US Patent 10949432 Method and system for recommending domain-specific content based on recent user activity within a software application

Aspects of the present disclosure provide techniques for recommending content to users of an application based on a probability distribution over a set of articles generated from user activity history within the application. Embodiments include receiving a content request from a user of the application. A system retrieves, using a predictive model, predictive scores for items of content in response to the content request and user activity history within the application prior to the content request. The predictive model may be trained using user clickstream data for sessions executed within the application by a plurality of users, and information about actions performed within the application prior to a request may be correlated with content viewed by a user subsequent to the request. The system selects a group of the items of content to display to the user and displays the selected group of the items of content to the user.