We developed AI algorithms to use available information on social media with the aim to enhance risk-modelling and improve compliance monitoring, data management and planning of inspections for regulatory delivery authorities:
Food safety compliance: An AI algorithm was developed to analyse restaurant reviews posted on a social network and identify potential hygiene issues in establishments. The algorithm was trained on a dataset of 5,000 reviews, achieving accuracy rates between 81% and 83%, showcasing its efficiency. The use of large language models (LLMs) allowed for automatic classification of the reviews, with results comparable to human performance, along with speed and significantly reduced costs.
Project: Reforming Regulatory Inspections in Italy at Regional and National Level (2021-2024)
Beneficiary: Regulatory delivery authorities of the Lazio region (Italy)
Project developed by: OECD – Organisation for Economic Co-operation and Development
Environmental authorisations compliance: In another assignment, a web scraping algorithm was developed to identify businesses online that should require environmental licensing – the Single Environmental Authorisation (AUA) for pollutant emissions but were not registered. The purpose was to address the lack of data held by authorising entities and the consequent lack of businesses known by inspection authorities. The algorithm extracts relevant data such as VAT numbers from businesses’ websites and cross-references them with the authorities’ registry to detect non-compliant activities.
Project: Reforming Regulatory Inspections in Italy at Regional and National Level (2021-2024)
Beneficiary: APPA (Provincial Agency for Environmental Protection) in Trentino-Alto Adige (Italy)
Project developed by: OECD – Organisation for Economic Co-operation and Development
These cases demonstrate how the use of AI can provide effective and efficient innovative tools to support regulatory delivery authorities in improving inspections and gathering crucial information for regulatory compliance.