Multiple output relaxation machine learning model

• US 8,352,389 B1
• Filed: 08/20/2012
• Issued: 01/08/2013
• Est. Priority Date: 08/20/2012
• Status: Active Grant

• Current Assignee: XANT Inc.
• Original Assignee: Insidesales.com
• Inventors: Martinez, Tony Ramon, Zheng, Xinchuan
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): CHANG, LI WU
• Application Number:
  US13/590,028
• Time in Patent Office: 141 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: H04L 67/10 in which an application is distributed across nodes in the network software deployment G06F8/60; multiprogramming arrangements G06F9/46

Abstract

A multiple output relaxation (MOR) machine learning model. In one example embodiment, a method for employing an MOR machine learning model to predict multiple interdependent output components of a multiple output dependency (MOD) output decision may include training a classifier for each of multiple interdependent output components of an MOD output decision to predict the component based on an input and based on all of the other components. The method may also include initializing each possible value for each of the components to a predetermined output value. The method may further include running relaxation iterations on each of the classifiers to update the output value of each possible value for each of the components until a relaxation state reaches an equilibrium or a maximum number of relaxation iterations is reached. The method may also include retrieving an optimal component from each of the classifiers.

Claims

1. A method for employing a multiple output relaxation (MOR) machine learning model to predict, using an input, multiple interdependent output components of a multiple output dependency (MOD) output decision, each output component having multiple possible values, the method comprising:

training a classifier for each of multiple interdependent output components of an MOD output decision to predict the output component based on the input and based on all of the other output components;

initializing each of the possible values for each of the output components to a predetermined output value;

running relaxation iterations on each of the classifiers to update the output value of each possible value for each of the output components until a relaxation state reaches an equilibrium or a maximum number of relaxation iterations is reached; and

retrieving an optimal output component from each of the classifiers.

2. The method as recited in claim 1, wherein each classifier comprises a multilayer perceptron (MLP) neural network, another multilayer neural network, a decision tree, or a support vector machine.

3. The method as recited in claim 1, wherein an equilibrium is reached when the difference between the output value of a possible value for an output component of a current iteration and the output value of a possible value for an output component of a previous iteration is less than or equal to a predetermined threshold value.

4. The method as recited in claim 1, wherein:

the MOD output decision is an LRM MOD output decision;

the input comprises an input feature vector that includes constant features about a lead and interactive features related to interactions between an agent and the lead, the features of the input feature vector including one or more of lead source, lead title, lead industry, lead state, lead created date, lead company size, lead status, number of previous dials, number of previous emails, previous action, hours since last action, response agent title, response method, response message type, response timing, agent or lead demographic profile, agent or lead histographic profile, agent or lead psychographic profile, agent or lead social network profile, agent or lead geographic profile, response frequency, response persistence, and data on current events; and

the output components include one or more of response agent title, response method, response message type, response timing, agent or lead demographic profile, agent or lead histographic profile, lead contact title, agent or lead psychographic profile, agent or lead social network profile, agent or lead geographic profile, response frequency, and response persistence.

5. The method as recited in claim 1, wherein the MOD output decision includes two or more interdependent output components.

6. The method as recited in claim 1, wherein the MOD output decision relates to sports, hostage negotiations, retail sales, online shopping carts, web content management systems, customer service, contract negotiations, or crisis management, or some combination thereof.

7. The method as recited in claim 1, wherein the method results in multiple correct MOD output decisions and each of the multiple correct MOD output decisions has the same output value or has an output value above a predetermined threshold.

8. The method as recited in claim 1, wherein the order of output components in the MOD output decision is determined simultaneously.

9. A non-transitory computer-readable medium storing a program that causes a processor to execute the method according to claim 1.

