Information-processing apparatus, information-processing method and storage medium Number:7,062,477 from the United States Patent and Trademark Office (PTO) owispatent

Title: Information-processing apparatus, information-processing method and storage medium

Abstract: In order to increase a response rate, a data mining server selects an advertisement mail among a plurality of advertisement mails to be sent in a test transmission to personal computers selected at random among those owned by registered customers. The data mining server then computes learning parameters for each of sent advertisement mails from learning data created from response results of the test transmissions. Then, the data mining server applies learning parameters of each advertisement mail to original assessment data of other personal computers each serving as an object of an actual transmission to find predicted values. Subsequently, the data mining server extracts assessment data with largest advertisement-mail predicted values from the original assessment data. The data mining server then synthesizes the extracted pieces of assessment data. Finally, the data mining server sorts the synthesized pieces of assessment data in an order of decreasing customer predicted values to create an assessment chart.

Patent Number: 7,062,477 Issued on 06/13/2006 to Fujiwara,   et al.

---

Inventors:	Fujiwara; Yoshihito (Kanagawa, JP); Matsubayashi; Masako (Tokyo, JP); Kurami; Naoya (Tokyo, JP); Nishioka; Hisao (Chiba, JP); Kaneko; Yasuyoshi (Kanagawa, JP)
Assignee:	Sony Corporation (Tokyo, JP)
Appl. No.:	975799
Filed:	October 11, 2001

---

Foreign Application Priority Data

---

Oct 12, 2000 [JP]			2000-312329

Current U.S. Class:	706/21 ; 725/46
Current International Class:	G06F 15/18 (20060101)
Field of Search:	706/21 725/46

---

References Cited [Referenced By]

---

U.S. Patent Documents

	5790801		August 1998		Funato
	5857073		January 1999		Tsukamoto et al.
	5864670		January 1999		Hayashi et al.
	5958059		September 1999		Oishi
	6433893		August 2002		Murayama
	6470337		October 2002		Nihei
	6530083		March 2003		Liebenow
	6549939		April 2003		Ford et al.
	6559964		May 2003		Tsukamoto et al.
	6667813		December 2003		Saruwatari et al.
	6731410		May 2004		Saito et al.

Primary Examiner: Davis; George
Attorney, Agent or Firm: Finnegan, Henderson, Farabow, Garrett & Dunner, L.L.P.

---

Claims

---

What is claimed is:

1. An information-processing apparatus comprising: computation means for computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; and select means for selecting some of the plurality of contents to be transmitted to each of said information-processing terminals based on said expected value computed by said computation means for each of said contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein said computation means computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a signoid function, a rule form of a conditional, a decision tree model, or a statistical technique based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denoting each user's independent preferences and a coefficient denoting the plurality of transmitted contents, and wherein said expected value can be expressed as a ratio of a maximum response rate and a minimum response rate.

2. The information-processing apparatus according to claim 1, wherein said information-processing apparatus further comprises transmission means for transmitting contents selected by said select means to any of said information-processing terminals.

3. The information-processing apparatus according to claim 1, wherein said computation means computes an expected value of any one of said information-processing terminals from results of a test transmission carried out for said information-processing terminal.

4. The information-processing apparatus according to claim 1, wherein, for any specific one of said information-processing terminals, said select means selects a content whose expected value computed by said computation means.

5. The information-processing apparatus according to claim 1, wherein said expected value is a probability of a response expected to be received from any one of said information-processing terminals or an expected response rate of responses received from said information-processing terminals.

6. The information-processing apparatus according to claim 1, wherein said expected value is a predicted probability of a response.

7. The information-processing apparatus according to claim 1, wherein said contents are different from each other because some text parts are modified.

8. The information-processing apparatus according to claim 1, wherein said contents are each an electronic mail or a web banner advertisement.

9. The information-processing apparatus according to claim 1, wherein said contents each include hyperlink information.

10. The information-processing apparatus according to claim 9, wherein said computation means computes said expected value on the basis of click information of said hyperlink information.

11. An information-processing method comprising: computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; and selecting some of the plurality of contents to be transmitted to each of said information-processing terminals based on said expected value computed for each of said contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein the computing step computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a signoid function, a rule form of a conditional, a decision tree model, or a statistical techniciue based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denoting each user's independent preferences and a coefficient denoting the plurality of transmitted contents, and wherein said expected value can be expressed as a ratio of a maximum response rate and a minimum response rate.

12. A method, stored on a computer-readable medium, comprising: computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; and selecting some of the plurality of contents to be transmitted to each of said information-processing terminals based on said expected value computed for each of said contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein the computing steps computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a sigmoid function, a rule form of a conditional, a decision tree model, or a statistical technique based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denoting each user's independent preferences and a coefficient denoting the plurality of transmitted contents, and wherein said expected value can be expressed as a ratio of a maximum response rate and a minimum response rate.

