Serendipity Engine

We increasingly rely on algorithmically generated recommendations to navigate in both online and offline contexts: listening to music on streaming platforms, reading news online, or following recommendations about activities and events in your favorite city. These recommender systems help us dealing with the abundance of available information, but at the same time raise questions about their impact on individual citizens and society.

​Many advocate for designs for serendipity in recommenders, but what does this mean in practice? While serendipity is generally understood as a beneficial design principle ought to deliver societal value, putting it into practice still presents major challenges. The Serendipity Engine project sets out to address these challenges and support societal stakeholders in designing recommender systems to foster serendipity in public contexts.

This four year SBO project consists of four Work Packages:

• Value of Serendipity – Serendipity is known as a vague concept, used by many in different ways. This work package aims to establish a thorough understanding of serendipity in recommender systems. More specifically, we aim to identify the value of serendipity for different stakeholders and in different contexts.
• Data Discoverability on the Web – The current web can be interpreted as a graph where every web page is a node that is connected with each other. But in the current state of things, the connections between those web pages are not semantically understandable by machines. This work package will semantically describe web pages so that a query engine could traverse them, which would leverage the innate serendipity of the web.
• Recommender Systems – This work package aims to increase experienced serendipity through changes at the level of the recommendation algorithms. Hereto we seek to define actionable metrics of experienced serendipity, and investigate how inductive biases in different recommendation algorithms affect experienced serendipity. These insights will be incorporated within the emerging paradigm of reinforcement learning.
• Designing and Evaluating  – Together with our project partners, we will implement and evaluate our research outcomes in different use cases. This will allow us to gather empirical insights in how design, contextual and personal factors influence users’ experiences of serendipity.