JP2002157394A - Network marketing system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To calculate merchandise sales trend or merchandise purchase prediction based on the purchase or retrieval of a user when selling merchandise on a Web. SOLUTION: This system is constituted of a marketing research means 130 generating the information on merchandise which sells well and future merchandise sales trend information from the history data of access from a user to a Web mail order site 110 and the purchase of desired merchandise by the user, and providing the information to a market. This marketing research means is provided with a means for increasing the number of users who perform access of the Web mail order site 110, a ranking means ranking merchandise and users for acquiring the present merchandise sales trend, a merchandise sales predicting means predicting the sales situation of merchandise from the merchandise purchase and access of the user, a merchandise recommending means selecting and recommending specific merchandise, a new merchandise trend predicting means predicting the sales trend of new merchandise based on the purchase history and access history of new merchandise, and a user basket analyzing means deriving representative tastes from the attributes of the user.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a network marketing system, and more particularly to a technique for increasing user access on a Web site, a product ranking, and a user ranking in order to appropriately provide future recommended products to customers (users). Method, product sales prediction method, product recommendation method, new product trend prediction method, user basket analysis method,
The present invention relates to a network marketing system provided with a means for recommending future hot selling products to customers.

[0002]

2. Description of the Related Art In order to predict consumer purchases in the prior art, merchandise recommendation means based on customer analysis must be considered.
It is well known as a product recommendation technique that the sales history data from S and the sales history data on the Web site are used to identify the preference and purchase motivation for each customer and to apply recommended products.

[0003]

However, the product recommendation method described in the prior art has a problem in that it is not possible to accurately judge the transition of customer interest. Because PO
This is because the data of S is limited to only those sold, and it is not possible to grasp the interest of the customer before the purchase. In order to capture the customer's interest in products, there is a method of conducting pre-sales activities such as questionnaires, but it is limited to customers and if it is performed on a large scale, there is a problem such as aggregation and time, There is also a problem that there is a limit in accurately grasping the time.

[0004] In addition, the data on the Web site can be used to collect the customer's interest in the product before the purchase in a short time. There is a problem in that it is difficult to recommend to the customer even products that can be purchased in the future only by deriving the desired preference.

[0005] Therefore, from the information on the current best-selling merchandise, the best-selling information on consumption that the customer is interested in and wants to purchase in the future, that is, the future recommended merchandise information for unlimited customers is quickly and accurately obtained. There is a problem that must be solved in a system that can be extracted.

[0006]

Means for Solving the Problems In order to solve the above problems, a network marketing system according to the present invention has the following configuration.

[0007] (1) A Web product sales site where a predetermined product can be purchased or browsed on a network, a user who can access the Web product sales site, that the user has accessed the Web product sales site, and Marketing research means for generating information on products sold in the future from the history data of the purchase of the desired product and future merchandise sales information and providing the information to the market;
means for increasing the number of users accessing the eb product sales site; ranking means for assigning a product ranking indicating the current sales of products and a user ranking; and product sales prediction for predicting the sales state of the product based on the purchase and access of the product by the user. Means, a product recommendation means for selecting and recommending a specific product from a user's purchase and access of a product, a new product trend predicting means for predicting a selling trend of the new product based on a purchase history and an access history of the new product, A network marketing system comprising user basket analysis means for deriving a representative preference from the attributes of a user. (2) The network marketing system according to (1), wherein the network is the Internet. (3) The means for increasing the number of users accessing the Web merchandise sales site is to access the merchandise product catalog provided by the Web merchandise site as points, add the points to obtain a pay-as-you-go score, and obtain the pay-as-you-go score. (1) The network marketing system according to (1), wherein a common prize is provided in response to the request. (4) The ranking means for assigning the product ranking and the user ranking indicating the best sellers of the current product is based on the user's access history and purchase history of the product for sale provided by the Web product sales site, and the overall ranking of the product for sale. The network marketing system according to (1), wherein the rankings are calculated and ranked. (5) The product sales prediction means predicts that the product is suitable for sale based on the purchase history of the product to be sold and the access history of the purchased product before purchase. A) a network marketing system described in (6) The product recommendation means extracts a category having a high frequency of access and purchase from the access history and purchase history of the product to be sold, creates a user group for each specific preference from the category, and recommends the product. The network marketing system according to (1), wherein: (7) The new product trend predicting means includes a time-series search history for each of the target new products for a certain period from the time of release,
Excerpt the time-series purchase history, calculate the statistic for each of the extracted time-series search history and time-series purchase history, compare the calculated statistic with the existing statistic, The network marketing system according to (1), wherein is predicted. (8) The network marketing system according to (1), wherein the user / basket analysis means derives a representative preference from the demographic of the user and selects a product according to the derived preference. . (9) The network marketing system according to (8), wherein the demographic of the user includes the age, gender, and area of residence of the user.

[0008]

Next, an embodiment of a network marketing system according to the present invention will be described with reference to the drawings.

As shown in FIG. 1, a network marketing system according to the present invention includes a Web product sales site 110 which allows a user to sell or watch a predetermined product on a Web site forming a network by network sales or access. Product sales site 11
0 and a user 120 who can access
The marketing research means 130 generates merchandise selling information and future merchandise trends from history data indicating that the user 120 has accessed the merchandise sales site 110 and purchased desired merchandise, and provides this to the market.