13. An information-processing apparatus comprising: computation means for computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; first producing means for producing a first assessment information on a set of largest expected values computed by said computation means for said responses transmitted by said information-processing terminals in response to said plurality of contents based on said set of largest expected values which are each computed by said computation means for one of said contents; and second producing means for producing a second assessment function of said set of largest expected values computed for all said contents including user specific information relating to each of said information-processing terminals by synthesizing pieces of said assessment information which are each produced by said first producing means for one of said contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein said computation means computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a sigmoid function, a rule form of a conditional, a decision tree model, or a statistical technique based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denotina each user's independent preferences and a coefficient denoting the plurality of transmitted contents, wherein said expected value can be expressed as a ratio of a maximum response rare and a minimum response rate, and wherein said response transmitted in response to each of the plurality of contents may be a selective transmission or a random transmission.

14. An information-processing method comprising: computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; producing assessment information on a set of largest expected values for said responses transmitted by said information-processing terminals in response to said contents based on said set of largest expected values each computed for one of said contents; and producing an assessment function of said set of largest expected values for all said contents by synthesizing pieces of said assessment information each produced for one of said contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein the computing step computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a sigmoid function, a rule form of a conditional, a decision tree model, or a statistical techniciue based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denoting each user's indenendent preferences and a coefficient denoting the plurality of transmitted contents. wherein said expected value can be expressed as a ratio of a maximum response rate and a minimum response rate, and wherein said response transmitted in response to each of the plurality of contents may be a selective transmission or a random transmission.

15. A method, stored on a computer-readable medium, comprising: computing an expected value of a response transmitted by a plurality of information-processing terminals in response to each of a plurality of contents transmitted to said information-processing terminals; producing assessment information on a set of largest expected values for said responses transmitted by said information-processing terminals in response to said contents based on said set of largest expected values each computed for one of said contents; and producing an assessment function of said set of largest expected values for all said plurality of contents by synthesizing pieces of said assessment information produced for each one of said plurality of contents, wherein said information-processing terminals comprise at least a pair of terminals used by independent users, each user having independent preferences, wherein said plurality of contents includes user specific information relating to each of said information-processing terminals, wherein the computing step computes said expected value by regular extraction based on a formula, wherein said formula is one of a linear association expression, a nueral network, a sigmoid function, a rule form of a conditional, a decision tree model, or a statistical technique based on a linear model, a discriminative analysis, a logistic recurssion/regression, or a cluster analysis, wherein said linear association expression is a linear expression of a sum of terms which are each a product of a numerical data denotina each user's independent preferences and a coefficient denoting the plurality of transmitted contents, wherein said expected value can be expressed as a ratio of a maximum response rate and a minimum response rate, and wherein said response transmitted in response to each of the plurality of contents may be a selective transmission or a random transmission.

---

Description

---

BACKGROUND OF THE INVENTION

The present invention relates to an information-processing apparatus, an information-processing method and a computer program. More particularly, the present invention relates to an information-processing apparatus and an information-processing method that are capable of increasing an advertising effect of distibution of advertisement contents to customers through a two-direction network, and relates to a computer program prescribing the information-processing method.

In recent years, there has been established a variety of services to distribute advertisement mails to registered customers. Such services to distribute advertisement mails each adopt an analysis method based on a data mining technique adopted in direct mailing by post.

The data mining technique is an advanced technique to search a large amount of data for a hidden cause-effect relation or a pattern, an advanced modeling technique or a decision-support technique which allows complex relations to be established among pieces of data and draws much attention for the past several years in fields such as artificial intelligence and data engineering.

It should be noted that there are various fields of the data mining technique. Examples of the data mining technique are a technique to determine what a customer will buy next by consideration of a past purchase record, a technique to predict when false credit cards will be used, a technique to determine a reason why a customer switches to a competitor, a technique to find a way to have such a customer give up the competitor and return to itself, an optimum technique to focus on potential customers in accordance with purchasing patterns and an aid to find solutions to these problems.

By the way, in a process to focus on customers, to whom advertisement mails are to be sent, by using the data mining technique, a rate of responses received from the customers can be estimated by creation of a graph called an assessment chart.

In addition, in order to increase the response rate, there is adopted a method whereby only most likely responding customers are selected on the basis of computed expected response probabilities and an advertisement mail is sent to only the selected customers. With this method, however, the number of sent advertisement mails decreases, raising a problem of a reduced response rate relative to all customers.

The following methods are adopted by an enterprise or the like making a request for distribution of an advertisement mail to determine an advertisement fee to be paid to an enterprise rendering a service to distribute advertisement mails:

(1) An exposure determination method based on the number of sent advertisement mails or the number of users inspecting a web page displaying the banner advertisement.

(2) A response-count determination method based on the number of accesses to a web page (an advertisement page) which are made by actually clicking a URL (Uniform Resource Locator) included in the advertisement mail or the number of accesses to a banner advertisement of a web page which are made by actual clicking operations.

Since the exposure determination method does not consider the number of users who actually click the URL to make accesses to the web page, however, a relation between the effect of the advertisement and the expense is not clear. Since the response-count determination method is based on an actual result indicated by the number of actual responses, on the other hand, there are raised problems that the advertisement cost cannot be determined in advance or, if the advertisement fee is set by making a contract based on a predetermined number of responses, the response count specified in the contract cannot be achieved or it takes a long time to achieve the response count.

