Data is becoming more and more valuable as technology advances. Through crowdsourcing organizations are able to collect large amounts of data at an effective rate with little cost. This paper proposes a crowdsourcing-based music playing system, where the next song to play is determined by the listening preferences of realtime online users. Unlike conventional radio play system, where songs are randomly selected, our system selects songs which satisfies the listening preferences of a majority of users, who are currently online. Our system consists of two important components: a predictor, which estimates a user's listening preferences based on his features, and a decision maker, which selects the next song to play based on a majority vote. In order to find out the relationship between user features and listening preferences, K-means algorithm is adopted. A large amount of real data is collected via online surveys, and extension simulations show that our system can effectively selects a proper song for its online audience.

We present SDA-Bayes, a framework for (S)treaming, (D)istributed, (A)synchronous computation of a Bayesian posterior. The framework makes streaming updates to the estimated posterior according to a user-specified approximation primitive function. We demonstrate the usefulness of our framework, with variational Bayes (VB) as the primitive, by fitting the latent Dirichlet allocation model to two large-scale document collections. We demonstrate the advantages of our algorithm over stochastic variational inference (SVI), both in the single-pass setting SVI was designed for and in the streaming setting, to which SVI does not apply.