Enhancing chatbot recognition of user intent through graph analysis

A computerized method of enhancing chatbot recognition of user intent through graph analysis, the method comprising:

• classifying, by a computing device, each of a plurality of user messages associated with a chat session between a user of a remote computing device and a chatbot application as either comprehensible user messages that are mapped to a user intent or incomprehensible utterances that are not mapped to a user intent;
• arranging, by the computing device, the comprehensible user messages into a first plurality of clusters and arranging the incomprehensible user messages into a second plurality of clusters;
• generating, by the computing device, a first sub-graph data structure for each cluster in the first plurality of clusters and generating a second sub-graph data structure for each cluster in the second plurality of clusters;
• calculating, by the computing device, a graph kernel matrix using the sub-graph data structures generated from the first plurality of clusters and the second plurality of clusters;
• training, by the computing device, a classification model using the graph kernel matrix, the trained classification model configured to predict an intent identifier for an input graph kernel matrix;
• and
• executing, by the computing device, the trained classification model using the graph kernel matrix as input to generate a predicted intent identifier for one or more of the incomprehensible user messages.

A computer system for enhancing chatbot recognition of user intent through graph analysis, the system comprising a computing device having a memory that stores computer-executable instructions and a processor that executes the computer-executable instructions to:

• classify each of a plurality of user messages associated with a chat session between a user of a remote computing device and a chatbot application as either comprehensible user messages that are mapped to a user intent or incomprehensible utterances that are not mapped to a user intent;
• generate a contextual embedding for each of the plurality of user messages;
• determine an intent classification score for each contextual embedding using a classification model;
• calculate a similarity measure between each contextual embedding and one or more embeddings generated from user messages in a chat log;
• match each of the incomprehensible user messages to one or more comprehensible user messages using a pattern matching algorithm;
• generate a predicted intent score for each incomprehensible user message based upon the intent classification score, the similarity measure, and the matched comprehensible user messages;
• and
• determine one or more existing user intents based upon the predicted intent score and assign the one or more existing user intents to the corresponding incomprehensible user message.

A computerized method of enhancing chatbot recognition of user intent through graph analysis, the method comprising:

• classifying, by a computing device, each of a plurality of user messages associated with a chat session between a user of a remote computing device and a chatbot application as either comprehensible user messages that are mapped to a user intent or incomprehensible utterances that are not mapped to a user intent;
• generating, by the computing device, a contextual embedding for each of the plurality of user messages;
• determining, by the computing device, an intent classification score for each contextual embedding using a classification model;
• calculating, by the computing device, a similarity measure between each contextual embedding and one or more embeddings generated from user messages in a chat log;
• matching, by the computing device, each of the incomprehensible user messages to one or more comprehensible user messages using a pattern matching algorithm;
• generating, by the computing device, a predicted intent score for each incomprehensible user message based upon the intent classification score, the similarity measure, and the matched comprehensible user messages;
• and
• determining, by the computing device, one or more existing user intents based upon the predicted intent score and assigning the one or more existing user intents to the corresponding incomprehensible user message.