When an advertisement mail is distributed after estimating a response rate by adoption of the conventional technique to create an assessment chart, however, there is raised a problem that a high actual response rate is not necessarily obtained.

In addition, the conventional technique to create an assessment chart is provided as a method to be used when a single content is distributed. There is also a problem that this conventional technique is not capable of creating an assessment chart, which is used for estimating a response rate with a high degree of accuracy when an optimum content is selected for each customer from a plurality of contents to be sent to the customer.

SUMMARY OF THE INVENTION

It is an object of the present invention addressing the problems described above to increase the accuracy of estimation of a response rate.

To achieve the above object, according to a first aspect of the present invention, there is provided an information-processing apparatus including:

computation means for computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals; and

select means for selecting some of a plurality of contents to be transmitted to each of the information-processing terminals on the basis of the expected value computed by the computation means for each of the contents.

In accordance to a second aspect of the present invention, there is provided an information-processing method including the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals; and

selecting some of a plurality of contents to be transmitted to each of the information-processing terminals on the basis of the expected value computed for each of the contents.

In accordance to a third aspect of the present invention, there is provided a program to be executed by a computer to carry out the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals; and

selecting some of a plurality of contents to be transmitted to each of the information-processing terminals on the basis of the expected value computed for each of the contents.

In accordance to a fourth aspect of the present invention, there is provided an information-processing apparatus including:

transmission means for transmitting a content to information-processing terminals;

response-rate-computing means for computing a response rate of responses transmitted by the information-processing terminals in response to the content transmitted by the transmission means;

storage means for storing a fee of transmitting the content for each response rate; and

acquirement means for acquiring a fee of transmitting the content for a response rate computed by the response-rate-computing means from the storage means.

In accordance to a fifth aspect of the present invention, there is provided an information-processing method including the steps of:

transmitting a content to information-processing terminals;

computing a response rate of responses transmitted by the information-processing terminals in response to the content; and

storing a fee of transmitting a content for each response rate in advance;

acquiring a stored fee of transmitting the content for the computed response rate.

In accordance to a sixth aspect of the present invention, there is provided a program to be executed by a computer to carry out the steps of:

transmitting a content to information-processing terminals;

computing a response rate of responses transmitted by the information-processing terminals in response to the content; and

storing a fee of transmitting a content for each response rate in advance;

acquiring a stored fee of transmitting the content for the computed response rate.

In accordance to a seventh aspect of the present invention, there is provided an information-processing apparatus including:

computation means for computing an expected value of a response transmitted by each of information-processing terminals in response to a content transmitted to the information-processing terminals;

setting means for setting a predetermined threshold value for the expected values computed by the computation means;

storage means for storing a fee of transmitting the content for each expected value; and

acquirement means for acquiring a fee of transmitting the content for the threshold value set by the setting means from the storage means.

In accordance to a eighth aspect of the present invention, there is provided an information-processing method including the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to a content transmitted to the information-processing terminals;

setting a predetermined threshold value for the computed expected values; and

storing a fee of transmitting a content in advance for each expected value;

acquiring a stored fee of transmitting the content for the predetermined threshold value.

In accordance to a ninth aspect of the present invention, there is provided a program to be executed by a computer to carry out the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to a content transmitted to the information-processing terminals;

setting a predetermined threshold value for the computed expected values; and

storing a fee of transmitting a content in advance for each expected value;

acquiring a stored fee of transmitting the content for the predetermined threshold value.

In accordance to a tenth aspect of the present invention, there is provided an information-processing apparatus including:

computation means for computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals;

first producing means for producing assessment information including largest expected values computed by the computation means for the responses transmitted by the information-processing terminals in response to the contents on the basis of the expected values which are each computed by the computation means for one of the contents; and

second producing means for producing an assessment function of the expected values computed for all the contents by synthesizing pieces of the assessment information which are each produced by the first producing means for one of the contents.

In accordance to a eleventh aspect of the present invention, there is provided an information-processing method including the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals;

producing assessment information including largest ones of the expected values for the responses transmitted by the information-processing terminals in response to the contents on the basis of the expected values each computed for one of the contents; and

producing an assessment function of the expected values for all the contents by synthesizing pieces of the assessment information each produced for one of the contents.

