Data generation system and data generation method

Abstract

To provide a data generating system, a computer program, and a data generating method by which a data set the information granularity of which is fine can be generated.SOLUTION: A data generating process in a data generating system includes acquiring a first data set S110, acquiring a second data set S120, generating a third data set by performing an analysis process on the basis of the first data set and the second data set S130, and outputting the generated third data set to an output destination designated by a user S140. In the first data set, a first feature amount about a single or multiple element groups in a first group in which a feature concerning the first group is expressed with a first granularity is written. In the second data set, a second feature amount about each of one or more elements in a second group in which a feature concerning the second group is expressed with a second granularity is written. The third data set is generated on the basis of the relationship between the first feature amount and a third feature amount and the relationship between the second feature amount and the third feature amount.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to data generation systems and methods. [Background technology]

[0002] Conventionally, a target has been analyzed by aggregating data (see, for example, Patent Document 1). For example, a technique is known that aggregates data within an area corresponding to a trade area with a specified radius centered on a store specified by a user, and displays the analysis results regarding the characteristics of the specified trade area. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2003-167880 A Summary of the Invention [Problem to be solved by the invention]

[0004] In general, collecting fine-grained data sets about large groups is more difficult than collecting fine-grained data sets about small groups. For example, while it is possible to collect fine-grained data sets about groups of people who cooperate with the collection of information, it is difficult to collect fine-grained data sets about a wide range of groups, including people who are uncooperative with the collection of information. For example, to collect detailed information about individuals, the individual's consent is often required due to privacy concerns.

[0005] For this reason, even if a detailed analysis of a large set is desired, there have been many cases in the past where the data set required for the analysis could not be obtained, making it impossible to perform the detailed analysis.

[0006] Therefore, according to one aspect of the present disclosure, it is desirable to provide a technique capable of generating a fine-grained data set of information about a set from a limited data set that can be acquired. [Means for solving the problem]

[0007] According to one aspect of the present disclosure, there is provided a data generation system including an acquisition unit and a generation unit. The acquisition unit is configured to acquire a first data set and a second data set. The first data set represents a first feature for a first set at a first granularity. The first data set describes a first feature amount for a group of elements or for each of a plurality of groups of elements corresponding to the first granularity in the first set.

[0008] The second data set represents second features at a second granularity that are different from the first features for a second set that is the same as the first set, or that are the same or different from the first features for a second set that is smaller than the first set. The second granularity is finer than the first granularity. The second data set describes second features for one or more elements in the second set that correspond to the second granularity.

[0009] The generator is configured to generate a third dataset based on the first dataset and the second dataset, the third dataset representing first features for the first set at a third granularity, the third granularity being finer than the first granularity and coarser than or equal to the second granularity, and the third dataset describing third features for each segment in the first set corresponding to the third granularity.

[0010] According to an aspect of the present disclosure, the second dataset may include, for each of one or more elements in the second set, information capable of identifying a segment of the corresponding one or more elements. The generator is configured to estimate a third feature for each segment based on the first dataset and the second dataset under a first constraint condition for estimating the third feature based on a relationship between the first feature and the third feature and a second constraint condition for estimating the third feature based on a relationship between the second feature and the third feature, and generate a third dataset.

[0011] According to one aspect of the present disclosure, by using a first dataset with coarse information granularity regarding a first feature together with a second dataset with fine information granularity regarding a second feature, and further using the relationship between the first feature amount and the third feature amount and the relationship between the second feature amount and the third feature amount, a third dataset describing the third feature amount with fine information granularity regarding the first feature can be generated. Thus, according to one aspect of the present disclosure, fine information granularity data regarding a set can be generated from limited obtainable data.

[0012] According to one aspect of the present disclosure, the first dataset may describe a first feature for each of a plurality of sequence elements in the predetermined sequence.The second dataset may describe a second feature for each of a plurality of sequence elements in the predetermined sequence.

[0013] According to an aspect of the present disclosure, the generator may estimate a third feature value for each segment for at least one sequence element in the predetermined sequence based on the first dataset and the second dataset. Examples of the predetermined sequence may include a time series and a location series.

[0014] According to an aspect of the present disclosure, the generation unit may estimate a third feature amount for each segment for each of a plurality of sequence elements in the predetermined sequence.

[0015] 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 sequence elements of a third feature amount.

[0016] The third feature amount of a first series element tends to exhibit a value close to the third feature amount of a second series element depending on the physical or conceptual closeness between the first series element and the second series element. Therefore, by making an estimation that takes into account constraints on the difference between the series elements in the third feature amount, it is possible to estimate a third feature amount with high reliability.

[0017] According to one aspect of the present disclosure, the first constraint may include a constraint that, for each of a plurality of sequence elements, a sum of the third feature for each segment in the corresponding sequence element corresponds to the first feature for the entire segment in the corresponding sequence element. Estimation using such a constraint makes it possible to estimate the third feature with high reliability.

[0018] According to an aspect of the present disclosure, the first data set may describe a first feature in a time series. The second data set may describe a second feature in a time series. The generator may estimate a third feature for each segment at at least one time point based on the first data set and the second data set.

