JPWO2005064511A1 - Campaign dynamic optimization system and method thereof, recording medium recording the method, and transmission medium transmitting the method - Google Patents

Abstract

Applying one or more types of analysis engines from the Bayesian network or MTS to the individual consumer data stored in the first storage means and the second storage means storing the media plan, Quantifying the dependence of the drafting plan, extracting the drafting plan that is estimated to be effective for the response based on the quantified dependence, and storing it in the first storage means according to the drafting plan determined to be valid Campaign dynamic optimization system and method with means for updating a published media plan. Selection diagram: Fig. 1

Description

  The present invention relates to a system and a method for dynamically optimizing a campaign, a recording medium recording the method, and a transmission medium for transmitting the method. Specifically, it is intended to optimize the advertising campaign by dynamically modifying the developing advertising campaign based on information obtained from available tracking tools.

  Today's advertising campaigns are highly customized, and planning them requires the instinct of the agency representatives and a great deal of manual work. It takes time and effort.

  To date, various advertising space selection support tools have already been proposed in order to reduce the work burden on the person in charge at the advertising agency. For example, a system for selecting an advertising space across a plurality of advertising media, based on contact probability data between consumers and each advertising media, within the range of acquisition cost for each advertising media type There has been proposed an advertisement space selection method and its system in which the advertisement space is sequentially selected and the value obtained by dividing the total contact probability by the acquisition cost of the advertisement space is maximized. (See Patent Document 1.)

  The advertising space selection method exemplified in Patent Document 1 is based on an empirical rule that if the number of times the advertisement reaches the consumer increases, it can be expected to indirectly improve the corresponding advertising effect for a specific brand. This system is effective as an advertising campaign planning support tool, and greatly reduces the workload of the person in charge at the site.

  However, even if an advertisement reaches a general consumer such as a viewer as expected, whether or not it is directly linked to a purchasing activity with a predetermined probability has not been subjected to a detailed verification. In addition, the advertisement campaign is usually premised on the progress according to the initial plan, and even if there is a review after the lapse of a predetermined period, a short-term or real-time correction is hardly assumed.

  On the other hand, for example, as disclosed in Patent Document 2, a probability tree in which a default is generated in a certain company is created based on the financial data of each company so far and Bayes' theorem is used based on the decision tree. A method for determining the probability of occurrence of an automobile accident, the probability of occurrence of a fire accident, and the probability of occurrence of an earthquake by a similar technique is disclosed. Further, Patent Document 3 discloses a method for realizing an automated diagnosis system for a printer fault inquiry from a customer using a method of probability inference using a directed acyclic graph representing a causal relationship between variables called a Bayesian network. ing.

  Non-Patent Document 1 explains the concept of Mahalanobis space and Mahalanobis distance in the Mahalanobis-Taguchi-System (hereinafter referred to as “MT system”), and 17 items regarding age, sex, and each inspection item are variables. As a result, the MT system is applied to the healthy group and the non-healthy group, and the accuracy of medical diagnosis is drastically compared between the case where all variables are used and the case where only the variables having a positive S / N ratio are used. Applied to the field of fire alarm systems with smoke detectors, and analyzed cases in the process of increasing Mahalanobis distance over time when cooking and in the event of a fire. It is disclosed that the MT system is effective in determining anomaly.

JP 2002-24525 A JP 2000-259719 A JP 2001-117776 A "The Indian Journal of Statistics 2000", Vol. 62, Series B, Pt. 2, pp. 233-248

  However, the level of demand from advertisers has become more of an interest in planning that allows them to expect purchase results with certain certainty without being bored with the indirect advertising effects that can be enjoyed by the above-mentioned arrival priority type advertising space selection method. Transitioning. For this reason, a system that can accurately analyze the propriety of a planned campaign plan and can correct a campaign under development in real time is expected.

  However, as described in Patent Document 1 described above, in the field of advertising, up to now, emphasis has been placed on the development of a campaign planning support tool, and the subsequent correction processing has been largely relied upon by human beings and has not been developed as a system.

  The main reason for this is that advertising campaigns are highly customized, unsuitable for machine processing, and often have to rely on analyst experiences, and to understand viewer trends. In contrast to advances in tracking technology and processing speed, the analysis method for the results of tracking is mainly due to the fact that cross tabulation is the mainstream based on a model that assumes the independence of variables. Like.

  On the other hand, consumers' purchasing behavior is generally considered to be affected by a number of factors that are intricately related to each other, and the manner of the impact is changing from time to time. The cross tabulation method based on modeling that assumed independence was limited in principle.

  On the other hand, as a data analysis technique in an environment that is dominated by factors that affect each other in a complicated manner, as in Patent Document 2, Patent Document 3, and Non-Patent Document 1, statistical inference such as a Bayesian network or an MT system is used. Although the method is known, it has been limited to limited use such as calculation of the probability of occurrence of bankruptcy, accidents, etc., failure diagnosis and medical diagnosis, and there has been no report on its application in the fields of advertising and publicity.

  In addition, the usage pattern so far is merely used as an analysis tool for evaluating the probability of occurrence of anomalies and determining the anomaly of individual situations, and determinism such as modification, planning, and execution of specific measures. Almost no consideration is given to the practical utilization method.