The marketing research means 130 is formed in a so-called four-layer structure as shown in FIG.
User 1 capable of driving input / output by HTML without processing such as login screen display, product input and search, calculation of ordered product output, result display, etc. by HTML
20, a second layer is a Web server 140 that manages an interface between a user and a Web application server such as a user search / purchase and input / output transaction processing and a user congestion countermeasure transaction processing, and a third layer is various applications. An application that drives means for increasing the number of users accessing the WEB product sales site, an application that drives a ranking unit that assigns a product ranking indicating the current sales of products and a user ranking, and an application that drives an application that drives a product sales prediction unit , An application for driving a product recommendation unit, an application for driving a new product trend prediction unit, and an application for driving a user basket analysis unit, and data stored in a DB server. A Web application server 150 to the desired operation or the like by appropriately accessing the over scan, 4th
"IA, IB, P-A1, P-B" which are databases used in the web application server of the layer
1, C-A1, C-B1, P-A2, P-B2, C-A
2, CB-B2, NP-A1, P, NP-B1, C, P
F, CF, CJ, PJ, CG, CP, CD, PD ”.

The user 120 is configured to operate on a GUI, and the browser operates on a Windows-based PC or WS, or a Macintosh or UNIX (registered trademark) -based WS by using the browser as a client. It has a structure that can be a client if it does.

Next, in the marketing research means 130 having such a configuration, the Web application server 150 can concentrate on the creation of an application.
An application that drives a means for increasing the number of users accessing the EB product sales site, (2) an application that drives a ranking unit that assigns a product ranking indicating the current sales of products and a user ranking, and (3) a product sales prediction unit that is driven. An application that drives an application to be driven, (4) an application that drives product recommendation means, (5) an application that drives new product trend prediction means, and (5) an application that drives user basket analysis means. This will be described with reference to FIG.

(1) Means for Increasing the Number of Users Accessing the Web Product Sales Site The means for increasing the number of users accessing the Web product sales site 110 is based on access to a product catalog for sale provided by the Web product sales site 110. This point is added to obtain a pay-as-you-go score, and a common prize is provided according to the pay-as-you-go score.

More specifically, the Web product sales site 11
Constantly increasing the history of user trends at zero is a very important factor in maintaining the quality of the analysis. The point here is how to constantly encourage user access. That is, the introduction of a pay-as-you-go score system for access to the product catalog targeted for sale at the web sale site 110. The introduction of the pay-as-you-go scoring system provides a common gift certificate as a common prize according to the total score. Specifically, when the user enters the Web product sales site 11
When the user accesses “0” and actually accesses information on a product to be sold posted on the web product sales site 110, this action is added to the total score of the user 120 as a fixed point.

(2) Ordering means for assigning a product ranking and a user ranking indicating the current sales of merchandise The ordering means for assigning a product ranking and a user ranking indicating the current sales of merchandise is provided by the Web sales site 110. This is to calculate and rank the overall rank of the target product for sale based on the access history and purchase history of the user 110 to the target product. Specifically, the market analysis on the web product sales site 110 is aggregated into the analysis of the product for sale and the analysis of the user 120 who has accessed the web product sales site 110. In particular, these two types of analytes are stored on the Web product sales site 11.
It is interesting to know how they are related to each other. Here, as one function, a method for explicitly indicating the overall ranking of the merchandise to be sold and the overall ranking of the user 120 will be provided below. As a feature of this ranking, ranking is performed in accordance with the current trend based on the history of product reference and product purchase. Therefore, according to this method, the ranking will be switched every time the ranking processing is performed. That is why it is a timely analysis.

First, as shown in FIG. 3, the product search history data (IA) and the product purchase history data (IB) at each Web product sales site 110 are obtained by searching for and purchasing all the Web product sales sites. For example, a product and a product b searched by the user U1, which are databases for the user, are stored in U1-IA, and purchased e products and f products are stored in U1-IB. G searched by user U2
The merchandise and h merchandise are accumulated in U2-IA, and the purchased e merchandise and h merchandise are accumulated in U2-IB. The x products and y products searched by the user Ux are stored in Ux-IA, and the purchased x products and z products are stored in Ux-IB. In this way, in each WEB product sales site, the products searched and purchased by each user are stored as data, and all the data are stored in the product search history data (IA) and the product purchase history data (IB). It is accumulating as.

FIG. 4 shows a method of ranking product rankings and user rankings. A first method for constructing a history data layer comprising product search and history data and user 120 product search and purchase history data is shown. , A second processing mode for constructing a statistical data layer composed of search statistics and purchase statistics of products and search and purchase statistics data of the user 120, and ranking composed of rank data of products and ranks of users And a third processing mode for constructing a data layer.

In the first processing mode, product search history data (I-
A), the product purchase history data (IB) is fetched, and a product search and purchase history data as viewed from the user side are constructed, and a search and purchase history data layer of the user 120 as viewed from the product side is created.

The history data layer includes search history data (P-A1) for all products, purchase history data (P-B1) for all products,
It is composed of merchandise search history data (C-A1) of all users and merchandise purchase history data (C-B1) of all users.

The search history data (P-A1) of all products is
The product search history data (IA) from all the web product sales sites 110 is totaled, and a search history of all products centering on products is created. The purchase history data (P-B1) of all products is obtained by summing up the purchase history data (IB) of the products from all Web merchandise sales sites 110 and creating a purchase history of all products centering on the products. . The product search history data (C-A1) of all users is obtained by summing up the product search history data (IA) from all Web product sales sites 110 and creating a search history of all products centering on the user 120. It is.

