Data generation system and data generation method

By obtaining the data sets of combining thickness and fineness, and generating fine-grained information based on constraints, the problem of difficulty in data collection in large sets is solved and high-reliability data generation is achieved.

CN118922826APending Publication Date: 2024-11-08HAKUHODO INC
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202280093985.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2022-03-24
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

The prior art is difficult to collect data sets of fine-grained information from large sets, especially due to personal information protection and other reasons, which makes it impossible to obtain detailed analytical data sets.

Method used

By acquiring the first data set of coarse-grained size and the second data set of fine-grained size, the third data set is generated based on the constraint condition calculating, and data generation of fine-grained information of the large set is realized.

Benefits of technology

Generating data sets with high reliability fine-grained information from a limited data set solves the problem of data collection difficulties in large sets and provides detailed analysis capabilities.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118922826A_ABST
    Figure CN118922826A_ABST
Patent Text Reader

Abstract

One aspect of the disclosure relates to generating a third data set based on a first data set and a second data set. The first data set expresses a first feature with respect to the first set at a first granularity, and the first data set describes a first feature quantity with respect to a single element group within the first set or a first feature quantity with respect to each of the plurality of element groups. The second data set expresses a second feature with respect to the second set at a second granularity, and the second data set describes a second feature quantity for each of the one or more elements within the second set. The third data set expresses a first feature with respect to the first set at a third granularity, and the third data set describes a third feature quantity for each subdivided population within the first set. The third data set is generated based on a relationship between the first feature quantity and the third feature quantity, and a relationship between the second feature quantity and the third feature quantity.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to a data generation system and a data generation method. Background Art

[0002] In the past, objects were analyzed by data aggregation (for example, see Patent Document 1). For example, a technique is known that displays analysis results on the characteristics of a specified business district by aggregating data in an area corresponding to a business district of a specified radius centered on a store specified by a user.

[0003] Prior art literature

[0004] Patent Literature

[0005] Patent Document 1: Japanese Patent Application Publication No. 2003-167880 Summary of the invention

[0006] Problem that the invention aims to solve

[0007] Generally speaking, it is more difficult to collect a dataset with fine-grained information about a large group than a dataset with fine-grained information about a small group. For example, it is possible to collect a dataset with fine-grained information about a group of people who cooperate with information collection, but it is difficult to collect a dataset with fine-grained information about a large group of people who do not cooperate with information collection. For example, in order to collect detailed information about an individual, it is often necessary to obtain the individual's consent for the sake of personal information protection.

[0008] Therefore, even if one wishes to perform detailed analysis on a large set, in the past there were many cases where the data set required for analysis was not available and therefore detailed analysis could not be performed.

[0009] Therefore, according to one aspect of the present disclosure, it is desirable to provide a technology that can generate a dataset of fine-grained information about a collection from an available limited dataset.

[0010] Technical solutions to the problem

[0011] According to an embodiment of the present disclosure, a data generation system is provided. An acquisition unit is configured to acquire a first data set and a second data set. The first data set expresses a first feature about a first set at a first granularity. The first data set describes a first feature quantity about a single element group or a first feature quantity about each of a plurality of element groups corresponding to the first granularity in the first set.

[0012] The second data set expresses a second feature at a second granularity, and the second feature is a feature of a second set that is the same as the first set and is different from the first feature, or the second feature is a feature of a second set that is smaller than the first set and is the same as or different from the first feature. The second granularity is finer than the first granularity. The second data set describes a second feature quantity of each of one or more elements corresponding to the second granularity in the second set.

[0013] The generating unit is configured to generate a third data set based on the first data set and the second data set. The third data set expresses the first feature of the first set at a third granularity. The third granularity is finer than the first granularity and coarser than the second granularity or the same as the second granularity. The third data set describes a third feature value of each sub-group corresponding to the third granularity in the first set.

[0014] According to one aspect of the present disclosure, for each of the one or more elements in the second set, the second data set may include information of a subdivided group that can identify the corresponding one or more elements. The generating unit is configured to, under the first constraint and the second constraint, infer the third feature quantity of each subdivided group based on the first data set and the second data set and generate the third data set. The first constraint is a constraint condition for inferring the third feature quantity based on the relationship between the first feature quantity and the third feature quantity. The second constraint is a constraint condition for inferring the third feature quantity based on the relationship between the second feature quantity and the third feature quantity.

[0015] According to one solution of the present disclosure, a third data set can be generated by jointly using a first data set having coarse-grained information about a first feature and a second data set having fine-grained information about a second feature, and further using the relationship between the first feature quantity and the third feature quantity and the relationship between the second feature quantity and the third feature quantity. The third data set is a data set that describes a third feature quantity having fine-grained information about the first feature. Therefore, according to one solution of the present disclosure, data having fine-grained information about a set can be generated from limited data that can be obtained.

[0016] According to one aspect of the present disclosure, the first data set may describe a first feature value for each of a plurality of series elements of a specific series. The second data set may describe a second feature value for each of a plurality of series elements of a specific series.

[0017] According to one aspect of the present disclosure, the generating unit may calculate a third feature quantity of each segmented group related to at least one series element in a specific series based on the first data set and the second data set. Examples of the specific series may include a time series and a series related to a place.

[0018] According to one aspect of the present disclosure, the generating unit may estimate the third feature value of each segmented group for each of a plurality of series elements in a specific series.

[0019] According to an aspect of the present disclosure, at least one of the first constraint condition and the second constraint condition may include a constraint condition regarding a difference between series elements of the third feature quantity.

[0020] The third characteristic quantity in the first series of elements tends to show a value close to the third characteristic quantity in the second series of elements according to the physical or conceptual proximity between the first series of elements and the second series of elements. Therefore, by making an inference with reference to the constraint conditions on the difference between the series elements regarding the third characteristic quantity, a third characteristic quantity with high reliability can be inferred.

[0021] According to an embodiment of the present disclosure, the first constraint condition may include the following constraint condition: for each series element in a plurality of series elements, the sum of the third feature quantities of each sub-group in the corresponding series element corresponds to the first feature quantity of all sub-groups in the corresponding series element. By using the above constraint condition for inference, a third feature quantity with high reliability can be inferred.

[0022] According to one solution of the present disclosure, the first data set may record the first feature quantity in time series. The second data set may record the second feature quantity in time series. The generation unit may estimate the third feature quantity of each segment group at at least one time based on the first data set and the second data set.

[0023] According to an embodiment of the present disclosure, the generation unit may calculate the third feature quantity of each segmented group at multiple times. At least one of the first constraint and the second constraint may include a constraint on the time variation of the third feature quantity. By defining the constraint taking into account the possible time variation of the feature quantity, a third feature quantity with high reliability can be calculated.

[0024] According to an embodiment of the present disclosure, the generation unit can calculate the third characteristic quantity of each segmented group at multiple times. The second constraint condition can include the following constraint condition, which is that for each of the multiple times, the third characteristic quantity at the corresponding time depends on the third characteristic quantity at a time earlier than the corresponding time and the second characteristic quantity at the corresponding time in a specific relationship. By calculating the third feature based on the above constraint condition, a highly reliable third characteristic quantity can be calculated for the characteristics of the state at a certain time that is affected by the state at a previous time.

[0025] According to an embodiment of the present disclosure, the generating unit may calculate the third characteristic value of each sub-group at multiple times. The first constraint condition may include the following constraint condition, which is that, for each of the multiple times, the sum of the third characteristic value of each sub-group at the corresponding time corresponds to the first characteristic value of all the sub-groups at the corresponding time.

[0026] According to an embodiment of the present disclosure, the first data set may associate and record each first feature quantity with a corresponding place, and the second data set may associate and record each second feature quantity with a corresponding place. The generation unit may infer a third feature quantity of each segmented group in at least one place based on the first data set and the second data set.

[0027] According to an embodiment of the present disclosure, the third data set may describe the third feature quantity of each sub-group in multiple places. The generation unit may calculate the third feature quantity of each sub-group of the multiple sub-groups based on the first data set and the second data set under the first constraint condition, the second constraint condition, and the third constraint condition. The third constraint condition is a constraint condition on the difference of the third feature quantity between multiple places.

[0028] The third feature quantity in the first place tends to show a value close to the third feature quantity in the second place according to the physical proximity between the first place and the second place. Therefore, by making an inference with reference to the constraint condition regarding the difference between the places of the third feature quantity, a third feature quantity with high reliability can be inferred.

