Selecting content objects for recommendation based on content object collections

摘要

Collections identifying associated content objects, such as user-defined collections, are stored. A request to display or play a seed content object is received. A plurality of seed collections are identified based on the seed object. A plurality of candidate content objects are identified based on the identified seed collections, for instance by identifying the content objects that are identified by the seed collections. Properties of the seed object are identified, such as the categorization of the seed object, or the tags associated with the seed object. The candidate objects are ranked based on the identified seed collections and the identified properties of the seed object, and a candidate object is selected for recommendation to the user based on the ranking. The ranking of candidate objects can also be determined based on the coherence of the seed collections and the centrality of the seed object to the seed collections.

主权项

1. A computer implemented method of selecting an image for recommendation to a user, the method comprising:
storing a plurality of image collections in a non-transitory computer-readable storage medium, wherein each image collection comprises a set of associated images; receiving a request for a seed image from a user; identifying a plurality of seed collections, wherein each seed collection comprises a stored image collection identifying the seed image; identifying a plurality of candidate images based on the identified seed collections; identifying properties of the seed image; determining, for each image collection, a coherence score representative of a similarity of the set of images within the image collection, the coherence score comprising a measure of entropy of object vectors associated with each of the set of images within the image collection, each entry of an object vector representative of a property of an image associated with the object vector; determining, for each candidate image in the plurality of candidate images, a corresponding result by dividing a first sum of the determined coherence scores for the image collections comprising seed collections to which the candidate image belongs by a second sum of the determined coherence scores for all of the plurality of image collections; ranking each candidate image in the plurality of candidate images based on the determined corresponding results and based on the identified properties of the seed image; and selecting a candidate image for recommendation to the user based on the ranking of candidate images.