The present invention has been made to solve such technical problems, and is specified by the following technical matters.
The present invention (1)
At least one type of consumer attribute information and at least one type of communication history information including any one of communication condition and response result information are stored in units of consumers or transactions each consisting of a plurality of fields. A consumer database that stores records so that they can be searched and updated,
First storage means for storing search condition information for each communication driver so as to be searchable and updatable;
The records stored in the consumer database are sequentially read out, and the data of the consumer attribute information and the communication history information are respectively normalized, and the standardized data is made the record as a row and the field as a column. A second storage means for storing each data in a searchable manner by assigning them to a matrix-like storage area and specifying the row and / or column;
The dependence on the combination of information of each field with respect to the response result is quantified with respect to the data stored in the second storage means by using one or a plurality of analysis engines of a Bayesian network or an MT system. Dependency quantification means,
An effective field pattern extracting means for extracting, as an effective field pattern, a combination of field information determined to be effective for the response result from the dependency;
A campaign comprising: posting condition information updating means for updating the posting condition information for each communication driver stored in the first storage means within the operable range so as to match the information of each field of the valid field pattern It is a dynamic optimization system.
The present invention (2)
An external database in which external information other than the information recorded in the consumer database is recorded based on at least one consumer attribute information in the consumer database.
Searching the external database using at least one consumer attribute information and / or communication history information common to the consumer database and the external database as searchers, and deriving standard external information for each searcher, The campaign dynamic optimization according to the present invention (1) further comprising information linking means for adding a new field to the record of the consumer database corresponding to the search element and additionally recording the derived standard external information. System.
The present invention (3)
The campaign dynamic optimization system according to any one of the present invention (1) or (2), wherein the consumer database is updated in real time or periodically.
The present invention (4)
Based on at least one consumer attribute information of the consumer database, mass media information database storing mass media information related to contact amount so as to be searchable;
Searching the mass media information database using at least one kind of consumer attribute information common to the consumer database and the mass media information database as a search element, deriving standard mass media information for each search element, Further comprising mass media information connecting means for providing a new field in the record of the consumer database corresponding to the searcher and additionally recording the derived standard external information.
Any one of the present inventions (1) to (3), wherein the second storage means sequentially reads records stored in the consumer database after the mass media information is connected. Is a campaign dynamic optimization system.
The present invention (5)
The mass media information database is a data update in real time or periodically, and is the campaign dynamic optimization system according to the present invention (4).
The present invention (6)
The dependency quantification means includes:
In the consumer database, the field corresponding to the communication history information related to the response result is set as a node related to the objective variable, and the field corresponding to the other consumer attribute information or the communication history information is set as a node related to the explanatory variable. A structure learning means for modeling a route from a node related to an explanatory variable to a node related to the target variable by a directed link so that there is no circulation;
Using the data recorded in the field corresponding to each node in the consumer database, a prior distribution for each specific value that can be taken by the data recorded in the field is calculated, and a conditional probability table ( CPT), and a conditional probability table setting means for setting to the node,
A communication chain extracting means for extracting all sets of specific values that can be taken by the nodes in the structure as a communication chain;
Posterior probability calculating means for calculating posterior probabilities for response results based on the prior distribution of each node and the conditional probability table (CPT) for each communication chain;
The campaign dynamic suitability according to any one of the present inventions (1) to (5), characterized in that the posterior probability value is a Bayesian network analysis engine having the dependency on the set of specific values. System.
The present invention (7)
The effective field pattern extraction means includes
On the premise of the structure modeled by the structure learning means, the simultaneous probabilities for each communication chain are compared, and a communication chain with a high rank in which the joint probability is equal to or greater than a predetermined value or the magnitude of the joint probability is the response result. The campaign dynamic optimization system according to the present invention (6) is characterized in that the effective field pattern is highly correlated with the communication history information relating to the above.
The present invention (8)
A communication contact database that stores searchable and updateable information about the cost of advertising for each communication driver and production cost information for promotional materials,
A transaction condition database that stores, in a retrievable manner, searchable / updateable consideration information of the product or service to be advertised, and consideration information of the related product or service derived from the transaction to be advertised;
The effective field pattern extraction means includes
Based on the transaction cost that can be searched from the communication contact database and the transaction price that can be expected when the response result is searchable from the transaction condition database. A field pattern that includes an investment consideration effect calculation unit that calculates an effect in consideration of an investment, and the investment consideration effect calculated by the investment consideration effect calculation unit is higher than a predetermined value or a higher communication chain is correlated with the response result The campaign dynamic optimization system according to the present invention (6) is characterized by being extracted as
The present invention (9)
Communication contact database that stores searchable and updatable contact information for each communication driver, information on cost of production of promotional materials, and contact information that can be communicated to consumers or transactions related to predetermined consumer attributes or communication history information Further including
The placement condition information update means includes:
Based on the effective field pattern, a probable customer detection means for searching a consumer or transaction corresponding to a record that matches the effective field pattern from a consumer database and identifying the consumer or transaction related to the record as a promising customer ,
A communication driver selection means for searching the communication contact database for a communication driver capable of directly communicating with the prospective customer;
The advertisement condition information relating to the selected communication driver in the first storage means is updated so as to execute a push-type advertisement at an appropriate timing for the prospective customer. 7. The campaign dynamic optimization system according to 7.
The present invention (10)
10. The campaign dynamic optimization system according to claim 8, wherein the communication contact database is updated in real time or periodically.
The present invention (11)
The dependence quantification means comprises:
A reference data group extracting means for extracting the record that has not resulted in a response from the consumer database and setting it as a reference data group;
While calculating the characteristic amount (m: average value of each feature amount, σ: standard deviation of each feature amount) of information of each field related to the reference data group, between the information of each field normalized by the characteristic amount Statistical processing means for calculating a correlation coefficient (r _ij ) and deriving a correlation matrix (R),
Pattern variation deriving means for deriving all combinations (y _l ) of values that each field can take as pattern variations;
Mahalanobis distance calculating means for calculating the Mahalanobis distance (D _l ² ) with respect to the reference data group using the following formula for each pattern variation (y _l ) derived by the pattern variation deriving means:
An MT system analysis engine comprising:
The campaign dynamic suitability according to any one of the present inventions (1) to (5), wherein each Mahalanobis distance ( _Dl ² ) is the dependency on the set of specific values. System.
The present invention (12)
The effective field pattern extraction means includes
A predetermined number of patterns in which the dependency (ie, Mahalanobis distance) value evaluated by the dependency evaluation means is equal to or higher than a predetermined value or higher in the ranking thereof is correlated with the communication history information related to the response result. The campaign dynamic optimization system according to the present invention (11) is characterized in that a high effective field pattern is obtained.
The present invention (13)
The placement condition information update means includes:
When updating the data in the consumer database, based on the updated data, update the data stored in the second storage means,
Based on the updated storage of the second storage means, the dependency quantification means and the effective field pattern extraction means are executed,
The campaign dynamic optimization system according to the present invention (11), wherein the placement condition information stored in the first storage means is updated to information corresponding to each newly extracted effective field pattern. It is.
The present invention (14)
For each communication driver, further includes a communication contact database for storing searchable / updateable contact information for a consumer or transaction related to predetermined consumer attributes or communication history information,
The placement condition information update means includes:
Based on the valid field pattern, a consumer or transaction of a record related to the valid field pattern is searched from the consumer database, and the consumer or transaction is extracted as a prospective customer.
Communication driver selection means for selecting a communication driver capable of communicating with the promising customer from the communication contact database;
The placement condition information stored in the storage means in the first storage means related to the selected communication driver is updated so as to be communicable at an appropriate timing with respect to the promising customer. Item 13. The campaign dynamic optimization system according to Item 12.
The present invention (15)
The communication contact database is a campaign dynamic optimization system according to the present invention (14), which is updated in real time or periodically.

  Moreover, this invention (16)-(30) is invention of the method corresponded to this invention (1)-(15), respectively. Furthermore, the present invention (31) is a recording medium on which the campaign dynamic optimization method of any one of the present inventions (16) to (30) is recorded so as to be readable by a computer. The present invention (32) is a transmission medium for transmitting the campaign dynamic optimization method of any one of the present inventions (16) to (30) in an executable manner by a computer.

Here, the “resident” is a collective term of individuals who are subject to advertising activities and who perform purchasing activities, and include “viewers” on TV and radio, “readers” in newspapers and magazines, and the like.
“Media” is a general term for mediating information transmission when introducing information about products and services to consumers. The present invention is not limited to so-called mass media such as TV, radio, and newspaper.
“Vehicle” is a general term for individual brands in the same type of media that need to be distinguished by media brand differences. (For example, A station / a channel, B station / b channel on TV, C paper, D paper, etc. in newspaper)
“Creative” refers to an advertising work itself that is mediated by media or vehicles, and is a term used when it is desired to distinguish between advertising works.

“Communication” is a general term for an act of introducing a company brand to consumers, such as advertisement, advertisement, and information provision. It is not limited to so-called “mass media”, but also includes individually promoting advertising to specific consumers, such as sending e-mails to addressable people.
“Communication driver” is a general term for mechanisms that control the placement of advertisements on media that can introduce their brands to consumers. For example, when e-mail is used for advertising activities, the mail server corresponds to a communication driver. In addition, mass media that are put out through many human hands such as TV and radio are work management tables for managing the advertisement placement schedule. It is assumed that the placement instruction is given according to the table schedule.

“Resident attribute information” is information relating to the characteristics or characteristics of a consumer or transaction such as gender, age, occupation, and address of a target consumer.
“Communication history information” is information relating to a history of actions such as viewing, reading, and purchasing performed by consumers.

The “record” is an information classification for collectively handling a set of information such as consumer attribute information and communication history information on the system, and a consumer or a transaction is a unit.
“Field” refers to an area allocated for specific information stored in a record, and one record includes a plurality of fields.
The “effective field pattern” is a data string of standardized consumer attribute information and communication history information that is effective for the response, and this effective field pattern is associated with the corresponding communication driver placement condition information. In comparison, each admission condition is judged as to suitability.
“Dependency” refers to the mutual dependency relationship between consumer attribute information and communication history information. In an embodiment using a Bayesian network, it is expressed with probability, whereas in an embodiment using an MT system, the Mahalanobis distance is used. Is displayed.

