Developing a Recommender System for a Public Service Organisation
The BBC is on a journey to make our audience experience more relevant and personalised, and a key part of our success lies in our ability to provide recommendations. At the BBC, we believe that recommendations should reflect the breadth and diversity of our content, while meeting our editorial guidelines. In this talk, I will describe how we develop a recommender system for BBC Sounds, including the Machine Learning model used, the architecture behind the engine, and the unique challenges we face to uphold the editorial policy and the values of the organisation.
Key Takeaways: • AI in a public service organisation brings unique challenges • ML is a very small part in building a recommender system for production • Interdisciplinary collaboration is key
Bettina is a Senior Data Scientist at the BBC. Her team aims to use machine learning algorithms to provide a better experience to their audiences, mainly through personalisation. She works very closely with the Data Engineering, Editorial and Product teams. Bettina has mostly been involved in building a recommender system for production use for one of the main BBC products. The Machine Learning algorithm used is hybrid and the code is developed in Python. Google Cloud Platform tools are used to manage the resources and to store the data, Airflow for the automation, and Redis for serving. Bettina has been involved in all of the steps: from the algorithm development, to the engine productionisation, but also in making sure that the recommendations are compliant with the editorial policies and company values.