Automated Due Diligence: Building Knowledge Graphs from News

Before a company enters a new business relationship it has to perform a background check, known as due diligence. It is commonly carried out by a human expert and involves screening a large amount of unstructured textual information (e.g. news articles), which is extremely labor intensive. We propose to automate this process, which would allow to, firstly, reduce the time needed for article screening, and, secondly, discover new insights about the network the company operates in. The solution includes (a) a classifier that detects articles containing negative events about the company of interest, and (b) a knowledge graph that combines the gained information with structured data sources. We report promising results of the novel approach to utilize semantic frames of the article’s predicates as features for the news article classification. Furthermore, we have successfully built a knowledge graph that combines information from different data sources. The proposed automated pipeline introduces a promising novel alternative for the commonly performed due diligence procedure.