| CPC G06F 40/174 (2020.01) [G06F 40/289 (2020.01); G06N 20/20 (2019.01); H04L 67/535 (2022.05)] | 13 Claims |
|
1. A method comprising:
obtaining user interaction data from which to generate a churn risk, wherein the user interaction data identifies a user interaction of a session and comprises values for one or more of:
a first number of times a user selected to go back, a second number of times the user selected to go forward, a third number of times the user selected to view a help file, a fourth number of times the user selected to request help, a fifth number of times the user selected to request a human agent, a sixth number of times the user responded negatively to a system, and a seventh number of times the user modified a predicted answer;
generating, using a machine learning model, the churn risk from the user interaction data;
receiving a structured document comprising a plurality of fields, wherein the structured document is stored as a structured language data object in a non-transitory computer readable storage medium;
selecting, when the churn risk satisfies a threshold, a field, from the plurality of fields, using a plurality of prediction confidences corresponding to a plurality of prediction values generated for the plurality of fields, wherein selecting the field comprises:
applying a plurality of machine learning models to the structured language data object to generate the plurality of prediction values and the plurality of prediction confidences for the plurality of fields,
identifying a prediction confidence, of the plurality of prediction confidences, that satisfies a criterion, and
identifying the field corresponding to the prediction confidence;
obtaining a prediction value, from among the plurality of prediction values, for the field;
obtaining a question, corresponding to the field, using the prediction value, wherein obtaining the question comprises:
generating the question by selecting, when the field has a specified number of possible values, the question from a plurality of question phrases stored in a data repository,
generating the question by updating, when the field has an unspecified number of possible values and using the prediction value, a question template stored in the data repository; and
generating the question is performed responsive to the churn risk satisfying the threshold; and
presenting the question by updating a graphical user interface to display the question.
|