[0019] According to an aspect of the present disclosure, the generating unit may estimate a third feature amount for each segment at a plurality of time points. At least one of the first constraint condition and the second constraint condition may include a constraint condition related to a time change of the third feature amount. By defining the constraint condition in consideration of a possible time change of the feature amount, it is possible to estimate the third feature amount with high reliability.

[0020] According to an aspect of the present disclosure, the generating unit may estimate a third feature amount for each segment at a plurality of time points. The second constraint condition may include a constraint condition that, for each of the plurality of time points, the third feature amount at a corresponding time point depends in a predetermined relationship on the third feature amount at a time point prior to the corresponding time point and the second feature amount at the corresponding time point. By estimating the third feature amount based on such a constraint condition, it is possible to estimate a third feature amount with high reliability with respect to a feature of a state in which a state at a certain time point is influenced by a state at a previous time point.

[0021] According to an aspect of the present disclosure, the generator may estimate a third feature amount for each segment at a plurality of time points. The first constraint condition may include a constraint condition that, for each of the plurality of time points, a sum of the third feature amount for each segment at a corresponding time point corresponds to a first feature amount of the entire segment at the corresponding time point.

[0022] According to an aspect of the present disclosure, the first dataset may describe each of the first feature amounts in association with a corresponding location, and the second dataset may describe each of the second feature amounts in association with a corresponding location. The generator may estimate a third feature amount for each segment in the at least one location based on the first dataset and the second dataset.

[0023] According to an aspect of the present disclosure, the third dataset may describe a third feature for each segment at the plurality of locations, and the generator may estimate the third feature for each of the plurality of segments based on the first dataset and the second dataset under a third constraint on a difference in the third feature between the plurality of locations in addition to the first constraint and the second constraint.

[0024] The third feature amount in the first location tends to have a value close to the third feature amount in the second location depending on the physical proximity between the first location and the second location. Therefore, according to an estimation that takes into account a constraint on the difference between the locations of the third feature amount, it is possible to estimate the third feature amount with high reliability.

[0025] According to one aspect of the present disclosure, the first dataset may represent, for each region, a feature amount related to behavioral characteristics of people in the corresponding region as a first feature amount. The second dataset may represent, for each area smaller than a region, a feature amount related to geographical characteristics of the corresponding area as a second feature amount. The generation unit may estimate, for each area as a segment, a feature amount related to behavioral characteristics of people in the corresponding area as a third feature amount based on the first dataset and the second dataset.

[0026] In cases where there is a correlation between human behavioral characteristics and the geographic characteristics of the corresponding areas, such estimation can estimate features for smaller areas with finer granularity than the first dataset.

[0027] According to one aspect of the present disclosure, the first dataset may represent, for each first period, a feature amount related to a behavioral characteristic of a person within the corresponding first period as a first feature amount. The second dataset may represent, for each second period shorter than the first period, a feature amount related to a geographical characteristic of the corresponding second period as a second feature amount. The generation unit may estimate, for each second period as a segment, a feature amount related to a behavioral characteristic within the corresponding second period as a third feature amount based on the first dataset and the second dataset.

[0028] In cases where there is a correlation between human behavioral characteristics and geographic characteristics for the corresponding time period, this estimation can provide feature estimates for shorter time periods with finer granularity than the first dataset.

[0029] 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 a variance in the third feature amount.

[0030] According to an aspect of the present disclosure, the generation unit may search for optimal values ​​of design parameters of an estimation model complying with a first constraint condition and a second constraint condition for estimating a third feature quantity from the first feature quantity and the second feature quantity, based on the first dataset and the second dataset, by statistical analysis taking into account a probability distribution related to variation, and estimate a third feature quantity complying with the estimation model corresponding to the optimal values ​​of the design parameters.

[0031] According to statistical analysis that takes into account probability distribution, it is possible to estimate a plausible feature as the third feature from limited data. An example of search using statistical analysis includes an example in which parameters are estimated from observed data using Bayesian estimation.

[0032] According to one aspect of the present disclosure, there may be provided a computer program for causing a computer to function as the acquisition unit and the generation unit in the data generation system described above. The computer program may be recorded in a non-transitory computer-readable recording medium.

[0033] According to one aspect of the present disclosure, there may be provided a data generation method corresponding to the data generation system described above.

[0034] According to one aspect of the present disclosure, there may be provided a data generation method including: acquiring a first dataset, the first dataset describing a first feature amount for a single element group or for each of a plurality of element groups corresponding to a first granularity in a first set in which a first feature for the first set is expressed at a first granularity; and a second dataset, the second dataset describing a second feature amount for one or more elements corresponding to a second granularity in a second set in which a second feature different from a first feature for a second set the same as the first set or a second feature the same or different from a first feature for a second set smaller than the first set is expressed at a second granularity finer than the first granularity; and generating a third dataset, based on the first dataset and the second dataset, describing a third feature amount for each segment corresponding to a third granularity in the first set in which the first feature for the first set is expressed at a third granularity finer than the first granularity, coarser than the second granularity, or the same as the second granularity. The data generation method may be executed by a computer.