Product purchase history data (C for all users 120)
-B1) is obtained by totalizing the product purchase history data (IB) from all of the Web product sales sites 110 and creating a purchase history of all the products centered on the user 120.

As described above, in the first processing mode, first, a search and purchase data layer for each product and a search and purchase data layer for each product are searched for the product searched and purchased by the user 120. To build.

In the second processing mode, from the data of the history data layer generated in the first processing mode, search statistical data (P-A2) of each product present for each product, and present for each product. Purchase statistical data of each product (P-B2), each user 1
The search statistical data (C
-A2), each user 120 existing for each user 120
Of purchase statistics data (C-B2).

The search statistical data (P-A2) of each product existing for each product is obtained by sorting the search history (P-A1) of all products sorted in the first processing mode for each product. The statistic is added to the number of time-series searches. Purchase statistics data for each product (P
-B2) is obtained by distributing the history data (P-B1) distributed in the first processing mode for each product, and adding a statistic including the number of purchases of each product in time series.

Each user 12 existing for each user 120
The search statistical data (C-A2) of 0 is the history data (C-A1) sorted in the first processing mode,
It is sorted for each 0, and a statistic including the number of time-series searches of each user 120 is added.

Each user 12 existing for each user 120
The purchase statistics data (C-B2) of No. 0 is obtained by combining the history data (C-B1) sorted in the first processing mode with each user 12.
It is further assigned to each of the 0s, and a statistic of time-series purchase of each user 120 is added.

The statistics for obtaining each of the statistical data are as follows:
The correlation cycle, the correlation duration, the correlation level, and the like are calculated based on the time series correlation coefficient, the first-order regression coefficient, and the time series correlation coefficient periodic function, which will be described later.

Next, based on the obtained data of the statistical data layer, each product is scored according to the first rule shown below with respect to the correlation cycle, correlation duration, and correlation level of the time-series search correlation of each product.

First rule: score = (time-series search correlation average value / correlation cycle) × the number of purchases per unit period

Further, the correlation cycle of the time-series purchase correlation of each product,
Each product is scored according to the second rule with respect to the correlation duration and the correlation level.

Second rule: score = (average value of time-series purchase correlation ÷ correlation period) × number of searches in a unit period

Next, based on the obtained data of the statistical data layer, each user is scored according to the third rule shown below with respect to the correlation cycle, correlation duration, and correlation level of the time-series search correlation of each user.

Third rule: user's search base score = (time-series search correlation average value ÷ correlation period) × number of purchases per unit period

Each product is scored according to the fourth rule with respect to the correlation cycle, correlation duration, and correlation level of the time-series purchase correlation of each user.

Fourth rule: user's purchase base score = (time-series purchase correlation average value ÷ correlation period) × number of searches in a unit period

The sum of the above two points is regarded as the total score of each user, and within the range of the score, the sum obtained by the first and second rules is regarded as the total score of each product.
The total value obtained by the fourth rule is used as the total score of each user. Product ranking data (P) in the ranking data layer stores products arranged in descending order of the total score, and user ranking data (C) in the ranking data layer stored in descending order of the total score. .

The statistics for obtaining each of the above statistical data are as follows:
It can be calculated based on the time series correlation coefficient, the primary regression coefficient, the time series correlation coefficient periodic function, the number of daily searches, and the number of daily purchases. Hereinafter, each will be described.

The time-series correlation coefficient is obtained by arranging the elapsed time (elapsed days, etc.) on the X-axis and the variable (sales per unit time, stock price, etc.) on the Y-axis in the two-dimensional XY coordinates. Is an index indicating whether or not the Y-axis data has a correlation with the change in the X-axis data with respect to the progress of. The time-series correlation coefficient having such a property can obtain the correlation by the equation shown in FIG. 5, which indicates that between the data of the X-axis and the data of the Y-axis, between −1 and +1. The magnitude of the amplitude gives the degree of the positive and negative functions,
This is obtained by replacing the correlation tendency of variables with time with a finite range variation. This makes it possible to grasp the visual change tendency at the next numerical value level.

For example, in the case of the data shown in FIG. 6A, the data on the Y-axis (the number of purchases, etc.) tends to increase as the data on the X-axis (elapsed days, etc.) change. The time-series correlation in this case is such that a positively correlated graph is drawn with respect to the Y-axis data and the X-axis data, so that the time-series correlation coefficient is limited to plus one. Approach without.

In the case of the data shown in FIG. 6B, no change is observed in the change in the data on the Y-axis with the change in the data on the X-axis. It is determined that there is no correlation between the X-axis data, and the time-series correlation coefficient indicates a value close to 0.

In the case of the data shown in FIG. 6 (C), the data on the Y-axis has a downward tendency to the right as the X-axis data changes. In the time-series correlation in this case, the data on the X-axis has a strong negative correlation with respect to the data on the Y-axis, and the time-series correlation coefficient approaches as much as minus one.

Next, the first-order regression coefficient will be described with reference to FIGS. The primary regression coefficient is the secondary space X
In the Y coordinate, when the elapsed time (elapsed days, etc.) is placed on the X axis, and the variable (sales per unit time, stock price, etc.) is placed on the Y axis,
It represents the gradient (first-order derivative) of the Y-axis variable at a certain point in the X-axis data, and its calculation formula can be calculated by the formula shown in FIG.

