Spotify Enhances 'Discover Weekly' Algorithm

- The 'Discover Weekly' algorithm is built on three core types of recommendation models: collaborative filtering (analyzing your behavior and others'), Natural Language Processing (NLP) for analyzing text, and audio models that analyze the raw audio tracks. - A key step in the evolution of Spotify's recommendation system was its 2014 acquisition of The Echo Nest, a music analytics firm that specialized in using machine learning and NLP to analyze songs and user behavior. - The algorithm processes both explicit user feedback, such as saving a song or adding it to a playlist, and implicit feedback, like song skips or the length of a listening session, to refine its user taste profiles. - To analyze the music itself, Spotify employs convolutional neural networks to evaluate the actual sound characteristics of a track, such as tempo, key, and loudness, rather than relying solely on user data or metadata. - The impact of personalization is significant, with personalized playlists like Discover Weekly accounting for over 30% of all listening time on the platform. - A recent update to Discover Weekly now allows users to customize their playlists by selecting from up to five offered genres, giving them more direct control over the recommendation outputs. - For a song to be picked up by the algorithm, key engagement metrics are crucial, especially within the first 72 hours of its release; these include a high completion rate (over 50%), a low skip rate, and a significant number of saves to user libraries. - Since its launch in 2015, Discover Weekly has become a major driver of music discovery, generating 5 billion streams in its first year and over 100 billion streams to date.