Method and system for generating conversation flows
A method and system for generating conversation flows is provided herein. The method and system comprise storing conversations between at least an agent and at least a user in logs. Further, the method and system comprise extracting topics from the logs. Also, the method and system comprise creating clusters from the extracted topics. The method and system further comprise generating the conversation flows from the clusters.
1 . A computer-implemented method for generating conversation flows, comprising:
storing conversations between at least an agent and at least a user in logs;
transforming the stored logs into text transcripts;
extracting topics from the text transcripts of the stored logs using a natural language processing (NLP) model;
creating topic clusters from the extracted topics using a topic modeling technique to group semantically similar topics identified across consecutive conversation turns of the stored logs, and assigning intent names to the topic clusters, the intent names representing intentions of user queries associated with the topic clusters, wherein the topic clusters include main topics and sub-topics based on a hierarchical relationship among the topics, each main topic being associated with at least two sub-topics;
generating the conversation flows, modeled as a graph, from the topic clusters, wherein the generating includes applying a process mining technique to the topic clusters to determine patterns, dependencies, and relationships among the topic clusters, and wherein the graph comprises nodes representing specific topics, and edges representing transitions between the nodes based on the determined patterns, dependencies, and relationships;
identifying anomalous conversation patterns based on the conversation flows; and
configuring operation of a virtual agent based on the identified anomalous conversation patterns and the conversation flow.
2 . The computer-implemented method of claim 1 , wherein the graph is a directed cyclic graph.
3 . The computer-implemented method of claim 1 , wherein the sub-topics and the main topics are modeled as a tree.
4 . The computer-implemented method of claim 1 , wherein the agent is a human agent.
5 . A computer system for generating conversation flows, the computer system comprising:
one or more computer processors, one or more computer readable memories, one or more computer readable storage devices, and program instructions stored on the one or more computer readable storage devices for execution by the one or more computer processors via the one or more computer readable memories, the program instructions comprising:
storing conversations between at least an agent and at least a user in logs;
transforming the stored logs into text transcripts;
extracting topics from the text transcripts of the stored logs using a natural language processing (NLP) model;
creating topic clusters from the extracted topics using a topic modeling technique to group semantically similar topics identified across consecutive conversation turns of the stored logs, and assigning intent names to the topic clusters, the intent names representing intentions of user queries associated with the topic clusters, wherein the topic clusters include main topics and sub-topics based on a hierarchical relationship among the topics, each main topic being associated with at least two sub-topics;
generating the conversation flows, modeled as a graph, from the topic clusters, wherein the generating includes applying a process mining technique to the topic clusters to determine patterns, dependencies, and relationships among the topic clusters, and wherein the graph comprises nodes representing specific topics, and edges representing transitions between the nodes based on the determined patterns, dependencies, and relationships;
identifying anomalous conversation patterns based on the conversation flows; and
configuring operation of a virtual agent based on the identified anomalous conversation pattern.
6 . The computer system of claim 5 , wherein the graph is a directed cyclic graph.
7 . The computer system of claim 5 , wherein the sub-topics and the main topics are modeled as a tree.
8 . The computer system of claim 5 , wherein the agent is a human agent.
9 . A non-transitory computer-readable storage medium having stored thereon computer executable instructions which, when executed by one or more processors, cause the one or more processors to carry out operations for generating conversation flows, the operations comprising:
storing conversations between at least an agent and at least a user in logs;
transforming the stored logs into text transcripts;
extracting topics from the text transcripts of the stored logs using a natural language processing (NLP) model;
creating topic clusters from the extracted topics using a topic modeling technique to group semantically similar topics identified across consecutive conversation turns of the stored logs, and assigning intent names to the topic clusters, the intent names representing intentions of user queries associated with the topic clusters, wherein the topic clusters include main topics and sub-topics based on a hierarchical relationship among the topics, each main topic being associated with at least two sub-topics;
generating the conversation flows, modeled as a graph, from the topic clusters, wherein the generating includes applying a process mining technique to the topic clusters to determine patterns, dependencies, and relationships among the topic clusters, and wherein the graph comprises nodes representing specific topics, and edges representing transitions between the nodes based on the determined patterns, dependencies, and relationships;
identifying anomalous conversation patterns based on the conversation flows; and
configuring operation of a virtual agent based on the identified anomalous conversation pattern.