Recommendation system using improved neural network

This invention describes a system that learns what items to recommend to users by studying their past interactions with an electronic catalog. It uses a special type of computer program called a neural network, which is trained by feeding it historical data about what users bought or looked at before a certain date, and then checking if it can predict what they interacted with after that date. The system then uses this trained network to score how likely a user is to interact with a second item after interacting with a first, and displays the highly-scored item as a recommendation on their device.