Neither in the Programs Nor in the Data: Mining the Hidden Financial Knowledge with Knowledge Graphs and Reasoning

Vadalog is a logic-based reasoning language for modern AI solutions, in particular for Knowledge Graph (KG) systems. It is showing very effective applicability in the financial realm, with success stories in a vast range of scenarios, including: creditworthiness evaluation, analysis of company ownership and control, prevention of potential takeovers of strategic companies, prediction of hidden links between economic entities, detection of family businesses, smart anonymization of financial data, fraud detection and anti-money laundering. In this work, we first focus on the language itself, giving a self-contained and accessible introduction to Warded Datalog+/-, the formalism at the core of Vadalog, as well as to the Vadalog system, a state-of-the-art KG system. We show the essentials of logic-based reasoning in KGs and touch on recent advances where logical inference works in conjunction with the inductive methods of machine learning and data mining. Leveraging our experience with KGs in Banca d’Italia, we then focus on some relevant financial applications and explain how KGs enable the development of novel solutions, able to combine the knowledge mined from the data with the domain awareness of the business experts.