System, method, and computer program for recommending items using a direct neural network structure

A non-transitory computer-readable medium comprising a computer program that, when executed by a computer system, enables the computer system to perform the following method for automatically recommending one or more items to a user using neural networks, the method comprising:

• performing the following with respect to a training phase:
• (a) for a training user in a training data set with a known rating on an item, receiving one or more input vectors with user data for the training user and one or more input vectors with item data for the item;
• (b) using a set of encoding neural networks to map each input vector to a lower-dimensional vector space, resulting in one or more lower-dimensional vectors for the user data and one or more lower-dimensional vectors for the item data, wherein a separate encoding neural network is used for each input vector, and wherein the lower-dimensional user and item vectors are lower dimensional as compared to the corresponding input vectors;
• (c) creating a single multidimensional vector representation of user and item data based on the one or more lower-dimensional user vectors and the one or more lower-dimensional item vectors;
• (d) using a prediction neural network to calculate a predicted rating for the item by the training user based on the single multidimensional vector representation of the training user and item data;
• (e) calculating an error value between the training user's predicted rating for the item and the training user's known rating for the item;
• (f) performing steps (a)-(e) with respect to a plurality of training users in the training data set;
• (g) updating parameters of the encoding neural networks and the prediction neural network to minimize the error values for the plurality of training users;
• (h) repeating steps (a)-(g) for a number of iterations to train the encoding neural networks and the prediction neural network;
• performing the following with respect to a prediction and recommendation phase:
• (i) receiving one or more input vectors with user data for a test user and one or more input vectors with item data for an item unrated by the test user;
• (j) using the set of encoding neural networks to map each input vector to a lower-dimensional vector space, resulting in one or more lower-dimensional vectors for the user data and one or more lower-dimensional vectors for the item data, wherein a separate encoding neural network is used for each input vector, and wherein the lower-dimensional user and item vectors are lower dimensional as compared to the corresponding input vectors;
• (k) creating a single multidimensional vector representation of user and item data based on the one or more lower-dimensional user vectors and the one or more lower-dimensional item vectors;
• (l) using the prediction neural network to calculate a predicted rating for the item by the test user based on the single multidimensional vector representation of user and item data;
• (m) repeating steps (i)-(l) with respect to the test user and a plurality of items unrated by the test user;
• and
• (n) using the test user's predicted ratings for the plurality of items to recommend one or more items to the test user.

A non-transitory computer-readable medium comprising a computer program that, when executed by a computer system, enables the computer system to perform the following method for training a neural network system to predict a user rating for an item, the method comprising:

• (a) for a training user in a training data set with a known rating on an item, receiving one or more input vectors with user data for the training user and one or more input vectors with item data for the item;
• (b) using a set of encoding neural networks to map each input vector to a lower-dimensional vector space, resulting in one or more lower-dimensional vectors for the user data and one or more lower-dimensional vectors for the item data, wherein a separate encoding neural network is used for each input vector, and wherein the lower-dimensional user and item vectors are lower dimensional as compared to the corresponding input vectors;
• (c) creating a single multidimensional vector representation of user and item data based on the one or more lower-dimensional user vectors and the one or more lower-dimensional item vectors;
• (d) using a prediction neural network to calculate a predicted rating for the item by the training user based on the single multidimensional vector representation of the training user and item data;
• (e) calculating an error value between the training user's predicted rating for the item and the training user's known rating for the item;
• (f) performing steps (a)-(e) with respect to a plurality of training users in the training data set;
• (g) updating parameters of the encoding neural networks and the prediction neural network to minimize the error values for the plurality of training users;
• and
• (h) repeating steps (a)-(g) for a number of iterations to train the encoding neural networks and the prediction neural network.

A non-transitory computer-readable medium comprising a computer program that, when executed by a computer system, enables the computer system to perform the following method for automatically recommending one or more items to a user using neural networks, the method comprising:

• (a) receiving one or more input vectors with user data for a user and one or more input vectors with item data for an item unrated by the user;
• (b) using a set of encoding neural networks to map each input vector to a lower-dimensional vector space, resulting in one or more lower-dimensional vectors for the user data and one or more lower-dimensional vectors for the item data, wherein a separate encoding neural network is used for each input vector, and wherein the lower-dimensional user and item vectors are lower dimensional as compared to the corresponding input vectors;
• (c) creating a single multidimensional vector representation of user and item data based on the one or more lower-dimensional user vectors and the one or more lower-dimensional item vectors;
• (d) using a prediction neural network to calculate a predicted rating for the item by the user based on the single multidimensional vector representation of user and item data, wherein the set of encoding neural networks and the prediction neural network were previously trained to perform steps (b)-(d);
• (e) repeating steps (a)-(d) with respect to the user and a plurality of items unrated by the user;
• and
• (f) using the user's predicted ratings for the plurality of items to recommend one or more items to the user.