A project that enables the company to include AI and Machine Learning in the constitution of its database of domains and URLs, and to optimize internal classificationperformance.

Olfeo offers a web security gateway to secure and optimize corporate Internet traffic. It can determine the risk posed by a website accessed by an Internet user, and take appropriate action: informing the user of the risk involved, blocking the site, etc.

With a database of over 20 million domains, corresponding to hundreds of millions of URLs, Olfeo is the only French player to offer a database of websites finely categorized according to French usage and legal framework. Olfeo is the number one web proxy in France, with more than 1,000 ETIs and government agencies.

The domain database is the nerve center of the Olfeo solution. Its accuracy and relevance are among Olfeo's major differentiators. It is therefore essential to ensure its quality and completeness over time. The current approach is based on keyword pre-ranking where possible, but always verified by manual ranking. This achieves unrivalled ranking quality, but with the exponential growth of new domains being created, this solution needs strengthening.

Indeed, with 70 million new domains created worldwide every month - including 13 million malicious ones - and a dynamic evolution of content over time, Olfeo has decided to implement additional techniques to continue ensuring irreproachable database quality despite the large volume of new content generated, while preparing a European extension of the solution.

The METIS project, with a budget of 1 million euros and co-financed by the RAPID scheme (Régime d'Appui à l'Innovation Duale) and supported by the French Ministry of Defence's Defense Innovation Agency and the French Armament Procurement Agency (Direction Générale de l'Armement), is therefore being launched in 2020. It aims to implement AI and Machine Learning algorithms to improve the performance of Olfeo's in-house classification tools. The project includes a key dimension of innovation around data classification methods and Machine Learning to identify the category to associate with a URL based on analysis of the web page content.

Carried out in collaboration with the Université de Reims Champagne-Ardenne and the computer science laboratory of the Université Grenoble Alpes over a 3-year period, the project combined exploration of the state of the art in semantic analysis and classification, the implementation of automated processing processes, the training of various Machine Learning models with existing Olfeo data, and finally the deployment of a Deep Learning model that optimized the content recognition rate.

The project was entirely successful in achieving its objective. Classification relevance was improved by up to 30% for certain types of content, while maintaining the quality of the existing database, with a URL recognition rate close to 100%. These results enabled us to integrate the classification algorithm into the Olfeo pre-qualification tool.

Olfeo continues its dual approach, combining classification using Machine Learning techniques, while allowing humans to retain control over the final classification. By increasing the accuracy of automatic classification, Olfeo is paving the way for automatic validation of certain types of content in the near future.

"The experience gained by Olfeo on such important issues as AI is excellent, and augurs well for the future. is excellent, and augurs well for future opportunities, The context is so promising. At a time when the AI bubble is bursting, chatGPT et al, it's crucial for a company that aims to become a French and European leader in French and European cybersecurity, it is crucial to master these challenges and take advantage of their strength," asserts Alexandre Souillé, CEO and founder of Olfeo.