“Normalization” is a process for preparing data suitable for the selected analysis engine. Specifically, in a system using a Bayesian network, 0, 1 or a finite number of categories, MT In a system using the system, 0 and 1 or 0 to 1 are consecutive numbers.
“Daily database” refers to individual consumer attribute information and information about viewer behavior history including viewing, reading, etc. for each consumer, and each or more of consumer, attribute information, and history information. It is a database that can be searched and tabulated using the combination of A group of the consumer attribute information and the action history information with the consumer as a unit is set as one record, and is composed of fields for storing each consumer attribute information and the action history information. The data is updated at any time or periodically.
The “mass media information database” is the total amount of contact with mass media information of sample consumers collected through the tracking tool, and the standard amount of contact with mass media information for each consumer attribute and / or contact opportunity. It is a database that can be calculated. Updated from time to time through the tracking tool.
“Communication contact database” stores communication unit prices for communication contacts for various consumers, restriction conditions including vacancy information in the submission, etc., and extracts contact points and placement methods that can be used under specified conditions. It is a database that can calculate the ad unit price at that time.

1 is a schematic diagram of a system configuration according to the present invention. It is the schematic of this processing flow of this invention. It is explanatory drawing of the data connection process and normalization process in this invention. It is a figure which shows an example of the memory content stored in the consumer database and communication contact database of this invention. It is a figure which shows an example of the memory content of the external database of this invention. It is the figure (the 1) which shows typically the process of integrating each data in the consumer database of this invention. It is FIG. (2) which shows typically the process of integrating each data in the consumer database of this invention. It is a figure which shows typically the process of connecting the external data of this invention to a consumer database. It is a figure which shows typically the process which connects the multimedia database of this invention with a consumer database. It is explanatory drawing about the arrangement | sequence of the data in the consumer database of this invention. It is explanatory drawing about the normalization system in the case of utilizing the Bayesian network of this invention. It is a figure which shows an example of the arrangement | sequence of the data in the 2nd memory | storage means in the case of utilizing the Bayesian network of this invention. It is explanatory drawing about the normalization system in the case of utilizing MT system of this invention. It is a figure which shows an example of the arrangement | sequence of the data in the 2nd memory | storage means in the case of utilizing MT system of this invention. It is a figure which shows an example of the memory content of the consumer database of this invention. It is a figure which shows the outline | summary of the Bayesian network by which the structure learning was carried out with respect to the data of FIG. 15 of this invention. It is a figure which shows an example of the restriction | limiting conditions in the connection of the link of this invention. It is a figure which shows an example of the probability distribution in each node of FIG. 16 of this invention. It is FIG. (1) which shows an example of the conditional probability table | surface in each node of FIG. 16 of this invention. It is FIG. (2) which shows an example of the conditional probability table | surface in each node of FIG. 17 of this invention. It is FIG. (3) which shows an example of the conditional probability table | surface in each node of FIG. 17 of this invention. It is a figure which shows an example of the communication chain concerning this invention. It is a figure which shows an example of the response probability ranking according to the field pattern concerning this invention. It is a figure which shows an example of the expected value ranking according to the field pattern concerning this invention. It is a figure which shows an example of the memory content of the communication contact database of this invention. It is a figure which shows an example of the return on investment (ROI) ranking according to the field pattern concerning this invention. It is a figure which shows an example of the memory content of the consumer database of this invention. It is a figure which shows an example of the ranking of Mahalanobis distance according to the field pattern concerning this invention. It is explanatory drawing about the update process of the drafting plan in the 1st memory | storage means at the time of employ | adopting MT system of this invention. It is explanatory drawing about the process of a push type | mold advertisement using the addressable medium concerning this invention.

  FIG. 1 shows an outline of the entire system configuration including a campaign dynamic optimization system according to the present invention. As the main part 1 of the system of the present invention, the central processing unit CPU is connected to the consumer database DB1 and the communication contact database DB2 via the external bus line B2.

  The central processing unit CPU is also connected to the first storage means and the second storage means via the internal bus line B1. On the other hand, as a dedicated processing unit, a standardization processing engine E1, an information connection engine E2, various dependency evaluation analysis engines E3, an effective field pattern extraction unit U1, and a placement condition update processing unit U2 are included.

  On the other hand, the central processing unit CPU is connected to the mass media information database DB3 and other external information database DB4 through the router R and the local area network LAN, and submits data in accordance with the posting conditions stored in the second storage means. It is also linked to the submission management server Server that issues the submission instruction.

Outside the premises, the clients CL1 to CLm and the communication terminals Tcl of the media M-1 to M-n are communicably connected through a network represented by the Internet.
Each client CL1 to CLm can contact each individual P1 to Pj by using various communication means such as telephone and direct mail as well as e-mail through the network. The media managers M-1 to M-n are assumed to be in an environment where information can be transmitted to consumers through respective information media such as newspapers and TVs.

(Step 1)
Next, FIG. 2 shows an outline of a flow according to the processing of the present invention. First, as a step 1, records in units of consumers or transactions are sequentially read from the consumer database DB1, and a standardization process is performed according to the analysis engine to be adopted.

  Prior to the processing of step 1, information in an external database such as mass media information such as how much a consumer is in contact with a brand advertisement at a predetermined time in order to reflect indirect factors such as mass media information in the analysis. It is desirable to perform a preliminary process for improving the analysis accuracy by linking to each record file of the consumer database.

  FIG. 3 shows an outline of the processing of the external data connection engine. Since the external information database is stored so that one or more of the attribute information or history information of the consumer database can be searched as a searcher, information common to the external information database and the consumer database The external database is searched based on the information, and the standard behavior pattern information of the consumer corresponding to the information (for example, information such as contact rate when the external database is a mass media information database) is derived. Then, a new item string is provided for each record in the consumer database and the corresponding specific value is assigned.

  Therefore, a common specific value is assigned as shown in FIG. 3 to a consumer or transaction having the same attribute information or history information used as a searcher. External information is not limited to information that is directly related to advertising / promotion such as mass media information, and may be, for example, weather and temperature on the date when a response is obtained. Any information stored in a database that can be searched using information common to the column information of the database may be used.

  The normalization process in step 1 in FIG. 2 is started for the consumer database that has been subjected to the preliminary process for connecting the external information. That is, in this step, the value is normalized to a value (typically 0 or 1) that can be used by the analysis engine. Note that the value that can be handled is not limited to 0 or 1, and a standardized value is also selected according to the analysis engine to be used. For example, when an analysis engine using a Bayesian network is adopted, the value may be a value that can be distinguished into a finite number of categories. On the other hand, when an analysis engine using an MT system is used, the number is continuous. Even available. The data after the normalization process is stored in the second storage means.

(Step 2)
Next, as step 2 in FIG. 2, the standardized data stored in the second storage means completed in step 1 is sequentially read and introduced into the analysis engine E3 to be used. Available analysis engines are Bayesian networks or MT systems.

  And this analysis engine E3 derives the dependence between the combination of the information of each column in consumer database DB1, and the information of the column of the response result made into object. For example, when an analysis engine using a Bayesian network is employed, the dependency between the combination of information in each column and the information in the response result column is displayed with probability. As a structure learning algorithm for constructing a Bayesian network, a known algorithm such as a “PC algorithm” can be used. Details will be described in Examples.

  On the other hand, also when the analysis engine using the MT system is employed, the dependency between the combination of information on each column in the consumer database DM1 and the information on the column of the response result is derived. However, in the analysis engine using the MT system, the dependency is displayed with the Mahalanobis distance.

