WO2000055790B1

WO2000055790B1 - Gradient criterion method for neural networks and application to targeted marketing

Info

Publication number: WO2000055790B1
Application number: PCT/US2000/006735
Authority: WO
Inventors: Yuri Galperin; Vladimir Fishman
Original assignee: Marketswitch Corp; Yuri Galperin; Vladimir Fishman
Priority date: 1999-03-15
Filing date: 2000-03-15
Publication date: 2001-02-22
Also published as: AU3884000A; WO2000055790A3; CA2403249A1; WO2000055790A2

Abstract

The present invention is drawn to a unique application of the Maximum Likelihood statistical method to commercial neural network technologies. The present invention utilizes the specific nature of the output in target marketing problems and makes it possible to produce more accurate and predictive results by minimizing a gradient criterion to produce model weights to get the maximum likelihood result. It is best used on 'noisy' data and when one is interested in determining a distribution's overall accuracy, or best general description of reality.

Claims

AMENDED CLAIMS

[received by the International Bureau on 8 December 2000 (08.12.00); new claims 7-11 added; remaining claims unchanged (3 pages)]

7. A system for training neural networks with a maximum likelihood utility function, comprising: a central application server; a modeling database connected to said central application server; at least one workstation networked to said central application server; at least one multithreaded calculation engine networked to said central application server; and software instructions on said central application server, at least one workstation and at least one multithreaded calculation engine so as to provide for: said at least one workstation to select an initial model function for a propensity score g(X, W), were F is a set of weights of the neural network and X is a vector of customer attributes from a modeling database; and said at least one multithreaded calculation engine to calculate propensity scores for the customers in the modeling database; calculate a training error Err, where

N £rr « -lnZ = ∑m(l + g, )- ∑lnte,)- ∑ln0-ft)

(■I imrap iuvm-nup measure the error to check for convergence below a desired value; obtain a new model and apply it to new data when convergence occurs; minimize the error to solve for new weights Why minimizing the gradient criterion defined by the formula:

£rr^' = (∑-^ ∑l/g, + ∑ -£— )*^' ; and

(=1 1 + gι Itrup lenan-nxp 1 "" §

11 begin a new iteration of the process by calculating new propensity scores for the customers in the modeling database. 8. The system for training neural networks with a maximum likelihood utihty function of claim 7, further comprising: a customer database connected to said central application server and said at least one multithreaded calculation engine; and software instructions to apply the new model to customer data from said customer database upon being selected by said at least one workstation. 9. The system for training neural networks with a maximum likelihood utility function of claim 7, further comprising: software instructions on said at least one multithreaded calculation engine to: define f as a normalized propensity score related to g(X, W) by the formula: g(X,W) = "(X,W) where/ is the output of the neural network; and choose the parameter τ in such a way that /may be of the order of 0.5; wherein R is an average response rate in the sample and the above condition is satisfied if:

r = l/ln^— ; and

R wherein:

12 Err = -lnR = ∑m(l + //^/' -i ∑ln ) - find -/¹") ;

1=1 ^» .erssp lenon_nιp and gradient criterion is computed as follows:

10- The system for training neural networks with a maximum likelihood utility function of claim 7, further comprising: a customer database connected to said central application server and said at least one multithreaded calculation engine; and software instructions to apply the new model to a top 20% of a targeted marketing sample customer pool selected from said customer database by said a least one workstation. 11. The system for training neural networks with a maximum likelihood utility function of claim 9, further comprising: a customer database connected to said central application server and said at least one multithreaded calculation engine; and software instructions to apply the new model to a top 20% of a targeted marketing sample customer pool selected from said customer database by said a least one workstation.

13 STATEMENT UNDER ARTICLE 19

New claims 7-12 added to define apparatus of invention . All the remaining claims are unchanged.

14