Generating a personalized preference ranking network for providing visually-aware item recommendations

A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computing device to:

• maintain an image training dataset of items comprising implicit feedback from a user for a subset of the items;
• generate triplets where each of the triplets comprises the user, a positive item image from the subset of the items based on the implicit feedback from the user, and a negative item image from the image training dataset of items;
• and
• jointly train, using the triplets, a Siamese convolutional neural network to determine latent item features that are favored by the user and a personalized ranking model to determine latent user features, wherein the joint training comprises maximizing correlations between the latent item features and the latent user features.

In a digital medium environment for recommending fashion items to a user, a computer-implemented method of determining latent user features based on implicit user feedback, comprising:

• generating triplets where each of the triplets comprises a user, a positive item image from an image training dataset of items based on the implicit feedback from the user, and a negative item image from the image training dataset of items;
• performing a step for jointly training, based on the triplets, a preference predictor using a Siamese convolutional neural network and personalized ranking model;
• and
• performing a step for determining item preferences for the user from an unobserved dataset of items using the preference predictor.