In accordance to a twelfth aspect of the present invention, there is provided a program to be executed by a computer to carry out the steps of:

computing an expected value of a response transmitted by each of information-processing terminals in response to each of a plurality of contents transmitted to the information-processing terminals;

producing assessment information including largest ones of the expected values for the responses transmitted by the information-processing terminals in response to the contents on the basis of the expected values each computed for one of the contents; and

producing an assessment function of the expected values for all the contents by synthesizing pieces of the assessment information each produced for one of the contents.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a block diagram showing a typical configuration of an advertisement-mail-distributing system provided by the present invention;

FIG. 2 is a diagram showing a typical configuration of the customer-information database employed in the advertisement-mail-distributing system shown in FIG. 1;

FIG. 3 is a block diagram showing a typical configuration of a mail server employed in the advertisement-mail-distributing system shown in FIG. 1;

FIG. 4 is a block diagram showing a typical configuration of a data mining server employed in the advertisement-mail-distributing system shown in FIG. 1;

FIG. 5 is a block diagram showing a typical configuration of a personal computer employed in the advertisement-mail-distributing system shown in FIG. 1;

FIG. 6 is a flowchart representing processing to collect data obtained from a test transmission;

FIGS. 7A to 7C show typical advertisement mails used in a test transmission 1;

FIGS. 8A to 8B show another typical advertisement mails used in the test transmission 1;

FIGS. 9A to 9C show typical advertisement mails used in a test transmission 2;

FIGS. 10A and 10B show another typical advertisement mails used in the test transmission 2;

FIGS. 11A to 11C show typical advertisement mails used in a test transmission 3;

FIG. 12 is an explanatory diagram used for describing results of responses to advertisement mails sent in the test transmission 1;

FIG. 13 is an explanatory diagram used for describing results of responses to advertisement mails sent in the test transmission 2;

FIG. 14 is an explanatory diagram used for describing results of responses to advertisement mails sent in the test transmission 3;

FIG. 15 is a flowchart representing processing carried out by the data mining server;

FIGS. 16A to 16C are diagrams showing typical learning models;

FIG. 17 is a flowchart representing other processing carried out by the data mining server;

FIG. 18 is an explanatory diagram used for describing an expected response probability;

FIGS. 19A to 19C are explanatory diagrams used for describing the number of customers each serving as a recipient for every advertisement mail and for every actual transmission;

FIG. 20 is a flowchart representing processing to collect data obtained from an actual transmission;

FIG. 21 is an explanatory diagram used for describing results of responses to advertisement mails sent in a test transmission 1;

FIG. 22 is an explanatory diagram used for describing results of responses to advertisement mails sent in a test transmission 2;

FIG. 23 is an explanatory diagram used for describing results of responses to advertisement mails sent in a test transmission 3;

FIG. 24 is an explanatory table showing a response rate for each customer profile and each of advertisement mails;

FIG. 25 is an explanatory table showing a response rate for each of other customer profiles and each of the advertisement mails;

FIG. 26 is an explanatory table showing a response rate for each of still other customer profiles and each of the advertisement mails;

FIG. 27 is an explanatory table showing a response rate for each of further customer profile s and each of the advertisement mails;

FIG. 28 is an explanatory table showing a response rate for each of still further customer profiles and each of the advertisement mails;

FIG. 29 is an explanatory table showing a response rate for each of still further customer profiles and each of the advertisement mails;

FIG. 30 is a diagram showing an assessment chart based on results of actual transmission A1;

FIG. 31 is a diagram showing an assessment chart based on results of actual transmission B1;

FIG. 32 is a diagram showing an assessment chart based on results of actual transmission 2;

FIG. 33 is a diagram showing an assessment chart based on results of actual transmission 3;

FIGS. 34A and 34B are explanatory diagrams showing histograms each representing a relative assessment chart based on results of an actual transmission;

FIG. 35 is a diagram showing a response pattern;

FIG. 36 is a diagram showing another response pattern;

FIG. 37 is a flowchart representing still other processing carried out by the data mining server;

FIG. 38 is a flowchart representing processing carried out by the mail server;

FIG. 39 is a diagram showing a still other response pattern;

FIG. 40 is a diagram showing a further response pattern;

FIG. 41 is a diagram showing a still further response pattern;

FIG. 42 is a flowchart representing further processing carried out by the data mining server;

FIG. 43 is an explanatory diagram used for describing a method of creating an assessment chart;

FIG. 44 is an explanatory diagram used for describing a method of creating the assessment chart as a continuation of the method shown in FIG. 43;

FIG. 45 is an explanatory diagram used for describing a method of creating the assessment chart as a continuation of the method shown in FIG. 44;

FIG. 46 is a flowchart representing still further processing carried out by the data mining server;

FIG. 47 is a diagram showing a typical assessment chart;

FIG. 48 is an explanatory diagram used for describing a method of creating an assessment chart;

FIG. 49 is an explanatory diagram used for describing a method of creating the assessment chart as a continuation of the method shown in FIG. 48;

FIG. 50 is an explanatory diagram used for describing a method of creating the assessment chart as a continuation of the method shown in FIG. 49;

FIG. 51 is an explanatory diagram used for describing a method of creating the assessment chart as a continuation of the method shown in FIG. 50;

FIG. 52 is a flowchart representing still further processing carried out by the data mining server;

FIG. 53 is a diagram showing another typical assessment chart;

FIG. 54 is a diagram showing a still other typical assessment chart;

FIG. 55 is a diagram showing a further typical assessment chart;

FIG. 56 is a diagram showing a still further typical assessment chart;

FIG. 57 is a flowchart representing still further processing carried out by the data mining server;

FIG. 58 is a flowchart representing still further processing carried out by the data mining server;

FIG. 59 is a flowchart representing still further processing carried out by the data mining server; and

FIG. 60 is a flowchart representing still further processing carried out by the data mining server.

