WO2005078616A1

WO2005078616A1 - System for predicting the number of customers by using bayesian network

Info

Publication number: WO2005078616A1
Application number: PCT/JP2005/001834
Authority: WO
Inventors: Toichiro Yamada
Original assignee: Inter-Db Co., Ltd.
Priority date: 2004-02-13
Filing date: 2005-02-08
Publication date: 2005-08-25
Also published as: JP2005228014A

Abstract

There is provided a system for predicting the number of customers coming to a retail shop or a restaurant by using the Bayesian network. The system for predicting the number of customers by using the Bayesian network includes: an external information input unit for receiving information from out of the system for predicting the number of customers; an experience data storage unit for storing the information received by the external information input unit as experience data; a Bayesian network creation unit for extracting the experience data stored and creating a probability table consisting of an event and a number-of-customers prediction probability distribution corresponding to it, according to the experience data; a Bayesian network calculation unit for calculating the number-of-customers prediction data according to the probability table created and the information indicating the event received from the external information input unit.

Description

Specification

Visitor Prediction System Using Bayesian Network

Technical field

The present invention relates to a system for predicting the number of visitors of a store such as a retail store or a restaurant, and more particularly to a system using a Bayesian network characterized by using a Bayesian network for the prediction. Related to the number of visitors forecast system.

Background art

[0002] For stores such as retail stores and restaurants, predicting the number of visitors is a very important factor. The reason is that the company purchases products and shifts employees (including employees) in accordance with the predicted number of customers, which leads to store operation. If the number of customers is incorrectly predicted, sales opportunities will be lost due to a shortage of products, conversely, costs will increase due to excess products, problems will occur such as a decrease in services due to a shortage of employees, and a rise in costs due to excess employees. I do. For this reason, the accuracy of the number of visitors prediction affects the profit of the store.

[0003] It is important to accurately predict the number of visitors as described above. Conventionally, it is often performed based on the experience and intuition of store managers and site managers! However, if such a reliance on individuals cannot be taken over when a store manager or site manager is replaced, various systematizations as shown in Patent Documents 1 to 4 below are required. Is planned.

Patent Document 1: Japanese Patent Application Laid-Open No. 08-314888

Patent Document 2: Japanese Patent Application Laid-Open No. 2002-24350

Patent Document 3: JP-A-2002-312527

Patent Document 4: JP 2003-114969 A

[0005] Non-Patent Document 1: Yoichi Motomura, "Probability Network and Its Application to Knowledge Information Processing", [online], January 24, 2001, Internet URL:

http: // staff, aist. go. jp / y. motomura / Db / DS. ntml />

Disclosure of the invention

Problems to be solved by the invention

[0006] However, a conventional number of visitors forecast is performed, as represented by Patent Documents 1 to 4 described above. In the case of such a system, a causal relationship between information given as parameters is defined in advance, and the number of customers is predicted based on the causal relationship.

[0007] Therefore, it is inevitable that the parameters for applying the external force must be fixed, and the causal relationship thereof must be carefully defined. If the causal relationship changes, the regulations must be changed again. In the present specification, the causal relationship refers to a relationship in which a past event affects a future event. For example, if the weather is fine, the number of visitors is large, and if the weather is rainy, the number of visitors is small, it can be said that there is a causal relationship between the weather and the number of visitors.

Means for solving the problem

[0008] Thus, the present inventors differ from the above-described conventional number-of-customer forecasts by using a large number of information power statistical methods without having to specify causal relationships between information given as parameters from the outside. By estimating the causal relationship, it is possible to execute the process of predicting the number of customers, and by systematizing it, it is possible to predict the number of customers without using the experience and intuition of store managers and site managers as in the past. Invented a visitor forecast system using Bayesian networks.

[0009] A Bayesian network is a stochastic model with a graph structure in which random variables are represented by nodes, and variables indicating dependencies such as causal relationships and correlations are linked to each other. This is a model represented by a V ヽ acyclic directed graph that has directionality in the direction of the relationship and does not circulate the path through the link (a Bayesian network is detailed in Non-Patent Document 1 above).

