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

A system, comprising:

• a processor;
• and
• a memory having instructions stored thereon which, when executed by the processor, perform an operation for presenting recommended content to a user of a software application, the operation comprising:
• detecting execution of a triggering event with a software application by a user of the software application;
• obtaining user activity history prior to execution of the triggering event for the user of the software application;
• generating, for each respective executed event in the obtained user activity history, a one-hot vector, wherein:
• the one-hot vector includes a number of bits corresponding to each of a plurality of events in the software application,
• one bit of the number of bits in the one-hot vector is set to high, and
• a position in the one-hot vector of the bit set to high corresponds to an identifier, in a set of unique identifiers, of the respective executed event in the software application:
• generating a two-dimensional representation of the obtained user activity history prior to execution of the triggering event by sequentially concatenating the one-hot vectors generated for each respective event in the obtained user activity history, wherein the two-dimensional representation of the obtained user activity history encodes timing relationships on a first axis and events executed within the software application on a second axis;
• requesting, from a predictive model, predictive scores for a set of content items based on the obtained user activity history prior to execution of the triggering event, wherein:
• the predictive model has been trained using vectors correlating user activity history within the software application correlated to content interaction history recorded for a plurality of users of the software application,
• each vector correlates information about user actions performed within the software application prior to the triggering event and content viewed by one of the plurality of users subsequent to the triggering event, and
• the predictive scores indicate a likelihood that each content item in the set of content items would be relevant to the one of the plurality of users based on the obtained user activity history prior to execution of the triggering event;
• ranking the content items in the set of content items based on the predictive scores for each content item;
• selecting a subset of the set of content items to display to the user based on the ranking of the content items;
• and
• outputting the selected subset of content items for display to the user of the software application.

A non-transitory computer-readable medium having instructions stored thereon which, when executed by one or more processors, perform an operation for presenting recommended content to a user of a software application, the operation comprising:

• detecting execution of a triggering event with a software application by a user of the software application;
• obtaining user activity history prior to execution of the triggering event for the user of the software application;
• generating, for each respective executed event in the obtained user activity history, a one-hot vector, wherein:
• the one-hot vector includes a number of bits corresponding to each of a plurality of events in the software application,
• one bit of the number of bits in the one-hot vector is set to high, and
• a position in the one-hot vector of the bit set to high corresponds to an identifier, in a set of unique identifiers, of the respective executed event in the software application;
• generating a two-dimensional representation of the obtained user activity history prior to execution of the triggering event by sequentially concatenating the one-hot vectors generated for each respective event in the obtained user activity history, wherein the two-dimensional representation of the obtained user activity history encodes timing relationships on a first axis and events executed within the software application on a second axis;
• requesting, from a predictive model, predictive scores for a set of content items based on the obtained user activity history prior to execution of the triggering event, wherein:
• the predictive model has been trained using vectors correlating user activity history within the software application correlated to content interaction history recorded for a plurality of users of the software application,
• each vector correlates information about user actions performed within the software application prior to the triggering event and content viewed by one of the plurality of users subsequent to the triggering event, and
• the predictive scores indicate a likelihood that each content item in the set of content items would be relevant to the one of the plurality of users based on the obtained user activity history prior to execution of the triggering event;
• ranking the content items in the set of content items based on the predictive scores for each content item;
• selecting a subset of the set of content items to display to the user based on the ranking of the content items;
• and
• outputting the selected subset of content items for display to the user of the software application.