For example, as shown in FIG. 8, a positive value is shown at the point A, and the primary regression coefficient indicates that the data on the Y-axis has an increasing tendency with respect to the change in the data on the X-axis. At the point B, a value near 0 is shown, and the primary regression coefficient moves up and down within a certain narrow value range without showing a response of the Y-axis data to a change in the X-axis data. At the point C, the value again shows a positive value. However, the number is larger than the first-order regression coefficient at the point A, and the momentum of the increase of the Y-axis data at the point C is larger than the momentum at the point A. Show. At point D, the value indicates a negative value and 1
The next regression coefficient indicates that the change in the data on the Y-axis has a downward trend with respect to the change in the data on the X-axis.

The above-mentioned time series correlation coefficient indicates the continuity of the change state of the Y axis with respect to the change of the X axis (increase or no change,
Although it is shown in detail), the time-series correlation coefficient cannot measure the level of rise and fall. On the other hand, the first-order regression coefficient is not good at measuring persistence, but is suitable for measuring upward and downward momentum.

Next, the time-series correlation periodic function will be described with reference to FIGS. The time-series correlation periodic function indicates that the time-series correlation function has a strong tendency to increase when the value calculated by the above-described time-series correlation coefficient is at a high level (close to plus 1). On the premise that a high level (close to minus 1) indicates that the time series descending tendency is strong, an index of which level is set to the high level is calculated. This index is, as shown in FIG.
Use the discrete Fourier transform of the time series correlation coefficient. The discrete Fourier transform expresses a periodic spectrum of an original function (in this case, a time series correlation function), and the strength of this spectrum can be used for setting the level of the time series correlation function.

FIG. 10 shows the relationship between the time series correlation coefficient (solid line) and the discrete Fourier transform spectrum (filled surface). When the spectrum is more than +1, the time series correlation is a positive strong correlation. By setting the time-series correlation to a negative strong correlation (ie, a strong downward trend) when the spectrum is -1 (ie, a strong upward trend), the level setting of the time-series correlation coefficient becomes clear. Specifically, points A to E serve as indices indicating a strong upward trend of plus one or more, and points F to H
The point is an index indicating a strong downward trend of -1 or less.

(3) Merchandise Sales Prediction Means The merchandise sales prediction method is to predict the sales number of the merchandise to be sold from the purchase history of the merchandise to be sold and the access history of the purchased merchandise before the purchase. .

That is, originally, the purpose of market analysis is to predict the future sales of each handled product and to reflect this in the management strategy. In particular, in recent product sales at the Web product sales site 110, as the environment in which the preferences of the user 120 can be grasped before the sales are established, higher development of more accurate sales prediction is required. As this sales prediction method, the user 120
The aim is to improve the accuracy of predictions by introducing a formula that takes into account not only the purchase history but also the pre-purchase products.

As shown in FIG. 11, the purchase prediction method is not only an analysis for each single product, but also a total support increase / decrease rate of the Web product sales site. A first processing mode for constructing a history data layer composed of 120 product search and purchase history data;
A second processing mode for constructing a statistical data layer composed of product search statistics and purchase statistics and user search and purchase statistical data, and a third processing mode for constructing a ranking data layer composed of the ranks of the users 120 And a fourth processing mode for deriving the degree of increase / decrease in the support of the Web product sales site from the ranking data layer of the user 120, and a fifth processing mode for performing purchase prediction. Among them, the first processing mode, the second processing mode, and the third processing mode relating to the statistical data on the user 120 are calculated by the same method as the above-described ranking method.

In the first processing mode, merchandise search history data (IA) and merchandise purchase history data (IB) of all users 120 possessed by all Web merchandise sales sites 110 are fetched and viewed from the user 120 side. The product search and purchase history data is constructed, and the search and purchase history data layer of the user 120 as viewed from the product side is created.

The history data layer includes search history data (P-A1) for all products, purchase history data (P-B1) for all products,
It is composed of product search history data (C-A1) of all users 120 and product purchase history data (C-B1) of all users 120.

The search history data (P-A1) of all products is
Product search history data (I
-A), and a search history of all products centered on products is created. Purchase history data of all products (P-
B1) collects the product purchase history data (IB) from the web product sales site 110 and creates purchase histories of all products centering on the product.

Product search history data (C for all users 120)
-A1) collects the product search history data (IA) from the Web product sales site 110 and creates a search history of all products centered on the user 120.

Product purchase history data (CB) for all users
In 1), the purchase history of all products centered on the user 120 is created by totalizing the product purchase history data (IB) from the web product sales site 110.

As described above, in the first processing mode, first, a search and purchase data layer for each product and a search and purchase data layer for each user are searched for the product searched and purchased by the user 120. To build.

In the second processing mode, the retrieval statistical data (P-A2) of each product that exists for each product from the data of the history data layer generated in the first processing mode, and exists for each product. Purchase statistical data of each product (P-B2), each user 1
The search statistical data (C
-A2), each user 120 existing for each user 120
Of purchase statistics data (C-B2).

The search statistical data (P-A2) of each product existing for each product is obtained by sorting the search history (P-A1) of all products sorted in the first processing mode for each product. The statistic is added to the number of time-series searches.

The purchase statistical data (P-B2) of each product, which is present for each product, is obtained by distributing the history data (P-B1) distributed in the first processing mode for each product, and chronologically purchasing each product. It is the result of adding a statistic consisting of numbers.

Each user 12 existing for each user 120
The search statistical data (C-A2) of 0 is the history data (C-A1) sorted in the first processing mode,
It is sorted for each 0, and a statistic including the number of time-series searches of each user 120 is added.