[0010] The invention according to claim 1 is a visitor number prediction system using a Bayesian network, which performs visitor number prediction using a Bayesian network, wherein the external information input unit receives information from outside the visitor number prediction system. An experience data storage unit for storing information received by the external information input means as experience data, and extracting the stored experience data, and, based on the experience data, events and the number of visitors corresponding thereto. Bayesian network creation unit that creates a probability table composed of forecast probability distributions, and calculates visitor forecast data based on the created probability table and information indicating events received from the external information input unit. A Bayesian network operation unit that performs And a visitor number prediction data output unit that outputs data.

[0011] As described above, after storing the information received from the external force as empirical data, a probability table is created based on the information and the Bayesian network is calculated, whereby the number of customers can be predicted. As a result, conventionally, it was possible to receive only data for which a causal relationship was previously defined, but even if the causal relationship was not defined, the causal relationship was estimated, and a probability table was automatically created. This makes it possible to forecast the number of customers. In addition, since it is not necessary to define the causal relationship, when the causal relationship is changed as in the related art, it is not necessary to change the system in the visitor forecast system. In particular, since the present invention uses a Bayesian network for estimating the causal relationship, it becomes an input / output of a probability distribution, and can handle not only numerical values but also data such as atmosphere (ambiguous data). In addition to being able to handle non-linear causal relationships such as “weather is dull when the weather is too good”, showing the causal relationships on a network (directed graph) makes it possible to correlate with intuition.

[0012] The invention according to claim 2, wherein the Bayesian network creation unit creates a probability table used in the Bayesian network by performing a multivariate analysis on the extracted empirical data, It is a prediction system.

[0013] When creating a probability table, the empirical data power may be obtained by multivariate analysis.

[0014] The invention according to claim 3, wherein the Bayesian network operation unit extracts the visitor number prediction corresponding to the information indicating the event from the probability table, and weights the information indicating the event in each case. This is a visitor forecast system using Bayesian networks, which calculates the visitor count after weighting and calculates the total.

[0015] The calculation of the number of visitors based on the probability table can be performed as in claim 3.

[0016] In the invention according to claim 4, the external information input unit is a daily report information input unit that receives input of daily report information, a labor information input unit that receives input of labor information, and sells the scale of the number of customers. Sales information input unit that receives input as information and receives input from weather sensors A visitor number prediction using a Bayesian network, including at least one of a weather sensor information input unit and a network information input unit that receives the presence / absence of an event and input of weather prediction information via a network. System.

[0017] The information to be received includes the daily report information, the labor information, the sales information, the weather sensor information, the network information, and the like as described above, but may be any information without being limited thereto. This is because, unlike the conventional system for predicting the number of visitors, processing can be performed by using a Bayesian network without explicitly specifying the causal relationship.

The invention's effect

[0018] With the system for predicting the number of visitors using the Bayesian network of the present invention, various kinds of information can be externally given as parameters. In other words, in the past, the causal relationship between pieces of information was previously grasped on the system designer side, and otherwise, it was impossible to execute the visitor forecast. In addition, it is possible to predict the number of visitors without having to specify causal relationships.

[0019] Furthermore, even if the causal relationship changes due to a change in the situation, it is possible to follow the change without changing the system of the visitor number prediction system. Furthermore, by changing the data input to the Bayesian network operation unit, it is possible to simulate how the number of visitors changes.

[0020] Furthermore, by systematizing in this way, there is no need to rely on the experience and intuition of the store manager or the site manager as in the past, so that no problem occurs even if the manager or the site manager changes. .

Brief Description of Drawings

FIG. 1 is a system configuration diagram showing an example of a system configuration of the present invention.

FIG. 2 is a flowchart showing an example of a process flow of the present invention.

FIG. 3 is an example of experience data stored in an experience data storage unit.

FIG. 4 is an example of a probability table created by a Bayesian network creation unit.

FIG. 5 is a conceptual diagram when a calculation is performed by a Bayesian network calculation unit.

Explanation of reference numerals [0022] 1: A visitor forecast system using Bayesian networks

2: External information input section

2a: Daily report information input section

2b: Labor information input section

2c: Sales information input section

2d: Weather sensor information input section

2e: Network information input section

3: Experience data storage

4: Bayesian network creation department

5: Bayesian network operation unit

6: Visitor count data output section

BEST MODE FOR CARRYING OUT THE INVENTION

FIG. 1 shows a system configuration diagram showing an example of a system configuration of a visitor number prediction system 1 (hereinafter, a visitor number prediction system 1) using the Bayesian network of the present invention. The visitor number prediction system 1 includes an external information input unit 2, an experience data storage unit 3, a Bayesian network creation unit 4, a Bayesian network operation unit 5, and a visitor number prediction data output unit 6.