[0035] In this data generation method, the generating may include estimating a third feature for each segment based on the first dataset and the second dataset under a first constraint on estimating the third feature based on the relationship between the first feature and the third feature and a second constraint on estimating the third feature based on the relationship between the second feature and the third feature, to generate a third dataset. [Brief description of the drawings]

[0036] [Figure 1] FIG. 1 is a block diagram illustrating a configuration of a data generation system. [Diagram 2] 11 is a flowchart of a data generation process executed by a processor. [Diagram 3] FIG. 2 is a diagram illustrating first, second, and third data sets according to the first embodiment. [Figure 4] FIG. 13 is a diagram illustrating the first, second, and third data sets according to the second embodiment. [Diagram 5]FIG. 13 is a diagram illustrating the first, second, and third data sets of the third embodiment. [Figure 6] FIG. 13 is a diagram illustrating the first, second, and third data sets according to the fourth embodiment. [Figure 7] FIG. 13 is a diagram illustrating the first, second, and third data sets according to the fifth embodiment. [Figure 8] FIG. 23 is a diagram illustrating the first, second, and third data sets of the sixth embodiment. [Figure 9] FIG. 23 is a diagram illustrating the first, second, and third data sets of the seventh embodiment. [Figure 10] FIG. 23 is a diagram illustrating the first, second, and third data sets of the eighth embodiment. [Figure 11] FIG. 23 is a diagram illustrating the first, second, and third data sets of the ninth embodiment. [Figure 12] FIG. 13 is a diagram illustrating the first, second, and third data sets of the tenth embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0037] Exemplary embodiments of the present disclosure will now be described with reference to the drawings. The data generation system 1 according to some embodiments described is configured to generate a third dataset describing the first characteristic of the first set of objects to be analyzed at a finer granularity than the first dataset, based on a first dataset with a coarse granularity describing a first characteristic of the first set of objects to be analyzed and a second dataset with a finer granularity describing a second characteristic of a second set of objects to be analyzed that is smaller than the first set of objects to be analyzed.

[0038] The second set may be a subset of the first set being analyzed, or another set of smaller size than the first set that is equivalent for analytical purposes to the subset. An example of an alternative set includes a set having many elements in the first set, but some elements not in the first set. The second feature may be the same feature as the first feature, or a different feature related to the first feature.

[0039] The data generation system 1 according to some further embodiments described above is configured to generate a third dataset describing the first characteristic of the set of objects to be analyzed at a finer granularity than the first dataset based on a first dataset describing a first characteristic of the set of objects to be analyzed and a second dataset describing a second characteristic of the set of objects to be analyzed that is different from the first characteristic.

[0040] An exemplary data generation system 1 is configured by installing a computer program for generating a third data set in a general-purpose information processing system. In the first to tenth embodiments described below, the configurations of the first, second, and third data sets, and the processing content related to the generation of the third data set will be described using specific examples. The hardware configuration of the data generation system 1 of the first embodiment described using FIG. 1 can also be applied to the data generation systems 1 of the other embodiments.

[0041] [First embodiment] As shown in FIG. 1, the data generation system 1 of this embodiment includes a processor 11, a memory 12, a storage 13, a display 15, an input device 17, a media reader / writer 18, and a communication device 19.

[0042] The processor 11 executes processing in accordance with a computer program stored in the storage 13. The memory 12 includes a RAM, and is used as a working area when the processor 11 executes processing in accordance with the computer program. The memory 12 temporarily stores the computer program read from the storage 13.

[0043] The storage 13 stores computer programs and various data. One of the computer programs stored in the storage 13 includes a computer program for generating a third data set. Examples of the storage 13 include a hard disk drive and a solid state drive.

[0044] 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.

[0045] The media reader / writer 18 is configured to be capable of reading information recorded on a recording medium such as a memory card and to be capable of writing new information to the recording medium. The communication device 19 is configured to be controlled by the processor 11 and to communicate with external devices within a local area network and / or a wide area network.

[0046] When a command to execute a data generation process is input from the user via the input device 17, the processor 11 starts the data generation process shown in FIG. 2 for generating a third data set in accordance with the computer program.

[0047] By executing this data generation process, the processor 11 generates a third dataset DS13 that explains sales by consumer attributes in the entire market for a specific product category, based on a first dataset DS11 that explains sales in the entire market for a specific product category, and a second dataset DS12 that explains purchasing volumes by consumer attributes of a specific consumer group.

[0048] Examples of consumer attributes include consumer demographic attributes. Examples of purchase amounts and sales by consumer attributes include purchase amounts and sales by gender and age group. Gender and age group here refers to consumer demographic attributes defined by a combination of gender and age group, examples of which include teenage males, teenage females, men in their twenties, women in their twenties, men in their thirties, and women in their thirties. Examples of purchase amounts include purchase amounts and number of items purchased.