[0029] According to one solution of the present disclosure, for each region, the first data set may represent a feature quantity related to the behavioral characteristics of people in the corresponding region as a first feature quantity. For each area smaller than the region, the second data set may represent a feature quantity related to the geographical characteristics of the corresponding area as a second feature quantity. Based on the first data set and the second data set, the generation unit may infer a feature quantity related to the behavioral characteristics of people in the corresponding area as a third feature quantity for each area as a segmented group.

[0030] In the case where there is a correlation between the behavioral characteristics of a person and the geographical characteristics of the corresponding area, according to the above calculation, it is possible to calculate the characteristic quantity of each area with finer information granularity and smaller area than that of the first data set.

[0031] According to an embodiment of the present disclosure, for each first time period, the first data set may represent a feature quantity related to the behavioral characteristics of people in the corresponding first time period as a first feature quantity. For a second time period shorter than the first time period, the second data set may represent a feature quantity related to the geographical characteristics of the corresponding second time period as a second feature quantity. The generation unit may, based on the first data set and the second data set, calculate a feature quantity related to the behavioral characteristics in the corresponding second time period as a third feature quantity for each second time period as a segmented group.

[0032] In the case where there is a correlation between the behavioral characteristics of a person and the geographical characteristics of a corresponding time period, according to the above calculation, it is possible to calculate the characteristic quantity of each time period with finer information granularity and shorter time period than that of the first data set.

[0033] According to an aspect of the present disclosure, at least one of the first constraint condition and the second constraint condition may include an error term related to dispersion of the third feature quantity.

[0034] According to one scheme of the present disclosure, a generation unit searches for appropriate values ​​of design parameters of an inference model based on a first data set and a second data set by using statistical analysis with reference to a probability distribution of dispersion, thereby inferring a third feature quantity that corresponds to the appropriate value of the design parameter and follows the inference model, wherein the inference model is an inference model for inferring the third feature based on the first feature quantity and the second feature quantity, and the inference model follows the first constraint and the second constraint.

[0035] According to the statistical analysis with reference to the probability distribution, a reasonable feature quantity can be estimated from limited data as the third feature quantity. The search example using the statistical analysis includes an example of estimating parameters from the observed data using Bayesian estimation.

[0036] According to one aspect of the present disclosure, a computer program can be provided for causing a computer to realize functions as an acquisition unit and a generation unit in the above data generation system. The computer program can be stored in a non-transitory recording medium readable by a computer.

[0037] According to one aspect of the present disclosure, a data generation method corresponding to the above-mentioned data generation system may be provided.

[0038] According to an embodiment of the present disclosure, a data generation method may be provided, which includes: obtaining a first data set and a second data set, wherein the first data set expresses a first feature about a first set at a first granularity, and the first data set describes a first feature quantity about a single element group or a first feature quantity about each of a plurality of element groups corresponding to the first granularity in the first set, and the second data set expresses a second feature at a second granularity finer than the first granularity, the second feature is a feature about a second set that is the same as the first set and is different from the first feature, or the second feature is a feature about a second set that is smaller than the first set and is the same as or different from the first feature, and the second data set describes a second feature quantity of each of more than one element corresponding to the second granularity in the second set; and generating a third data set based on the first data set and the second data set, wherein the third data set expresses the first feature about the first set at a third granularity that is finer than the first granularity and coarser than the second granularity or at a third granularity that is the same as the second granularity, and the third data set describes a third feature quantity of each subdivided group corresponding to the third granularity in the first set. The data generation method is executed by a computer.

[0039] In the data generation method, the generation may include: under the first constraint condition and the second constraint condition, based on the first data set and the second data set, inferring the third characteristic quantity of each segmented group and generating a third data set, wherein the first constraint condition is a constraint condition for inferring the third characteristic quantity based on the relationship between the first characteristic quantity and the third characteristic quantity, and the second constraint condition is a constraint condition for inferring the third characteristic quantity based on the relationship between the second characteristic quantity and the third characteristic quantity. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a block diagram showing the configuration of a data generation system.

[0041] Figure 2 is a flow chart of a data generation process executed by a processor.

[0042] Figure 3 It is a diagram for explaining the first data set, the second data set, and the third data set in the first embodiment.

[0043] Figure 4 It is a diagram for explaining the first data set, the second data set, and the third data set in the second embodiment.

[0044] Figure 5 It is a diagram for explaining the first data set, the second data set, and the third data set according to the third embodiment.

[0045] Figure 6 It is a diagram for explaining the first data set, the second data set, and the third data set according to the fourth embodiment.

[0046] Figure 7 It is a diagram for explaining the first data set, the second data set, and the third data set according to the fifth embodiment.

[0047] Figure 8 It is a diagram for explaining the first data set, the second data set, and the third data set according to the sixth embodiment.

[0048] Fig. 9 It is a diagram for explaining the first data set, the second data set, and the third data set according to the seventh embodiment.

[0049] Fig.10 It is a diagram for explaining the first data set, the second data set, and the third data set according to the eighth embodiment.

[0050] Fig.11 It is a diagram for explaining the first data set, the second data set, and the third data set in the ninth embodiment.

[0051] Fig.12 It is a diagram for explaining the first data set, the second data set, and the third data set in the tenth embodiment.

[0052] Description of Reference Numerals

[0053] 1…data generation system; 11…processor; 12…memory; 13…storage;

[0054] 15…display; 17…input device; 18…media reader / writer; 19…communication device;

[0055] DS11, DS21, DS31, DS41, DS51, DS61, DS71, DS81, DS91, DS101…the first data set;

[0056] DS12, DS22, DS32, DS42, DS52, DS62, DS72, DS82, DS92, DS102… the second data set;

[0057] DS13, DS23, DS33, DS43, DS53, DS63, DS73, DS83, DS93, DS103…the third data set. DETAILED DESCRIPTION

[0058] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings.

[0059] The data generation system 1 of some embodiments described is configured to generate a third data set based on the first data set and the second data set described below. The first data set is a data set with coarse-grained information that describes a first feature related to a first set of analysis objects. The second data set is a data set with fine-grained information that describes a second feature related to a second set of analysis objects, wherein the second set is smaller than the first set. The third data set is a data set that describes the first feature of the first set of analysis objects with finer information granularity than the first data set.

[0060] The second set may be a subset of the first set of analysis objects, or other sets that are equivalent to the subset and smaller than the first set according to the analysis purpose. Examples of other sets include: a set in which many elements belong to the first set but some elements do not belong to the first set. The second feature may be the same feature as the first feature or other features related to the first feature.

[0061] The data generation system 1 of some other embodiments described is configured to generate a third data set based on the first data set and the second data set described below. The first data set is a data set with coarse-grained information that describes a first feature related to a set of analysis objects. The second data set is a data set with fine-grained information that describes a second feature related to a set of analysis objects, wherein the second feature is different from the first feature. The third data set is a data set that describes the first feature of the set of analysis objects with finer information granularity than the first data set.

[0062] The data generation system 1 is configured by installing a computer program for generating a third data set in a general information processing system. In the first to tenth embodiments described below, specific examples are used to illustrate the configuration of the first data set, the configuration of the second data set, the configuration of the third data set, and the processing content related to the generation of the third data set. Figure 1 The hardware configuration of the data generation system 1 of the first embodiment described above is also applicable to the data generation systems 1 of other embodiments.

[0063] [First embodiment]

[0064] like Figure 1 As shown, the data generation system 1 of the present embodiment includes a processor 11 , a memory 12 , a storage 13 , a display 15 , an input device 17 , a medium reader / writer 18 , and a communication device 19 .

[0065] The processor 11 executes processing in accordance with a computer program stored in the memory 13. The memory 12 includes a RAM and is used as a work area when the processor 11 executes processing in accordance with a computer program. The memory 12 temporarily stores a computer program read from the memory 13.

[0066] The memory 13 stores computer programs and various data. One of the computer programs stored in the memory 13 includes a computer program for generating a third data set. Examples of the memory 13 include a hard disk drive and a solid state drive.

[0067] The display 15 is, for example, a liquid crystal display, and is configured to display various information to the user. The input device 17 includes a keyboard and a pointing device that can be operated by the user, and is configured to input an operation signal from the user to the processor 11 .

[0068] The media reader / writer 18 is configured to read information recorded on a recording medium such as a memory card and to write new information to the recording medium. The communication device 19 is controlled by the processor 11 and is configured to communicate with external devices within a local area network and / or a wide area network.

