Product idea - an episode-focused podcast player
-
Podcast app user interfaces work at the level of a show - you subscribe to a show, it recommends shows.
-
But this is an old-world way of thinking.
-
Really, I want a podcast app that focuses on episodes - that recommends me the best episodes of a given show, and the episodes from other shows that I might like.
-
Recommendation engines usually rely on two main signals:
-
- social, i.e. what we can learn from the behaviour of other users, e.g. "other users like you liked this"
-
- content, i.e. what we can learn from the substance of the content itself, e.g. "this episode is about X and Y, and you listen to a lot of stuff about both X and Y".
-
Viewed through this lens, what data can tell us which are the best episodes of a given show?
-
- Social - if you're a popular enough app, you can use behavioural analytics (e.g. the proportion of listeners that listen all the way to the end, or share the episode).
-
- Content - ask an LLM to read the transcripts, and ask it which episodes seem useful/interesting/funny, which involved famous guests, which involve trending topics, etc.
-
What about the episodes from other shows that I might like?
-
- Social - standard cross-user collaborative filtering, to identify episodes that other users like me liked. This might not work that well at an episode level, because so much of our listening behaviour currently is driven by show-level patterns.
-
- Content - build up a profile of the user's interests, and then look for episodes that seem relevant.
-
The tricky part, as always with recommendation engines, is to balance relevance with serendipity, e.g. just because I've listened to a couple of interviews with Demis Hassabis doesn't mean that I only want to listen to interviews with him...