Granted Patent B1

US 12,657,482 · App. 19/371,142 · Granted Jun 16, 2026

Inventors: Dean A Marek (Columbus, OH); Jason W. Black (Columbus, OH)

Assignee: The Huntington National Bank

G06N5/022G06F18/22G06F40/274G06N3/0475

Abstract

Claims (82)

1 . A computer-implemented method comprising:

receiving a request from an edge device;

providing, to a generative model in a prompt, the request;

determining, based at least in part on an output of the generative model, a negative response condition in accordance with a comparison of a first similarity score with a first pre-determined threshold;

comparing natural language information to a first graph network in accordance with a first pre-defined degree of separation, wherein the first graph network comprises a first set of nodes that represent a plurality of entities and a first set of edges that represent relationships between the plurality of entities and the natural language information comprises at least one of a portion of the output or a portion of the request;

determining a second similarity score based at least in part by comparing natural language text of a second set of nodes with the natural language information, wherein the second set of nodes do not occur in the first graph network and the second similarity score represents similarity between at least the natural language information and first metadata associated with the second set of nodes;

extracting second metadata from the second set of nodes based at least in part on the second similarity score being in accordance with a second pre-determined threshold, wherein the second set of nodes is within a second pre-defined degree of separation different from the first pre-defined degree of separation;

creating a second graph network based at least in part by creating edges between i) nodes of the second set of nodes, ii) the first set of nodes, or both i) and ii), that represent relationships between nodes of the second set of nodes and/or the first set of nodes, wherein the second graph network is different than the first graph network;

generating a response to the request by accessing particular metadata of the second metadata associated with at least one node of the second set of nodes by extracting and formatting contextual natural language from the particular metadata; and

providing the response including the particular metadata to the edge device.

2 . The computer-implemented method of claim 1 , further comprising:

deploying the second graph network to a cloud environment that hosts the first graph network or the second graph network; and

updating one or more skills associated with a chat bot that received the request based at least in part on the second graph network.

3 . The computer-implemented method of claim 1 , wherein the first pre-defined degree of separation includes two or fewer edges, and the second pre-defined degree of separation includes at least one edge more than the first pre-defined degree of separation, and wherein the negative response condition indicates a lack of a response or a non-response condition.

4 . The computer-implemented method of claim 1 , wherein the request is received by a first chat bot, the computer-implemented method further comprising:

creating a new skill for the first chat bot based at least in part on the second metadata or the response; and

storing the new skill in a cloud environment for retrieval by a second chat bot different from the first chat bot.

5 . The computer-implemented method of claim 1 , further comprising:

extracting portions of information from the second metadata;

transforming the portions of information into plain English summary with a context associated with the request; and

inserting one or more hyperlinks into the plain English summary in accordance with documents associated with the second metadata, wherein the response comprises the plain English summary, the one or more hyperlinks, or both.

6 . The computer-implemented method of claim 1 , further comprising:

creating a production generative model by updating the generative model, the second graph network, the second metadata, the response, or combinations thereof, as unlabeled training data; and

deploying the production generative model in a cloud environment.

7 . The computer-implemented method of claim 1 , further comprising:

receiving, from the edge device, a second request associated with the response, the second request associated with controlling an asset associated with a first entity of the plurality of entities and the response;

controlling the asset associated with the first entity to transfer to a second entity of the plurality of entities; and

updating the first graph network or the second graph network in accordance with the transfer.

8 . A computing device, comprising:

one or more processors; and

one or more memories storing computer-executable instructions that, when executed by the one or more processors, causes the one or more processors to:

receive a request from an edge device;

provide, to a generative model in a prompt, the request;

determine, based at least in part on an output of the generative model, a negative response condition in accordance with a comparison of a first similarity score with a first pre-determined threshold;

compare natural language information to a first graph network in accordance with a first pre-defined degree of separation, wherein the first graph network comprises a first set of nodes that represent a plurality of entities and a first set of edges that represent relationships between the plurality of entities and the natural language information comprises at least one of a portion of the output or a portion of the request;

determine a second similarity score based at least in part by comparing natural language text of a second set of nodes with the natural language information, wherein the second set of nodes do not occur in the first graph network and the second similarity score represents similarity between at least the natural language information and first metadata associated with the second set of nodes;

extract second metadata from the second set of nodes based at least in part on the second similarity score being in accordance with a second pre-determined threshold, wherein the second set of nodes is within a second pre-defined degree of separation different from the first pre-defined degree of separation;

create a second graph network based at least in part by creating edges between i) nodes of the second set of nodes, ii) the first set of nodes, or both i) and ii), that represent relationships between nodes of the second set of nodes and/or the first set of nodes, wherein the second graph network is different than the first graph network;

generate a response to the request by accessing particular metadata of the second metadata associated with at least one node of the second set of nodes by extracting and formatting contextual natural language from the particular metadata; and

provide the response including the particular metadata to the edge device.