[0069] When the processor 11 receives an execution command from the user through the input device 17 to execute the data generation process, the processor 11 starts to execute the data generation process according to the computer program. Figure 2 The data generation process for generating the third data set is shown.

[0070] By executing the data generation process, the processor 11 generates a third data set DS13 based on the first data set DS11 and the second data set DS12. The first data set DS11 is a data set that describes the sales of a specific commodity category in the market as a whole. The second data set DS12 is a data set that describes the purchases of a specific consumer group by consumer attributes. The third data set DS13 is a data set that describes the sales of a specific commodity category in the market as a whole by consumer attributes.

[0071] Examples of consumer attributes include demographic attributes of consumers. Examples of purchases and sales by consumer attributes include purchases and sales by gender and age. Gender and age here refer to demographic attributes of consumers defined by a combination of gender and age, such as males aged 10 to 19, females aged 10 to 19, males aged 20 to 29, females aged 20 to 29, males aged 30 to 39, and females aged 30 to 39. Examples of purchases include purchase amounts and purchase quantities.

[0072] When the data generation process is started, the processor 11 obtains the first data set DS11 and the second data set DS12 (S110, S120) specified by the user through the input device 17. The processor 11 can obtain the first data set DS11 and the second data set DS12 from the memory 13, or from the recording medium via the media reader / writer 18, or from an external device via the communication device 19, for example.

[0073] Specifically, the first data set DS11 obtained in this embodiment is a time series data set about market sales conditions, which summarizes the sales conditions of a specific commodity category in the overall market by date. The overall market sales conditions correspond to the purchase conditions of all consumers.

[0074] like Figure 3 As shown, the exemplary first data set DS112 has the following records, wherein for each date, the overall sales situation of the market on the corresponding date and the date are recorded. In S110, the processor 11 can obtain the data set before aggregation, and generate the following data set according to the obtained data set: Figure 3 The aggregated first data set DS11 is shown.

[0075] The second data set DS12 obtained in this embodiment is generated based on the purchase history records of a specific consumer group among all consumers who cooperate in collecting purchase history records. The purchase history records can be collected through monitoring surveys.

[0076] Specifically, the second data set DS12 is data on purchases related to specific commodity categories of specific consumer groups, summarized by consumer attributes and on a daily basis, and is time series data on purchases divided by consumer attributes (for example, by gender and age group).

[0077] like Figure 3 As shown, the exemplary second data set DS12 has the following records, wherein for each combination of date and consumer attribute, the purchase situation of the consumer group with the corresponding consumer attribute on the corresponding date, the date, and the consumer attribute information are recorded. In S120, the processor 11 can obtain the data set before aggregation, and generate the data set according to the obtained data set. Figure 3 The aggregated second data set DS12 is shown.

[0078] The processor 11 generates a third data set DS13 by performing analysis processing based on the acquired first data set DS11 and second data set DS12 ( S130 ).

[0079] The third data set DS13 generated in this embodiment is data that records the estimated sales value of a specific commodity category by consumer attribute and by day for all consumers not limited to a specific consumer group. That is, the third data set DS13 is time series data on market sales divided by consumer attribute.

[0080] like Figure 3 As shown, the exemplary third data set DS13 has the following records, wherein for each combination of date and consumer attribute, there are recorded the estimated value of the purchase situation of the consumer group with corresponding consumer attributes in the market as a whole on the corresponding date, the date, and the consumer attribute information.

[0081] The processor 11 outputs the third data set DS13 generated in S130 to an output destination specified by the user through the input device 17 (S140). For example, the processor 11 can output to the memory 13, or to a recording medium via the media reader / writer 18, or to an external device via the communication device 19. The processor 11 can also cause the display 15 to display the output third data set DS13. Thereafter, the processor 11 ends Figure 2 The data generation process shown.

[0082] Next, the analysis process (S130) is described in detail. In the analysis process, the third data set DS13 is generated using Bayesian statistics. According to the present embodiment, the analysis process includes a parameter estimation process using a Markov Chain Monte Carlo (MCMC) method.

[0083] To explain the details of the processing using mathematical expressions, let variable y t Assigned to the daily sales in the first data set DS11. t is the time index in days. That is, y t Represents the sales situation on the date corresponding to time t.

[0084] The variable x mt The purchase situation of each consumer attribute assigned to each day in the second data set DS12. m is the index of the consumer attribute. That is, x mt Represents the purchase situation of the consumer group with the corresponding consumer attribute m on the date corresponding to time t.

[0085] Set the variable u mt The purchase situation of each consumer attribute assigned to each day in the third data set DS13. That is, u mt Represents the sales situation of the consumer group with the corresponding consumer attribute m on the date corresponding to time t.

[0086] In this embodiment, the parameter estimation process is performed under the following constraint conditions.

[0087] <First constraint>

[0088] Sales u by consumer attributes at time t mt The sum of ∑u mt At the same time t, the sales situation y t is within the error range of the benchmark.

[0089] <Second constraint>

[0090] Sales situation u at time t and consumer attribute m mt The sales situation u of consumer attribute m at the previous time (i.e., the previous day) at time t-1 mt-1 and the purchase situation x related to consumer attribute m at time t mt There is a linear relationship between them. The constraint condition includes the sales situation u mt time-varying constraints.

[0091] In order to estimate the parameters according to the first constraint and the second constraint, a simulation model can be established according to the following mathematical formulas (1) and (2). However, the index m of the consumer attribute takes a value from 1 to M.

[0092] [Number 1]

[0093]

[0094] According to mathematical formula (2), u mt The definitions are as follows.

[0095] [Number 2]

[0096] u mt =α+u mt-1 ρ+x mt β+e mt , e mt ~N(0,ψ 2 )…Eq.(3)

[0097] The error term ε in equation (1) t It means that the center is 0 and the variance is σ 2 The normal distribution N(0,σ 2 ) is the probability distribution of u. mt The scattered error term e mt It means that the center is 0 and the variance is ψ 2 The normal distribution N(0,ψ 2 )’s probability distribution.

[0098] According to this model, the parameter of the estimated object is u mt, α, ρ, β, σ 2 , 2 When estimating parameters through MCMC, it is possible to predict u in advance. mt , α, ρ, β, σ 2 , 2 The prior probability distribution (hereinafter referred to as "prior distribution") is set as follows.

[0099] <α, ρ, β, σ 2 , 2 Prior distribution of>

[0100] Assume a conjugate prior distribution that is designed to be essentially uninformative. For example, assume that α, ρ, and β have a normal distribution with a large variance, and σ 2 and ψ 2 Sets the inverse gamma distribution parameterized to 1.

[0101] mt Prior distribution of>

[0102] The following prior distribution (ie, normal distribution) was set following the model.

[0103] [Number 3]

[0104] u mt ~N(α+u mt-1 ρ+x mt β,ψ 2 )…Eq.(4)

[0105] For parameter estimation based on MCMC, for example, the Gibbs sampling method can be applied. According to this parameter estimation, α, ρ, β, σ as design parameters are searched through statistical analysis with reference to probability distribution. 2 , 2 At the same time, we search for a reasonable u mt Through this search (i.e., Bayesian inference), we obtain u mt The posterior probability distribution of (hereinafter referred to as “posterior distribution”).

[0106] Processor 11 is capable of calculating u mt The posterior mean or median of the posterior distribution is taken as u mt The estimated value of the third data set DS13 can be generated, and the third data set DS13 records the daily sales situation u divided by consumer attributes. mt The estimated value of .

[0107] Mathematical formula (2) can be generalized to the following mathematical formula (5) having an arbitrary order K with respect to time. The order K may be an integer value greater than or equal to 1.

[0108] [Number 4]​

[0109]

[0110] In this case, u can be set according to the following mathematical formula (6): mt The prior distribution of is obtained and the parameters are estimated.

[0111] [Number 5]

[0112]

[0113] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to all consumers, the "second set" corresponds to a specific consumer group, the "first feature" corresponds to sales conditions, and the "second feature" corresponds to purchase conditions.

[0114] According to the data generation system 1, the following effects can be obtained. That is, when only a data set with coarse-grained information can be obtained as a data set describing the characteristics of a set of analysis objects, a data set with fine-grained information describing the characteristics of a set smaller than the analysis object can be used as a basis to generate a data set with fine-grained information describing the characteristics of a set of analysis objects, and this data set has high reliability from a statistical point of view.