The purchase statistical data (C-B2) of each user existing for each user 120 is obtained by further distributing the history data (C-B1) allocated in the first processing mode to each user 120. A statistic consisting of 120 time-series purchases is added.

The statistics for obtaining each of the statistical data are as follows:
The correlation cycle, the correlation duration, the correlation level, and the like are calculated based on the time series correlation coefficient, the first-order regression coefficient, and the time series correlation coefficient periodic function. .

Next, in the third processing mode, the data of the obtained statistical data layer is used to determine the correlation period, the correlation duration, and the correlation level of the time-series search correlation of each user 120. Points are scored according to the first rule shown.

First rule: score = (time-series search correlation average value ÷ correlation period) × the number of purchases in a unit period

The correlation period, the correlation duration, and the correlation level of the time-series purchase correlation of each user 120 are related to each user 1.
20 is scored according to the second rule.

Second rule: score = (time-series purchase correlation average value / correlation period) × the number of searches within a unit period

The sum of the above two scores is regarded as the total score of each user, and the user is ranked within the range of the score.
The result is assigned to the user ranking data (C) in the ranking data layer.
Save as

The fourth processing mode is the Web product sales site 11
The total increase / decrease of the users 120 at 0 is extracted from the user ranking data (C). Specifically, user 1
From the 20 total scores, a primary regression coefficient K of the total score average value in the past N measurements including this time, and a time series correlation coefficient R are calculated, and the degree of increase / decrease is indexed. Assuming that the indexed value is UD, UD = EXP (K × R), which is an important index indicating the direction of the total score transition. The user increase / decrease index data (C
Save as F).

The fifth processing mode comprises a first-order regression coefficient (A1), a time-series correlation coefficient periodic function (R1), and a maximum number of purchases (MAX)
1) and the minimum purchase number (MIN1) are extracted from the time-series search correlation data (P-B2) of each product existing for each product. Here, the unit period is one day, one week, one month, and the like, and the certain period N is an integral multiple (1, 2, 3,...) Of the period specified in the unit period.

The number of searches for each product for each unit period in the fixed period N is represented by a linear regression coefficient (A2), a time-series correlation coefficient periodic function (R2), a maximum search number (MAX2), and a minimum search number (MAX2). MIN2) is extracted from the time-series search correlation data (P-A2) of each product present for each product.

Here, the purchase acceleration coefficient AF (Access
ratio Factor) can be obtained by the following equation.

AF = a × (A1 × R1 × log (MAX
1 / MIN1) × UD) + b × (A2 × R2 × log
(MAX2 / MIN2) × UD) Here, both a and b are variation coefficients for deriving a purchase acceleration coefficient adjusted for each product.

The predicted purchase number range can be calculated by (the number of purchases in the latest unit time ± the number of purchases in the latest unit time × AF). The calculated predicted purchase number range is stored in the product purchase prediction data (PF) existing for each product classification. This forecast period depends on the unit period,
For example, if the unit period is one month, the forecast is one month ahead. If this is three months, it is expected three months ahead. The reason for using the latest data is that the average is not included, so that a delay due to averaging can be eliminated.

Although the purchase prediction can be performed in this way, the simulation relating to the actual purchase prediction is performed using the pseudo data of only the purchase history, and the predicted purchase number calculation model adopted in the fifth processing mode. It is calculated by substituting the following data by the formula. First, the first-order regression coefficient (A
1) The time series correlation coefficient periodic function (R1), the maximum purchase number (MAX1) and the minimum purchase number (MIN1) are set for each user 1.
20 time-series purchase correlation data (P-B2). Next, the number of searches for each product in a fixed period per unit period, the primary regression coefficient (A2), the time-series correlation coefficient periodic function (R2), the maximum search number (MAX2), and the minimum search number (M
IN2) to time-series search correlation data of each product (P-A2)
Extract from The above equation of the acceleration coefficient AF (AF = a
× (A1 × R1 × log (MAX1 / MIN1) × U
D) + b × (A2 × R2 × log (MAX2 / MIN)
2) x UD)), the search-related weight "b" is calculated under the conditions of zero, "a" = 5, and "UD" = 1 because only the purchase history is used. The expected purchase number range is calculated using the above formula (number of purchases in the latest unit period ± number of purchases in the latest unit period ×
AF).

As shown in FIG. 12, the result obtained in this way can indicate the expected purchase range at the time of the number of purchases with respect to the time-series change of the number of purchases of the product. It can be determined how the number of purchases changes.

(4) Product Recommending Means The product recommending means is one of the elements that have been desired in the field of marketing, like the above-mentioned future sales prediction of the handled goods, and the next purchase of each user 120. By increasing the accuracy of selecting the right product group to promote potential,
It becomes easy to increase the dependence of the user 120 on the Web product sales site 110. Therefore, as a product recommendation means,
A product group that matches the taste of each user 120 or that is in line with the trend is selected, and the user 120's Web product sales site 1 is selected.
The aim is to provide sustained interest in 10. Specifically, a category with a high frequency of access and purchase is extracted from the access history and purchase history of the product for sale, and a user group for each specific preference is created from this category.

FIG. 13 shows a product recommendation method. A first processing mode for constructing a history data layer comprising search and history data of a product and search and purchase history data of the user 120 is described. Search and purchase statistics,
A second processing mode for constructing a statistical data layer including search and purchase statistical data of the user 120, a third processing mode for constructing a ranking data layer including rank data of products, and a fourth processing mode , A fifth processing mode, a sixth processing mode, and a seventh processing mode.