[0024] The external information input unit 2 is a means for receiving an input of information as a parameter when predicting the number of visitors from outside the number-of-customers prediction system 1. Based on the information input by the external information input unit 2, a causal relationship between the parameters is created by a Bayesian network creation unit 4, which will be described later, and a probability table is created. As an example of the external information input unit 2, this specification describes a case where a daily report information input unit 2a, a labor information input unit 2b, a sales information input unit 2c, a weather sensor information input unit 2d, and a network information input unit 2e are provided. However, means for receiving input of other information may be provided.

[0025] The daily report information input unit 2a receives a daily report database (not shown) input by the store manager, a direct mail flyer, and information indicating the size of the sales promotion activity, and stores the experience data storage unit 3 and the Bayesian network. This is a means for inputting to the arithmetic unit 5. Here, as the data format to be input, the case where sales promotion activities that increase the number of customers are performed is indicated by “1”. [0026] In addition to mechanically reading information related to sales promotion activities from the daily report (reading from a predetermined column of the daily report), a means for performing Japanese syntax analysis is provided in the daily report information input section 2a, so that the daily report information is input. After receiving the information, the above-mentioned information may be obtained by performing Japanese syntax analysis and extracting information indicating the scale of the sales promotion activity from the parsing. When Japanese parsing is used, various events can be input, not limited to sales promotion activities. Of course, the above information may be obtained by performing a syntax analysis of a language other than Japanese.

[0027] In addition to the information described above, the information received by the daily report information input unit 2a includes the quantity of flyers, the quantity of leaflets described by type, the quantity of leaflets described by product, the quantity of direct mail, and the amount of direct mail by type. Includes the amount of e-mail, the amount of direct mail for each product, the presence or absence of direct mail for each customer, the degree of discount, the degree of discount by type, the degree of discount by product, and so on.

[0028] The labor information input unit 2b receives information indicating the scale (number of people, number of groups, etc.) of the arranged clerks from a personnel database (not shown) of the store, and inputs the information to the experience data storage unit 3. is there. The input data format is indicated by “1” when the number is larger than the average number of clerks.

[0029] In the present specification, although one data is input, a plurality of data may be input by classifying according to factors such as a time zone and a sales floor.

[0030] The information received by the labor information input unit 2b includes, in addition to the above, the number of clerks, the number of clerks by department, and the attendance / department department of clerks.

[0031] The sales information input unit 2c is a means for receiving the scale of the number of customers (the number of people) from the accounting database (not shown) of the store and inputting it to the experience data storage unit 3. The input data format here is based on the average number of visitors, based on the average number of visitors: “0” if it is less than 80%, “1” if it is 80-100%, “2” if it is 100-120%, 120 If it exceeds%, it is indicated by "3".

[0032] In the present specification, four levels of data format may be used. Different levels may be used, or more detailed classification may be made based on factors such as a time zone and a sales floor in stores other than stores. .

The information received by the sales information input unit 2c includes, in addition to the above, total sales, total sales volume, total sales customers, sales by product, sales volume by product, sales customers by product, shelves Sales by position, sales volume by shelf position, number of customers by shelf position, sales by customer, sales volume by customer, There is presence / absence of visits for each customer.

[0034] The weather sensor information input unit 2d is means for receiving weather information from a weather sensor provided outside the number-of-customers prediction system 1 and inputting the information to the experience data storage unit 3. In the data format to be entered, “1” indicates weather that is expected to increase the number of visitors, for example, comfortable weather (fine weather, etc.).

[0035] In this specification, a single data input force may be applied at a plurality of levels, may be classified by factors such as a time zone, and may be inputted with temperature, wind speed, sunshine, or the like. good.

[0036] The information received by the weather sensor information input unit 2d includes, in addition to the above, temperature, humidity, sunshine, amount of ultraviolet rays, wind direction, wind speed, air pressure, air smell, and the like.

[0037] The network information input unit 2e is used to transmit event information and weather forecast information to be held around the store to a network (preferably a closed LAN such as a power LAN that is the Internet) to which the visitor forecast system 1 is connected. (May be a network) and input it to the empirical data storage unit 3 and the Bayesian network operation unit 5. The input data format indicates the direction in which the number of visitors is expected to increase, for example, event holding or comfortable weather (fine weather, etc.) with “1”.

