Machine learning for intent matching engine
A server obtains a natural language query from a first user device. The server matches an intent to the natural language query using an intent matching engine. The intent represents predicted data associated with the natural language query. The server transmits the natural language query and the intent to a second user device. The server receives, from the second user device, a response indicating whether the natural language query is properly matched to the intent. The server trains the intent matching engine based on a machine learning technique and the response.
1. A method, comprising:
obtaining, by a server, a natural language query from a first user device within a chatbot;
storing a set of intents that are matched to natural language queries based on previous natural language queries and network activities of user devices in response to previous workflow outputs;
matching an intent to the natural language query using an intent matching engine, wherein the intent represents predicted data associated with the natural language query, wherein each intent is mapped to a workflow in a data repository accessed by the server, and wherein the set of intents includes a set of global intents that are globally applicable and a set of custom intents that are only applicable to the server;
transmitting the natural language query and the intent to a second user device in response to matching the intent to the natural language query;
receiving, from the second user device, a response indicating whether the natural language query is properly matched to the intent;
transmitting, to the first user device within the chatbot, a workflow output associated with the workflow to which the intent is mapped;
monitoring network activity within the chatbot of the first user device with respect to the workflow output; and
training, using online learning applied to a convolutional neural network of the intent matching engine, the intent matching engine based on the monitored network activity of the first user device and the response.
2. The method of claim 1 ,
wherein the workflow output comprises at least one of: a hyperlink, a webpage, an article, or a list of actions.
3. The method of claim 1 , wherein the natural language query is one of a plurality of queries, wherein each query of the plurality of queries is matched to an intent from a set of intents, the method comprising:
monitoring a number of queries from the plurality of queries that are associated with each intent during multiple time periods; and
transmitting, to the second user device, a dashboard display indicating the number of queries associated with a given intent during a subset of the multiple time periods.
4. The method of claim 1 , wherein the response comprises a selection of either a first option indicating that the matching is accurate or a second option indicating that the matching is inaccurate.
5. The method of claim 1 , wherein receiving from the second user device the response indicating whether matching the intent to the natural language query is accurate comprises:
transmitting a page associated with the intent to second user device, wherein the page displays multiple queries, including the natural language query that matched to the intent; and
receiving the response from the second user device via the page.
6. The method of claim 1 , comprising:
determining that the natural language query is in a natural language different from a natural language associated with the intent; and
translating the natural language query into the natural language associated with the intent prior to matching the intent to the natural language query.
7. The method of claim 1 , comprising:
determining that the natural language query is in a natural language different from a natural language associated with the intent, wherein the intent is matched to the natural language query without translating the natural language query.
8. One or more non-transitory computer readable media storing instructions operable to cause one or more processors to perform operations comprising:
obtaining, by a server, a natural language query from a first user device within a chatbot;
storing a set of intents that are matched to natural language queries based on previous natural language queries and network activities of user devices in response to previous workflow outputs;
matching an intent to the natural language query using an intent matching engine, wherein the intent represents predicted data associated with the natural language query, wherein each intent is mapped to a workflow in a data repository accessed by the server, and wherein the set of intents includes a set of global intents that are globally applicable a set of custom intents that are only applicable to the server;
transmitting the natural language query and the intent to a second user device in response to matching the intent to the natural language query;
receiving, from the second user device, a response indicating whether the natural language query is properly matched to the intent;
transmitting, to the first user device within the chatbot, a workflow output associated with the workflow to which the intent is mapped;
monitoring network activity within the chatbot of the first user device with respect to the workflow output; and
training, using online learning applied to a convolutional neural network of the intent matching engine, the intent matching engine based on the monitored network activity of the first user device and the response.
9. The one or more computer readable media of claim 8 ,
wherein the workflow output comprises at least one of: a hyperlink, a webpage, an article, or a list of actions.
10. The one or more computer readable media of claim 8 , wherein the natural language query is one of a plurality of queries, wherein each query of the plurality of queries is matched to an intent from a set of intents, the operations comprising:
monitoring a number of queries from the plurality of queries that are associated with each intent during a set of discrete time periods; and
transmitting, to the second user device, a dashboard display indicating the number of queries associated with a given intent during a subset of the discrete time periods.
11. The one or more computer readable media of claim 8 , wherein the response comprises a selection of either a first graphical user interface icon indicating that the matching is accurate or a second graphical user interface icon indicating that the matching is inaccurate.
12. The one or more computer readable media of claim 8 , wherein receiving from the second user device the response indicating whether matching the intent to the natural language query is accurate comprises:
transmitting a graphical user interface associated with the intent to second user device, wherein the graphical user interface displays multiple queries, including the natural language query that matched to the intent; and
receiving the response from the second user device via the graphical user interface.
13. The one or more computer readable media of claim 8 , the operations comprising:
determining that the natural language query is in a first natural language, wherein the intent is associated with a second natural language different from the first natural language; and
translating the natural language query into the second natural language prior to matching the intent to the natural language query.
14. The one or more computer readable media of claim 8 , the operations comprising:
determining that the natural language query is in a first natural language, wherein the intent is associated with a second natural language different from the first natural language, wherein the intent is matched to the natural language query without translating the natural language query into the second natural language.
15. A system, comprising:
one or more memories; and
one or more processors configured to execute instructions stored in the one or more memories to:
obtain, by a server, a natural language query from a first user device within a chatbot;
store a set of intents that are matched to natural language queries based on previous natural language queries and network activities of user devices in response to previous workflow outputs;
match an intent to the natural language query using an intent matching engine, wherein the intent represents predicted data associated with the natural language query, wherein each intent is mapped to a workflow in a data repository accessed by the server, and wherein the set of intents includes a set of global intents that are globally applicable and a set of custom intents that are only applicable to the server;
transmit the natural language query and the intent to a second user device in response to matching the intent to the natural language query;
receive, from the second user device, a response indicating whether the natural language query is properly matched to the intent;
transmit, to the first user device within the chatbot, a workflow output associated with the workflow to which the intent is mapped;
monitor network activity within the chatbot of the first user device with respect to the workflow output; and
train, using online learning applied to a convolutional neural network of the intent matching engine, the intent matching engine based on the monitored network activity of the first user device and the response.
16. The system of claim 15 ,
wherein the workflow output comprises at least one of: a hyperlink, a webpage, an article, or a list of actions.
17. The system of claim 15 , wherein the natural language query is one of a plurality of queries, wherein each query of the plurality of queries is matched to an intent from a set of intents, the one or more processors configured to execute the instructions stored in the one or more memories to:
monitor a number of queries from the plurality of queries that are associated with each intent over time; and
transmit, to the second user device, a graphical dashboard display indicating the number of queries associated with a given intent over time.
18. The system of claim 15 , wherein the response comprises a selection of either a first user interface element indicating that the matching is accurate or a second user interface element indicating that the matching is inaccurate.