Methods and systems for implementing dynamic-action systems in real-time data streams

A system comprising:

• one or more processors;
• and
• a non-transitory machine-readable storage medium storing instructions that when executed by the one or more processors, cause the one or more processors to perform operations including:
• generating a feature vector from communications of a real-time data stream, wherein the real-time data stream is a representation of a communication session between a user of a first device and a service of a second device, wherein the communications include natural language communications provided by the user;
• executing a first machine-learning model using the feature vector, wherein of first machine-learning model is configured to generate a conversation score, wherein the conversation score indicates a quality of natural language communications between the user and the service;
• determining that a condition of a trigger of a set of triggers is satisfied based on the conversation score being below a threshold, wherein the conversation score being below the threshold indicates poor quality natural language communications between the user and the service;
• facilitating a connection of a third device to the communication session in response to executing a function associated with the trigger, wherein a user of the third device communicates with the user of the first device using natural language communications;
• generating a training vector for a second machine-learning model using natural language communications of the real-time data stream that occur after execution of the first machine-learning model, wherein the second machine-learning model is trained using supplemental training data, and wherein the training vector is configured to replace at least a portion of the supplemental training data to improve an accuracy of the second machine-learning model with respect to the communication session;
• updating the second machine-learning model using the training vector, wherein the second machine-learning model is configured to manage the set of triggers;
• and
• modifying, using the second machine-learning model, one or more triggers of the set of triggers based on the conversation score and the real-time data stream, wherein the one or more triggers are modified in response to executing the function associated with the trigger, and wherein modifying the one or more triggers reduces a probability that a condition of the one or more triggers will be satisfied during a subsequent time interval of the communication session.

A non-transitory machine-readable storage medium storing instructions that when executed by one or more processors, cause the one or more processors to perform operations including:

• generating a feature vector from communications of a real-time data stream, wherein the real-time data stream is a representation of a communication session between a user of a first device and a service of a second device, wherein the communications include natural language communications provided by the user;
• executing a first machine-learning model using the feature vector, wherein of first machine-learning model is configured to generate a conversation score, wherein the conversation score indicates a quality of natural language communications between the user and the service;
• determining that a condition of a trigger of a set of triggers is satisfied based on the conversation score being below a threshold, wherein the conversation score being below the threshold indicates poor quality natural language communications between the user and the service;
• facilitating a connection of a third device to the communication session in response to executing a function associated with the trigger, wherein a user of the third device communicates with the user of the first device using natural language communications;
• generating a training vector for a second machine-learning model using natural language communications of the real-time data stream that occur after execution of the first machine-learning model, wherein the second machine-learning model is trained using supplemental training data, and wherein the training vector is configured to replace at least a portion of the supplemental training data to improve an accuracy of the second machine-learning model with respect to the communication session;
• updating the second machine-learning model using the training vector, wherein the second machine-learning model is configured to manage the set of triggers;
• and
• modifying, using the second machine-learning model, one or more triggers of the set of triggers based on the conversation score and the real-time data stream, wherein the one or more triggers are modified in response to executing the function associated with the trigger, and wherein modifying the one or more triggers reduces a probability that a condition of the one or more triggers will be satisfied during a subsequent time interval of the communication session.