A New Approach to Detecting Irregular Behavior in the Network Structure of Public Contracts

Abstract

Corruption scandals are a major concern worldwide. Situations of instability are usually a breeding ground for new forms of corruption. The COVID-19 pandemic has not been an exception creating new opportunities for fraud and corruption. Hence, there are no few cases where irregularities have been uncovered in different countries since the beginning of the pandemic. Based on 213,729 public contracts granted in Spain in the years 2020 and 2021, this paper proposes an empirical approach to detect irregularities in public procurement. The proposed approach is mainly based on the Node2Vec algorithm, a graph embedding algorithm that automatically learns the complex latent relationships among public contracting authorities and awarded companies. By using Node2Vec each node of the network of public contracts is transformed into low-dimensional dense vectors which are then clustered using the Self-Organizing maps (SOM) algorithm. Five different groups of contracts have been detected. Each group is finally labeled according to their risk of corruption by considering uncovered cases of irregularities that have appeared in the news. This paper contributes to the literature by proposing a new approach to detecting corruption practices by exploiting the network relationships among the different participants in public contracts representing a complementary approach to other traditional or machine learning methods.

Keywords: corruption, public procurement, self-organizing maps, fraud detection, graph embedding algorithm, Node2Vec