The first processing mode is to retrieve and purchase a product from the user side by taking in the product search history data (IA) and the product purchase history data (IB) of all users possessed by the Web product sales site. Construction of history data and creation of a history data layer of user search and purchase from the product side.

The history data layer includes search history data (P-A1) for all products, purchase history data (P-B1) for all products,
It is composed of merchandise search history data (C-A1) of all users and merchandise purchase history data (C-B1) of all users.

The search history data (P-A1) for all products is
Product search history data (I
-A), and a search history of all products centered on products is created. Purchase history data of all products (P-
B1) collects the product purchase history data (IB) from the web product sales site 110 and creates purchase histories of all products centering on the product.

Product search history data (C for all users 120)
-A1) collects the product search history data (IA) from the Web product sales site 110 and creates a search history of all products centered on the user 120.

Product purchase history data (CB) for all users
In 1), the purchase history of all products centered on the user 120 is created by totalizing the product purchase history data (IB) from the web product sales site 110.

As described above, in the first processing mode, for the product searched and purchased by the user 120, first, the search and purchase data layer for each product and the product search and purchase data for each user 120. Build layers.

In the second processing mode, retrieval statistical data (P-A2) of each product, which exists for each product, exists from the data of the history data layer generated in the first processing mode, and exists for each product. Purchase statistical data of each product (P-B2), each user 1
The search statistical data (C
-A2), each user 120 existing for each user 120
Of purchase statistics data (C-B2).

The search statistical data (P-A2) of each product present for each product is obtained by sorting the search histories (P-A1) of all products sorted in the first processing mode for each product. The statistic is added to the number of time-series searches.

The purchase statistical data (P-B2) of each product, which is present for each product, is obtained by distributing the history data (P-B1) allocated in the first processing mode for each product, and chronologically purchasing each product. It is the result of adding a statistic consisting of numbers.

Each user 12 existing for each user 120
The search statistical data (C-A2) of 0 is the history data (C-A1) sorted in the first processing mode,
It is sorted for each 0, and a statistic including the number of time-series searches of each user 120 is added.

Each user 12 existing for each user 120
The purchase statistics data (C-B2) of No. 0 is obtained by combining the history data (C-B1) sorted in the first processing mode with each user 12.
This is obtained by further distributing the data for each 0 and adding the statistics of the time series purchase of each user 120.

The statistics for obtaining each of the statistical data are as follows:
The correlation cycle, the correlation duration, the correlation level, and the like are calculated based on the time series correlation coefficient, the first-order regression coefficient, and the time series correlation coefficient periodic function. .

Next, in the third processing mode, each product is described below with respect to the correlation cycle, correlation duration, and correlation level of the time-series search correlation of each product based on the obtained data of the statistical data layer. Points are scored by the rule of 1.

First rule: score = (time-series search correlation average value / correlation period) × the number of purchases in a unit period

Further, the correlation cycle of the time-series purchase correlation of each product,
Each product is scored according to the second rule with respect to the correlation duration and the correlation level.

Second rule: score = (average value of time-series purchase correlation / correlation period) × number of searches in unit period

The sum of the above two points is used as the total score of each product, the products are ranked within the range of the scores, and the result is stored as product ranking data (P) in the ranking data layer.

The fourth processing mode is based on the merchandise history data (IA) of all users on the Web sales site,
b From the product purchase history data (IB) on the product sales site,
A category that is frequently searched and purchased by each user 120 is extracted, and a user group (CJ) for each specific preference is created.

In the fifth processing mode, a user group (CJ) for each specific preference is further created, and the time-series search statistical data (C-A2) of each user 120 existing for each user 120 created in the second processing mode ), Time-series purchase search statistical data of each user 120 existing for each user 120 (C-B2)
Thus, the user group is finely grouped according to the following rules, and the result is created as a similar preference user group (CG) based on the user trend existing for each user group.

Rules: The user 120 belonging to each user group of the user group data (CJ) for each specific preference genre is classified into time-series search statistical data (C-A2) of each user 120, and time-series purchase statistics of each user 120. The data (C-B2) is finely grouped according to the similarity of the time series search correlation, the time series purchase correlation, and the time series search correlation periodic function.

In the sixth processing mode, the product ranking data (P) generated in the third processing mode is sorted into product genres. At this time, the product ranking within each genre is maintained. This is stored in the product ranking data (PJ) of each preference (genre) for each genre.

In the seventh processing mode, a high-ranking product ranking of each preference (highly supported by the market) is extracted from the product ranking data (PJ) of each preference (genre), and the similarity based on user trends is extracted. Matching is performed for each user 120 grouped by the same preference of the preference user group (CG). This result is stored in the product group data (CP) recommended for each user 120 for each user.

(5) New Product Trend Prediction Method In the new product trend prediction method, when a new product is put on the market, the manufacturer conducts various market researches, distributes R & D prototypes, and looks at the reaction of the market. Improve and appear as a new product. However, the risks that come out of the market solely based on the effects of the sampling market and that it will be accepted by all markets cannot be wiped out. Therefore, a method is provided in which a new product comes out into the world, and in a certain period (short period), the market tendency of the product as a latent characteristic is statistically determined, and a sales prediction method is provided.

That is, the new product trend predicting means extracts the time-series search history and the time-series purchase history of each of the target new products for a certain period from the time of release, and extracts the extracted time-series search history and time-series purchase history. , A statistic is calculated for each of them, and the calculated statistic is compared with the statistic of an existing product to predict the trend of a new product.

