End-to-end deep collaborative filtering

Abstract

A recommendation system generates recommendations for an online system using one or more neural network models that predict preferences of users for items in the online system. The neural network models generate a latent representation of a user and of a user that can be combined to determine the expected preference of the user to the item. By using neural network models, the recommendation system can generate predictions in real-time for new users and items without the need to re-calibrate the models. Moreover, the recommendation system can easily incorporate other forms of information other than preference information to generate improved preference predictions by including the additional information to generate the latent description of the user or item.

What is claimed is: 
     
         1 . A method for predicting a preference of a user in a plurality of users for an item in a plurality of items, the method comprising:
 determining a set of types of user descriptors for the user, the set of types of user descriptors including a user preference vector indicating preferences of the user for other items in the plurality of items, and a user content vector that indicates a set of identified characteristics for the user;   applying a first neural network model to the set of user descriptors to generate a latent user vector for the user;   determining a set of types of item descriptors for the item, the set of types of item descriptors including an item preference vector indicating preferences of the item from other users in the plurality of users, and an item content vector that indicates a set of identified characteristics for the item;   applying a second neural network model to the set of item descriptors to generate a latent item vector for the item; and   combining the latent user vector and the latent item vector to determine the predicted preference of the user for the item.   
     
     
         2 . The method of  claim 1 , wherein determining the set of types of user descriptors further comprises:
 setting values of at least one of the user preference vector or the user content vector as a zero vector.   
     
     
         3 . The method of  claim 1 , wherein determining the set of types of item descriptors further comprises:
 setting values of at least one of the item preference vector or the item content vector as a zero vector.   
     
     
         4 . The method of  claim 1 , wherein the set of types of user descriptors further includes a rated item vector that describes characteristics of items in the plurality of items for which there is preference information for the user. 
     
     
         5 . The method of  claim 4 , wherein the rated item vector is a weighted sum of one or more item content vectors, each item content vector in the one or more item content vectors describing characteristics of a corresponding item for which there is preference information for the user. 
     
     
         6 . The method of  claim 1 , wherein the set of types of item descriptors further include a rating user vector that describes characteristics of users in the plurality of users that are associated with preference information for the item. 
     
     
         7 . The method of  claim 6 , wherein the rating user vector is a weighted sum of one or more user content vectors, each user content vector in the one or more user content vectors describing characteristics of a corresponding user that has interacted with the item and has preference information associated with the item. 
     
     
         8 . The method of  claim 1 , wherein the first neural network and the second neural network are identical. 
     
     
         9 . The method of  claim 1 , wherein combining the latent user vector the latent user vector further comprises performing a dot product between the latent user vector and the latent item vector to determine the predicted preference. 
     
     
         10 . The method of  claim 1 , further comprising:
 modifying the set of user descriptors for the user; and   applying the first neural network model to the modified set of user descriptors to generate a second latent user vector for the user.   
     
     
         11 . The method of  claim 10 , wherein modifying the set of user descriptors comprises modifying the user preference vector based on a user interaction with another item in the plurality of items. 
     
     
         12 . The method of  claim 1 , further comprising:
 modifying the set of item descriptors for the item; and   applying the second neural network model to the modified set of item descriptors to generate a second latent item vector for the item.   
     
     
         13 . A non-transitory computer-readable medium containing instructions for execution on the processor, the instructions comprising:
 determining a set of types of user descriptors for the user, the set of types of user descriptors including a user preference vector indicating preferences of the user for other items in the plurality of items, and a user content vector that indicates a set of identified characteristics for the user;   applying a first neural network model to the set of user descriptors to generate a latent user vector for the user;   determining a set of types of item descriptors for the item, the set of types of item descriptors including an item preference vector indicating preferences of the item from other users in the plurality of users, and an item content vector that indicates a set of identified characteristics for the item;   applying a second neural network model to the set of item descriptors to generate a latent item vector for the item; and   combining the latent user vector and the latent item vector to determine the predicted preference of the user for the item.   
     
     
         14 . The non-transitory computer-readable medium of  claim 13 , wherein determining the set of types of user descriptors further comprises:
 setting values of at least one of the user preference vector or the user content vector as a zero vector.   
     
     
         15 . The non-transitory computer-readable medium of  claim 13 , wherein determining the set of types of item descriptors further comprises:
 setting values of at least one of the item preference vector or the item content vector as a zero vector.   
     
     
         16 . The non-transitory computer-readable medium of  claim 13 , wherein the set of types of user descriptors further includes a rated item vector that is a weighted sum of one or more item content vectors, each item content vector in the one or more item content vectors describing characteristics of a corresponding item for which there is preference information for the user. 
     
     
         17 . The non-transitory computer-readable medium of  claim 13 , wherein the set of types of item descriptors further include a rating user vector that is a weighted sum of one or more user content vectors, each user content vector in the one or more user content vectors describing characteristics of a corresponding user that has interacted with the item and has preference information associated with the item. 
     
     
         18 . The non-transitory computer-readable medium of  claim 13 , wherein the first neural network and the second neural network are identical. 
     
     
         19 . The non-transitory computer-readable medium of  claim 13 , further comprising:
 modifying the set of user descriptors for the user; and   applying the first neural network model to the modified set of user descriptors to generate a second latent user vector for the user.   
     
     
         20 . The non-transitory computer-readable medium of  claim 13 , further comprising:
 modifying the set of item descriptors for the item; and   applying the second neural network model to the modified set of item descriptors to generate a second latent item vector for the item.