[0038] In the present specification, there are five different types of data, namely, the presence / absence of today's event, the presence / absence of tomorrow's event, today's weather, forecast of tomorrow's weather, and the weather of tomorrow's day. Is also good. Further, as for the information of the event of today, the event information received on the previous day may be stored and used. Similarly, as for today's weather, the information of the weather forecast received the day before may be stored and used. For information on the weather today, the information received by the weather sensor information input unit 2d may be used.

[0039] In addition to the above information, the information received by the network information input unit 2e includes, as event information, whether or not there is a holiday, whether or not a holiday, whether or not there is a holiday, whether or not there is an exhibition, whether or not there is an athletic meet, and whether or not there is a cultural festival. And seminars.

[0040] In addition to the above, a calendar date information input unit (not shown) is provided as information to be received by the external information input unit 2, and the month, day, week, day of the week, lunar calendar, six days of the week, month phase, The sunrise / sunset time may be received, or a product information input unit (not shown) may be provided to advertise the mass media for each product, introduce the mass media for each product, and store each product at the store. Advertising volume, product-specific You can receive the amount of topics on one net!

The experience data storage unit 3 is a means for storing information received from the external information input unit 2 as experience data.

As described above, when the external information input unit 2 is provided with the daily report information input unit 2a, the labor information input unit 2b, the sales information input unit 2c, the weather sensor information input unit 2d, and the network information input unit 2e , Daily report information (“1” when sales promotion activities that increase the number of customers are performed), labor information (“1” when the number of clerks is more than average), and sales information (less than 80% based on the average number of customers). When turning, set to "0", when 80-100% is "1", when 100-120% is "2", when it exceeds 120% "3"), weather sensor information (for comfortable weather) "1") and network information ("1" if there is an event today, "1" if the weather forecast is comfortable today, "1" if the weather forecast is comfortable tomorrow).

If the Bayesian network creation unit 4 described below has a storage function, the external information input unit 2 may directly input and store the data into the Bayesian network creation unit 4. In that case, Need not be provided with the experience data storage unit 3.

The Bayesian network creation unit 4 extracts the experience data stored in the experience data storage unit 3, creates a Bayesian network probability table by a method such as multivariate analysis using the extracted experience data, and performs Bayesian network calculation. This is a means for outputting to section 5.

[0045] The probability table is a table showing the probability value (probability distribution) of the number of visitors in a certain event. Here, the event is information (for example, daily report information, labor information, sales information, weather sensor information, network information) received by the external information input unit 2 from outside the number-of-customers prediction system 1. For example, if daily report information is `` 1 '', labor information is `` 0 '', sales information is `` 0 '', weather sensor information is `` 0 '', and network information is all `` 0 '', that is, customers who do only sales promotion activities This is a table in which the numbers (either as a ratio to the average number of visitors or as an approximate number of visitors) may be shown as probability values (probability distribution).

In order to create a Bayesian network probability table from empirical data, a known multivariate analysis method can be used. For example, a Bayesian network construction support system (Bayesian Network Construction) sold by Mathematical Systems Co., Ltd.

System: BavoNet) (http://www.msi.co.jp/BAYONET) and Hugin Expert HUGIN Explorer (

http://www.hugin.com/Products_Services/Products/Commercial/Explorer/) can be used.

The Bayesian network operation unit 5 outputs the number-of-customers prediction data to the number-of-customers prediction data output unit 6 based on the probability table created by the Bayesian network creation unit 4 and the information received from the external information input unit 2. Means.

[0048] In the case of the above example, the daily report information ("1" when a sales promotion activity that increases the number of customers is performed) and the network information are obtained from the daily report information input section 2a and the network information input section 2e of the external information input section 2. (If there is tomorrow's event, "1" if the weather forecast for tomorrow is comfortable, "1" if the weather forecast for tomorrow is comfortable, then "1"), and output tomorrow's visitor forecast data. However, for example, when data that affects the number of visitors tomorrow can be obtained from the daily report information, such information may be input.