10. A method of employing a multiple output relaxation (MOR) machine learning model to predict multiple interdependent output components of a multiple output dependency (MOD) output decision, each output component having multiple possible values, the method comprising:

training a first classifier to predict a first of two interdependent output components of an MOD output decision based on an input and based on the second output component;

training a second classifier to predict the second of the two output components of the MOD output decision based on the input and based on the first output component;

initializing each of the possible values for each of the output components to a predetermined output value;

running relaxation iterations on each of the classifiers to update the output value of each possible value for each of the output components until a relaxation state reaches an equilibrium or a maximum number of relaxation iterations is reached; and

retrieving an optimal output component from each of the classifiers.

11. The method as recited in claim 10, where the first and second classifiers each comprises a multilayer perceptron (MLP) neural network, another multilayer neural network, a decision tree, or a support vector machine.

12. The method as recited in claim 10, wherein the input comprises an input feature vector including constant features about a lead and interactive features related to interactions between an agent and the lead, the features of the input feature vector including one or more of lead source, lead title, lead industry, lead state, lead created date, lead company size, lead status, number of previous dials, number of previous emails, previous action, hours since last action, response agent title, response method, response message type, response timing, agent or lead demographic profile, agent or lead histographic profile, agent or lead psychographic profile, agent or lead social network profile, agent or lead geographic profile, response frequency, response persistence, and data on current events.

13. The method as recited in claim 12, wherein the MOD output decision comprises an LRM MOD output decision and the multiple interdependent output components include one or more of response agent title, response method, response message type, response timing, agent or lead demographic profile, agent or lead histographic profile, lead contact title, agent or lead psychographic profile, agent or lead social network profile, agent or lead geographic profile, response frequency, and response persistence.

14. The method as recited in claim 10, wherein an equilibrium is reached when the difference between the output value of a possible value for an output component of a current iteration and the output value of a possible value for an output component of a previous iteration is less than or equal to a predetermined threshold value.

15. The method as recited in claim 10, where one or more additional classifiers is trained to predict one or more additional interdependent output components of the MOD output decision based on the input and based on the first output component, the second output component, and each other additional output component;

the input, the second output component, and each additional output component are used to predict the first output component; and

the input, the first output component, and each additional output component are used to predict the second output component.

16. A non-transitory computer-readable medium storing a program that causes a processor to execute the method according to claim 10.

17. A method of employing a multiple output relaxation (MOR) machine learning model to predict multiple interdependent output components of a lead response management (LRM) multiple output dependency (MOD) output decision, each output component having multiple possible values, the method comprising:

training a first classifier to predict a first of two interdependent output components of an LRM MOD output decision based on an input and based on the second output component;

training a second classifier to predict the second of the two output components of the LRM MOD output decision based on the input and based on the first output component;

initializing each of the possible values for each of the output components to a predetermined output value;

running relaxation iterations on each of the classifiers to update the output value of each possible value for each of the output components until a relaxation state reaches an equilibrium or a maximum number of relaxation iterations is reached; and

retrieving an optimal output component from each of the classifiers.

18. The method as recited in claim 17, where the first and second classifiers each comprises a multilayer perceptron (MLP) neural network, another multilayer neural network, a decision tree, or a support vector machine.

19. The method as recited in claim 17, wherein the output components include one or more of response agent title, response method, response message type, response timing, agent or lead demographic profile, agent or lead histographic profile, lead contact title, agent or lead psychographic profile, agent or lead social network profile, agent or lead geographic profile, response frequency, and response persistence.

20. The method as recited in claim 17, wherein an equilibrium is reached when the difference between the output value of a possible value for an output component of a current iteration and the output value of a possible value for an output component of a previous iteration is less than or equal to a predetermined threshold value.

21. The method as recited in claim 17, where one or more additional classifiers is trained to predict one or more additional interdependent output components of the LRM MOD output decision based on the input and based on the first output component, the second output component, and each other additional output component;

the input, the second output component, and each additional output component are used to predict the first output component; and

the input, the first output component, and each additional output component are used to predict the second output component.

22. A non-transitory computer-readable medium storing a program that causes a processor to execute the method according to claim 17.