[0049] When the data generation process starts, the processor 11 acquires the first data set DS11 and the second data set DS12 specified by the user through the input device 17 (S110, S120). The processor 11 can acquire the first data set DS11 and the second data set DS12 from, for example, the storage 13, or from a recording medium via the media reader / writer 18, or from an external device via the communication device 19.

[0050] Specifically, the first data set DS11 acquired in this embodiment is a time-series data set on market sales, which is calculated by aggregating the daily sales of the entire market of a specific product category. The sales of the entire market corresponds to the purchase amount of all consumers.

[0051] The exemplary first dataset DS11 has records in which the sales in the entire market on the corresponding day are described together with the date, as shown in Fig. 3. The processor 11 may acquire a dataset before aggregation in S110, and generate the first aggregated dataset DS11 shown in Fig. 3 from the acquired dataset.

[0052] The second data set DS12 acquired in this embodiment is generated based on the purchase history of a specific consumer group among all consumers who are cooperative in collecting the purchase history. The purchase history can be collected by a monitor survey.

[0053] Specifically, the second dataset DS12 is data on purchasing volumes of a specific product category by a specific consumer group, aggregated by consumer attribute and by day, and is time-series data on purchasing volumes by consumer attribute (e.g., by gender and age group).

[0054] The exemplary second dataset DS12 has a record in which, for each combination of a day and a consumer attribute, the purchase amount of a group of consumers having the corresponding consumer attribute on the corresponding day is described together with the date and consumer attribute information, as shown in Fig. 3. The processor 11 may acquire the dataset before aggregation in S120, and generate the second aggregated dataset DS12 shown in Fig. 3 from the acquired dataset.

[0055] The processor 11 generates a third dataset DS13 by executing an analysis process based on the acquired first dataset DS11 and second dataset DS12 (S130).

[0056] The third dataset DS13 generated in this embodiment is data describing estimated values ​​of daily sales by consumer attribute for a specific product category for all consumers, not limited to a specific consumer group. In other words, the third dataset DS13 is time-series data on market sales by consumer attribute.

[0057] As shown in FIG. 3, the exemplary third dataset DS13 has records in which, for each combination of day and consumer attribute, an estimate of sales for a group of consumers having the corresponding consumer attribute on the corresponding day in the entire market is described along with the date and consumer attribute information.

[0058] The processor 11 outputs the third data set DS13 generated in S130 to an output destination specified by the user via the input device 17 (S140). For example, the processor 11 can output the data to the storage 13, to a recording medium via the media reader / writer 18, or to an external device via the communication device 19. The processor 11 may cause the display 15 to display and output the third data set DS13. Thereafter, the processor 11 ends the data generation process shown in FIG.

[0059] Next, the analysis process (S130) will be described in detail. In the analysis process, a 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.

[0060] To explain the details of the process using a formula, we use the variable y t where t is the index of the time in days, i.e., y t represents the sales on the day corresponding to time t.

[0061] For the purchase amount by day and consumer attribute in the second data set DS12, the variable x mt m is the index of the consumer attribute, i.e., x mt represents the purchase amount of a consumer group having the corresponding consumer attribute m on the day corresponding to time t.

[0062] For the sales by day and consumer attribute in the third dataset DS13, the variable u mt That is, u mt represents the sales of a consumer group having the corresponding consumer attribute m on the day corresponding to time t.

[0063] In this embodiment, the parameter estimation process is performed under the following constraints. <First constraint> Sales by consumer attribute at time t u mt The sum of Σu mt is the sales y at the same time t t This is within the margin of error based on the standard.

[0064] <Second constraint> Sales u at time t and consumer attribute m mt is the sales u of consumer attribute m at the previous time point t-1 (i.e. the previous day) mt-1 , and the purchase amount x of consumer attribute m at time t mt There is a linear relationship between sales and mtIncludes constraints on time-varying

[0065] For parameter estimation according to the first and second constraints, a simulation model according to the following equations (1) and (2) can be established. Here, the index m of the consumer attribute takes values ​​from 1 to M.

[0066]

number

[0067] According to equation (2), u mt is defined as follows:

[0068]

number

[0069] The error term ε in Eq. (1) t has center 0 and variance σ 2 Normal distribution of N(0,σ 2 ) is the probability distribution according to Eqs. (2) and (3). mt The error term for the variability of mt is centered at 0 and has variance ψ 2 Normal distribution N(0,ψ 2 ) shows the probability distribution according to

[0070] According to this model, the parameters to be estimated are u mt ,α,ρ,β,σ 2 , ψ 2 In MCMC parameter estimation, u mt ,α,ρ,β,σ 2 , ψ 2 For this, a prior probability distribution (hereinafter simply referred to as "prior distribution") can be set as follows:

[0071] <α,ρ,β,σ 2 , ψ 2 Prior distribution of > We set a conjugate prior distribution designed to be a substantially uninformative prior distribution. For example, we set a normal distribution with large variance for α, ρ, and β, and σ 2 , ψ 2 For , we set an inverse gamma distribution with the parameter adjusted to 1.

[0072] mt Prior distribution of > We set the following prior distribution (normal distribution) according to the model.