The visitor number prediction data output section 6 is means for outputting the visitor number prediction data output by the Bayesian network operation section 5 to the outside of the visitor number prediction system 1. At the time of this output, the number of visitors forecast data may be output as it is, or the number of probabilities may be calculated and output.

Next, an example of a process flow of the visitor number prediction system 1 of the present invention will be described with reference to a flowchart of FIG. 2 and a system configuration diagram of FIG.

The visitor number prediction system 1 using the Bayesian network also receives an input of information as a parameter when the external information input unit 2 predicts the number of customers with respect to the external force of the system (S100). The information received at this time is not limited to information such as the weather and the number of customers, and the time zone and the number of visitors as in the past, for which the causal relationship is empirically determined. Good. That is, information can be arbitrarily input as a parameter. The received information is transmitted to the experience data storage unit 3 and stored as experience data in the experience data storage unit 3 (S110).

As described above, when the external information input unit 2 is provided with the daily report information input unit 2a, the labor information input unit 2b, the sales information input unit 2c, the weather sensor information input unit 2d, and the network information input unit 2e. The daily report information input unit 2a transmits a sales promotion activity from a daily report database (not shown). Receives information indicating the scale (indicating that a sales promotion activity that increases the number of customers is indicated by `` 1 ''), and the labor information input section 2b uses the personnel database (not shown) to determine the size of the , The number of groups, etc.) (more than the average number of clerks !, the case is indicated by “1”), and the sales information input unit 2c reads the scale of the number of customers from the accounting database (not shown). Information indicating the number of people (based on the average number of visitors, `` 0 '' if the number is below 80%, `` 1 '' if 80-100%, `` 2 '' if 100-120%, 120% The weather sensor information input unit 2d receives the weather information from the weather sensor (for example, weather that is expected to increase the number of visitors, for example, comfortable weather (sunny weather, etc.) as “1”). ), The network information input unit 2e receives event information and weather forecast information (the number of visitors increases) from the network. Direction considered, for example, receives an event held and comfort weather the (sunny, etc.) indicated by "1") is stored in the empirical data storage unit 3.

The various types of information described above are merely examples, and any type of information may be used.

The Bayesian network creation unit 4 extracts the experience data stored in the experience data storage unit 3 and creates a Bayesian network probability table by a known method such as multivariate analysis using the extracted experience data ( S120), and outputs the result to the Bayesian network operation unit 5.

To create a Bayesian network probability table from empirical data, a known method such as multivariate analysis can be used. For example, a Bayesian network construction support system (Bayesian Network Construction) sold by Mathematical Systems Co., Ltd.

System: BayoNet) (http://www.msi.co.jp/BAYONET) and HUGIN Explorer (HUGIN Expert sold by Hugin Expert)

An example in which the Bayesian network creation unit 4 creates a probability table of the empirical data force by multivariate analysis (factor analysis in the following case) in S120 will be described below. In this case, it is assumed that the data stored in the experience data storage unit 3 as the experience data is shown in FIG. Although the input data received from the external information input unit 2 is a probability value, it is represented by a real number. However, in this embodiment, for the sake of explanation, the case of an integer will be described. In FIG. 3, the information of the item indicated as “sales promotion” is the daily report information received from the daily report information input unit 2a, and the information of the item indicated as “many clerks” is The labor information received from the information input unit 2b is shown as “comfortable weather”, and the information of the item is the weather sensor information received from the weather sensor information input unit 2d, and includes “event”, “tomorrow” The information of the items shown as "weather forecast" and "weather forecast for tomorrow" is the network information received from the network information input unit 2e, and the information of the item shown as "number of customers" is the sales information input. This is the sales information received from the unit 2c.

[0058] Here, the Bayesian network operation unit 5 finally calculates and the output number data from the visitor number prediction data output unit 6 is the visitor number prediction. Therefore, the visitor number prediction changes from "0" to "3". This creates a probability table that looks like this. As with sales information, the forecast for the number of visitors is “0” when the number is below 80%, “1” when 80-100%, and 100-120%, based on the average number of visitors. Is shown as “2”, and the case where it exceeds 120% is shown as “3”. Also, when predicting the number of customers using this probability table, daily information (“sales promotion”) and network information (“event”, “tomorrow's weather forecast”, “weather tomorrow's weather forecast”) are input from outside. ) Is used. Therefore, the probability table used in the Bayesian network consists of daily report information, network information, and the corresponding number of visitors forecast. Therefore, when information other than the daily report information and the network information, such as labor information and weather sensor information, is also taken into account as the information to be input by the external force, the probability table includes the daily report information, the network information, the labor information, and the weather sensor. It consists of information and the corresponding number of visitors forecast. Therefore, the probability table created by the Bayesian network creation unit 4 is composed of a certain event (input information used for predicting the number of visitors) and the corresponding probability distribution of the number of visitors prediction.