FIG. 14 shows a method of predicting a new product trend, which comprises a first processing mode, a second processing mode, a seventh processing mode, and an eighth processing mode.

The first processing mode includes merchandise search history data (IA) of all users 120 possessed by the Web product sales site,
User purchase history data (IB) is fetched and user 120
A search and purchase history data of a product as viewed from the side is constructed, and a search and purchase history data layer of the user 120 as viewed from the product is created.

The history data layer includes search history data (P-A1) for all products, purchase history data (P-B1) for all products,
It is composed of product search history data (C-A1) of all users 120 and product purchase history data (C-B1) of all users 120, of which search history data (PA) of all products.
1) Use purchase history data (P-B1) of all products.

The search history data (P-A1) for all products is
Product search history data (I
-A), and a search history of all products centered on products is created. Purchase history data of all products (P-
B1) collects the product purchase history data (IB) from the web product sales site 110 and creates purchase histories of all products centering on the product.

As described above, in the first processing mode, for a product searched and purchased by the user 120, first, a search and purchase data layer for each product is used.

In the second processing mode, based on the data of the history data layer generated in the first processing mode, the retrieval statistical data (P-A2) of each product present for each product and the search statistical data for each product exist. It is composed of purchase statistical data (P-B2) of each product.

The search statistical data (P-A2) of each product existing for each product is obtained by sorting the search history (P-A1) of all products sorted in the first processing mode for each product. The statistic is added to the number of time-series searches.

The purchase statistical data (P-B2) of each product present for each product is obtained by distributing the history data (P-B1) allocated in the first processing mode for each product, and chronologically purchasing each product. It is the result of adding a statistic consisting of numbers.

The statistic for obtaining each of the statistic data is:
The correlation cycle, the correlation duration, the correlation level, and the like are calculated based on the time series correlation coefficient, the first-order regression coefficient, and the time series correlation coefficient periodic function. .

The seventh processing mode is based on the search history (P-A1) of all products sorted in the first processing mode and the purchase history data (P-B1) of all products, from the point of sale of the target new product. A time-series search history and a time-series purchase history for a certain period are extracted, and a statistic is calculated for each of them. The time-series search statistical data of the target new product (NP-A1) and the time-series purchase statistical data of the target new product (NP-A1) -B1) is generated and stored.
Here, the fixed period from the time of release is a period of one to three months, and is adjusted depending on the product. The statistics are calculated from the time-series correlation coefficient, the first-order regression coefficient, the first-order regression constant term, the time-series correlation coefficient periodic function, the number of daily searches, and the number of daily purchases. Is the same as the statistic described in the section, and a description thereof will be omitted.

In the eighth processing mode, the time-series search correlation data (NP-) of the target new product for a certain period from the release time of the already-released product (the same period as the release period of the target new product).
A1), Time-series purchase correlation data of the target new product (NP-B
The statistics of the target new product stored in 1) are compared, and the product having the highest statistical similarity is selected. First-order regression of the purchase transition after a certain period from the release point of the selected already-released product is performed, and a first-order regression coefficient is applied to the target new product to derive a purchase prediction curve of the new product.

(6) User / Basket Analysis Method The user / basket analysis method is based on demographic data of each user 120 (user attributes; age, gender, residence, annual income, married / unmarried, occupation, etc.). Derive typical preferences to be derived. That is, a basket analysis method based on similar user attributes. This allows
It is possible to specify a product to be promoted in a limited attribute area. The feature of this method is that the derived favorite product group is not aggregated into a specific product category, and thus serves as a good guideline for packaging a plurality of products called cross-sale.

The user basket analysis means that embodies this method is to derive a representative preference from the demographics of the user 120 and to select products according to the derived preference.

FIG. 15 shows a user basket analysis method, which is a first processing mode for constructing a history data layer including product search and history data and product search and purchase history data of the user 120. A second processing mode for constructing a statistical data layer including search statistics and purchase statistics of products, and search and purchase statistical data of the user 120;
A third processing mode, a ninth processing mode, and a tenth processing mode for constructing a ranking data layer including the ranking data of the user 120.
And the processing form of (1).

The first processing mode is the Web product sales site 11
0 has product search history data (I-
A), the product purchase history data (IB) is fetched, and a product search and purchase history data as viewed from the user 120 are constructed, and a search and purchase history data layer of the user 120 as viewed from the product 120 is created.

The history data layer is composed of product search history data (C-A1) of all users 120 and product purchase history data (C-B1) of all users 120.

Product search history data (C for all users 120)
-A1) collects the product search history data (IA) from the Web product sales site 110 and creates a search history of all products centered on the user 120. The product purchase history data (C-B1) of all users 120 is obtained by totaling the product purchase history data (IB) from the web product sales site 110 and creating a purchase history of all products centered on the user 120. It is.

As described above, in the first processing mode, for the product searched and purchased by the user 120, first, a data layer for searching and purchasing the product for each user 120 is constructed.

The second processing mode is based on the search statistical data (C-A2) of each user 120 existing for each user from the data of the history data layer generated in the first processing mode. It consists of purchase statistics data (C-B2) of each existing user 120.

Each user 12 existing for each user 120
The search statistical data (C-A2) of 0 is the history data (C-A1) sorted in the first processing mode,
It is sorted for each 0, and a statistic including the number of time-series searches of each user 120 is added.