(Step 3)
Next, as step 3, based on the dependency derived in step 2, a combination of information in each column that is expected to be likely to reach the expected response, that is, a specific value of each field in the second storage means A pattern constituted by is derived.

  Here, when an analysis engine that uses a Bayesian network is employed, a probability of obtaining an expected response result or a combination of information in each column that has a large investment consideration effect is extracted. On the other hand, when using the MT system, a combination of information of each column having a large Mahalanobis distance is extracted. The combination of information of each column obtained in this way (that is, the arrangement of the data in the record) is called an “effective field pattern”.

(Step 4)
Further, in step 4, only the column information that can be controlled as a part of the drafting plan in the effective field pattern is focused on and collated with the drafting plan stored in advance in the first storage means. If the current drafting plan stored in the first storage means and the derived valid field pattern do not match, match the valid field pattern within the range that can be operated under the limited conditions such as the drafting budget. Update process is executed.

  Note that it is not always necessary to modify the current drafting plan itself, including the case where the current drafting plan remains as it is and added to a drafting plan such as advertising through other media such as e-mail or direct mail.

(Step 5)
Each communication driver inquires the stored contents of the first storage means modified in the above-mentioned step 4 in real time or periodically, and corrects the subsequent advertising campaign according to the stored contents, or transmits the instruction for the submission. To do. The steps on the media driver side are not necessarily integrated with the system of the present invention.

Specifically, a placement management server (Server) is assumed, and the placement management server directly sends an e-mail for advertising to a target consumer, or the content and placement of advertisements displayed through each site. Condition change control. Further, like the clients (CL1 to CLm) in FIG. 1, the communication driver of the present invention is not limited to the one that performs direct placement control like the Internet banner advertisement management server, and manages each client or media. Including indirect placement control such as presenting a revised plan in response to inquiries about advertisements and advertisements from the user terminal, and is uniformly managed in a form that does not involve personal factors. Any form is acceptable.
It should be noted that all prior art documents cited in the present specification are incorporated herein by reference.

  Next, the details of the processing in each step will be described as a case where the present system is applied to specific data. In a system suitable for practical operation, as illustrated in FIG. 4, each database stores data independently in accordance with each input form or tracking form, so step 1 in FIG. Prior to standardization processing, it is desirable to perform necessary preliminary processing such as data connection.

  First, the storage status of individual data in each database in FIG. 4 will be described. In the consumer database, as data related to consumer attribute information, information about the area of each consumer's residence (eg, “Tokyo”, “Hokkaido”,..., Etc.), gender (“male” and “ Information about “woman”), information about age, and the like are stored based on the consumer ID. In addition, as consumer attribute information, it is not restricted to these information, The method of classification is not restricted to this.

  In addition, communication history information is also stored in the consumer database. Here, “TV DB”, “News DB”, “Banner DB”, “Web DB”, etc. are used for each media unit. The system which stores separately based on consumer ID is illustrated. From the viewpoint of the current method of tracking the consumer's communication history, it is desirable to store each medium as a unit in addition to the attribute information as shown in FIG. 4 and recalculate each consumer ID as a preliminary process.

  For example, as shown in FIG. 4, the consumer database sequentially stores information about how each consumer has contacted an advertisement such as CF through which medium and how and when the consumer responded. Time, vehicle, program name, contacted creative, offer, response method, response content, as well as TV, the number of seconds of advertising space, newspaper, the placement side, number of steps, banner It is desirable to store information such as a location, a space, and a page in a web as far as possible.

  On the other hand, as shown in FIG. 5, the mass media information database stores information about a large number of consumers about contact with advertisements such as CF. Among the consumer attribute information and communication history information of the consumer database, It is possible to tabulate the average number of contacts with CF and the probability of contact with advertisements for consumers whose conditions match for information items such as contact time.

  The number of times of contact and the probability of contact are derived in accordance with the minimum category of consumers in the consumer database. In addition, since the contact timing is also taken into consideration here, the number of times of contact varies depending on the timing of contact during the campaign period, even for records with the same consumer attribute.

  Furthermore, in the present invention, if it is external data having items in common with the consumer database, it is possible to analyze the data in consideration of the data. Here, as shown in DB4 of FIG. 5, when there is a possibility that the weather will affect the response, an example is given of the case where weather data is procured from outside data in addition to mass media information and analyzed. . Specifically, the weather (sunny, cloudy, rain, ...) in that area on the day of contact was used from an external database.

  The communication contact database DB2 uses the contact point based on the conditions such as the area, date, time when the user contacts the consumer, vehicle, program, display surface or location, number of seconds, number of steps or space. For example, information such as unit price can be determined.

  The first storage means M1 stores management information such as what kind of advertisements, such as what media, vehicle, etc., and when only the program name and time when contacted are known. Even if it exists, it is desirable to be configured so that detailed information such as the number of seconds and the creative can be supplemented by referring to the information in the first storage means with the information.

  Based on the above accumulated data, in order to make a data group suitable for use in each analysis engine, the following steps are taken for processing. First, since there is a possibility that the same consumer communicates a plurality of times in the consumer database, these data are collected and totaled for each consumer ID. FIG. 6 schematically shows this process.

  In the example of FIG. 6, since the consumer with ID 00001 has accessed the web twice during the tracking period and has two histories, these histories are rearranged for each ID.

  Similarly, since attribute information and a plurality of communication history information are separately counted, FIG. 7 illustrates a process of grouping for each consumer ID. In short, each of the prime tables is sequentially connected in the horizontal direction using the ID as a searcher.

  Next, FIG. 8 schematically shows the process of linking information in the external database with specific values. In this case, the items common to the external information database DB4 and the consumer database DM1 are “area” and “year / month / day” information. Calculate weather information and add weather information to each response history. That is, on the system, a field for storing the weather information is added for each response type.

  Similarly, FIG. 9 is a diagram schematically showing a process of further connecting the data in the mass media information database. In this case, the items common to the mass media information database and the consumer database are “Area”, “Date”, “Time”, “Gender”, and “Age”. As described above, the number of times of contact with the TVCF until the average response time of the group corresponding to each condition and contact probability data as the possibility of contact with the magazine advertisement are added.

  Here, external information such as mass media information is expected to affect a wide range of similar consumer groups and constitute a part of the response. Is added. That is, for example, it is assumed that the same consumer requests a material for product A by watching TV, and purchases product A by watching TV again on another day. Since the number of contact times of the TVCF at the time of purchase is naturally different, a contact field is added separately for each response.

  Through the above processing, the information in the consumer database is expanded, and the data group that is the basis for the analysis engine is completed. An example of the data arrangement in the file with the consumer ID at that time as a unit is shown in FIG. In this example, “information about consumer attributes”, “history information about TVCF”, “history information about newspaper advertisements”, “history information about banner advertisements”, “history information when requesting product materials”, “ They are connected and arranged in the order of “history information”. In addition, it does not restrict to this order, and the information connected is not restricted to these.

  Even at this time, the information stored in the consumer database is a specific value such as the name of the prefecture or the actual age, so it cannot be introduced into the analysis engine as it is. Therefore, step 1 in FIG. 2 is performed, and a normalization process is performed for conversion into a data format suitable for the analysis engine to be used.

  For example, as described above, the normalization process in Step 1 of FIG. 2 is different between the case where an analysis engine using a Bayesian network is used and the case where an analysis engine using an MT system is used.

  First, the case where an analysis engine using a Bayesian network is used will be described. Since each node in the Bayesian network needs to have at least a finite number of events, each column of the consumer database needs to be expanded into a finite number.