[0073]

number

[0074] For example, the Gibbs sampling method can be applied to parameter estimation by MCMC. According to this parameter estimation, α, ρ, β, σ as design parameters can be obtained by statistical analysis considering probability distribution. 2 , ψ 2 At the same time as searching for the optimum value of , mt is searched. By this search (Bayesian estimation), u mt The posterior probability distribution (hereinafter simply referred to as the "posterior distribution") of

[0075] Processor 11 is u mt As an estimate of u mt Calculate the posterior mean or median of the posterior distribution of u mt A third data set DS13 describing estimates of

[0076] Equation (2) can be generalized to the following equation (5) with any order in time, K. The order K can be an integer value equal to or greater than 1.

[0077]

number

[0078] In this case, u mt ​The prior distribution of can be set as shown in the following equation (6) to perform parameter estimation.

[0079]

number

[0080] According to this embodiment, the above-mentioned "first set of analysis targets" corresponds to all consumers, the "second set" corresponds to a specific consumer group, the "first characteristic" corresponds to sales, and the "second characteristic" corresponds to purchasing volume.

[0081] According to this data generation system 1, the following effect can be obtained: When only a dataset with coarse information granularity is available as a dataset explaining the characteristics of a set to be analyzed, it is possible to generate a dataset with fine information granularity that explains the characteristics of the set to be analyzed and that is highly reliable from a statistical point of view, based on a dataset with fine information granularity that explains the characteristics of a set smaller than the set to be analyzed.

[0082] For example, when it is not possible to obtain a dataset that explains sales by consumer attributes for all consumers, it is possible to generate a dataset that explains sales by consumer attributes with finer information granularity for all consumers based on the purchasing history of a specific consumer group obtained through a monitor survey.

[0083] In this embodiment, in particular, the sales u mt Since the estimated value is calculated, the estimated value is highly reliable.

[0084] A comparative example would be to adjust the aggregated value of a specific consumer group by a coefficient according to the size ratio between the specific consumer group and the consumer set in the entire market, but this comparative example includes the assumption that the attribute distribution in the specific consumer group is the same as the attribute distribution in the entire market.

[0085] The assumptions in the comparative example are usually very different from the truth. Therefore, even if a data set with fine information granularity is generated in the comparative example, the reliability is low. In contrast, according to the present embodiment using Bayesian statistics, a data set with finer information granularity and higher reliability than the conventional one can be generated.

[0086] Various modifications of this embodiment are possible. For example, the third dataset DS13 does not have to explain sales for each category of consumer attributes as in the second dataset DS12. In other words, the third dataset DS13 may be a dataset that explains sales for each attribute group that combines two or more consumer attributes. The estimated value of sales for each attribute group may be calculated as the sum of the estimated values ​​of sales for each consumer attribute belonging to the group.

[0087] For example, when the second dataset DS12 describes purchase amounts by gender and age group, the third dataset DS13 may be generated as a dataset describing sales by age group regardless of gender.

[0088] The third dataset DS13 may be data describing sales by segments of the same or greater granularity as the second dataset DS12. The segments may be defined by classifying the set at the same division as the second dataset DS12 or at a coarser division than the second dataset DS12 and at a finer division than the first dataset DS11.

[0089] Additionally, in this embodiment, sales by segments across an entire market are estimated to generate the third dataset DS13, however, the technical ideas of this embodiment may be used to estimate sales by segments across one or more retailers of interest instead of across the entire market.

[0090] In this case, the first dataset DS11 may be a dataset describing sales by day for one or more retailers, and the second dataset DS12 may be a dataset describing purchase volumes by day and combination of consumer attributes for a specific group of consumers corresponding to a portion of the customer group of one or more retailers.

[0091] The third dataset DS13 may be a dataset describing daily sales by consumer attribute for one or more retailers. Moreover, the second dataset DS12 may be a dataset created based on the purchase history of a different set of consumers that does not completely overlap with the set of consumers corresponding to the first dataset DS11.

[0092] [Second embodiment] 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 and subsequent embodiments is the same as that of the data generation system 1 of the first embodiment. In the second and subsequent embodiments, the data set to be handled is different from that of the first embodiment. In the second embodiment, when an execution command is input, the processor 11 starts the data generation process shown in FIG. 2 in the same manner as in the first embodiment.

[0093] By executing this data generation process, the processor 11 is configured to generate a third dataset DS23 that describes sales of a specific product category in the market by product based on a first dataset DS21 regarding sales of a specific product category in the market and a second dataset DS22 regarding sales by product in the specific product category based on sales records of a limited number of one or more retailers.

[0094] When the data generation process starts, the processor 11 acquires the first data set DS21 and the second data set DS22 specified by the user via the input device 17 (S110, S120).

[0095] The first dataset DS21 acquired in this embodiment is a time series dataset on market sales, which is a daily aggregate of sales of a specific product category in the market. As shown in Fig. 4, the exemplary first dataset DS21 has records in which the sales of a specific product category in the market on a corresponding day are described together with the date.

