the first example of company-risk relationship extraction.