Serendipity in Recommender Systems: updates of 2023

Recommender systems are ubiquitous. Netflix recommends movies and tv series to users. Spotify recommends audio tracks. Facebook recommends friends. Although these recommendations are often precise, they are also often boring. For example, these systems often suggest items, such as movies or audio recordings that users are already familiar with or would find anyway by themselves. To mitigate this problem, researchers started optimising recommender systems not only for accuracy (items that users like), but also for serendipity. There is no consensus on the definition of the term serendipity in recommender systems (RecSys), but most often researchers consider an item serendipitous if it is relevant, novel and unexpected to the user. Relevance indicates that the user is interested in the item, novelty - the user had been unaware of the item prior to consuming it, while unexpectedness has a number of definitions with the most common being that the item is dissimilar to what the user usually consumes.

In the recent article [1], the authors indicated a mismatch between the dictionary definition of the term and the RecSys one. The dictionary definition of the term or generalised serendipity (how it is referred in the article) does not require the item to be novel or unexpected. According to the generalised definition of serendipity, an item is serendipitous to the user only if it helps them achieve at least one goal, which is different from goals the user wanted to achieve, when they started using the recommender system.

The authors of the article [1] provide details on each serendipity definition and examples how these definitions apply in real life. The authors also present an experimental setting, which allows to measure generalised serendipity in a field experiment.

For more details, feel free to check the article, which is also available on ResearchGate: https://www.researchgate.net/publication/366445545_Rethinking_Serendipity_in_Recommender_Systems

Referenced

1. Denis Kotkov, Alan Medlar, and Dorota Głowacka. 2023. Rethinking Serendipity in Recommender Systems. In ACM SIGIR Conference on Human Information Interaction and Retrieval (CHIIR ’23), March 19–23, 2023, Austin, TX, USA. ACM, New York, NY, USA, 5 pages. https://doi.org/10.1145/3576840.3578310