PREFERRED EMBODIMENTS OF THE INVENTION

FIG. 1 is a block diagram showing a typical configuration of an advertisement-mail-distributing system provided by the present invention. In the advertisement-mail-distributing system, a mail server 1 transmits an advertisement mail to personal computers 7-1 to 7-3 and 8-1 to 8-3 owned by registered customers by way of a network 3 represented by the Internet. A customer-information database 2 is used for storing information on customers to whom an advertisement mail is to be distributed.

FIG. 2 is a diagram showing a typical configuration of the customer-information database 2. The typical customer-information database 2 is used for storing profile items for each customer identified by a customer ID. The profile items include demographic information, a product/service-purchase & utilization history and personal characteristic data.

The demographic information stored in the typical database shown in FIG. 2 includes a name, a phone number, a gender, an age, a family code and an income code. Some of the pieces of demographic information are coded. It should be noted that the demographic information is information on basic attributes as well as information usable as the so-called personal information.

The product/service-purchase & utilization history is a behavioral history including a purchase data and product codes. The product codes are each a coded item.

The personal-characteristic data is information on the customer's personality. The personal-characteristic data comprises answers to predetermined questions 1, 2, 3 and so on.

An advertisement mail transmitted by a mail server 1 includes a URL (Uniform Resource Locator) as additional information. A customer receiving an advertisement mail is capable of making an access to a web page stored in a web server 4 as indicated by a URL included in the advertisement mail. As an alternative, an advertisement mail may include the web server 4 as a destination to which a response to the advertisement can be sent by the customer. The web server 4 stores information on an access made by a customer in response to an advertisement mail received by the customer, and notifies a data mining server 5 of the response (the access).

The data mining server 5 controls an analysis database 6 on the basis of response information received from the web server 4, carrying out regular-extraction processing by adoption of a data mining technique. To put it concretely, the data mining server 5 is capable of picking up potential customers for a specific product or a specific service from a customer database with a structure similar to the structure of the database shown in FIG. 2. The customer's degree of potentiality is explained as follows.

A customer's degree of potentiality is an indicator or an predicted value representing the customer's interest in a specific product or a specific service or the customer's need for the specific product or the specific service. A degree of potentiality can be found typically by regular extraction based on a formula or the like.

A typical simple formula for computing a potentiality degree F is a linear associative expression of an equation like Eq. (1) including an expression of a sum of terms which are each a product of numerical data and a coefficient. F=a.times.Q1+b.times.Q2+c.times.Q3+d.times.Q4+e.times.Q5+ . . . (1) where notations Q1, Q2, Q3, Q4, Q5 and so on each denote customer profile data whereas notations a, b, c, d, e and so on each denote a constant or a coefficient set for a specific product or a specific service.

It should be noted that equations for computing a potentiality degree F are not limited to such a linear associative equation but may also be expressed as a variety of nonlinear computation formulas. For example, a formula for computing a potentiality degree F can be a neural network model using a sigmoid function.

In addition, a degree of potentiality can be derived from a rule form of a condition such as an IF statement as follows. IF (Q1>a AND Q2>b AND Q3>c AND Q4>d AND Q5>e), THEN F=X

It should be noted that, as for a statistical technique based on a linear model, a discriminative analysis, logistic recursion/regression, a cluster analysis or the like is appropriate and suitable for responses to the query words such as `why` and `how`.

In addition, a tree model which is also known as an induction technique is one of nonlinear models. A tree model is a decision tree formed from data. This tree model is appropriate for a case in which important variables are selected and unnecessary predicted elements are eliminated.

On the other hand, the neural network which is a nonlinear model is capable of predicting a future result based on history data and thus suitable for a response to the query word `what`.

The data mining server 5 adopts the data mining technique based on such formulas or the like to compute each customer's degree of potentiality with respect to a specific product or a specific service.

The data mining server 5 picks up potential customers for a specific product or a specific service on the basis of customers' degrees of potentiality which are found with respect to the product or the service. Thus, for example, marketing activities can be carried out effectively. To be more specific, an advertisement mail can be transmitted to customers desiring to purchase a specific product or a specific service.

In addition, a formula for computing a degree of potentiality can be derived by adoption of any arbitrary technique. For example, a formula for computing a customer's degree of potentiality can be derived from an existing relation among pieces of information received from the customer. In general, a variety of parameters can be obtained by application of the models described above to a data set of known variables (or target variables) for a relation between profile data of a customer and an expected interest of the customer. The data set is typically a database for a learning purpose.

FIG. 3 is a block diagram showing a typical configuration of the mail server 1. A CPU (Central Processing Unit) 21 employed in the mail server 1 executes programs stored in a ROM (Read Only Memory) 22 or programs loaded into a RAM (Random Access Memory) 23 from a storage unit 28 in order to carry out various kinds of processing. The RAM 23 is also used for properly storing data required by the CPU 21 in the execution of the processing.