[0096] The second data set DS22 is data that summarizes the sales of each product in a specific product category on a daily basis based on sales records stored in the POS systems of a limited number of retailers, and is time-series data that explains the sales of each product. Hereinafter, this sales will be referred to as POS sales. For example, when the specific product category is "beer," the sales of each product correspond to the sales of each brand of beer.

[0097] As shown in FIG. 4, the exemplary second dataset DS22 has, for each combination of day and product, a record in which the POS sales of the corresponding product on the corresponding day are described together with the date and product information.

[0098] The processor 11 generates a third dataset DS23 by executing an analysis process based on the acquired first dataset DS21 and second dataset DS22 (S130), and outputs the generated third dataset DS23 (S140).

[0099] The third dataset DS23 generated in this embodiment is time-series data that explains market sales by product in a specific product category. As shown in Fig. 4, the exemplary third dataset DS23 has a record in which, for each combination of day and product, an estimated value of the sales of the corresponding product in the market on the corresponding day is described together with the date and product information.

[0100] Next, the analysis process (S130) will be described in detail. For the purpose of this description, the variable y t t is the index of the time corresponding to the day, as in the first embodiment, and y trepresents the market sales of a particular product category on the day corresponding to time t.

[0101] For the POS sales by day and product in the second data set DS22, the variable x mt m is the index of the product, i.e., x mt represents the POS sales of the corresponding product m on the day corresponding to time t.

[0102] Furthermore, for the sales by day and product in the third data set DS23, the variable u mt That is, u mt represents the market sales of the corresponding product m on the day corresponding to time t.

[0103] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> Sales by product at time t mt The sum of Σu mt is the sales y at the same time t t This is within the margin of error based on the standard.

[0104] <Second constraint> Sales u of product m at time t mt is the sales u of product m at the previous time t-1 (i.e. the previous day) mt-1 , and POS sales of product m at time t x mt There is a linear relationship.

[0105] As can be understood from the similarity of the first and second constraint conditions to the first embodiment, according to this embodiment, a model according to the above-mentioned formulas (1) and (2) is defined, a prior distribution is set in the same manner as in the first embodiment, and u is estimated by parameter estimation using MCMC. mt The posterior distribution of can be obtained.

[0106] Processor 11 calculates the calculated u mt Based on the posterior distribution of mtA third dataset DS23 describing an estimated value of may be generated. The third dataset DS23 may be generated as a dataset describing sales for each segment that groups together two or more products in a particular product category.

[0107] According to this embodiment, the above-mentioned "first set of analysis objects" corresponds to a set of transactions in the market, the "second set" corresponds to a set of transactions recorded in the POS systems of one or more retailers, the "first characteristic" corresponds to sales, and the "second characteristic" corresponds to POS sales.

[0108] [Third embodiment] According to the third embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and acquires a first data set DS31 shown in Fig. 5 related to the line usage status of all users in S110. Furthermore, a second data set DS32 related to the line usage status by user attribute of a specific user group is acquired in S120. The line usage status by user attribute may be, for example, line usage status by gender and age group.

[0109] The processor 11 generates and outputs a third data set DS33 that explains the line usage status by user attribute for all users, based on the acquired first data set DS31 and second data set DS32 (S130, S140).

[0110] Specifically, the first dataset DS31 acquired in this embodiment is a time series dataset in which the usage amount (e.g., usage time) of all users on each line is collected on a daily basis. As shown in Fig. 5, the exemplary first dataset DS31 has, for each day and line, a record in which the usage amount (e.g., total usage time) of all users on the corresponding line on the corresponding day is described together with the date and line information.

[0111] The second dataset DS32 is a time series dataset that aggregates the amount of use of a specific user group on each line by day and by user attribute. The specific user group may be a group of users who are cooperative in collecting the usage history among all users.

[0112] As shown in FIG. 5, the exemplary second dataset DS32 has records that, for each combination of day, route, and user attribute, describe the usage volume of the corresponding route for a group of users having the corresponding user attribute on the corresponding day, together with the date, route information, and user attribute information.

[0113] The third dataset DS33 is a time series dataset describing estimated values ​​of usage for each route by day and by user attribute for all users. As shown in Fig. 5, the exemplary third dataset DS33 has records in which, for each combination of day, route, and user attribute, estimated values ​​of usage for the corresponding route for a group of users having the corresponding user attribute on the corresponding day are described together with the date, route information, and user attribute information.

[0114] Here, the variable y nt n is the line index, i.e., y nt represents the usage of line n by all users on the day corresponding to time t.

[0115] For the amount of use for each combination of day, route, and user attribute in the second dataset DS32, the variable x nmt m is the index of the user attribute, i.e., x nmt represents the usage volume of route n for a group of users having the corresponding user attribute m on a day corresponding to time t.

[0116] For the volume of use for each combination of day, route, and user attributes in the third dataset DS33, the variable u nmt That is, u nmt represents the usage volume of route n for a group of users having the corresponding user attribute m on a day corresponding to time t.

[0117] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> The amount of use of route n by user attributes at time t, u nmt The sum of Σu nmt is the total number of users of line n at the same time t nt Furthermore, the amount of usage y nt When is greater than zero, the sum Σu nmt is also greater than zero.