[0115] For example, when it is impossible to obtain a data set describing the sales situation of all consumers by consumer attributes, it is possible to generate a data set describing the sales situation of all consumers by consumer attributes with fine-grained information based on the purchase history of a specific consumer group obtained through monitoring surveys.

[0116] In this embodiment, Bayesian statistics is used in particular, and as can be understood from equations (2) and (3), time series information is also used to calculate the sales situation u mt Therefore, the estimated value has a higher reliability.

[0117] As a comparative example, an example of correcting the summary value of a specific consumer group by a coefficient based on the scale ratio between the specific consumer group and the consumer set of the market as a whole can be considered. However, this comparative example includes the assumption that the attribute distribution in the specific consumer group is the same as the attribute distribution of the market as a whole.

[0118] The assumptions of the comparative examples are usually significantly different from the facts. Therefore, in the comparative examples, even if a data set with fine-grained information is generated, its reliability is low. In contrast, according to the present embodiment using Bayesian statistics, a highly reliable data set with finer information granularity than before can be generated.

[0119] Various modifications may be made to this embodiment. For example, the third data set DS13 may not need to describe the sales situation of each consumer attribute category that is the same as the second data set DS12. That is, the third data set DS13 may be a data set that describes the sales situation of each attribute group including more than two consumer attributes. The estimated value of the sales situation of each attribute group can be calculated by the sum of the estimated values ​​of the sales situation of each consumer attribute belonging to the group.

[0120] For example, when the second dataset DS12 describes the purchase situation divided by gender and age group, the third dataset DS13 can be generated as a dataset that describes the sales situation divided by age group regardless of gender.

[0121] The third data set DS13 may be data describing the sales situation for each segment group with the same or greater granularity as the second data set DS12. The segment groups may be defined by classifying the set by the same division as the second data set DS12, or by a division coarser than the second data set DS12 and finer than the first data set DS11.

[0122] In addition, in this embodiment, the third data set DS13 is generated by estimating the sales of the entire market by segmented groups, but the technical idea of ​​this embodiment can also be applied to estimating the sales of one or more retailers of interest by segmented groups instead of estimating the market as a whole.

[0123] In this case, the first data set DS11 may be a data set that describes the sales of one or more retailers on a daily basis. The second data set DS12 may be a data set that describes the purchases of a specific consumer group corresponding to a part of the customer group of one or more retailers on a date and for each combination of consumer attributes.

[0124] The third dataset DS13 may be a dataset describing the sales of one or more retailers by consumer attributes on a daily basis. In other words, the second dataset DS12 may be a dataset created based on the purchase history records of another consumer group that does not completely overlap with the consumer set corresponding to the first dataset DS11.

[0125] [Second Embodiment]

[0126] Next, the data generation system 1 of the second embodiment will be described. As described above, the hardware configuration of the data generation system 1 of the second embodiment and its subsequent embodiments is the same as that of the data generation system 1 of the first embodiment. In the second embodiment and its subsequent embodiments, the data set to be processed is different from the data set of the first embodiment. In the second embodiment, when an execution command is input, the processor 11 starts to execute the execution command as in the first embodiment. Figure 2 The data generation process shown.

[0127] The processor 11 is configured to generate a third data set DS23 based on the first data set DS21 and the second data set DS22 by executing the data generation processing. The first data set DS21 is a data set about the sales situation of a specific commodity category in the market. The second data set DS22 is a data set about the sales situation of a specific commodity category described based on the sales situation records of one or more limited retailers, divided into commodities. The third data set DS23 is a data set that describes the sales situation of a specific commodity category in the market by commodity.

[0128] When the data generation process starts, the processor 11 acquires the first data set DS21 and the second data set DS22 designated by the user through the input device 17 ( S110 , S120 ).

[0129] The first data set DS21 obtained in this embodiment is a time series data set about market sales conditions, which summarizes the sales conditions of specific commodity categories in the market on a daily basis. Figure 4 As shown, the exemplary first data set DS21 has the following records, in which the sales situation of a specific commodity category in the market on a corresponding date and the date are recorded on a daily basis.

[0130] The second data set DS22 is data that summarizes the sales situation of each commodity in a specific commodity category on a daily basis based on the sales situation records accumulated in the POS system of more than one limited retailer, and the second data set DS22 is time series data that describes the sales situation of each commodity. Hereinafter, the sales situation is referred to as POS sales situation. For example, when the specific commodity category is "beer", the sales situation of each commodity corresponds to the sales situation of each brand of beer.

[0131] like Figure 4 As shown, the exemplary second data set DS22 has the following records, wherein for each combination of date and commodity, the POS sales of the corresponding commodity on the corresponding date and the date and commodity information are recorded.

[0132] The processor 11 generates a third data set DS23 by performing an analysis process based on the acquired first data set DS21 and second data set DS22 ( S130 ), and outputs the generated third data set DS23 ( S140 ).

[0133] The third data set DS23 generated in this embodiment is time series data describing the market sales situation of a specific commodity category by commodity. Figure 4 As shown, the exemplary third data set DS23 has the following records, in which, for each combination of date and commodity, the estimated value of the sales situation of the corresponding commodity in the market on the corresponding date and the date and commodity information are recorded.

[0134] Next, the details of the analysis process (S130) will be described. t Assigned to the daily sales in the first data set DS21. As in the first embodiment, t is the index of the time corresponding to the date, y t Represents the market sales of a specific commodity category on the date corresponding to time t.

[0135] The variable x mt The POS sales of each commodity in the second data set DS22 are assigned to each day. m is the index of the commodity. That is, x mt Indicates the POS sales of the corresponding product m on the date corresponding to time t.

[0136] In addition, the variable u mt The sales of each commodity per day in the third data set DS23. mt Represents the market sales situation of the corresponding commodity m on the date corresponding to time t.

[0137] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0138] <First constraint>

[0139] Sales u by product at time t mt The sum of ∑u mt At the same time t, the sales situation y t is within the error range of the benchmark.

[0140] <Second constraint>

[0141] The sales of product m at time t is u mt The sales situation u of product m at the previous time (i.e., the previous day) at time t-1 mt-1 and the POS sales x of product m at time t mtThere is a linear relationship between them.

[0142] From the similarity between the first constraint and the second constraint and the first embodiment, it can be understood that according to the present embodiment, a model following the above mathematical formulas (1) and (2) is defined, and a priori setting distribution is set in the same manner as in the first embodiment, so that u can be calculated by performing parameter estimation using MCMC. mt The posterior distribution of .

[0143] The processor 11 can calculate u based on mt The posterior distribution of is used to generate the third data set DS23, which records the daily sales situation u divided by commodity in the market. mt The third data set DS23 may also be generated as a data set that describes the sales status of each segment group including two or more commodities in a specific commodity category.

[0144] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to a set of commercial transactions in the market, and the "second set" corresponds to a set of commercial transactions recorded in the POS system of one or more retailers. The "first feature" corresponds to the sales situation, and the "second feature" corresponds to the POS sales situation.

[0145] [Third Embodiment]

[0146] According to the third embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Figure 5 The first data set DS31 about route usage of all users is shown. In S120, the processor 11 also obtains a second data set DS32 about route usage of a specific user group divided by user attributes. The route usage divided by user attributes can be, for example, route usage divided by gender and age.

[0147] The processor 11 generates and outputs a third data set DS33 ( S130 , S140 ) based on the acquired first data set DS31 and second data set DS32 . The third data set DS33 ( S130 , S140 ) describes the route usage of all users divided by user attributes.

[0148] Specifically, the first data set DS31 obtained in this embodiment is a time series data set that summarizes the usage (for example, usage time) of each route by all users on a daily basis. Figure 5As shown, the exemplary first data set DS31 has the following records, wherein for each date and route, the usage amount (eg, total usage time) of all users on the corresponding date and date and route information are recorded.

[0149] The second data set DS32 is a time series data set that summarizes the usage of each route by a specific user group according to date and user attributes. The specific user group can be a set of users among all users who cooperate in collecting usage history records.

[0150] like Figure 5 As shown, the exemplary second data set DS32 has the following records, wherein for each combination of date, route, and user attribute, the usage of the corresponding route by the user group with the corresponding user attribute on the corresponding date as well as the date, route information, and user attribute information are recorded.

