On Building a Podcast Collection with User Interactions

The podcast is a growing listening medium that has surged in popularity in recent years. Despite the great research opportunities, it has only attracted limited attention from the community so far. This is mainly due to the lack of available data collections that have considerably restricted research in academia. To facilitate it, in 2020, the Spotify Podcast Dataset was released, a corpus of 100k episodes with associated text transcript and metadata. However, no user interactions are available, hence making its usability challenging for certain domains, such as recommendation, personalisation, and user behaviour and consumption analysis. In this position paper, we present various approaches to augment such collection with user interactions, together with their respective
strengths and weaknesses. If developed further, this work has the potential of a broader impact on the research community.