Continuing with recommender systems, I wanted to build something for podcasts. I’ve found new podcasts by looking through the charts on iTunes but I prefer a personalized solution. I follow a few podcasts on twitter and I have had success with association analysis in the past, so I wondered if I could identify similar podcasts by looking at overlapping twitter followers. To test the concept, I downloaded the follower lists for two similar podcasts and found that approximately 50% of the followers were overlapping.
Using the podcasts database from Listen Notes and the Twitter API, I extended this concept to 1,000 podcasts and built a website using Django. I received feedback that people weren’t finding their favorite podcasts which led me to extend it to 3,000 podcasts. I received the same feedback but without a paid Twitter Developer account, one can only download a certain number of follower lists at one time, which makes it hard to scale to additional podcasts, or keep recommendations up-to-date. So you might not find your favorite podcast, but the recommendations work pretty well.