[0151] The third data set DS33 is a time series data set that records the estimated values ​​of the usage of each route by all users divided by user attributes every day. Figure 5 As shown, the exemplary third data set DS33 has the following records, wherein for each combination of date, route, and user attribute, there are recorded the estimated value of usage of the corresponding route by the user group with the corresponding user attribute on the corresponding date, as well as the date, route information, and user attribute information.

[0152] Here, the variable y nt The usage amount assigned to each combination of date and route in the first data set DS31. n is the index of the route. That is, y nt Indicates the usage of route n by all users on the date corresponding to time t.

[0153] The variable X nmt The usage amount allocated to each combination of date, route, and user attribute in the second data set DS32. m is the index of the user attribute. That is, x nmt It represents the usage of route n by the user group with corresponding user attribute m on the date corresponding to time t.

[0154] Set the variable u nmt The usage amount allocated to each combination of date, route, and user attribute in the third data set DS33. nmt It represents the usage of route n by the user group with corresponding user attribute m on the date corresponding to time t.

[0155] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0156] <First constraint>

[0157] Usage u of route n by user attributes at time t nmt The sum of ∑u nmt is within the error range based on the usage ynt of route n by all users at the same time t. nt When greater than zero, the sum ∑u nmt Also greater than zero.

[0158] <Second constraint>

[0159] For each combination of route n and user attribute m, the usage u at time t nmt The rate of increase u relative to time t-1 of the previous day nmt / u nmt-1 At the observed usage x nmt The rate of increase x nmt / x nmt-1 is within the error range of the benchmark.

[0160] <Third constraint>

[0161] For each combination of route n and user attribute m, the usage u at time t nmt The usage u at time t-7 a week ago nmt-7 is within the error range of the benchmark.

[0162] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. Error ε nt ,σ1 t ,σ2 t All obey normal distribution. The index m of user attribute takes values ​​from 1 to M.

[0163] [Number 6]

[0164]

[0165] u nmt =u nmt-7 +σ2 t

[0166] The processor 11 performs parameter estimation based on MCMC according to the model, thereby calculating the usage u nmt The posterior distribution of , and can generate a third data set DS33, which records the usage u divided by daily routes and user attributes nmt The third data set DS33 may also be generated as a data set that describes the usage amount divided by date and route for each segment group including two or more user attributes.

[0167] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to the set of all users, the "second set" corresponds to the set of a specific user group, and the "first feature" and the "second feature" correspond to the usage amount.

[0168] [Fourth Embodiment]

[0169] According to the fourth embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Figure 6 The first data set DS41 shown in the figure shows the number of stayers in each area of ​​a specific region. The areas are defined by dividing the specific region into, for example, a grid shape.

[0170] In S120, the processor 11 further obtains a second data set DS42, which describes the number of stayers divided by region and attribute for a specific survey subject group. The number of stayers divided by attribute can be, for example, the number of stayers divided by occupation for a specific survey subject group. The specific survey subject group corresponds to, for example, a set of people who have undergone street surveys.

[0171] The processor 11 generates and outputs a third data set DS43 based on the acquired first data set DS41 and second data set DS42, wherein the third data set DS43 describes the number of residents in a specific area divided by region and attribute (S130, S140).

[0172] Figure 6 The first data set DS41 shown has the following records, in which for each region, the number of residents in the corresponding region and the region information are recorded. The second data set DS42 has the following records, in which for each combination of region and attribute, the number of residents with the corresponding attribute in the corresponding region, as well as the region information and the attribute information are recorded. The third data set DS43 has the following records, in which for each combination of region and attribute, the estimated value of the number of residents with the corresponding attribute in the corresponding region, as well as the region information and the attribute information are recorded.

[0173] Here, the variable y r The number of stayers assigned to each region in the first data set DS41. r is the index of the region. That is, y r Indicates the number of people staying in region r.

[0174] The variable x mr The number of stayers assigned to each combination of region and attribute in the second data set DS42. m is the index of the attribute. That is, x mr represents the number of people with attribute m in region r.mr The number of stayers assigned to each combination of region and attribute in the third data set DS43. mr represents the number of people staying in region r with attribute m.

[0175] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0176] <First constraint>

[0177] Total number of stayers ∑u mr , that is, in region r, by attribute u mr The sum of the number of residents divided into different groups is the number of residents y in the same region r r is within the error range of the benchmark.

[0178] <Second constraint>

[0179] The proportion u of residents with attribute m in region r mr / ∑u mr The proportion x of residents with attribute m in the observed region r mr / ∑x mr is within the error range of the benchmark.

[0180] <Third constraint>

[0181] Total number of stayers ∑u mr Gradually changes between adjacent regions. Specifically, the total number of residents ∑u in the first region r1 mr1 The total number of people staying in the second region r2 ∑u mr2 The difference between |∑u mr1 -∑u mr2 | is a threshold value H=D(r1, r2)·H0 that increases as the distance D(r1, r2) between regions r1 and r2 increases. This constraint condition is about the number of people u staying between regions r1 and r2 corresponding to multiple places. mr The difference constraints.

[0182] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. However, the error ε r and σ r All obey normal distribution.

[0183] [Number 7]

[0184]

[0185] The processor 11 performs parameter estimation based on MCMC according to the model, thereby calculating the number of staying persons u mrThe posterior distribution of , and the third data set DS43 can be generated, which records the number of stayers u in each area A divided by attributes mr The third data set DS43 may also be generated as a data set that describes the number of residents for each segment group including two or more attributes in each region.

[0186] According to this embodiment, the number of people u staying between the regions r1 and r2 is used. mr The number of stayers u in each region is estimated based on the constraints of the difference between mr Therefore, the number of people staying with high reliability u can be calculated according to the actual trend mr The estimated value of .

[0187] According to this embodiment, the "first set of analysis objects" corresponds to the set of all residents in a specific area, and the "second set" corresponds to the set of a specific survey subject group. The "first feature" and the "second feature" correspond to the number of residents.

[0188] [Fifth Embodiment]

[0189] According to the fifth embodiment, the processor 11 starts based on the execution command Figure 2 In S110, the processor 11 obtains Figure 7 The first data set DS51 shown in the figure shows the number of movers for each combination of a source region and a destination region in a specific region and the number of stayers in each source region. As in the fourth embodiment, the region is defined by dividing the specific region into, for example, a grid shape. The destination region can be limited to a region adjacent to the source region.

[0190] The processor 11 further acquires a second data set DS52 in S120. The second data set DS52 is a data set related to the number of mobile persons of a specific survey subject group, and Figure 7 As shown, it is a data set that describes the number of movers for each combination of a moving source area and a moving destination area in a specific region, according to whether or not they have purchased a specific product. In the following, whether or not a specific product has been purchased is represented as a purchase attribute.

[0191] The processor 11 generates and outputs a third data set DS53 based on the acquired first data set DS51 and second data set DS52 (S130, S140). The third data set DS53 is a data set that describes the number of movers divided by the combination of the moving source area and the moving destination area in a specific region and the number of movers divided by the purchasing attribute.

[0192] Here, the variable yr1→r2 The number of movers assigned to each combination of the source region and the destination region in the first data set DS51. r1 is the index of the source region, and r2 is the index of the destination region. That is, y r1→r2 represents the number of movers moving from the source region r1 to the destination region r2. r1 The number of people staying in the movement source region r1 in the first data set DS51 is allocated.

[0193] x mrl→r2 The number of movers divided by purchase attributes for each combination of the moving source area and the moving target area in the second data set DS52. m is the index of the purchase attribute. For example, when a specific commodity is purchased, m takes a value of 1, and when no specific commodity is purchased, m takes a value of 2.

[0194] For example, when the purchase attribute m = 1, x mr1→r2 represents the number of people who purchased a specific product from the source region r1 to the destination region r2. When the purchase attribute m = 2, x mr1→r2 It indicates the number of people who did not purchase a specific product among those who moved from the movement source area r1 to the movement destination area r2.

[0195] In addition, mr1→r2 The number of movers by purchasing attribute assigned to each combination of the moving source area and the moving destination area in the third data set DS53. mr1→r2 The description of the index in X mr1→r2 Same description as for the index in .

[0196] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0197] <First constraint>

[0198] Total number of movers ∑u mr1→r2 , that is, the number of movers u from the source region r1 to the destination region r2 divided by purchasing attributes mr1→r2 The sum of the observed number of movers y from the source region r1 to the target region r2y t-r2 is within the error range of the benchmark.

