Assessment of user price sensitivity

A system for training a predictive model to assess a user's price sensitivity, the system comprising:

• one or more processors;
• and
• at least one memory coupled to the one or more processors and storing instructions that, when executed by the one or more processors, cause the system to perform operations including:
• retrieving traversal sequences associated with users, each traversal sequence including a chronological list of pages accessed by the associated user during a session, and each traversal sequence representative of the associated user's navigational pattern during the session prior to being presented with a notification about a price to continue the session;
• assigning a first label or a second label to each traversal sequence based on whether the associated user terminated the session upon being presented with the notification about the price to continue the session, the first label assigned if the associated user terminated the session upon being presented with the notification about the price to continue the session, and the second label assigned if the associated user accessed one or more additional pages after being presented with the notification about the price to continue the session;
• transforming each traversal sequence into a corresponding graph including a plurality of nodes and a plurality of edges connecting pairs of the nodes, each of the nodes mapped to an identifier assigned to one of the pages accessed by the associated user during the session prior to being presented with the notification about the price to continue the session, and each of the edges representing an instance in which the associated user successively accessed the pages assigned the identifiers mapped to the nodes connected to the edge;
• identifying structural attributes of the graphs based on the nodes and edges;
• defining a number of predictive features based on the structural attributes, each of the predictive features suggesting an extent to which the structural attributes of a graph corresponding to a given user's navigational patterns during a session impact a likelihood that the given user will terminate the session upon being presented with a notification about a price to continue the session;
• generating model training data from the labeled traversal sequences and the defined predictive features;
• and
• training, using a machine learning algorithm in conjunction with the model training data, a predictive model to assess a user's sensitivity to being presented with a notification about a price to continue a session, the predictive model incorporating a set of weighting factors determined for the defined predictive features using one or more loss functions, and the predictive model trained to predict, given a particular sequence of pages traversed by the user during a session prior to being presented with a notification about a price to continue the session, a likelihood that the user will terminate the session upon being presented with the notification about the price to continue to session.

A non-transitory computer-readable medium storing instructions that, when executed by at least one or more processors of a system, cause the system to perform operations including:

• retrieving traversal sequences associated with users, each traversal sequence including a chronological list of pages accessed by the associated user during a session, and each traversal sequence representative of the associated user's navigational pattern during the session prior to being presented with a notification about a price to continue the session;
• assigning a first label or a second label to each traversal sequence based on whether the associated user terminated the session upon being presented with the notification about the price to continue the session, the first label assigned if the associated user terminated the session upon being presented with the notification about the price to continue the session, and the second label assigned if the associated user accessed one or more additional pages after being presented with the notification about the price to continue the session;
• transforming each traversal sequence into a corresponding graph including a plurality of nodes and a plurality of edges connecting pairs of the nodes, each of the nodes mapped to an identifier assigned to one of the pages accessed by the associated user during the session prior to being presented with the notification about the price to continue the session, and each of the edges representing an instance in which the associated user successively accessed the pages assigned the identifiers mapped to the nodes connected to the edge;
• identifying structural attributes of the graphs based on the nodes and edges;
• defining a number of predictive features based on the structural attributes, each of the predictive features suggesting an extent to which the structural attributes of a graph corresponding to a given user's navigational patterns during a session impact a likelihood that the given user will terminate the session upon being presented with a notification about a price to continue the session;
• generating model training data from the labeled traversal sequences and the defined predictive features;
• and
• training, using a machine learning algorithm in conjunction with the model training data, a predictive model to assess a user's sensitivity to being presented with a notification about a price to continue a session, the predictive model incorporating a set of weighting factors determined for the defined predictive features using one or more loss functions, and the predictive model trained to predict, given a particular sequence of pages traversed by the user during a session prior to being presented with a notification about a price to continue the session, a likelihood that the user will terminate the session upon being presented with the notification about the price to continue to session.