The CPU 21, the ROM 22 and the RAM 23 are connected to each other by a bus 24. The bus 24 is also connected to an input/output interface unit 25. The input/output interface unit 25 is connected to an input unit 26, an output unit 27, the storage unit 28 and a communication unit 29. The input unit 26 includes a keyboard and a mouse. The output unit 27 comprises a speaker and a display unit, which can be a CRT or an LCD unit. The storage unit 28 is typically a hard disk. The communication unit 29 includes a modem and a terminal adaptor. The communication unit 29 acquires customer information from the customer-information database 2 as instructed by a command issued by the CPU 21 and transmits an advertisement mail stored in the storage unit 28 to a customer indicated by the customer information by way of the network 3.

If necessary, the input/output interface unit 25 is also connected to a drive 30, on which a magnetic disk 31, an optical disk 32, a magneto-optical disk 33, a semiconductor memory 34 or another storage medium is mounted. A computer program read out from a storage medium mounted in the drive 30 is installed in the storage unit 28 if necessary.

FIG. 4 is a block diagram showing a typical configuration of the data mining server 5. As shown in the figure, the data mining server 5 comprises components ranging from a CPU 41 to a semiconductor memory 54. This configuration is basically the same as the configuration of the mail server 1 which comprises components ranging from the CPU 21 to a semiconductor memory 34 as described above. Since block components of the data mining server 5, which have the same names as their counterparts employed in the mail server 1, have functions identical with the counterparts, their explanation is not repeated.

A storage unit 48 is used for storing a variety of programs for analysis purposes. The CPU 41 executes the programs, which are required for analyses.

A communication unit 49 receives response information from the web server 4 through the network 3, to which the communication unit 49 is connected. On the contrary, the communication unit 49 informs the mail server 1 of information on customers selected by an analysis carried out by the CPU 41. The customers are each selected as a recipient of an advertisement mail.

FIG. 5 is a block diagram showing a typical configuration of each of the personal computers 7-1 to 7-3 and 8-1 to 8-3. As shown in the figure, each of the personal computers 7-1 to 7-3 and 8-1 to 8-3 comprises components ranging from a CPU 61 to a semiconductor memory 74. This configuration is basically the same as the configuration of the mail server 1 which comprises components ranging from the CPU 21 to a semiconductor memory 34 as described above. In each of the personal computers 7-1 to 7-3 and 8-1 to 8-3, block components, which have the same names as their counterparts employed in the mail server 1, have functions identical with the counterparts. It is thus unnecessary to repeat their explanation.

By using an advertisement-mail-distributing system having the configuration described above, a rate of responses or the number of responses received from customers can be increased. In order to increase the rate of responses, the data mining server 5 selects most likely responding customers on the basis of customer information reported by the mail server 1. Then, an advertisement mail is sent only to selected customers. In this case, however, the total number of advertisement mails sent by the mail server 1 decreases. Thus, the total number of responses also decreases as well. In order to solve this problem, in the advertisement-mail-distributing system provided by the present invention, the mail server 1 composes a plurality of sentences for specific information and sends an advertisement mail to customers based on a customer analysis which is carried out by the data mining server 5 and capable of increasing the number of responses and, hence, the response rate.

The following description begins with an explanation of a test transmission for obtaining customer information to be analyzed by the data mining server 5 with reference to a flowchart shown in FIG. 6.

As shown in the figure, the flowchart begins with a step S1 at which the mail server 1 transmits customer information acquired from the customer-information database 2 to the data mining server 5 by way of the network 3.

Then, at the next step S2, the data mining server 5 randomly selects customers each to serve as a target of the test transmission on the basis of the customer information received from the mail server 1, and stores information on the selected customers in an analysis database 6. It should be noted that terminals used by customers each selected by the data mining server 5 to serve as a target of the test transmission are the personal computers 7-1 to 7-3. If it is not necessary to distinguish the personal computers 7-1 to 7-3 from each other in the following description, the personal computers 7-1 to 7-3 are denoted by a generic reference numeral of 7.

Subsequently, at the next step S3, the data mining server 5 transmits the information on the selected customers each selected in the processing carried out at the step S2 to serve as a target of the test transmission to the mail server 1 by way of the network 3. The information includes the mail addresses of the personal computers 7.

Then, at the next step S4, the communication unit 29 employed in the mail server 1 sends an advertisement mail to the customers each selected to serve as a target of the test transmission as indicated by the information received from the data mining server 5.

FIGS. 7A, 7B and 7C to 11A, 11B and 11C are diagrams each showing typical advertisement mails stored in the storage unit 28 employed in the mail server 1 to be sent in the test transmission. In data collection processing, 3 kinds of test transmission, namely, test transmissions 1, 2 and 3, are implemented. In test transmission 1, five different advertisement mails shown in FIGS. 7 and 8 are sent. In test transmission 2, five different advertisement mails shown in FIGS. 9 and 10 are sent. In test transmission 3, three different advertisement mails shown in FIGS. 11A to 11C are sent.