[0199] <Second constraint>

[0200] For the number of movers who move from the moving source area r1 to the moving destination area r2, the proportion u of the number of movers with the purchase attribute m mr1--r2 / ∑u mr1→r2 The proportion of observed movers x mr1→r2 / ∑xmr1→r2 is within the error range of the benchmark.

[0201] <Third constraint>

[0202] The number of people staying in the source region r1 is z r1 Not less than 0.

[0203] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. However, the error ε r1→r2 and σ r1→r2 All obey normal distribution.

[0204] [Number 8]

[0205]

[0206] The processor 11 performs parameter estimation based on MCMC according to the model, thereby calculating the number of movers u mr1→r2 The posterior distribution of , and the third data set DS53 can be generated, which records the number of movers u divided by purchase attributes for each combination of the moving source area and the moving target area mr1→r2 The estimated value of .

[0207] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to a set of people in a specific region, the "second set" corresponds to a set of specific survey object groups, and the "first feature" and "first feature" correspond to the number of mobile people. As a variation, other "attributes" can be used as "purchase attributes". For example, the second data set DS52 and the third data set DS53 can be data sets that describe the number of mobile people divided by demographic attributes (for example, by gender and age stage).

[0208] [Sixth Embodiment]

[0209] According to the sixth embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data generated in S110 is Figure 8 The first data set DS61 shown. The first data set DS61 has records indicating the daily sales of a specific commodity to consumer groups including Japanese and foreigners. In addition, the first data set DS61 has records indicating the annual sales of the specific commodity to foreigners. In addition, the first data set DS61 has records indicating the number of foreigners who purchase the specific commodity every day.

[0210] In S120, the processor 11 further obtains Figure 8The second data set DS62 about the purchase situation of a specific consumer group selected from Japanese is shown. The second data set DS62 has records showing the purchase situation of the specific consumer group on a daily basis and for each consumer attribute. The purchase situation of each consumer attribute can be, for example, the purchase situation divided by gender and age stage. The purchase situation can be the purchase amount.

[0211] The processor 11 generates and outputs the first data set DS61 and the second data set DS62 based on the acquired data set DS61. Figure 3 The third data set DS63 (S130, S140) shown describes the daily sales situation of Japanese people according to consumer attributes.

[0212] According to this embodiment, in the analysis process of S130, the processor 11 first organizes the first data set DS61 and calculates the daily sales of the specific product to the Japanese consumer group. Specifically, the daily sales of the Japanese are calculated by subtracting the daily sales of foreigners from the daily sales. The daily sales of foreigners are calculated by allocating the annual sales of foreigners to each day according to the daily population ratio.

[0213] The processor 11 generates a third data set DS63 based on the daily sales of specific commodities by the Japanese consumer group derived from the first data set DS61 and the second data set DS62.

[0214] Here, the variable yt is assigned to the daily sales of a specific commodity to the Japanese consumer group derived from the first data set DS61. t is an index of time corresponding to a date.

[0215] The variable x mt Assign the purchase situation of each consumer attribute to each day in the second data set DS62. m is the index of the consumer attribute. mt The sales situation assigned to each day and each consumer attribute in the third data set DS63.

[0216] In this embodiment, u is calculated in S130 using the same method as in the first embodiment. mt The posterior distribution of , and the third data set DS63 can be generated, which records the sales situation u divided by daily and consumer attributes for the Japanese consumer group mt The third data set DS63 may also be generated as a data set that describes the daily sales situation of each segment group including more than two consumer attributes.

[0217] In this embodiment, the above-mentioned "first set of analysis objects" corresponds to the set of all consumers, the "second set" corresponds to the set of specific consumer groups, the "first feature" corresponds to the sales situation, and the "second feature" corresponds to the purchase situation.

[0218] [Seventh Embodiment]

[0219] According to the seventh embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Fig. 9 The first data set DS71 shown. The first data set DS71 represents the sales ranking of a group of commodities belonging to a specific commodity category in the market. Specifically, the first data set DS71 has a record representing the sales amount of the corresponding commodity for each sales ranking in the market.

[0220] In S120, the processor 11 further obtains Fig. 9 The second data set DS72 related to the purchases of the specific consumer group shown. The second data set DS72 represents the ranking of the purchase amount of the specific consumer group for the commodity group belonging to the specific commodity category. Specifically, the second data set DS72 represents the purchase amount of the specific consumer group for the corresponding commodity, divided by consumer attributes, for each purchase ranking. The purchase amount divided by consumer attributes can be the purchase amount divided by age stage.

[0221] The processor 11 generates and outputs a third data set DS73 based on the acquired first data set DS71 and second data set DS72 (S130, S140). Fig. 9 As shown, for a commodity group of a specific commodity category, the third data set DS73 is a data set that describes the sales situation of the corresponding commodity according to consumer attributes for each sales ranking in the market.

[0222] Here, the variable y r The sales situation assigned to each ranking in the first data set DS71. r is the index of the sales ranking. That is, y r Indicates the sales status of the product corresponding to rank r.

[0223] In addition, the variable x mr The purchase amount divided by consumer attributes assigned to each ranking in the second data set DS72. That is, x mr It represents the purchase amount of the commodity corresponding to the purchase ranking r by the consumer group with consumer attribute m in the specific consumer group.

[0224] In addition, the variable umr is assigned to the sales situation classified by consumer attributes for each ranking in the third data set DS73. That is, umr represents the market sales situation of the commodity corresponding to the sales ranking r to the consumer group with consumer attribute m.

[0225] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0226] <First constraint>

[0227] The sales situation u of the product corresponding to the sales ranking r according to consumer attributes mr The sum of ∑u mr The sales situation y of the product corresponding to the sales ranking r observed r is within the error range of the benchmark.

[0228] <Second constraint>

[0229] For each consumer attribute m, the sales situation u of the product corresponding to the sales ranking r mr The sales of the product with the previous sales ranking r-1 is u mr-1 , the sales situation u of the product with the next sales ranking r+1 mr+1 , and the observed purchase amount x of the product corresponding to the purchase ranking r mr There is a linear relationship between them. The second constraint condition includes the sales situation u between the rankings mr The difference constraints.

[0230] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. However, the error ε r and e mr All obey normal distribution.

[0231] [Number 9]

[0232]

[0233] u mr =α+u mr-1 ρ -1 +u mr+1 ρ +1 +x mr β+e mr

[0234] The processor 11 performs parameter estimation based on MCMC following the model, thereby calculating the sales situation u mr The posterior distribution of can generate a third data set DS73, which records the sales situation u divided by sales ranking and consumer attributes in the market mrThe third data set DS73 may also be generated as a data set that describes the sales situation divided by segment groups including more than two consumer attributes for each sales ranking.

[0235] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to the set of all consumers, the "second set" corresponds to the set of specific consumer groups, the "first feature" corresponds to the sales situation, and the "second feature" corresponds to the purchase amount.

[0236] The first data set DS71 is a series of elements in the sales ranking, and the sales situation y r The second data set DS72 records the purchase amount x for each purchase ranking as a series element in the purchase amount ranking. mr The third data set DS73 records the sales situation u for each sales ranking as a series element in the sales ranking. mr It is described as the third feature quantity.

[0237] [Eighth Embodiment]

[0238] According to the eighth embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Fig.10 The first data set DS81 shown in FIG. 1 has the following records, wherein for a specific commodity, for each consumer age, it indicates the weekly sales situation for the consumer group of the corresponding age in the market. In the following, the weekly sales situation is referred to as weekly sales situation.

[0239] In addition, in S120, the processor 11 further obtains Fig.10 The second data set DS82 about the purchase amount of a specific consumer group is shown. The second data set DS82 has records of the purchase amount of the specific consumer group divided by age for each weekday and each holiday in the above-mentioned week.

[0240] The processor 11 generates and outputs the first data set DS81 and the second data set DS82 based on the acquired data set DS81. Fig.10 The third data set DS83 (S130, S140) shown describes the sales situation divided by consumer age on each weekday and holiday in the entire market.

[0241] Here, the variable y r Assigned to the weekly sales by consumer age in the first data set DS81. r is the index of age. That is, y rRepresents the weekly sales of a specific product to a consumer group of age r.

[0242] In addition, the variable x mr The purchase amount by age on each weekday and each holiday assigned to the specific consumer group in the second data set DS82. m is the index of the time attribute. The time attribute here refers to the attribute of the date classified by weekday and holiday. For example, when the date corresponding to the purchase amount is a weekday, m takes the value of 1, and when the corresponding date is a holiday, m takes the value of 2. mr Represents the purchase amount of a specific product purchased by a consumer group of age r on the date corresponding to the time attribute m.

