Implement a recommendation engine to promote access to culture

Context

The Pass Culture provides young people aged 15 to 18 with a budget to explore the diversity of culture in France. In collaboration with the Pass Culture data team, we optimized the application's recommendation engine to better target cultural offerings and enrich the cultural practices of young people.

The Challenge

The collaboration between Pass Culture and Theodo Data & AI aimed to leverage our expertise in Artificial Intelligence to improve the application's recommendation algorithm. Our experts were able to address the two challenges presented by Pass Culture: increasing the diversification of offers booked by users while improving the quality of recommendations.

The Solution

We utilized our MLOps expertise to make model training scalable. Effective use of tools such as Airflow, Google Cloud Platform, and TensorFlow enabled us to significantly increase the volume of data used to train the model.

After optimizing the training pipeline, we implemented a new recommendation model inspired by architectures used by e-commerce platforms like eBay and video-sharing platforms like YouTube.

Tech Stack

Results

We utilized user click data on the application as implicit feedback for the recommendation engine to address the disparity in booking data previously used.