Abstract: Systems and methods for estimating a long-term effect in the presence of unobserved confounding are provided. For example, a long-term effect of a change in user interface of a media playback application may be estimated. An experimental dataset and an observational dataset may be compiled using data from a plurality of computing devices. The observational dataset may include unobserved confounding. Using the experimental dataset, a short-term effect may be determined. Using the short-term effect and samples from the observational dataset, an instrumental variable may be computed that may be used in instrumental variable regression to estimate the long-term effect.

Abstract: The present application describes various methods and devices for providing content to users. In one aspect, a method includes, for each content item of a set of content items, obtaining a score for the content item using a recommender system, the score corresponding to a calculation of subsequent repeated engagement by a user with the content item. The method also includes ranking the set of content items based on the respective scores and providing recommendation information to the user for one or more highest ranked content items in the set of content items.

Abstract: An electronic device stores a plurality of vector representations for respective media content items in a vector space, where each vector represents a media content item. The electronic device receives a first set of input parameters representing a previous session of a user of the media-providing service where the previous session included two or more of the respective media content items. The electronic device then receives a second set of input parameters representing a current context of the user and provides the first set of input parameters and the second set of input parameters to a neural network to generate a prediction vector for a current session. The prediction vector is embedded in the vector space.

Abstract: A method comprises the following steps: providing a Gaussian process variational autoencoder (GP-VAE) including a Gaussian process (GP) encoder and a neural network decoder; selecting a plurality of inducing points in a data space; generating a mapping of the plurality of inducing points in a latent space; and training the GP-VAE using a training dataset.

Abstract: A reinforcement learning ranker can take into account previously-recommended media content items to produce a ranked list of media content items to recommend next. The ranker finds a policy that gives the probability of sampling a media content item given a state. The policy is learned such that it maximizes a reward. A reward function associated with the media content item can be defined with respect to whether the user finds the media content item relevant (likelihood that the user will like the media content item) and a diversity score of the media content item.

Abstract: An electronic device associated with a media-providing service obtains metadata for a collection of media content items. The metadata specifies an initial value for a language of the audio of a respective media content item. The electronic device obtains a listening history for users of the media-providing service. The listening history specifies which media content items of the collection of media content items a respective user has listened to. The electronic device determines, for a first user, one or more languages corresponding to the first user based on the initial values of the languages of the audio of the media content items that the respective user has listened to. The electronic device determines, for the respective media content item, an updated value for the language of the audio based on the one or more languages corresponding to the users that have listened to the respective media content item.

Abstract: An electronic device stores a plurality of vector representations for respective media content items in a vector space, where each vector represents a media content item. The electronic device receives a first set of input parameters representing a previous session of a user of the media-providing service where the previous session included two or more of the respective media content items. The electronic device then receives a second set of input parameters representing a current context of the user and provides the first set of input parameters and the second set of input parameters to a neural network to generate a prediction vector for a current session. The prediction vector is embedded in the vector space.