Each user 12 existing for each user 120
The purchase statistics data (C-B2) of No. 0 is obtained by combining the history data (C-B1) sorted in the first processing mode with each user 12.
It is further assigned for each 0, and a statistic including the number of time-series purchases of each user 120 is added.

The statistics for obtaining each of the statistical data are as follows:
The correlation cycle, the correlation duration, the correlation level, and the like are calculated based on the time series correlation coefficient, the first-order regression coefficient, and the time series correlation coefficient periodic function. .

Next, in the third processing mode, each user 120 is set next to the correlation cycle, correlation duration, and correlation level of the time-series search correlation of each user 120 based on the obtained data of the statistical data layer. Points are scored according to the first rule shown.

First rule: score = (time-series search correlation average value / correlation period) × the number of purchases in a unit period

The correlation period, correlation duration, and correlation level of the time-series purchase correlation of each user 120 are set for each user 120.
20 is scored according to the second rule.

Second rule: score = (time-series purchase correlation average value / correlation period) × the number of searches within a unit period

The sum of the above two scores is used as the total score of each user 120, and the user is ranked within the range of the scores. The result is ranked by the product ranking data (P) and the user ranking data (C) in the ranking data layer. And user rank data (C) is used.

The ninth processing mode is based on the user ranking data (C) derived in the third processing mode.
User 120 corresponding to the demographic designated by 0
Is extracted. The extracted result is stored in a designated user group (CD) for each demographic, which exists for each demographic.

The tenth processing mode accesses the user group (CD) for each designated demographic, which exists for each user group and is created in the ninth processing mode. The preference item ranking data (PD) corresponding to each user demographic existing for each user group is stored.

[0130]

As described above, the network marketing system according to the present invention ranks users, predicts product sales, and the like based on purchase histories and search histories of users who access the Web product sales site and purchase products. By doing so, it is possible to obtain a product trend based on the user's preference, gender, etc. without actually purchasing the product, and it is possible to always keep the products desired by the user on the WEB.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of a network marketing system of the present invention.

FIG. 2 is an overall configuration diagram of a four-layer structure of the marketing research means.

FIG. 3 is a block diagram showing a database on the user side.

FIG. 4 is a block diagram showing a ranking method in the Web application server.

FIG. 5 is a formula for calculating a simultaneous series correlation coefficient.

FIG. 6 is an index calculated by a time-series correlation coefficient in the statistic.

FIG. 7 is a formula for calculating the same linear regression coefficient.

FIG. 8 is an example showing a gradient in the first-order regression coefficient.

FIG. 9 is a formula calculation formula in the discrete Fourier transform.

FIG. 10 is an example showing a simultaneous series correlation coefficient and a discrete Fourier variable spectrum.

FIG. 11 is a block diagram showing the purchase prediction method.

FIG. 12 is an example showing the number of purchases and an expected purchase range.

FIG. 13 is a block diagram of the product recommendation method.

FIG. 14 is a block diagram of the new product trend method.

FIG. 15 is a block diagram of the user basket analysis method.

[Explanation of symbols]

110; Web product sales site, 120; user, 130;
Marketing research means, 140; web server,
150; Web application server, 160; DB
server

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G06F 17/15 G06F 17/15

Claims (9)

[Claims] 1. A Web product sales site that allows a user to purchase or browse a predetermined product on a network, a user who can access the Web product sales site, and whether or not the user has accessed the Web product sales site. And marketing research means for generating product information and future product sales information from the history data of the purchase of the product and providing the information to the market, and the marketing research means comprises the Web
Means to increase the number of users accessing the merchandise site;
A ranking unit that ranks a product ranking and a user ranking indicating the current product sales, a product sales forecasting unit that predicts the sales status of a product from a user's purchase and access to a product, and a specific product from a user's purchase and access to a product A product recommendation means for selecting and recommending, a new product trend prediction means for predicting a trend of a new product based on a purchase history and an access history of the new product, and a user basket for deriving typical preferences from user attributes. Network marketing system consisting of analysis means. 2. The network marketing system according to claim 1, wherein said network is the Internet. 3. The means for increasing the number of users who access the Web merchandise sales site includes using points as access to a merchandise product catalog provided by the Web merchandise sales site, adding the points to obtain a pay-as-you-go score, The network marketing system according to claim 1, wherein a common prize is provided according to the pay-as-you-go score. 4. A ranking means for assigning a product ranking and a user ranking indicating the best sellers of the current product, based on a user's access history and purchase history of the product for sale provided by the Web product sales site, 2. The network marketing system according to claim 1, wherein the total ranking is calculated and ranked. 5. The product sales prediction means predicts that the product is suitable for sale based on the purchase history of the product to be sold and the access history of the purchased product before purchase. The network marketing system according to claim 1. 6. The product recommendation means extracts a category having a high frequency of access and purchase from the access history and purchase history of the product for sale, and creates a user group for each specific preference from the category to recommend the product. The network marketing system according to claim 1, wherein 7. The new product trend forecasting means extracts a time-series search history and a time-series purchase history of each of the target new products for a certain period from the time of release, and extracts the extracted time-series search history and time-series purchase history. Calculate statistics for each of the histories, compare the calculated statistics with existing statistics,
The network marketing system according to claim 1, wherein a trend of a new product is predicted. 8. The network according to claim 1, wherein the user basket analysis means derives a representative preference from the demographic of the user and selects a product according to the derived preference. Marketing system. 9. The network marketing system according to claim 8, wherein the demographic of the user includes the age, gender, and area of residence of the user.