In the case of the empirical data shown in FIG. 3, since the number of cases where the individual input data is “1” and the case of “0” are the same, the average is calculated for each to obtain the difference. You can see the contribution of the data to the output data. Therefore, in this example, a probability table is created by this method. However, if the input data is as large as 1,000 or 10,000, there is no problem even if other multivariate analysis methods are used.

[0060] First, experience data A "experience data D" that is "sales promotion = 0 (no sales promotion activity)" Calculate the average of the number of visitors for experience data E to experience data H, which is "sales promotion = 1 (performed sales promotion)".

That is,

(Experience data A + Experience data B + Experience data C + Experience data D) / 4 = 3/4 = 0.75 (Experience data E + Experience data F + Experience data G + Experience data H) / 4 = 7/4 = 1.75.

[0061] Next, experience data A, experience data C, experience data with "event = 0 (no event)"? Calculate and compare the average number of visitors for experience data H, experience data B, experience data 0, experience data E, and experience data G, which are “event = 1 (there is an event)”.

That is,

(Experience data A + Experience data C + Experience data F + Experience data H) / 4 = 5/4 = 1.25 (Experience data B + Experience data D + Experience data E + Experience data G) / 4 = 5/4 = 1.25.

[0062] Next, experience data A, experience data, experience data G, experience data H, which is "Tomorrow's weather forecast = 0 (not comfortable weather)", and "Expected tomorrow's weather forecast = 1 (comfortable weather) )), The average number of visitors is calculated and compared with experience data C, experience data D, experience data E, and experience data F.

That is,

(Experience data A + Experience data B + Experience data G + Experience data H) / 4 = 2/4 = 0.5 (Experience data C + Experience data D + Experience data E + Experience data F) / 4 = 8/4 = 2.

[0063] Furthermore, experience data A, experience data C ヽ experience data E, and experience data G, which are "weather forecast of the day after tomorrow = 0 (not comfortable weather)", and "experience data of day after tomorrow = 1 (comfortable weather) The average of the number of visitors is calculated for the experience data B, experience data D, experience data F, and experience data H, which are “weather”.

(Experience data A + Experience data C + Experience data E + Experience data G) / 4 = 6/4 = 1.5 (Experience data B + Experience data D + Experience data F + Experience data H) / 4 = 4/4 = 1 It becomes.

[0064] By using the first item "sales promotion" as a reference value and adding / subtracting the influence of the remaining items, a probability table of the number of visitors forecast for all combinations is created. Figure 4 shows the probability table in this case.

For example, in the case of “sales promotion = 0, event = 0, tomorrow's weather = 0, tomorrow's weather = 0”, 0.75— (1.25-1.25) / 2- (2-0.5 ) /2-(1-1.5) / 2 = 0.25

And the probability distribution (3: 2: 1: 0) of the visitor forecast is “0: 0: 0: 1”.

[0066] Also, in the case of "sales promotion = 0, event = 0, tomorrow's weather = 0, tomorrow's weather = 1,"

0.75— (1.25-1.25) / 2- (2-0.5) / 2 + (1-1.5) /2=-0.25

[0067] Further, in the case of "sales promotion = 0, event = 0, tomorrow's weather = 1, tomorrow's weather = 0,"

0. 75— (1. 25-1. 25) / 2 + (2-0. 5) — (1—1.5) / 2 = 1.75

And the probability distribution (3: 2: 1: 0) of the number of visitors forecast is “0: 1: 0: 0”.

[0068] As described above, when a probability table is created for all combinations of four pieces of information of "sales promotion", "event", "weather forecast of tomorrow", and "weather forecast of tomorrow", as shown in FIG. Become. Note that, in the probability table of FIG. 4, the items indicated by “calculation” are only those that show the calculation process in order to easily explain the present embodiment, and therefore are not included in the actual probability table. Absent.