[0118] <Second constraint> For each combination of route n and user attribute m, the usage volume u at time t nmt The increase rate u from the point t-1 one day before nmt / u nmt-1 is the observed usage x nmt Increase rate x nmt / x nmt-1 The error is within the range based on .

[0119] <Third constraint> For each combination of route n and user attribute m, the usage volume u at time t nmt is the usage amount u at time t-7 one week ago nmt-7 The error is within the range based on .

[0120] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: nt ,σ1 t ,σ2 t Each of them follows a normal distribution. The index m of the user attribute takes values ​​from 1 to M.

[0121]

number

[0122] The processor 11 estimates the usage amount u by MCMC parameter estimation according to this model.nmt , and calculate the daily route and user attribute usage volume u nmt A third dataset DS33 describing an estimated value of may be generated. The third dataset DS33 may be generated as a dataset describing the amount of use by day and route for each segment that combines two or more user attributes.

[0123] According to this embodiment, the above-mentioned "first set of analysis targets" corresponds to the set of all users, the "second set" corresponds to a set of a specific group of users, and the "first characteristic" and "second characteristic" correspond to the amount of usage.

[0124] [Fourth embodiment] According to the fourth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and acquires a first data set DS41 shown in Fig. 6 that explains the number of visitors for each area in a specific region in S110. The areas are defined by partitioning the specific region into, for example, a mesh shape.

[0125] The processor 11 further obtains a second data set DS42 in S120 that describes the number of visitors by area and attribute for a specific survey subject group. The number of visitors by attribute can be, for example, the number of visitors by occupation type for a specific survey subject group. The specific survey subject group corresponds, for example, to a set of people who have been surveyed on the street.

[0126] The processor 11 generates and outputs a third data set DS43 that explains the number of visitors by area and attribute in a specific region, based on the acquired first data set DS41 and second data set DS42 (S130, S140).

[0127] 6 has a record in which, for each area, the number of visitors in the corresponding area is described together with the area information. The second dataset DS42 has a record in which, for each combination of area and attribute, the number of visitors having the corresponding attribute in the corresponding area is described together with the area information and attribute information. The third dataset DS43 has a record in which, for each combination of area and attribute, an estimate of the number of visitors having the corresponding attribute in the corresponding area is described together with the area information and attribute information.

[0128] Here, the variable y r r is the index of the area, i.e., y r represents the number of visitors in area r.

[0129] For the number of visitors for each combination of area and attribute in the second dataset DS42, the variable x mr m is the index of the attribute, i.e., x mr represents the number of visitors with attribute m in area r. For the number of visitors for each combination of area and attribute in the third data set DS43, the variable u mr That is, u mr represents the number of visitors with attribute m in area r.

[0130] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> Number of visitors by attribute in area r mr The total number of visitors, Σu, is the sum of mr is the number of visitors in the same area r, y r This is within the margin of error based on the standard.

[0131] <Second constraint> Proportion u of visitors with attribute m in area r mr / Σu mris the proportion of visitors with attribute m in observed area r, x mr / Σx mr The error is within the range based on .

[0132] <Third constraint> Total number of guests Σu mr changes gradually between adjacent areas. Specifically, the total number of visitors in the first area r1, Σu mr1 and the total number of visitors in the second area r2, Σu mr2 Difference with |Σu mr1 -Σu mr2 | is the threshold H = D(r1, r2) H that increases as the distance D(r1, r2) between areas r1 and r2 increases. 0 The constraint is as follows: The number of visitors between areas r1 and r2 corresponding to multiple locations is u mr This is a constraint on the difference between the

[0133] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: r ,σ r respectively follow normal distribution.

[0134]

number

[0135] The processor 11 estimates the number of visitors u by MCMC parameter estimation according to this model. mr Calculate the posterior distribution of the number of visitors by attribute for each area, u mr A third dataset DS43 describing an estimated value of may be generated. The third dataset DS43 may be generated as a dataset describing the number of visitors for each segment that combines two or more attributes for each area.

[0136] According to this embodiment, the number of visitors between areas r1 and r2 is u mr Using the constraints on the difference between the attributes of each area, the number of visitors u mrTherefore, we can estimate the number of visitors u with high confidence according to the actual trend. mr It is possible to obtain an estimate of

[0137] According to this embodiment, the above-mentioned "first set of analysis subjects" corresponds to the set of all visitors in a specific area, the "second set" corresponds to the set of a specific group of survey subjects, and the "first characteristic" and "second characteristic" correspond to the number of visitors.

[0138] [Fifth embodiment] According to the fifth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and in S110, obtains a first data set DS51 shown in Fig. 7 that explains the number of movers for each combination of source area and destination area in a specific region, and the number of visitors for each source area. As in the fourth embodiment, the area is defined by partitioning the specific region, for example, into a mesh shape. The destination area can be limited to an area adjacent to the source area.

[0139] The processor 11 further acquires, in S120, a second data set DS52 relating to the number of people who moved among a specific survey subject group, which is shown in Fig. 7 and explains the number of people who moved for each combination of origin area and destination area in a specific region, according to whether or not a specific product was purchased. Hereinafter, whether or not a specific product was purchased is referred to as a purchase attribute.

