PROJECTS

RISK GUARD: Effective Risk Assessment in Public Procurement 

FCT Project Reference: 2024.07601.IACDC – RiskGuard: Effective Risk Assessment in Public Procurement (DOI:10.54499/2024.07601.IACDC)

The RISK GUARD project aims to strengthen risk detection and oversight mechanisms in public procurement by developing a robust, data-driven analytical framework. Funded by FCT Portugal, the project focuses on improving institutional capacity to identify irregularities and potential integrity breaches, particularly through the use of red flag indicators, aligned with international typologies from the OECD, SIGMA, and Transparency International.

The project integrates two main scientific components:

1. Natural Language Processing (NLP) to automatically analyze justifications and contract object descriptions in tender documents, flagging vague or non-compliant language;

2. Machine Learning (ML) based Risk Detection, using structured datasets and rule-based indicators to identify patterns of concern such as direct awards, repeated suppliers, or anomalous bidding behavior.

This project builds on prior collaborations with the Portuguese Court of Auditors (Tribunal de Contas) and develops a modular, scalable infrastructure for processing large public procurement datasets. The methodology involves a combination of statistical analysis, indicator engineering, and AI-driven insights, aiming to support both auditors and policymakers in improving transparency and accountability in public spending.

KNOW-NET-COMPET: Embedded Knowledge and Networks for Complexity in Antitrust Collusion Mitigation

FCT Project Reference: 2024.07378.IACDC – Embedded Knowledge and Networks: Complexity in Antitrust Collusion Mitigation (DOI:10.54499/2024.07378.IACDC)

The KNOW-NET-COMPET project explores the use of advanced data-driven techniques to improve the detection and analysis of collusive behaviors in markets, such as cartels and other anti-competitive practices. Funded by FCT Portugal, the project addresses the limitations of traditional antitrust tools by integrating network analysis, natural language processing (NLP), and knowledge graph methodologies.

Its core objective is to develop an interpretable decision-support tool capable of combining institutional knowledge embedded in previous decisions with external data sources. This unified knowledge structure enables the identification of potential collusive patterns, missing links between entities, and the assessment of risks in specific market segments.

The project unfolds in three key phases:

1. Data Mapping and Theoretical Foundations: identifying and cataloguing relevant data sources and conducting a comprehensive literature review on computational antitrust and AI/ML models;

2. Graph Construction and Integration: building a flexible knowledge graph architecture that integrates structured and unstructured data and allows continuous updates;

3. Applied Case Studies: testing the system through concrete use cases such as link prediction and market segmentation, demonstrating its potential to support antitrust analysis and policy.
   
   

This project contributes to the broader effort of making AI-based regulatory tools more transparent, explainable, and adaptable to the complexity of real-world market dynamics.