Abstract: In recommender systems, the goal is to learn a model from a small set of interacted users and items, and identify positively-related user item pairs among a large number of pairs with unknown interactions. Known methods for training the model rely on both similar and dissimilar user-item pairs. Using dissimilar pairs introduces several challenges, such as increasing training time or labeling pairs with unknown interactions as dissimilar even though the user might like the item if presented with it. If only similar pairs are used in the known methods, the result is a collapsed solution in which all users and items are mapped to the same representations. The methods disclosure herein overcome these challenges by using only similar pairs but adding two terms to the objective function that prevent a collapsed or partially-collapsed solution.

Abstract: The present disclosure relates to an improved machine learning-based recommender system and method for cold-start predictions on an ecommerce platform. The improved system predicts user-item interactions with respect to cold-start items in which only side information is available. Item representations generated by an item neural network encoder from item side information are shared with a user neural network. The item representations are used, along with user feedback history, to generate user representations. Specifically, a weight matrix in the first layer of the user neural network encoder is fixed with the shared item embeddings. The effect of this is that, when the user neural network encoder is applied to an input user-item interaction vector, the output of the first layer of the user neural network encoder is a function of the item representations of the items for which the user provided positive feedback.

Abstract: The present disclosure relates to a system, method, and computer program for recommending products using a neural network architecture that directly learns a user's predicted rating for an item from user and item data. A set of encoding neural networks maps each input source for user and item data to a lower-dimensional vector space. The individual lower-dimensional vector outputs of the encoding neural networks are combined to create a single multidimensional vector representation of user and item data. A prediction neural network is trained to predict a user's rating for an item based on the single multidimensional vector representation of user and item data. The neural network architecture allows for more efficient optimization and faster convergence that recommendations systems that rely on autoencoders. The system recommends items to users based on the users' predicted ratings for items.

Abstract: The present disclosure relates to a recommender system and method. Unlike known systems, which learn neural network parameters during training and fix the input vectors, the recommender system learns both the input vectors and machine learning model parameters during training. In one embodiment, the initial user and item input vectors are interaction vectors that are based on known and unknown user feedback. The non-zero elements of the interaction vectors correspond user-item pairs for which feedback is known, and the zero elements corresponding to user-item pairs for which feedback is unknown. The non-zero elements of the interaction vectors are learnable parameters during the training phase. The user and item vectors, as well as the model parameters, learned during the training phase are used in a prediction and recommendation phase to make product recommendations for a user.

Abstract: The present disclosure relates to a system, method, and computer program for recommending products using a neural network architecture that directly learns a user's predicted rating for an item from user and item data. A set of encoding neural networks maps each input source for user and item data to a lower-dimensional vector space. The individual lower-dimensional vector outputs of the encoding neural networks are combined to create a single multidimensional vector representation of user and item data. A prediction neural network is trained to predict a user's rating for an item based on the single multidimensional vector representation of user and item data. The neural network architecture allows for more efficient optimization and faster convergence that recommendations systems that rely on autoencoders. The system recommends items to users based on the users' predicted ratings for items.