Pandora Media, Inc.

Gordon Parker Rios

Tao Ye

Eric Kenson Bieschke

Kyle George Lind

Fabien Gouyon

Mingxuan Sun

Systems and methods are provided for personalizing new song suggestions for a user. A plurality of users provides thumbs up and thumbs down ratings for a pair of songs (song A and B). A first aggregate rating for song B is calculated based on ratings given to song A by all of the users that gave a like rating to song A. Similarly, a second aggregate rating is calculated for song B based on ratings given to song B by all users that gave a dislike rating to song A. Then a user that has not previously rated song B, supplies a rating for song A. A prediction is performed of how the user will rate song B based on the user's rating of song A. For example, the first aggregate rating for song B is used if the user rated song A positively.

System and method of personalizing playlists using memory-based collaborative filtering

A content server system provides content items to a plurality of users in a content stream. The content server system provides an interface to collect feedback from the plurality of users on the content items in the content stream. Based on the collected feedback the content server system generates a feedback model that predicts the likelihood of receiving feedback on a content item at any given time in a streaming session. The content server system then selects a content item in need of feedback and, using the model, determines a time at which the content item is more likely to receive feedback. The content server system then provides the content item to the user at the determined time.

Increasing the likelihood of receiving feedback for content items

A content server system provides a user of a client device with a playlist of content, such as streamed audio. The provided playlist is customized for the particular country of the user, even if the content server system has only recently begun to provide content to that country. To customize the playlist for a country lacking a sufficient amount of country-specific statistical data, the content server system generates a number of non-country-specific channels, each channel recommending content according to its own criteria. The content server system generates a non-content specific playlist by combining the channels. The content server system also generates, using the statistics for the country of the user, a country-specific channel that is specific to that country. The content server system generates a country-specific playlist by combining the non-country-specific playlist with the country-specific channel, e.g., according to the amount of country-specific statistics that are available.

Country-specific content recommendations in view of sparse country data

A content server generates sponsored audio including procedurally generated background music. The content server obtains reference music features describing musicological characteristics of reference songs as well as sponsored audio information received from a third-party. The content server determines music generation parameters based on the sponsored audio information and based on a music model mapping the reference music features to music generation parameters. The music model may incorporate machine learning techniques to improve the mapping. The content server generates background music by using the determined music generation parameters as input to a procedural music algorithm, and generates sponsored audio by combining the generated background music concurrently with an audio voiceover obtained from the sponsored audio information. The sponsored audio is provided to a client device, which presents the sponsored audio to a user.

Procedurally generating background music for sponsored audio

Systems, methods, and computer readable storage mediums are provided for selecting a media content object for a user using a combination of inputs. A media input seed associated with a user is obtained. A plurality of channels of media content objects is obtained. At least one of the plurality of channels is associated with the media input seed. Also, in some embodiments, each media content object of each of those channels has a score specific to that channel. A combination score for a respective media content object is calculated based at least in part on that respective media content object's channel specific score for each of at least two of the plurality of channels. Then at least some of the media content objects are ranked based at least in part on their respective combination scores. Finally, at least one ranked media content object is then selected for transmission.

Growing community of inventors

Fort Bragg, CA, United States of America

Gordon Parker Rios