The five advertisement mails sent in test transmission 1 are each an advertisement mail describing renewal of a meeting room. To be more specific, an advertisement mail A1 shown in FIG. 7A has a title of `Peace of Mind.` An advertisement mail B1 shown in FIG. 7B has a title of `Excitements.` An advertisement mail C1 shown in FIG. 7C has a title of `Touching Hearts of Each Other.` An advertisement mail D1 shown in FIG. 8A has a title of `Stylish.` An advertisement mail E1 shown in FIG. 8B has a title of `Making Profits.` The advertisement mails A1, B1, C1, D1 and E1 are each a text written as a catch copy. Each of the advertisement mails includes a URL for making an access to the renewed meeting room. By merely clicking the URL, a customer inspecting the advertisement mail displayed on a personal computer 7 is capable of making an access to the renewed meeting room's web page, which is stored in the web server 4.

In test transmission 1, each of the five different advertisement mails is sent to 20,000 customers selected at random. Thus, the five different advertisement mails are sent to a total of 100,000 customers.

The five advertisement mails sent in test transmission 2 are each an advertisement mail regarding renewal of a web page describing movie (cinema) information. To be more specific, an advertisement mail A2 shown in FIG. 9A has a title of `Fashion.` An advertisement mail B2 shown in FIG. 9B has a title of `Real Things.` An advertisement mail C2 shown in FIG. 9C has a title of `Convenience.` An advertisement mail D2 shown in FIG. 10A has a title of `Peace of Mind.` An advertisement mail E2 shown in FIG. 10B has a title of `Excitements.` The advertisement mails A2, B2, C2, D2 and E2 are each a text written as a catch copy. Each of the advertisement mails includes a URL for making an access to the renewed web page describing movie information. By merely clicking the URL, a customer inspecting the advertisement mail displayed on a personal computer 7 is capable of making an access to the web page, which is stored in the web server 4.

In test transmission 2, each of the five different advertisement mails is sent to 15,000 customers selected at random. Thus, the five different advertisement mails are sent to a total of 75,000 customers.

The three advertisement mails sent in test transmission 3 are each an advertisement mail regarding renewal of a web page describing magazine information. To be more specific an advertisement mail A3 shown in FIG. 11A has a title of `Peace of Mind.` An advertisement mail B3 shown in FIG. 11B has a title of `Handle.` An advertisement mail C3 shown in FIG. 11C has a title of `Be Provided.` The advertisement mails A3, B3 and C3 are each a text written as a catch copy. Each of the advertisement mails includes a URL for making an access to the renewed web page describing movie information. By merely clicking the URL, a customer inspecting the advertisement mail displayed on a personal computer 7 is capable of making an access to the web page, which is stored in the web server 4.

In test transmission 3, each of the three different advertisement mails is sent to 20,000 customers selected at random. Thus, the three different advertisement mails are sent to a total of 60,000 customers.

Refer back to the flowchart shown in FIG. 6. At a step S5, the web server 4 stores information on a customer using a personal computer 7 from which a response to the sent advertisement mail has been received. The information is known as response information.

Then, at the next step S6, the web server 4 transmits the response information to the data mining server 5 by way of the network 3. The response information includes a registered ID of the customer and a mail address of the personal computer 7 used by the customer.

Subsequently, at the next step S7, the data mining server 5 identifies responding customers among all those serving as targets of the test transmission on the basis of the response information received from the web server 4. The data mining server 5 then analyzes information on each of the responding customers, that is, the profile of each of the responding customers. The customer-profile analysis carried out by the data mining server 5 will be described later.

FIGS. 12 to 14 are tables showing typical response results for test transmissions 1 to 3 respectively. Each of the tables shown in FIGS. 12 to 14 includes the number of customers each serving as a target of the advertisement-mail transmission, the number of customers serving as targets of the advertisement-mail transmission and making accesses to the web page and a response rate. The number of customers serving as targets of the advertisement-mail transmission and making accesses to the web page is referred to hereafter as the number of responses or a response count. The response rate is defined as a ratio of the response count to the number of customers each serving as a target of the advertisement-mail transmission.

As shown in FIG. 12, the response results for test transmission 1 indicate that the total number of customers each serving as a target of the advertisement-mail transmission is 102,563 and the number of responses is 1,024. Thus, the response rate is 1.00%.

As shown in FIG. 13, the response results for test transmission 2 indicate that the total number of customers each serving as a target of the advertisement-mail transmission is 76,644 and the number of responses is 660. Thus, the response rate is 0.86%.

As shown in FIG. 14, the response results for test transmission 3 indicate that the total number of customers each serving as a target of the advertisement-mail transmission is 61,517 and the number of responses is 389. Thus, the response rate is 0.63%.

