Transformation of chat logs for chat flow prediction
Methods and systems for automating user interactions include annotating chat logs to identify user intent in question-answer pairs. A classifier is trained, using the annotated chat log, to identify user intent in automated conversations. Chat flows are formed, using the annotated chat logs, that provide responses to user statements based on identified user intent. An automated conversation is conducted with a user, using the chat flows and the classifier, to provide automated responses to user statements.
1. A computer-implemented method for automating user interactions, comprising:
annotating a plurality of chat logs to identify user intent in question-answer pairs;
training a classifier, using the plurality of annotated chat logs, to identify user intent in automated conversations;
forming chat flows, using the plurality of annotated chat logs, that provide responses to user statements based on identified user intent, including merging chat flows that represent individual chat logs of the plurality of annotated chat logs into a branching chat flow; and
conducting an automated conversation with a user, using the chat flows and the classifier, to provide automated responses to user statements.
2. The computer-implemented method of claim 1 , wherein annotating the plurality of chat logs is performed automatically using named entity recognition.
3. The computer-implemented method of claim 1 , wherein annotating the plurality of chat logs further comprises identifying question-answer pairs that include a question portion from a first conversant and an answer portion from a second conversant.
4. The computer-implemented method of claim 3 , further comprising clustering the plurality of annotated chat logs according to a first-layer clustering based on the answer portions of the question-answer pairs.
5. The computer-implemented method of claim 4 , further comprising clustering the clustered chat longs according to a second-layer clustering based on the question portions of the question-answer pairs.
6. The computer-implemented method of claim 1 , wherein each chat flow is represented as a directed graph, with each node representing an answer portion from a particular question-answer pair that is indicated based on a question portion.
7. The computer-implemented method of claim 6 , wherein conducting the automated conversation comprises applying the classifier to a question input by a user to determine a user intent, identifying a next node in a conversational chat flow based on the identified user intent, and issuing a response indicated by the identified next node.
8. The computer-implemented method of claim 1 , further comprising normalizing the plurality of chat logs before annotating, wherein the plurality of chat logs are generated by human operators in conversation with users.
9. A non-transitory computer readable storage medium comprising a computer readable program for automating user interactions, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:
annotating a plurality of chat logs to identify user intent in question-answer pairs;
training a classifier, using the plurality of annotated chat logs, to identify user intent in automated conversations;
forming chat flows, using the plurality of annotated chat logs, that provide responses to user statements based on identified user intent, including merging chat flows that represent individual chat logs of the plurality of annotated chat logs into a branching chat flow; and
conducting an automated conversation with a user, using the chat flows and the classifier, to provide automated responses to user statements.
10. A system for automating user interactions, comprising:
an annotation module configured to annotate a plurality of chat logs to identify user intent in question-answer pairs;
a training module configured to train a classifier, using the plurality of annotated chat logs, to identify user intent in automated conversations;
a chat flow module comprising a processor configured to form chat flows, using the plurality of annotated chat logs, that provide responses to user statements based on identified user intent, including merger of chat flows that represent individual chat logs of the plurality of annotated chat logs into a branching chat flow; and
an automated chat module configured to conduct an automated conversation with a user, using the chat flows and the classifier, to provide automated responses to user statements.
11. The system of claim 10 , wherein the annotation module is further configured to annotate the plurality of chat logs automatically using named entity recognition.
12. The system of claim 10 , wherein the annotation module is further configured to identifying question-answer pairs that include a question portion from a first conversant and an answer portion from a second conversant.
13. The system of claim 12 , further comprising a clustering module configured to cluster the plurality of annotated chat logs according to a first-layer clustering based on the answer portions of the question-answer pairs.
14. The system of claim 13 , wherein the clustering module is further configured to cluster the clustered chat longs according to a second-layer clustering based on the question portions of the question-answer pairs.
15. The system of claim 10 , wherein each chat flow is represented as a directed graph, with each node representing an answer portion from a particular question-answer pair that is indicated based on a question portion.
16. The system of claim 15 , wherein the automated chat module is further configured to apply the classifier to a question input by a user to determine a user intent, to identify a next node in a conversational chat flow based on the identified user intent, and to issue a response indicated by the identified next node.
17. The system of claim 10 , wherein the annotation module is further configured to normalize the plurality of chat logs before annotating, wherein the plurality of chat logs are generated by human operators in conversation with users.