9 . The computing device of claim 8 , wherein the instructions further cause the processor to:

deploy the second graph network to a cloud environment that hosts the first graph network or the second graph network; and

update one or more skills associated with a chat bot that received the request based at least in part on the second graph network.

10 . The computing device of claim 8 , wherein the first pre-defined degree of separation includes two or fewer edges, and the second pre-defined degree of separation includes at least one edge more than the first pre-defined degree of separation, and wherein the negative response condition indicates a lack of a response or a non-response condition.

11 . The computing device of claim 8 , wherein the request is received by a first chat bot, wherein the instructions further cause the processor to:

create a new skill for the first chat bot based at least in part on the second metadata or the response; and

store the new skill in a cloud environment for retrieval by a second chat bot different from the first chat bot.

12 . The computing device of claim 8 , wherein the instructions further cause the processor to:

extract portions of information from the second metadata;

transform the portions of information into plain English summary with a context associated with the request; and

insert one or more hyperlinks into the plain English summary in accordance with documents associated with the second metadata, wherein the response comprises the plain English summary, the one or more hyperlinks, or both.

13 . The computing device of claim 8 , wherein the instructions further cause the processor to:

create a production generative model by updating the generative model, the second graph network, the second metadata, the response, or combinations thereof, as unlabeled training data; and

deploy the production generative model in a cloud environment.

14 . The computing device of claim 8 , wherein the instructions further cause the processor to:

receive, from the edge device, a second request associated with the response, the second request associated with controlling an asset associated with a first entity of the plurality of entities and the response;

control the asset associated with the first entity to transfer to a second entity of the plurality of entities; and

update the first graph network or the second graph network in accordance with the transfer.

15 . A non-transitory computer-readable storage medium storing computer-executable instructions that, when executed with one or more processors of a computing device, causes the one or more processors to:

receive a request from an edge device;

provide, to a generative model in a prompt, the request;

determine, based at least in part on an output of the generative model, a negative response condition in accordance with a comparison of a first similarity score with a first pre-determined threshold;

compare natural language information to a first graph network in accordance with a first pre-defined degree of separation, wherein the first graph network comprises a first set of nodes that represent a plurality of entities and a first set of edges that represent relationships between the plurality of entities and the natural language information comprises at least one of a portion of the output or a portion of the request;

determine a second similarity score based at least in part by comparing natural language text of a second set of nodes with the natural language information, wherein the second set of nodes do not occur in the first graph network and the second similarity score represents similarity between at least the natural language information and first metadata associated with the second set of nodes;

extract second metadata from the second set of nodes based at least in part on the second similarity score being in accordance with a second pre-determined threshold, wherein the second 20 set of nodes is within a second pre-defined degree of separation different from the first pre-defined degree of separation;

create a second graph network based at least in part by creating edges between i) nodes of the second set of nodes, ii) the first set of nodes, or both i) and ii), that represent relationships between nodes of the second set of nodes and/or the first set of nodes, wherein the second graph network is different than the first graph network;

generate a response to the request by accessing particular metadata of the second metadata associated with at least one node of the second set of nodes by extracting and formatting contextual natural language from the particular metadata; and

provide the response including the particular metadata to the edge device.

16 . The non-transitory computer-readable storage medium of claim 15 , wherein the instructions further cause the processor to:

deploy the second graph network to a cloud environment that hosts the first graph network or the second graph network; and

update one or more skills associated with a chat bot that received the request based at least in part on the second graph network.

17 . The non-transitory computer-readable storage medium of claim 15 , wherein the first pre-defined degree of separation includes two or fewer edges, and the second pre-defined degree of separation includes at least one edge more than the first pre-defined degree of separation, and wherein the negative response condition indicates a lack of a response or a non-response condition.

18 . The non-transitory computer-readable storage medium of claim 15 , wherein the request is received by a first chat bot, wherein the instructions further cause the processor to:

create a new skill for the first chat bot based at least in part on the second metadata or the response; and

store the new skill in a cloud environment for retrieval by a second chat bot different from the first chat bot.

19 . The non-transitory computer-readable storage medium of claim 15 ,

extract portions of information from the second metadata;

transform the portions of information into plain English summary with a context associated with the request; and

insert one or more hyperlinks into the plain English summary in accordance with documents associated with the second metadata, wherein the response comprises the plain English summary, the one or more hyperlinks, or both.

20 . The non-transitory computer-readable storage medium of claim 15 , wherein the instructions further cause the processor to:

create a production generative model by updating the generative model, the second graph network, the second metadata, the response, or combinations thereof, as unlabeled training data; and

deploy the production generative model in a cloud environment.

Assignments (1)

References Cited (28)