[0069] Based on the probability table created in S120 and the information received from the external information input unit 2, the Bayesian network operation unit 5 calculates the probability table (S130), and predicts the number of visitors. The data is output to the number-of-customers prediction data output unit 6.

Here, as an example using the probability table of FIG. 4, a case will be described in which the daily report information and the network information input unit 2e also receive the network information from the daily report information input unit 2a. For example, from the daily report information input section 2a, the daily report information is “1” (performing activities to increase the number of customers), from the network information input section 2e, as network information, “0” as the probability of an event tomorrow, Suppose that “0.5” is received as the probability that the prediction is comfortable, and “0.2” is received as the probability that the weather prediction of the day after tomorrow is comfortable.

In this case, since the presence or absence of a sales promotion and the presence or absence of an event are indicated by binary values, “sales promotion” = 1) and 'Event = 0', but the weather tomorrow and the day after tomorrow are indicated by probability values, so 'Tomorrow = 0, Tomorrow = 0', 'Tomorrow = 0, Tomorrow = 1', 'Tomorrow' = 1, the day after tomorrow = 0 ", and the four patterns of" tomorrow = 1, the day after tomorrow = 1 ". Therefore,

The Bayesian network operation unit 5 calculates the probability distribution in each case as follows.

The probability of promotion = 1, event = 0, tomorrow = 0, the day after tomorrow = 0,

(1-0. 5) X (1-0. 2) = 0.4

Probability = 1, Event = 0, Tomorrow = 0, Tomorrow = 1

(1-0. 5) X0. 2 = 0.1

The probability of promotion = 1, event = 0, tomorrow = 1, and tomorrow = 0 is

0.5X (1-0.2) = 0.4

The probability that promotion = 1, event = 0, tomorrow = 1, and tomorrow = 1 = 1

0.5X0.5.2 = 0.1

[0072] Therefore, the Bayesian network operation unit 5 refers to the probability table created by the Bayesian network creation unit 4, calculates the corresponding visitor forecast, and weights it in each case. Can be calculated.

[0073] Specifically, in the probability table created by the Bayesian network creation unit 4 by the Bayesian network calculation unit 5, "sales promotion = 1, event = 0, tomorrow = 0, day after tomorrow = 0" Visitor forecast (3: 2: 1: 0) is “0: 0: 0: 1: 0”, and the number of visitors when “promotion = 1, event = 0, tomorrow = 0, day after tomorrow = 1” The forecast is “0: 0: 1: 0”, and the number of visitors when “promotion = 1, event = 0, tomorrow = 1, tomorrow = 0” is “1: 0: 0: 0” In the case of “sales promotion = 1, event = 0, tomorrow = 1, tomorrow = 1”, the number of visitors is “0: 1: 0: 0”.

[0074] Since the weighting in each case is "0.4", "0.1", "0.4", and "0.1", the prediction of the number of customers after weighting is performed in each case. If “sales promotion = 1, event = 0, tomorrow = 0, tomorrow = 0”, the weighted visitor forecast is “(0X0.4): (0X0.4): (1X0.4): (0X0. 4) = 0: 0: 0.4: 0 ”, and the weighted visitor forecast when“ sales promotion = 1, event = 0, Akira = 0, the day after tomorrow = 1 ”is“ (0X0.1) : (0X0.!) :( 1X0 1) :( 0X0. 1) = 0: 0: 0. 1: 0 "and weighted visitor forecast when" promotion = 1, event = 0, tomorrow = 1, tomorrow = 0 " Is "(1X0.4): (0X0.4): (0X0.4): (0X0.4) = 0.4: 0: 0: 0" and "Promotion = 1, Event = 0, Tomorrow = 1, Day after tomorrow" In the case of “= 1”, the weighted visitor forecast is “(0X0.1) :( 1X0.1) :( 0X0.1) :( 0X0.1) = 0: 0.1: 0: 0”.

[0075] Therefore, since the visitor count prediction data to be output is the total value in each of these cases, "0.4

: 0. 1: 0.5: 0 ". Therefore, the Bayesian network operation unit 5 outputs the number-of-customer forecast data calculated in this way to the number-of-customer forecast data output unit 6. FIG. 5 shows a conceptual diagram of the operation in the above example in the Bayesian network operation unit 5.

