Student creates impact on viewer recommendation systems, with BBC R&D

Electronic and Electrical Engineering undergraduate student, Iason Chaimalas’ recently completed BEng project has been recognised for its impact, being featured on the BBC R&D website and will be presented at the Association for Computing Machinery (ACM) 17th Recommender Systems Conference (ACM RecSys 2023).

Supervised by Dr Laura Toni and Edoardo Gruppi from UCL, and Dr Ben Clark and Dr Duncan Walker from the BBC, Iason developed machine learning algorithms for recommender systems, like BBC iPlayer. Iason’s placement was enabled by the BBC's Data Science Research Partnership, of which, UCL is a member.

Iason developed a novel machine learning algorithm to address the issue of the “cold start problem”. The “cold start problem” highlights issues in recommendations for new and nearly-new users. Balancing the accuracy and diversity of recommendations is crucial for streamlined user experience and ensuring a fair promotion of items on the platform. With reduced metadata available for new users, maintaining accuracy and diversity of recommendations is more difficult.

Recommender systems are the machine learning models responsible for curating content for users on digital platforms, from streaming services, online shipping to dating apps. The systems draw from a range of meta data including user data on watch history, location and age to ensure dynamic and relevant recommendations.

Writing for BBC’s R&D blog Iason explains:

We tested our model on a real-world dataset with millions of user interactions on iPlayer, which included a representative proportion of new and nearly new users. This has resulted in higher accuracy and competitive diversity when compared to other successful recommender models from academic literature!

Within the blog post, Iason expands on his project, detailing how data on popularity fluctuations and metadata were used to develop a “metadata-infused popularity model”, boosting the diversity of existing recommender model “EASER” by using Metadata infusion.

The project has resulted in a conference paper on the findings being accepted at the upcoming 17th ACM conference on Recommender Systems, where researchers and practitioners across academia and industry share the latest challenges and trends within recommender systems.

Collaborating with industry

Iason's project was possible due to UCL Electronic and Electrical Engineering's proactive approach to engaging with industry across our education and research. The department actively encourages industry to work with our academics in providing student projects, this allows students to experience engineering in realworld environments and where their studies can have direct impact. It also allows creativity and diversity of thought when tackling industry problems and exposes our industry partners to the world class talent found in our undergraduate cohort of students.

For industry interested in exploring collaborations with UCL Electronic and Electrical Engineering, or providing student projects please explore our collaboration pages.

Links

• BBC Research and Development Blog: Addressing the cold start problem in recommender systems
• Dr Laura Toni's profile
• Learning and Signal Processing (LASP) research group
• ACM Conference on Recommender Systems
• BBC Data Science Research Partnership
• EEE Industry eXchange Network