  That is, as illustrated in FIG. 11, the column relating to “gender” may be 1 for males and 0 for females, and therefore, there is no need to divide the gender column. On the other hand, for a column in which a continuous number such as “age” is stored, it is necessary to introduce a convenient division and expand it into a finite number of divisions. For example, processing such as “up to 19 years old”, “20 to 35 years old”, “36 to 50 years old”, and “51 years old or older” is required.

  Similarly, “Date” is divided into four seasons and “early”, “mid” and “late” of the month. Time was also divided into “morning”, “daytime”, and “night”. Furthermore, “average number of contacts” and “average contact probability” were classified into “many”, “medium”, and “small”. However, it is not limited to this division.

  As shown in FIG. 11, all information for each consumer ID in the consumer database is converted to “1” or “0”, and the converted data is stored in the first storage means in step 1. It is. It should be noted that the sequence of numerical values “1” or “0” for each consumer ID is referred to as a “field pattern”. FIG. 12 shows an example of the contents stored in the second storage means when an analysis engine that uses a Bayesian network is applied.

  For simplicity of explanation, “1” or “0” is used, but it is not always necessary to binarize “1” or “0” in order to suppress the explosion of the calculation amount in the analysis engine. If the number is finite, data can be compressed by multi-leveling. Furthermore, a clustering process in which several related items are collected and processed as one node is also effective.

  On the other hand, in the case of using an analysis engine that uses the MT system, a normalization process that is partially similar to the case of using a Bayesian network but slightly different is executed. FIG. 13 schematically shows the process.

  In the analysis engine using the MT system, even if it is a continuous number, it can be applied as it is. For example, “age”, “year”, “month”, “day”, “hour”, “second” For data that is originally a numerical value such as “number”, “average number of times of contact”, “average contact probability”, the numerical values can be applied as they are. This is a point that is greatly different from the standardization processing when an analysis engine using a Bayesian network is applied.

  The actual data thus stored in the consumer database is subjected to normalization processing and stored in the second storage means in the form of “0” or “1” or a continuous number. An example of the stored contents of the second storage means after storage is shown in FIG. The arrangement of data in the same row is data relating to the same consumer, and the data in the same column is data relating to the same attribute information or history information.

  Next, the processing step for obtaining the dependency between the combination of information in step 3 in FIG. 2 and the response result will be described using the record group illustrated in FIG. That is, as explanatory variables, whether it is “end of month”, “weekdays”, whether it is other than holidays, whether it is “α newspaper”, “β newspaper”, “B product” “Advertisement”, whether the size of the printed paper is “15 steps”, “TVCF contact amount is small”, “A product advertisement”, “Newspaper advertisement contact amount is 10 items, “less” or not, while the objective variable is the presence or absence of “insurance contract” related to the response result.

Processing when the Bayesian network analysis engine is applied to each file in the second storage means shown in FIG. 15 will be described. A Bayesian network is a probability network composed of nodes representing variables and directed paths representing the causal relationship between the nodes. The probability network can be expressed by the following joint probability distribution.
Here, FIG. 16 shows a Bayesian network in which communication chains are constructed by applying the structure learning means.

In structural learning that determines the graph structure of Bayesian networks from data, the well-known PC algorithm (Spirtes, P., C. Glymour, and R. Scheines: Causation, Prediction and Search, No. 81 in Lecture Notes in Statistics. New York: Springer-Verlag (1993) P116-123). The PC algorithm is an algorithm for constructing the structure of a Bayesian network from data based on a statistical test for dependency. However, the algorithm used for determining the graph structure is not particularly limited to the PC algorithm. Here, the Bayesian network in FIG. 16 is described by the following equation.
At this time, x ₁ : end of month, x ₂ : other than holidays, x ₃ : A product advertisement, x ₄ : number of steps, x ₅ : weekdays, x ₆ : B product advertisement, x ₇ : newspaper, x ₈ : newspaper advertisement contact Amount, x ₉ : TVCF contact amount, x ₁₀ : corresponds to insurance contract.

  Here, prior to learning the graph structure using the PC algorithm, the limiting conditions between the variables shown in FIG. 17 were specified. This is because it is possible to introduce known knowledge about the causal relationship between variables in the advertising field, and a realistic graph structure can be obtained. In this case, it is known that the presence or absence of an insurance contract does not affect the date regulation items (month-end, non-holiday, weekdays), newspaper and TVCF ad contact volume, as well as the printed paper, number of columns, and advertising product name. Therefore, it is forbidden to connect links in this direction.

  On the other hand, it is known that the printed paper, the number of columns, the name of the advertisement product, and the contact amount of the newspaper or TVCF do not affect bidirectionally. The solid line with x in FIG. 17 is a bidirectional link between these nodes. Is not connected.

  When learning CPT in a Bayesian network after the graph structure is determined, if the dataset used for learning exists for all combinations of random variables, it is possible to learn all terms of CPT. If the graph structure does not include all possible combinations of graph structures or includes variables that are not observed, it is necessary to estimate the probability distribution for unobserved data. This time, a known EM algorithm (AP Dempster, NM Laird, and Rubin DB: Maximum likelihood from incomplete data via the EM algorithm, Journal of the Royal Statistical Society B, vol. 39, p. 1-38 (1977)). However, the algorithm used for learning the CPT after the graph structure is determined is not particularly limited to the EM algorithm.

  Here, the probability distribution for each node learned as a result of the EM algorithm is as shown in FIG. Further, conditional probability tables for each learned node are shown in FIGS. For example, as shown in FIG. 16, in the “published paper” node, the structure learning is performed so that the “end of month” node, the “weekday” node, and the “stage number” node are involved as parent nodes. Is a combination of “α Newspaper”, “β Newspaper”, “Weekend”, “Other than End of Month”, “Saturday and Sunday”, “Weekday”, “30th Level”, “15th Level” Since there may be a combination of “others”, a table of 8 rows × 3 columns is obtained as shown in FIG.

  On the other hand, the “insurance contract” node has five parent nodes heading to that node, four of which are binary types, and the other one is classified into three newspaper brands. Thus, as shown in FIG. 21, the table becomes 48 rows × 2 columns.

  When a Bayesian network is constructed in this way, it becomes possible to infer posterior probabilities in each pattern using the probability distribution learned in the above as a prior probability. In the posterior probability inference, the probability is calculated by tracing the entire network from parent to child and from child to parent, and belief propagation (J.Pearl: Probabilistic reasoning in intelligent systems, Morgan Kaufmann) (1988) P150-177) has been established. However, as shown in FIG. 16, in the case of a network called multiply connected where there is a portion where one path is looped when the network is viewed without considering the direction of the link, the accuracy of the Bayesian network These updates have proven to be NP-hard (see Cooper, GF: The Computational Complexity of Probabilistic Inference using Bayesian Belief Networks, Artificial Intelligence, 42: p.393-405, (1990)). In order to deal with this problem, this time the junction tree algorithm (Jensen, FV, Lauritzen, SL, and Olesen, KG: Bayesian updating in causal probabilistic networks by local computations.Computational Statistics Quarterly, 4, p.269-282 (1990 )). The junction tree algorithm solves the computational problem in believe propagation by converting the network structure into a multiple tree by creating a cluster that merges nodes after making the directed graph structure an undirected graph. However, the probability inference of the Bayesian network is not particularly limited to the junction tree.

  When the junction tree algorithm is used in the Bayesian network in which the graph structure and conditional probabilities have already been learned, as illustrated in FIG. 22, the “β newspaper” is posted at “other than the end of the month” and “newspaper advertisement contact amount” The probability of simultaneous occurrence of communication with a small number of consumers and not reaching an “insurance contract” is calculated as “96.9%”.