The following description explains processing carried out by the data mining server 5 to analyze customer profiles. The data mining server 5 stores information on responses, which is received from the web server 4, and information on customers each selected as a target of a test transmission, with each of the responses associated with one of the selected customers, as learning data in the analysis database 6. Then, the data mining server 5 analyzes the profile of each customer giving a response by adopting a variety of data mining techniques for every catch copy. A customer giving a response is also referred to as a responding customer. Finally, the data mining server 5 determines which catch copy is a most suitable catch copy to be transmitted to customers not receiving the advertisement mails on the basis of analysis results in order to increase the response rate most. In the advertisement-mail-distributing system shown in FIG. 1, the customers not receiving the advertisement mails are represented by customers using the personal computers 8-1 to 8-3. If it is not necessary to distinguish the personal computers 8-1 to 8-3 from each other in the following description, the personal computers 8-1 to 8-3 are denoted by a generic reference numeral of 8.

The following description explains processing carried out by the data mining server 5 to compute learning parameters by referring to a flowchart shown in FIG. 15.

As shown in the figure, the flowchart begins with a step S21 at which the CPU 41 selects pieces of data at random from the customer database and uses the selected pieces of data as learning data. The CPU 41 stores the learning data in a learning database with response information used as a target variable serving as a dependent variable.

Then, at the next step S22, the CPU 41 creates an independent variable serving as a variable for predicting a characteristic of a customer. For example, the CPU 41 extracts a profile item to be used as an independent variable from items common to the learning database and the customer database. Then, the CPU 41 carries out deficiency processing to compensate for lost data, an abnormal value and the like. In addition, the CPU 41 carries out variable formation such as editing and syntheses on the profile item to create a final independent variable.

Subsequently, at the next step S23, the CPU 41 splits the learning database into rule discovery data and assessment data for assessment of a rule.

Then, at the next step S24, the CPU 41 selects a learning model and applies the model to the learning data.

FIGS. 16A to 16C are diagrams showing a typical learning model applied by the CPU 41 to learning data in the processing carried out at the step S24.

Learning model 1 shown in FIG. 16A is a learning model applied by the CPU 41 to learning data for each advertisement mail used in test transmission 1. As shown in the figure, this learning model comprises decision trees linked to each other in a cascade connection. In this learning model, the cascade connection of the decision trees comprises 2 stages for first and second decision trees respectively. The CPU 41 carries out a learning process by adopting 2 types of analysis method. In an analysis based on the first decision tree, a response rate for each advertisement mail is predicted. In an analysis based on the second decision tree, a response rate is again predicted with respect to customers who less likely respond to the advertisement mail. In this way, it is possible to improve the precision of prediction of a response rate predicted with respect to customers who less likely respond to the advertisement mail.

Learning model 2 shown in FIG. 16B is a learning model applied by the CPU 41 to learning data for each advertisement mail used in test transmission 2. As shown in the figure, this learning model comprises a decision tree and a neural network arranged in an ensemble form. The CPU 41 adds a predicted response rate obtained from an analytical technique based on the decision tree to a predicted response rate obtained from another analytical technique based on the neural network in accordance with a weighted expression with both weights set at 1. In this way, one of the analytical techniques can compensate the prediction precision provided by the other analytical technique for its deficiency and vice versa.

Learning model 3 shown in FIG. 16C is a learning model applied by the CPU 41 to learning data for each advertisement mail used in test transmission 3. As shown in the figure, this learning model comprises an additional contrivance as selection of variables at the beginning and an analytical method based on a neural network for the selected variables.

Refer back to the flowchart shown in FIG. 15. At a step S25, the CPU 41 applies learning models (learning parameters) to assessment data. From results of the application of the learning models to the assessment data, a most effective learning model (most effective learning parameters) is selected. It should be noted that, from the results of the application of the learning models to the assessment data, an assessment chart of the assessment data can be created.

Then, at the next step S26, the CPU 41 determines and stores learning parameters based on results of assessment in the processing carried out at the step S25. The pieces of processing described above are carried out on all advertisement mails.

With reference to a flowchart shown in FIG. 17, the following description explains processing carried out by the data mining server 5 to compute an expected response probability by using learning parameters selected in the processing represented by the flowchart shown in FIG. 15. The data mining server 5 applies learning parameters extracted from a learning database to a customer database of customers each used as a target of the transmission. Then, the data mining server 5 computes an expected response probability for each customer and for every advertisement mail in case the mail is transmitted to the customer. Finally, in order to increase a response rate, the data mining server 5 determines which advertisement mail is to be transmitted to give a most effective result on the basis of the computed response probabilities.

As shown in FIG. 17, the flowchart begins with a step S41 at which the CPU 41 carries out name collect processing on demographic information. To be more specific, names of customers are put in order so that each customer is not treated as if the same customer were different customers and, if necessary, the customers are grouped into families so that different customers of a family can be put in the same group for the family.

Then, at the next step S42, the CPU 41 creates an independent variable in the same way as the processing carried out on the learning database. To put it in detail, the CPU 41 extracts a profile item from the customer database. Then, the CPU 41 carries out deficiency processing and variable formation to create the same independent variable as the independent variable created for the learning database.

Subsequently, at the next step S43, the CPU 41 applies the learning parameters saved in the processing carried out at the step S26 of the flowchart shown in FIG. 15 to the customer database.

Then, at the next step S44, the CPU 41 computes an expected response probability for every customer and for each advertisement mail in case the mail is transmitted to the custom