Machine Learning System Implementation of Education Podcast Recommendations on Spotify Applications Using Content-Based Filtering and TF-IDF

Abstract

Spotify, this popular music and podcast streaming service, has a fundamental problem in assisting clients in finding podcasts that fit their interests. Thus, the goal of this project is to develop a podcast recommendation system that would enhance users' capacity to identify pertinent content, particularly in the educational genre. By using content-based filtration techniques, this system analyzes the user's listening preferences and interests before recommending educational podcasts. The podcast data source is Spotify, and the suggestions are produced using the TF-IDF and Cosine Similarity techniques. The recommendations provide a list of educational podcasts catered to the user's specific interests. The Confusion Matrix Classification Report was tested to assess system performance during the review phase. Precision values show how accurate the system was at recommending educational podcasts; on average, they range from 0.52 to 0.74. Additionally, the recall value showed a mean of 0.51 and a mean of 0.79, indicating that the algorithm successfully located the relevant content. To put it briefly, this custom recommendation engine enhances the listening experience for Spotify customers by suggesting educational podcasts based on their preferences. The system's ability to match users with material that aligns with their interests was demonstrated by the metrics used to assess its performance. With more user interactions with the system, it was anticipated by Cosine Similarity, a statistic used to determine the quality of recommendations, will continue to improve. To improve user experience and personalize the podcast listening experience on Spotify, this research addresses the challenge of locating suitable podcasts.