  Assuming a Bayesian network that has already been trained in this way, the posterior probability of each pattern can be inferred from the prior probability even from uncertain information. Effective use for decision-making becomes possible.

  As a result of executing the above calculation for all communication chains assuming that they are placed in the “α newspaper”, which is a restriction condition, and rearranging in order from the highest probability of reaching the “insurance contract” ( Response probability ranking is shown in FIG. The media plan for “this month” is the seventh pattern from the top of the table of FIG. 23, and shows that a more effective communication chain has appeared with the progress of the campaign.

  Therefore, if the pattern at the top of the ranking is adopted as the media plan for “next month”, which is the next plan period, instead of the pattern for “current month”, it is expected that the probability of closing will increase. Therefore, in this embodiment, the storage contents of the first storage means are updated to the first rank pattern in FIG.

  Here, in order to simplify the processing, the top pattern of the response probability ranking is selected as the “effective field pattern”. However, the present invention is not limited to this. Various selection methods are conceivable, such as selecting an aspect that takes into account the balance with the submission frame, and selecting all those with a predetermined response probability or higher.

  In the second embodiment, the effective field pattern stored in the first storage unit is obtained using the response probability or its ranking as it is, but here, the effective field pattern is obtained with an effect in consideration of the investment amount.

  In other words, if you run large-scale advertising campaigns by using high-pay media and creatives or presenting expensive offers without making money, you can naturally expect a good response opportunity. However, it is not expected to be realistic in terms of cost effectiveness.

  Therefore, after considering the cost required for the submission plan when following the field pattern, the product of the profit obtained by the response and the response probability is used to obtain the investment consideration effect, and based on the amount or its ranking It is desirable to derive an effective field pattern.

  In FIG. 25, the stored contents of the communication contact database (DB3) are illustrated, and the advertising cost unit price for communication with consumers is stored in a database. Further, an expected profit unit price when a response is obtained is also stored. The profit unit price may refer to another external database.

  Then, the analysis engine using the Bayesian network exemplified in Example 2 is applied to the record group of the consumer database exemplified in FIG.

  Further, the investment consideration effect is calculated and ranked as shown in FIG. 26 based on the drawing unit price information stored in the communication contact database exemplified in FIG. 25 and the profit unit price information obtained by the response. As a specific process, the investment consideration effect is calculated by summing up the products of the posterior probabilities of the nodes and the unit price information corresponding to the nodes.

  At this time, as for the item “shampoo purchase”, the data of the past consumer database is aggregated or the data of another general database is adopted to determine the corresponding value.

  The field pattern having a high investment consideration effect obtained in this way can be determined as an effective field pattern, and can be adopted as a publishing pattern for the next planning period.

  On the other hand, an analysis engine using the MT system is applied to the record group shown in FIG. 27 similar to the record group shown in FIG. 15, and an effective field pattern can be obtained based on the Mahalanobis distance.

Specifically, the average value of the information of each field related to the reference data group is extracted as a reference data group from the record group after the normalization process of FIG. A characteristic quantity such as (m _i ) and standard deviation (σ _i ) is obtained, a correlation coefficient (r _ij ) between information of each field normalized by the characteristic quantity is obtained, and a correlation matrix (R) is calculated. To do.

After all combinations (y _l ) of values that can be taken by each field are derived as pattern variations, the Mahalanobis distance (D _l ² ) is calculated for each pattern variation (y _l ) by the following equation.
However, in practice, in many cases, it is assumed that a problem occurs in the inverse matrix (R ⁻¹ ) of the correlation matrix due to multicollinearity. Multi-collinearity means that when there is a strong correlation between fields, the determinant of RR ^-1 is close to zero, so that the inverse matrix (R ^-1 ) cannot be calculated, or an abnormal solution is calculated. It is a phenomenon that happens. In order to cope with such a problem, it is desirable to include a device for calculating an approximate inverse matrix such as a method for avoiding multicollinearity by introducing noise on the diagonal line of the correlation matrix (Jittering method).

  FIG. 28 shows the results arranged in the order of the Mahalanobis distance. In addition, the draft plan currently in progress is indicated by an arrow as “current month”. Since a field pattern indicating a Mahalanobis distance greater than that of the current month's submission plan was detected, in order to replace the media plan of “next month” with the field pattern of the greater Mahalanobis distance in FIG. The contents of the first storage means were updated.

  This process is schematically shown in FIG. If such a process is performed at any time or periodically, it becomes possible to select a media and a placement form that are highly effective and appropriate at the time of the currently ongoing media plan.

  Here, the feature amount can be narrowed down by comparing the case where each explanatory variable (called the feature amount in the MT system) is used for evaluation with the case where it is not used for evaluation. Specifically, the record that has reached the response is regarded as an abnormal value for the record group that has not reached the response, the Mahalanobis distance for the abnormal value is calculated, the SN ratio for each set of explanatory variables is calculated, and the explanatory variable It is desirable to obtain the difference between the case where the value is used and the case where the value is not used, and evaluate using only the explanatory variable indicating a large value.

  In addition, it is possible to track whether it is necessary to evaluate whether each feature value is effective in determining the presence or absence of a response, and to calculate the Mahalanobis distance using only highly effective feature values. Even if all the feature values are adopted and the determination is based on the Mahalanobis distance, the determination can be made with sufficient accuracy.

  The effective field pattern obtained by the above-described embodiment 3 or 4 can be used as the placement pattern in the first storage means in the next planning period, so that the ongoing advertising campaign can be optimized and the explanatory variables or features can be further improved. If the pattern concerning the quantity is the same as this effective field pattern, and it is a consumer or transaction that has not yet responded, it can be assumed that it is a “promising customer”.

  Accordingly, a record that matches the pattern of the field related to the explanatory variable of the obtained effective field pattern is retrieved from the consumer database, estimated as a “promising customer”, and a communication contact that is accessible to a person having the pattern Is retrieved from the communication contact database DM2 and the stored contents are updated to activate the communication driver corresponding to the contact in the first storage means M1.

  This makes it possible to carry out push-type advertisements using addressable media such as e-mail for consumers or transactions that are presumed to be promising customers.

  As another example, as illustrated in FIG. 30, for example, a field pattern that is considered to be highly effective for responses collected in the previous month (for example, an analysis engine using a Bayesian network, a response probability Is an analysis engine that uses the MT system, if the Mahalanobis distance is estimated to be 1.2 or more, etc.), records that have not yet been responded to by the corresponding consumer or transaction Searching from the consumer database, the record is regarded as a “promising customer”, and push-type advertisement is developed using addressable media such as e-mail.

  In this way, the target of push-type advertisements such as e-mail distribution can be narrowed down, and the total advertising costs can be saved or the budget generated by the savings can be concentrated on the costs for push-type advertisements to prospective customers. It is possible to increase advertising efficiency.

  According to the present invention, it is possible to accurately analyze the propriety of an advertising campaign under development based on the influence on consumer's consumption behavior, and to automatically correct the campaign plan in real time or at short intervals.

  Further, according to the present invention, it is possible to select an effective placement format and guide a promising customer in real time in inducing consumers to consumption behavior.

  Furthermore, in the present invention, non-addressable mass media information can be handled in a unified manner by associating with individual consumers in the same way as addressable media information, and advertising tools of different dimensions such as media, vehicles, and creatives can be used. It has become possible to evaluate with the same indicators and handle them all together.

