Case-based reasoning system using normalized weight vectors

Patent metadata

Field	Value
Publication number	US-9330358-B1
Application number	US-201314037640-A
Country	US
Kind code	B1
Filing date	Sep 26, 2013
Priority date	Jan 4, 2013
Publication date	May 3, 2016
Grant date	May 3, 2016

Abstract

A system and method include comparing a context to cases stored in a case base, where the cases include Boolean and non-Boolean independent weight variables and a domain-specific dependency variable. The case and context independent weight variables are normalized and a normalized weight vector is determined for the case base. A match between the received context and each case of the case base is determined using the normalized context and case variables and the normalized weight vector. A skew value is determined for each category of domain specific dependency variables and the category of domain specific dependency variables having the minimal skew value is selected. The dependency variable associated with the selected category is then displayed to a user.

First claim

I claim:

1. A computer-implemented method comprising the steps of:
providing a system including a computer having a computer input device connected thereto, a display device connected thereto, and a plurality of distributed processors communicatively coupled to the computer, wherein the computer is configured to coordinate the activities of the distributed processors, wherein each of the distributed processors is configured to maintain a case base comprising a plurality of cases arranged in a tabular format, each case comprising one or more case base independent weight variables and a domain-specific dependency variable, wherein the case base independent weight variables comprise case base Boolean variables and case base non-Boolean variables;
normalizing the non-Boolean variables for each case of the case base;
determining a normalized weight vector for the case base;
receiving a context comprising one or more contextual independent weight variables, wherein the contextual independent weight variables comprise contextual Boolean variables and contextual non-Boolean variables;
normalizing the contextual non-Boolean variables;
determining a match between the received context and each case of the case base using the normalized non-Boolean variables for each case of the case base, the normalized contextual non-Boolean variables, and the normalized weight vector for the case base, using the equation

match
⁡

(
i
)

=

∑

k
=
1

n

⁢

w
k

⁢



c

i
,
k

-

c

j
,
k





participating
⁢

⁢
situational
⁢

⁢
variables



,

i
≠
j

,

where c i is a case in the case base, c j is a received context, and w is the normalized weight vector for the case base;
determining a skew value for each category of domain specific dependency variables;
selecting the category of domain specific dependency variables having a minimal skew value;
comparing the minimal skew value to a user-defined squelch value;
returning feedback to a user of the system that the system is unsure of the result if the minimal skew value exceeds the user-defined squelch value; and
returning feedback to a user of the system that the system is sure of the result if the minimal skew value does not exceed the user-defined squelch value.

2. The computer-implemented method of claim 1 , wherein the step of determining a skew value for each category of domain specific dependency variables uses a 3-2-1 skew.

3. The computer-implemented method of claim 1 , wherein the step of determining a skew value for each category of domain specific dependency variables uses a uniform skew.

4. The computer-implemented method of claim 1 , wherein the step of determining a normalized weight vector for the case base comprises the steps of:
comparing the case base Boolean variables and case base non-Boolean variables of each case of the case base to each other to determine a weight vector for each comparison;
normalizing the weight vector for each comparison;
summing, across each comparison, the weights of each of the case base Boolean variables and case base non-Boolean variables;
dividing each of the sums of the weights of each of the case base Boolean variables and case base non-Boolean variables by the number of participating situational variables to determine an average weight vector for each of the case base Boolean variables and case base non-Boolean variables; and
normalizing the average weight vector to determine the normalized weight vector.

5. The computer-implemented method of claim 1 further comprising the step of displaying the value of the selected category of domain specific dependency variables on a display.

6. A system comprising:
a computer having a computer input device and a display device connected thereto; and
a plurality of distributed processors communicatively coupled to the computer wherein the computer is configured to coordinate the activities of the distributed processors, wherein each of the distributed processors is configured to maintain a case base, wherein each of the plurality of distributed processors are configured to perform a method including the steps of:
providing a case base comprising a plurality of cases arranged in a tabular format, each case comprising one or more case base independent weight variables and a domain-specific dependency variable, wherein the case base independent weight variables comprise case base Boolean variables and case base non-Boolean variables;
normalizing the non-Boolean variables for each case of the case base;
determining a normalized weight vector for the case base;
receiving a context comprising one or more contextual independent weight variables, wherein the contextual independent weight variables comprise contextual Boolean variables and contextual non-Boolean variables;
normalizing the contextual non-Boolean variables;
determining a match between the received context and each case of the case base using the normalized non-Boolean variables for each case of the case base, the normalized contextual non-Boolean variables, and the normalized weight vector for the case base, using the equation

match
⁡

(
i
)

=

∑

k
=
1

n

⁢

w
k

⁢

Assignees

• Us Navy

Inventors

• Rubin Stuart H

Classifications

• G06N5/022 Primary

  Knowledge engineering; Knowledge acquisition · CPC title

• G06N5/04 Primary

  Inference or reasoning models · CPC title