[0243] In addition, the variable u mr Assigned to the third data set DS83, sales by consumer age on each weekday and each holiday. mr The description of the index in x mr Same description as for the index in .

[0244] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0245] <First constraint>

[0246] Sales u by time attribute based on consumer age r mr The sum of ∑u mr In the observed sales situation y r is within the error range of the benchmark.

[0247] <Second constraint>

[0248] The sales situation u of the consumer age r on the date corresponding to the time attribute m mr The sales situation u to the consumer group whose age is the previous age r-1 mr-1 , the sales situation u to the consumer group whose age is the next age r+1 mr+1 , and the observed purchase amount x of consumers of age r mr There is a linear relationship between them.

[0249] <Third constraint>

[0250] The sales amount between adjacent age groups is similar. Specifically, the sales amount of consumers of age r u mr Sales to adjacent consumers of age r-1 u mr- The difference between 1 |∑u mr -∑u mr-1 | is below the threshold C.

[0251] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. However, the error ε r and e mr All obey normal distribution.

[0252] [Number 10]

[0253]

[0254] u mr =α+u mr-1 ρ -1 +u mr+1 ρ +1 +x mr β+e mr

[0255] |u mr -u mr-1 |≤C

[0256] The processor 11 performs parameter estimation based on MCMC following the model, thereby calculating the sales situation u mr The posterior distribution of , and can generate a third data set DS83, which records the sales situation u divided by consumer age on each weekday and each holiday market mr The estimated value of .

[0257] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to the set of purchase events of all consumers, the "second set" corresponds to the set of purchase events of a specific consumer group, the "first feature" corresponds to the sales situation, and the "second feature" corresponds to the purchase amount.

[0258] The first data set DS81 has weekly sales y for each age as a series element r The second data set DS82 records the purchase amount x on each weekday and each holiday for each age as a series element. mr The third data set DS83 records the sales situation u on each weekday and each holiday for each age as a series element. mr It is described as the third feature quantity.

[0259] [Ninth Embodiment]

[0260] According to the ninth embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Fig.11The first data set DS91 shown in FIG. 1 includes records showing the consumption of specific products in each region (for example, prefecture) throughout the country. The specific products may be specific foods such as tuna, for example.

[0261] In S120, the processor 11 further obtains Fig.11 The second data set DS92 shown describes the population divided by age in each town, which is a region smaller than the above-mentioned area, as the geographical characteristics of the corresponding town for the entire country.

[0262] The processor 11 generates and outputs the first data set DS91 and the second data set DS92 based on the acquired data set DS91. Fig.11 The third data set DS93 (S130, S140) shown describes the consumption of specific commodities in each town in the entire country.

[0263] Here, the variable y r The consumption situation assigned to each region in the first data set DS91. r is the index of the region. That is, y r Represents the consumption of a specific commodity in region r.

[0264] The variable x ma The population by age assigned to each town in the second data set DS92. m is the index of the town and a is the index of age. ma represents the population of town m with age a. m The consumption allocated to each town in the third dataset DS93. m Represents the consumption of a specific commodity in town m.

[0265] In this embodiment, the parameter estimation process is performed in S130 by setting the following constraints.

[0266] <First constraint>

[0267] Consumption by town in region r u m The sum of ∑u m The consumption situation y in the observed region r r is within the error range of the benchmark.

[0268] <Second constraint>

[0269] Consumption by town m The age-specific population x of the corresponding town ma Related.

[0270] In order to estimate the parameters according to the above constraints, a simulation model can be established according to the following mathematical formula. However, the error ε r and e m All obey normal distribution.

[0271] [Mathematical formula 11]

[0272]

[0273] The processor 11 performs parameter estimation based on MCMC following the model, thereby calculating the consumption situation u m The posterior distribution of , and can generate a third data set DS93, which records the consumption of specific commodities divided by town u m The third data set DS93 may also be generated as a data set that describes the consumption situation of each sub-group including two or more towns.

[0274] According to this embodiment, the above-mentioned "set of analysis objects" corresponds to the set of towns, the "first feature" corresponds to the consumption situation, and the "second feature" different from the first feature corresponds to the population divided by age, in other words, corresponds to the population composition. According to this embodiment, the first data set DS91 and the second data set DS92 respectively record the different first and second features of the same set of towns.

[0275] According to the present embodiment, the first data set DS91 is a data set that describes a first feature of a set. The second data set DS92 is a data set that describes a second feature of the same set as the first data set DS91 with fine-grained information. The third data set DS93 is a data set with fine-grained information about the first feature. The third data set DS93 can be generated based on the first data set DS91 and the second data set DS92.

[0276] [Tenth embodiment]

[0277] According to the tenth embodiment, the processor 11 starts based on the execution command Figure 2 The data generation process shown in FIG. 1 is performed, and the data is obtained in S110 Fig.12 The first data set DS101 shown in FIG. DS101 is for a specific region and has records indicating the annual consumption of specific commodities in the specific region each year.

[0278] In S120, the processor 11 also obtains, for each combination of year and month, Fig.12The second data set DS102 shown in the figure describes the average temperature of the corresponding month in the corresponding year as the geographical feature of the year and the month according to the region in the specific region. The word "month" used below should be understood as a month distinguished by a combination of year and month.

[0279] The processor 11 generates and outputs the first data set DS101 and the second data set DS102 based on the acquired data set DS101. Fig.12 The third data set DS103 (S130, S140) shown here describes the monthly consumption of a specific commodity in a specific region.

[0280] Here, the variable y r The annual consumption of a specific commodity in the first data set DS101. r is the index of the year. That is, y r Represents the annual consumption in the corresponding year r.

[0281] The variable x ma The average temperature assigned to each month in the second data set DS102. m is the index of the month, and a is the index of the region. That is, x ma represents the average temperature of region a in month m. m Monthly consumption of specific commodities assigned to each month in the third data set DS103. m Represents the monthly consumption of a specific commodity in month m.

[0282] In this embodiment, the parameter estimation process is performed by setting the following constraint conditions.

[0283] <First constraint>

[0284] Monthly consumption in a given year r u m The sum of ∑u m The annual consumption situation y observed in year r r is within the error range of the benchmark.

[0285] <Second constraint>

[0286] Consumption by month m The average temperature x by region in the corresponding month ma Related.

[0287] In order to estimate the parameters in accordance with the above constraints, a simulation model that follows the same mathematical formula as that in the ninth embodiment can be established. The processor 11 estimates the parameters based on MCMC in accordance with the model, thereby calculating the monthly consumption situation u mThe posterior distribution of , and can generate a third data set DS103, which records the monthly consumption u divided by month in a specific region m The third data set DS103 may also be generated as a data set that describes the consumption situation of each segment group in more than two months.

[0288] According to this embodiment, the above-mentioned "set of analysis objects" corresponds to a set of time in months, the "first feature" corresponds to consumption conditions, and the "second feature" different from the first feature corresponds to the average temperature by region.

[0289] It is conceivable that there is a case where the type of geographical feature that affects the behavior characteristics of consumers may be different depending on the type of goods. For example, precipitation and sunshine hours may affect the behavior characteristics of consumers. Therefore, as a variation of this embodiment, it is conceivable to use precipitation by region or sunshine hours by region instead of the average temperature by region.

[0290] Geographical features other than temperature, precipitation, and sunshine hours can also be used for analysis. Examples of geographical features include meteorological features such as air pressure in the corresponding region. As described in the ninth embodiment, examples of geographical features include the population composition of the corresponding region. Examples of geographical features can also include the number of specific facilities or stores in the corresponding region, the characteristics of television broadcast content in the corresponding region, the interests, personality and temperament of residents, and the characteristics of the classification and distribution of land in the corresponding region.

[0291] In addition, consumption is one of the characteristic quantities that represent the behavioral characteristics of consumers. The technical idea of ​​this embodiment can also be used to replace consumption and to infer other characteristic quantities that represent the behavioral characteristics of consumers. Examples of behavioral characteristics may include, for example, characteristics related to one or more of purchasing behavior, media contact behavior, facility use behavior, entertainment behavior, indoor and outdoor movement behavior, and web browsing behavior. Media contact behavior may include contact behavior of content provided by information transmission media such as television broadcasts, radio broadcasts, newspapers, magazines, outdoor advertising, and the Internet.