Explanation of symbols

1 Main part of this system 2 Management center including this system CPU Central processing unit B1 Primary bus line B2 Secondary bus line
E1 Standardization processing engine
E2 External data connection engine
E3 Dependency evaluation analysis engine U1 Effective field pattern extraction processing unit U2 Placement condition update processing unit M1 First storage means M2 Second storage means DB1 Consumer database DB2 Communication contact database DB3 Mass media information database DB4 Other external database R Router P1 ~ Pj Consumer CL1 ~ CLm Client terminal M-1 ~ M-n Media side terminal
Server Trafficking management server
Internet Internet
LAN local area network
DM Direct Mail Postage delivery service including postal mail

Claims (32)

At least one type of consumer attribute information and at least one type of communication history information including any one of communication condition and response result information are stored in units of consumers or transactions each consisting of a plurality of fields. A consumer database that stores records so that they can be searched and updated,
First storage means for storing search condition information for each communication driver so as to be searchable and updatable;
The records stored in the consumer database are sequentially read out, and the data of the consumer attribute information and the communication history information are respectively normalized, and the standardized data is made the record as a row and the field as a column. A second storage means for storing each data in a searchable manner by assigning them to a matrix-like storage area and specifying the rows and / or columns;
Using the analysis engine of one or more types of Bayesian network or MT system for the data stored in the second storage means, the dependence on the set of specific values of each field with respect to the response result is determined. Dependency quantification means to quantify,
Effective field pattern extraction means for extracting, as an effective field pattern, a set of concrete field values determined to be valid for the response result from the dependency;
A campaign dynamic optimizing system comprising a placement condition information update means for updating placement condition information for each communication driver stored in the first storage means within a range operable to the effective field pattern. An external database in which external information other than the information recorded in the consumer database is recorded based on at least one consumer attribute information in the consumer database.
Searching the external database using at least one consumer attribute information and / or communication history information common to the consumer database and the external database as searchers, and deriving standard external information for each searcher, 2. The information linking unit according to claim 1, further comprising information linking means for additionally recording the derived standard external information by providing a new field in the record of the consumer database corresponding to the searcher. Campaign dynamic optimization system. The campaign dynamic optimization system according to claim 1, wherein the consumer database is updated in real time or periodically. Based on at least one consumer attribute information of the consumer database, mass media information database storing mass media information related to contact amount so as to be searchable;
Searching the mass media information database using at least one kind of consumer attribute information common to the consumer database and the mass media information database as a search element, deriving standard mass media information for each search element, Further comprising mass media information connecting means for providing a new field in the record of the consumer database corresponding to the searcher and additionally recording the derived standard external information.
The campaign movement according to any one of claims 1 to 3, wherein the second storage means sequentially reads records stored in a consumer database after the mass media information is connected. Optimization system. 5. The campaign dynamic optimization system according to claim 4, wherein the mass media information database is updated in real time or periodically. The dependency quantification means includes:
In the consumer database, the field corresponding to the communication history information related to the response result is set as a node related to the objective variable, and the field corresponding to the other consumer attribute information or the communication history information is set as a node related to the explanatory variable. A structure learning means for modeling a route from a node related to an explanatory variable to a node related to the target variable by a directed link so that there is no circulation;
Using the data recorded in the field corresponding to each node in the consumer database, a prior distribution for each specific value that can be taken by the data recorded in the field is calculated, and a conditional probability table ( CPT), and a conditional probability table setting means for setting to the node,
A communication chain extracting means for extracting all sets of specific values that can be taken by the nodes in the structure as a communication chain;
Posterior probability calculating means for calculating posterior probabilities for response results based on the prior distribution of each node and the conditional probability table (CPT) for each communication chain;
The campaign dynamic optimization system according to any one of claims 1 to 5, wherein the campaign dynamic optimization system is a Bayesian network analysis engine in which the value of each posterior probability is the dependency on the set of specific values. The effective field pattern extraction means includes
An effective field having a high correlation with the communication history information related to the response result by comparing the communication chains for each communication chain and comparing the communication chain having the dependency higher than the predetermined value or the rank of the magnitude of the dependency. The campaign dynamic optimization system according to claim 6, wherein the campaign dynamic optimization system is extracted as a pattern. A communication contact database that stores searchable and updateable information on at least the cost of advertising for each communication driver and production cost information for promotional materials, and
A transaction condition database that stores, in a retrievable manner, searchable / updateable consideration information of the product or service to be advertised, and consideration information of the related product or service derived from the transaction to be advertised;
The effective field pattern extraction means includes
Based on the transaction cost that can be searched from the communication contact database and the transaction price that can be expected when the response result is obtained that can be searched from the transaction condition database. A field pattern that includes an investment consideration effect calculation unit that calculates an effect considering the investment, and the investment consideration effect calculated by the investment consideration effect calculation unit is greater than or equal to a predetermined value or a higher communication chain with the response result The campaign dynamic optimizing system according to claim 6, wherein the campaign dynamic optimizing system is extracted. Communication contact database that stores searchable and updatable contact information for each communication driver, information on cost of production of promotional materials, and contact information that can be communicated to consumers or transactions related to predetermined consumer attributes or communication history information Further including
The placement condition information update means includes:
Based on the effective field pattern, a consumer or transaction related to the record corresponding to the effective field pattern is searched from a database of consumers, and a prospective customer extracting means for extracting the consumer or transaction related to the record as a promising customer;
A communication driver selecting means for selecting a communication driver capable of directly communicating with the prospective customer from the communication contact database;
8. The campaign operation according to claim 7, wherein the placement condition information in the first storage means related to the selected communication driver is updated so as to be communicable with the promising customer at an appropriate timing. Optimization system. 10. The campaign dynamic optimization system according to claim 8, wherein the communication contact database is updated in real time or periodically. The dependence quantification means comprises:
A reference data group extracting means for extracting the record that has not resulted in a response from the consumer database and setting it as a reference data group;
While calculating the characteristic amount (m: average value of each feature amount, σ: standard deviation of each feature amount) of information of each field related to the reference data group, between the information of each field normalized by the characteristic amount Statistical processing means for calculating a correlation coefficient (r _ij ) and deriving a correlation matrix (R),
Pattern variation deriving means for deriving all combinations (y _l ) of values that each field can take as pattern variations;
Mahalanobis distance calculating means for calculating the Mahalanobis distance (D _l ² ) with respect to the reference data group using the following formula for each pattern variation (y _l ) derived by the pattern variation deriving means:
An MT system analysis engine comprising:
The campaign dynamic optimization system that characterized in that said it is an independent, any one of claims 1 to 5 for the set of the specific values of each Mahalanobis distance (D l ^_2). The effective field pattern extraction means includes
Correlation of a predetermined number of patterns in which the dependency (ie, Mahalanobis distance) value evaluated by the dependency quantification means is greater than or equal to a predetermined value or higher in ranking with the communication history information related to the response result The campaign dynamic optimizing system according to claim 11, wherein the effective field pattern is extracted as a high effective field pattern. The placement condition information update means includes:
When updating the data in the consumer database, based on the updated data, update the data stored in the second storage means,
Based on the updated storage of the second storage means, the dependency quantification means and the effective field pattern extraction means are executed,
12. The campaign dynamic optimization system according to claim 11, wherein each placement condition information stored in the first storage means is updated to a newly extracted effective field pattern. For each communication driver, further includes a communication contact database for storing searchable / updateable contact information for a consumer or transaction related to predetermined consumer attributes or communication history information,
The placement condition information update means includes:
Based on the effective field pattern, a consumer or transaction of a record related to the effective field pattern is searched from the consumer database, and the consumer or transaction is extracted as a promising customer;
Communication driver selection means for selecting a communication driver capable of communicating with the promising customer from the communication contact database;