The guest number prediction data output unit 6 outputs the customer number prediction data received from the Bayesian network operation unit 5 to outside the customer number prediction system 1 (S140). In the case of the above example, “0.4: 0.1: 0.5: 0” is output. Alternatively, the number-of-customer forecast data output unit 6 does not output the probability distribution in this way, but the information meaning that, that is, in this case, `` the probability that the average number of visitors falls below 80% is 0 (0 %), The probability of being 80-100% for the average number of visitors is 0.5 (50%), the probability of being 100-100% for the average number of visitors is 0.1 (10%), and the probability of being For example, the probability of exceeding 120% is 0.4 (40%) ".

[0077] Each means and table in the present invention are only logically distinguished in their functions, and may have the same physical or practical area. It goes without saying that a database and a data file may be used instead of a table, and the description of a table includes a database and a data file.

In implementing the present invention, a storage medium storing software programs for realizing the functions of the present embodiment is supplied to the system, and the computer of the system reads out and executes the program stored in the storage medium. Of course

In this case, the program itself read from the storage medium realizes the functions of the above-described embodiment, and the storage medium storing the program naturally constitutes the present invention. Become. [0080] As a storage medium for supplying the program, for example, a magnetic disk, a hard disk, an optical disk, a magneto-optical disk, a magnetic tape, a nonvolatile memory card, and the like can be used. Further, the program may be downloaded via a network such as the Internet in addition to the recording on the storage medium.

[0081] Further, the functions of the above-described embodiments are not only realized by the computer executing the readout program, and the operating system or the like that runs on the computer and runs on the computer in accordance with the instructions of the program. It goes without saying that a part or all of the processing is performed, and the function of the above-described embodiment is realized by the processing. At this time, a server or the like on the network may perform part or all of the processing.

[0082] Furthermore, the storage medium power is also read from the non-volatile or volatile storage means provided in the function expansion board inserted into the computer or the function expansion unit connected to the computer, and then stored in the memory. Based on the instructions of the program, the arithmetic processing unit or the like provided in the function expansion board or the function expansion unit may perform part or all of the actual processing, and the processing may realize the functions of the above-described embodiments. Of course it is included.

Industrial applicability

[0083] As described above, according to the visitor number prediction system using the Bayesian network of the present invention, external force can also give various information as parameters. In other words, if the causal relationship between the pieces of information is not known in advance by the system designer, it is impossible to execute the visitor forecast. It is possible to predict the number of visitors without having to specify causal relationships.

[0084] Even if the causal relationship changes due to a change in the situation, the change can be followed without changing the system of the visitor number prediction system. Furthermore, by changing the data input to the Bayesian network operation unit, it is possible to simulate how the number of visitors changes.

[0085] Furthermore, by systematizing in this way, unlike the conventional method, it is no longer necessary to rely on the experience and intuition of the store manager or the site manager, so even if the store manager or the site manager changes, a problem occurs.

',

l7C8l00 / S00Zdf / X3d L V 9T98.0 / S00Z OAV

Claims

The scope of the claims

[1] A visitor number forecasting system using a Bayesian network, which predicts the number of visitors using a Bayesian network,

An external information input unit that receives information from outside the number of visitors prediction system,

An experience data storage unit for storing information received by the external information input means as experience data;

A Bayesian network creation unit that extracts the stored experience data and creates a probability table based on the experience data and a probability distribution of the number of visitors corresponding to the event based on the experience data;

A Bayesian network operation unit that calculates visitor count prediction data based on the created probability table and information indicating an event received from the external information input unit; and a visitor count that outputs the calculated visitor count prediction data. A prediction data output unit, and a visitor number prediction system using a Bayesian network, comprising:

[2] The Bayesian network creation unit:

By performing a multivariate analysis on the extracted empirical data, a probability table is created using a Bayesian network.

2. A system for predicting the number of visitors using the Bayesian network according to claim 1.

[3] The Bayesian network operation unit includes:

The number of visitors forecast corresponding to the information indicating the event is extracted from the probability table, the weight in each case of the information indicating the event is calculated, and the number of visitors forecast after the weighting is calculated. Sum,

[4] The external information input unit includes:

A daily report information input unit that receives input of daily report information, a labor information input unit that receives input of labor information, a sales information input unit that receives input of the number of customers as sales information, and a sensor A weather sensor information input section that receives input and an event No, has at least one of a network information input unit that receives input of weather forecast information via a network,