System and method for supporting the utilization of machine learning

• US 8,200,610 B1
• Filed: 09/19/2011
• Issued: 06/12/2012
• Est. Priority Date: 05/22/2006
• Status: Active Grant

• Current Assignee: Concentrix CVG Customer Management Delaware LLC (Synnex Corporation)
• Original Assignee: Convergys Corporation (Synnex Corporation)
• Inventors: Birch, Robert D., Craig, Brian D., Irwin, Scott A., Weagraff, Stephen D.
• Primary Examiner(s): FERNANDEZ RIVAS, OMAR F
• Application Number:
  US13/236,079
• Time in Patent Office: 267 Days
• Field of Search: None
• US Class Current: 706/62
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A system and method is disclosed which integrates a machine learning solution into a large scale, distributed transaction processing system using a supporting architecture comprising a combination of computer hardware and software. Methods of using a system comprising such supporting architecture provide application designers access to the functionality included in a machine learning solution, but might also provide additional functionality not supported by the machine learning solution itself.

Claims

1. A non-transitory computer readable medium having stored thereon a set of instructions operable to configure a computer system to perform a set of acts comprising:

a) receiving a learning event;

b) sending said learning event to a model management component, wherein said model management component is configured to determine a machine learning model associated with said learning event;

c) using said machine learning model, accessing a synchronization policy associated with said model;

d) sending said learning event to a prototypical machine learning model determined from said synchronization policy;

e) updating said prototypical machine learning model according to said learning event to produce an updated prototypical machine learning model; and

f) propagating said updated prototypical machine learning model to a plurality of machine learning models comprising said machine learning model and at least one other machine learning model hosted remotely from said updated prototypical machine learning model and said machine learning model.

2. The non-transitory computer readable medium, as claimed in claim 1, wherein said set of instructions is operable to configure the computer system to perform the propagating step in real time.

3. A computerized method comprising:

a) receiving a request from an application program through an application interface;

b) communicating the request received from the application program to a model management component, wherein the model management component comprises;

i) a model pool;

ii) a model pool manager; and

iii) a synchronization manager;

wherein a model is stored within said model pool and wherein a synchronization policy is associated with said model;

wherein said model pool manager is configured to create, retrieve, update and/or delete said model within the model pool based on said request received through said application interface component;

wherein said synchronization manager executes said synchronization policy associated with said model;

wherein said synchronization manager is configured to update a prototypical model only if said request contains an appropriate learning event;

wherein said appropriate learning event comprises that said model, associated with said request, and said prototypical model comprise an identical vendor, an identical algorithm structure, and a set of identical model parameter attribute types.

4. A non-transitory computer readable medium having stored thereon a set of instructions operable to configure a transaction processing system to perform a set of steps comprising:

a) sending a message to a machine learning controller in response to a predefined event;

b) based on the message, requesting a model manager and an algorithm manager from said machine learning controller for an instance of an associated algorithm and an instance of an associated model; and

c) responding to the predefined event based on processing of the instance of the associated algorithm, the instance of the associated model, and an output from a classification method.

5. The non-transitory computer readable medium of claim 4, wherein the set of steps further comprises:

a) calling a learn method when a second predefined event takes place;

b) passing a second predefined event message, regarding said second predefined event, to said machine learning controller as a learning event;

c) passing said learning event to an update method exposed by a synchronization policy interface of said instance of said associated model via said model manager;

d) sending said learning event to a prototypical model; and

e) creating an updated version of said prototypical model.

6. The non-transitory computer readable medium of claim 5, wherein the set of steps further comprises propagating said updated prototypical model to a plurality of distributed servers according to a propagation method exposed by said synchronization policy interface.

7. The non-transitory computer readable medium of claim 4, wherein:

a) the predefined event comprises a customer visiting a web site; and

b) responding to the predefined event comprises determining a recommendation from a set of available products.

8. The non-transitory computer readable medium of claim 7, wherein the second predefined event comprises a purchase taking place.

9. The non-transitory computer readable medium of claim 4, wherein:

a) the predefined event comprises receiving a natural language input from a user; and

b) responding to the predefined event comprises performing at least one action taken from the set of actions consisting of;

i) classifying the natural language input; and

ii) determining a response by a computer in an automated dialog.

10. The non-transitory computer readable medium of claim 4, wherein

a) the predefined even comprises receiving a call from a user;

and

b) responding to the predefined event comprises making a routing determination for the call.