Music playlist prediction via detecting song moods

Modern internet technologies make people easily access millions of songs. However it also forms a huge barrier between customers and those songs people truely want, due to the difficulty to explore this large collections. In this paper, we propose a novel music playlist prediction algorithm to facilitate this process for users. This method captures the moods expressed by songs in playlist context and also models the nature of composing a playlist. With the help of captured moods, personalized predictions can be achieved. We offer two ways to represent these hidden song moods and the transitions between them. The empirical evaluations show that our method outperforms other state-of-the-art methods in terms of perplexity.