[0292] According to this embodiment, it is possible to estimate the feature quantity of human behavior characteristics in finer time units based on the data set of geographical features divided by region and time as described above. The same effect can also be obtained through the ninth embodiment. According to the ninth embodiment, it is possible to estimate the feature quantity of human behavior characteristics in finer regional units.

[0293] [Other embodiments]

[0294] The present disclosure is not limited to the above-mentioned embodiments, and can be in various forms. In the first to tenth embodiments, data sets related to human behavior are processed, but the technology of the present disclosure can also be applied to generate data sets with fine-grained information about non-human organisms, non-biological objects or physical phenomena.

[0295] In addition, the above-mentioned data sets, such as the first data set, etc., may not be data sets based on a comprehensive survey of the set of corresponding analysis objects, but may be data sets based on a sample survey. For example, the above-mentioned data sets may be data sets generated by inferring the characteristics of the whole through a sample. Data obtained from a large-scale fixed sample (panel) survey with a large number of samples may have a value substantially equivalent to that of complete data depending on the scope of the analysis purpose.

[0296] In addition, the function of one constituent element in the above-mentioned embodiment can be dispersed and arranged in multiple constituent elements. The functions of multiple constituent elements can also be integrated into one constituent element. Part of the structure of the above-mentioned embodiment can be omitted. All modes included in the technical idea determined by the phrases recorded in the claims are embodiments of the present disclosure.

Claims

1. A data generation system, characterized in that: have: an acquisition unit configured to acquire a first data set and a second data set, wherein the first data set expresses a first feature about a first set at a first granularity, and the first data set describes a first feature quantity about a single element group or a first feature quantity about each of a plurality of element groups corresponding to the first granularity in the first set, and the second data set expresses a second feature at a second granularity finer than the first granularity, the second feature is a feature about a second set that is the same as the first set and is different from the first feature, or the second feature is a feature about a second set that is smaller than the first set and is the same as or different from the first feature, and the second data set describes a second feature quantity for each of one or more elements corresponding to the second granularity in the second set; and a generating unit configured to generate a third data set based on the first data set and the second data set, wherein the third data set expresses the first feature about the first set at a third granularity that is finer than the first granularity and coarser than the second granularity or at a third granularity that is the same as the second granularity, and the third data set describes a third feature value of each sub-group corresponding to the third granularity in the first set; The generating unit calculates the third characteristic value of each segmented group and generates the third characteristic value based on the first data set and the second data set under the first constraint condition and the second constraint condition, wherein the first constraint condition is a constraint condition for calculating the third characteristic value based on the relationship between the first characteristic value and the third characteristic value, and the second constraint condition is a constraint condition for calculating the third characteristic value based on the relationship between the second characteristic value and the third characteristic value.

2. The data generation system according to claim 1, characterized in that The first data set describes the first feature value for each of a plurality of series elements of a specific series, The second data set describes the second feature value for each of a plurality of series elements of the specific series, The generating unit estimates the third feature quantity of each segmented group related to at least one series element in the specific series based on the first data set and the second data set.

3. The data generation system according to claim 2, characterized in that: The generating unit estimates the third feature value of each sub-group for each of the plurality of series elements in the specific series, At least one of the first constraint condition and the second constraint condition includes a constraint condition regarding a difference between the series elements of the third feature quantity.

4. The data generation system according to claim 2, characterized in that: The generating unit estimates the third feature value of each sub-group for each of the plurality of series elements in the specific series, The first constraint condition includes the following constraint condition, which is that, for each series element in the multiple series elements, the sum of the third feature quantities of each sub-group in the corresponding series element corresponds to the first feature quantity of all sub-groups in the corresponding series element.

5. The data generation system according to claim 1, characterized in that: The first data set describes the first feature quantity in time series. The second data set describes the second feature quantity in time series, The generating unit estimates the third feature value of each segment group at at least one time based on the first data set and the second data set.

6. The data generation system according to claim 5, characterized in that: The generating unit estimates the third feature value of each segment group at multiple times, At least one of the first constraint condition and the second constraint condition includes a constraint condition regarding a temporal change of the third feature quantity.

7. The data generation system according to claim 5, characterized in that: The generating unit estimates the third feature value of each segment group at multiple times, The second constraint condition includes the following constraint condition, which is that, for each of the multiple times, the third feature quantity at the corresponding time depends on the third feature quantity at a time earlier than the corresponding time and the second feature quantity at the corresponding time in a specific relationship.

8. The data generation system according to claim 5, characterized in that: The generating unit estimates the third feature value of each segment group at multiple times, The first constraint condition includes the following constraint condition, which is that, for each of the multiple times, the sum of the third feature quantities of each segment group at the corresponding time corresponds to the first feature quantities of all segment groups at the corresponding time.

9. The data generation system according to claim 1, characterized in that: The first data set associates and records each of the first feature quantities with a corresponding place, and the second data set associates and records each of the second feature quantities with a corresponding place. The generating unit estimates the third feature value of each segmented group in at least one place based on the first data set and the second data set.

10. The data generation system according to claim 9, characterized in that: The third data set describes the third feature value of each segment group in multiple places. The generating unit calculates the third characteristic value of each of the multiple sub-groups based on the first data set and the second data set under the first constraint condition, the second constraint condition, and the third constraint condition, wherein the third constraint condition is a constraint condition regarding the difference in the third characteristic value between the multiple places.

11. The data generation system according to claim 1, characterized in that: The first data set represents, for each region, a feature quantity related to a behavior feature of people in the corresponding region as the first feature quantity. The second data set represents, for each area smaller than the region, a feature quantity related to a geographical feature of the corresponding area as the second feature quantity, The generating unit estimates, for each of the regions as the segmented groups, a feature quantity related to a behavior characteristic of people in the corresponding region as the third feature quantity based on the first data set and the second data set.

12. The data generation system according to claim 1, characterized in that: The first data set represents, for each first time period, a feature quantity related to a behavior feature of a person in the corresponding first time period as the first feature quantity, For a second time period shorter than the first time period, the second data set represents a feature quantity related to a geographical feature of the corresponding second time period as the second feature quantity, The generating unit calculates, based on the first data set and the second data set, a feature quantity related to the behavior feature in the corresponding second time period as the third feature quantity for each of the second time periods as the segmented group.

13. The data generation system according to any one of claims 1 to 12, characterized in that: At least one of the first constraint condition and the second constraint condition includes an error term associated with dispersion of the third feature quantity.

14. The data generation system according to any one of claims 1 to 13, characterized in that: The generating unit searches for appropriate values ​​of design parameters of an inference model based on the first data set and the second data set by using statistical analysis with reference to probability distribution of dispersion, thereby inferring the third feature quantity corresponding to the appropriate value of the design parameter and following the inference model, wherein the inference model is an inference model for inferring the third feature based on the first feature quantity and the second feature quantity, and the inference model follows the first constraint condition and the second constraint condition.

15. A computer program, characterized in that A method for causing a computer to function as the acquisition unit and the generation unit in the data generation system according to any one of claims 1 to 14.

16. A data generation method, which is executed by a computer, wherein the data generation method is characterized in that it includes: A first data set and a second data set are obtained, wherein the first data set expresses a first feature about a first set at a first granularity, and the first data set describes a first feature quantity about a single element group or a first feature quantity about each of a plurality of element groups corresponding to the first granularity in the first set, and the second data set expresses a second feature at a second granularity finer than the first granularity, the second feature is a feature about a second set that is the same as the first set and is different from the first feature, or the second feature is a feature about a second set that is smaller than the first set and is the same as or different from the first feature, and the second data set describes a second feature quantity for each of one or more elements corresponding to the second granularity in the second set, and The third data set is generated based on the first data set and the second data set, wherein the third data set expresses the first feature about the first set at a third granularity that is finer than the first granularity and coarser than the second granularity or at a third granularity that is the same as the second granularity, and the third data set describes a third feature quantity of each segment group corresponding to the third granularity in the first set, The generating means that, under a first constraint condition and a second constraint condition, based on the first data set and the second data set, the third characteristic quantity of each segmented group is inferred and the third data set is generated, wherein the first constraint condition is a constraint condition for inferring the third characteristic quantity based on the relationship between the first characteristic quantity and the third characteristic quantity, and the second constraint condition is a constraint condition for inferring the third characteristic quantity based on the relationship between the second characteristic quantity and the third characteristic quantity.