13. The placement condition information stored in the first storage means related to the selected communication driver is updated so as to be communicable with the promising customer at an appropriate timing. Campaign dynamic optimization system. The campaign dynamic optimization system according to claim 14, wherein the communication contact database is updated in real time or periodically. A record with a unit of a consumer or a transaction consisting of fields for storing at least one kind of consumer attribute information and at least one kind of communication history information including any one of communication condition and response result information. A campaign dynamic optimization method characterized by causing a computer coupled with a consumer database stored in a searchable / updatable manner via an electric signal to execute the following steps.
A first storage step for storing search condition information for each communication driver in a searchable and updatable manner;
The records stored in the consumer database are sequentially read out, and the data of the consumer attribute information and the communication history information are respectively normalized, and the standardized data is made the record as a row and the field as a column. A second storage step of storing each data in a searchable manner by assigning them to a matrix-like region and specifying the row and / or column;
Using the analysis engine of one or more types of Bayesian network or MT system for the data stored in the second storage step, the dependence on the set of specific values of each field with respect to the response result is determined. Quantification, dependency evaluation quantification process,
An effective field pattern extraction step of extracting a set of specific values of fields determined to be valid for the response result from the dependency as an effective field pattern;
A publishing condition information update step of updating the publishing condition information for each communication driver stored in the first storage step within a range operable to the effective field pattern; The computer
Based on at least one consumer attribute information of the consumer database, an external database that records external information other than the information recorded in the consumer database is also connectable via an electrical signal,
The external database is searched using at least one consumer attribute information common to the consumer database and the external database as a search element, and standard external information is derived for each search element. 17. The campaign dynamic optimizing method according to claim 16, further comprising an information linking step in which external information is additionally recorded by providing a new field in the record of the consumer database corresponding to the searcher. 18. The campaign dynamic optimization method according to claim 16, wherein the consumer database is updated in real time or periodically. The computer
Based on at least one type of consumer attribute information in the consumer database, the mass media information database storing the searchable mass media information related to the contact amount is coupled via an electrical signal,
Searching the mass media information database using at least one kind of consumer attribute information common to the consumer database and the mass media information database as a search element, deriving standard mass media information for each search element, Further including a mass media information connecting step of adding a new field to the record of the consumer database corresponding to the search element and additionally recording the derived standard external information,
The campaign dynamic appropriateness according to any one of claims 16 to 18, wherein in the second storage step, records stored in a consumer database after the mass media information is connected are sequentially read out. Method. 20. The campaign dynamic optimization system according to claim 19, wherein the mass media information database is updated in real time or periodically. The dependency quantification step includes:
In the consumer database, the field corresponding to the communication history information related to the response result is set as a node related to the objective variable, and the field corresponding to the other consumer attribute information or the communication history information is set as a node related to the explanatory variable. A structure learning step for modeling a route from a node related to an explanatory variable to a node related to the explanatory variable by a directed link so that there is no circulation of a path;
Using the data recorded in the field corresponding to each node in the consumer database, a prior distribution for each specific value that can be taken by the data recorded in the field is calculated, and a conditional probability table ( CPT) and a prior distribution calculation step for setting the node
A communication chain extraction step of extracting all sets of specific values that can be taken by the nodes in the structure as a communication chain;
Bayesian network analysis characterized by including a posterior probability calculation step of calculating posterior probabilities for response results based on the prior distribution of each node and the conditional probability table (CPT) for each communication chain. The campaign dynamic optimization method according to any one of claims 16 to 20, wherein an engine is used. The effective field pattern extracting step is
On the premise of the structure modeled by the structure learning step, the simultaneous probability for each communication chain is compared, and the communication chain having a high rank in which the joint probability is equal to or greater than a predetermined value or the magnitude of the joint probability is the response result. The campaign dynamic optimizing method according to claim 21, wherein the effective field pattern is highly correlated with the communication history information. The computer
A communication contact database that stores searchable and updateable information about the cost of advertising for each communication driver and production cost information for promotional materials,
At least the consideration information of the product or service to be advertised and the consideration information of the related product or service derived from the transaction to be advertised are stored in a searchable and updatable manner and further coupled via an electric signal. ,
The effective field pattern extraction step includes:
Based on the transaction cost that can be searched from the communication contact database and the transaction price that can be expected when the response result is obtained that can be detected from the transaction condition database, for each communication. Including an investment-considering effect calculation process for calculating the effect considering the investment,
The investment consideration effect calculated in the investment consideration effect calculation step is a step of extracting, as a field pattern having a high correlation with the response result, a communication chain having an investment consideration effect equal to or higher than a predetermined value. Campaign dynamic optimization method. The computer
Communication contact database that stores searchable and updatable contact information for each communication driver, information on cost of production of promotional materials, and contact information that can be communicated to consumers or transactions related to predetermined consumer attributes or communication history information Both via electrical signals,
The placement condition information update step includes:
Based on the effective field pattern, a consumer or transaction corresponding to a record that matches the effective field pattern is searched from the consumer database, and a consumer or transaction relating to the record is identified as a prospective customer. ,
Including a communication driver selection step of selecting from the communication contact database a communication driver capable of directly communicating with the promising customer,
The first storage step includes a step of updating the placement condition information relating to the selected communication driver so as to execute a push-type advertisement at an appropriate timing for the prospective customer. Item 22. The campaign dynamic optimization method according to Item 22. 25. The campaign dynamic optimization method according to claim 23, wherein the communication contact database is updated in real time or periodically. The dependency quantification step comprises:
A reference data group creation step for extracting the record that has not resulted in a response result from the consumer database and making it a reference data group,
A characteristic amount (m: average value of each feature amount, σ: standard deviation) of information of each field related to the reference data group is calculated, and a correlation coefficient between the information of each field normalized by both of the characteristics A statistical processing step of calculating (r _ij ) and deriving a correlation matrix (R);
A pattern variation derivation step of deriving all combinations (y _l ) of values that each field can take as pattern variations;
A Mahalanobis distance calculating step for calculating each Mahalanobis distance (D _l ² ) with respect to the reference data group using the following formula for each pattern variation (y _l ):
Characterized in that said each Mahalanobis distance (D l ^₂₎ is a step of said dependence for a set of the specific value, of any one of claims 16 to 20 using the MT system analysis engine Campaign dynamic optimization method. The effective field pattern extraction step includes:
In the dependency quantification step, a predetermined number of patterns whose dependency values are equal to or higher than a predetermined value or higher in the ranking are used as effective field patterns having high correlation with communication history information related to the response result. 27. The campaign dynamic optimization method according to claim 26, wherein: The placement condition information update step includes:
Updating the data stored in the second storage means based on the updated data when updating the data in the consumer database;
Based on the updated second storage means data, the dependency evaluation step and the effective field pattern extraction step are executed,
27. The campaign dynamic optimization method according to claim 26, further comprising the step of updating each placement condition information stored in the first storage means to a newly extracted effective field pattern. The computer
For each communication driver, it is also connected via an electrical signal to a communication contact database that stores searchable / updateable contact information that can be communicated to a consumer or transaction related to predetermined consumer attributes or communication history information,
The placement condition update step includes:
Based on the effective field pattern, a consumer or transaction of a record related to the effective field pattern is searched from the consumer database, and the consumer or transaction is extracted as a promising customer;
Including a communication driver selection step of selecting a communication driver capable of communicating with the promising customer from the communication contact database,
28. The placement condition information stored in the first storage means related to the selected communication driver is updated so as to be communicable with the prospective customer at an appropriate timing. Campaign dynamic optimization method. 30. The campaign dynamic optimization method according to claim 29, wherein the communication contact database is updated in real time or periodically. The recording medium which recorded the campaign dynamic optimization method of any one of Claims 16-30 so that reading by a computer was possible. The transmission medium which transmits the campaign dynamic optimization method of any one of Claims 16-30 so that execution is possible with a computer.