[0140] Based on the acquired first dataset DS51 and second dataset DS52, the processor 11 generates and outputs a third dataset DS53 that explains the number of movers by combination of source area and destination area in a specific region and by purchasing attributes (S130, S140).

[0141] Here, the variable y r1→r2 r1 is the index of the source area, and r2 is the index of the destination area. That is, y r1→r2represents the number of people moving from the source area r1 to the destination area r2. Furthermore, the variable z r1 Assign.

[0142] For the number of people moving by purchase attribute for each combination of origin area and destination area in the second data set DS52, x mr1→r2 m is an index of the purchase attribute. For example, m takes a value of 1 when a specific product is purchased, and takes a value of 2 when a specific product is not purchased.

[0143] For example, when purchase attribute m=1, x mr1→r2 represents the number of people who purchased a specific product among those who moved from the source area r1 to the destination area r2. When the purchase attribute m=2, x mr1→r2 represents the number of people who have not purchased a specific product among those who moved from source area r1 to destination area r2.

[0144] Furthermore, for the number of people moving by purchase attribute for each combination of origin area and destination area in the third dataset DS53, u mr1→r2 Assign u mr1→r2 The interpretation of the index in x mr1→r2 is the same as:

[0145] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> The number of people who move from source area r1 to destination area r2 by purchase attribute, u mr1→r2 The total number of people moving is Σu mr1→r2 is the observed number of people moving from source area r1 to destination area r2, y r1→r2 This is within the margin of error based on the standard.

[0146] <Second constraint> The ratio of people who have purchase attribute m to the number of people who move from source area r1 to destination area r2 is u mr1→r2 / Σu mr1→r2is the proportion of observed migrants x mr1→r2 / Σx mr1→r2 This is within the margin of error based on the standard.

[0147] <Third constraint> Number of visitors in the source area r1 z r1 cannot be less than 0.

[0148] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: r1→r2 ,σ r1→r2 respectively follow normal distribution.

[0149]

number

[0150] The processor 11 estimates the number of travelers u by MCMC parameter estimation according to this model. mr1→r2 Then, for each combination of origin area and destination area, the number of people who move by purchase attribute u mr1→r2 A third data set DS53 describing estimates of

[0151] According to this embodiment, the above-mentioned "first set of analysis subjects" corresponds to a group of people in a specific area, the "second set" corresponds to a group of specific survey subjects, and the "first characteristic" and the "second characteristic" correspond to the number of migrants. As a modified example, an "attribute" other than the "purchase attribute" may be used. For example, the second dataset DS52 and the third dataset DS53 may be datasets that explain the number of migrants by demographic attribute (e.g., by gender and age group).

[0152] [Sixth embodiment] According to the sixth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and in S110, acquires a first dataset DS61 shown in Fig. 8. This first dataset DS61 has records representing sales of a specific product to a consumer group including Japanese and foreigners for each day. Furthermore, the first dataset DS61 has records representing annual sales of a specific product to foreigners for each year. Furthermore, the first dataset DS61 has records representing the number of foreigners who purchased a specific product for each day.

[0153] The processor 11 further acquires, in S120, a second data set DS62 shown in Fig. 8, which is related to the purchase volume of a specific consumer group selected from Japanese people. The second data set DS62 has records indicating the purchase volume for each day and each consumer attribute for the specific consumer group. The purchase volume for each consumer attribute may be, for example, the purchase volume by gender and age group. The purchase volume may be the purchase amount.

[0154] The processor 11 generates and outputs a third dataset DS63 explaining daily sales by consumer attributes for Japanese people shown in Figure 8 based on the acquired first dataset DS61 and second dataset DS62 (S130, S140).

[0155] According to this embodiment, in the analysis process of S130, the processor 11 first organizes the first data set DS61 and calculates daily sales of a specific product to Japanese consumers. Specifically, the daily sales of Japanese consumers 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 in proportion to the number of foreigners on each day.

[0156] The processor 11 generates a third data set DS63 based on the daily sales of a specific product to a group of Japanese consumers derived from the first data set DS61 and the second data set DS62.

[0157] Here, the variable y represents the daily sales of a particular product to a group of Japanese consumers derived from the first data set DS61. t where t is the time index corresponding to the day.

[0158] For the purchase amount by day and consumer attribute in the second data set DS62, the variable x mt m is the index of the consumer attribute. For the sales by day and consumer attribute in the third dataset DS63, we assign the variable u mt Assign.

[0159] In this embodiment, in S130, u is calculated in the same manner as in the first embodiment. mt , and calculate the posterior distribution of sales by day and consumer attribute for the Japanese consumer group. mt A third dataset DS63 describing estimated values ​​of may be generated. The third dataset DS63 may be generated as a dataset describing sales by segment that combines two or more consumer attributes on a daily basis.

[0160] According to this embodiment, the above-mentioned "first set of analysis targets" corresponds to the set of all consumers, the "second set" corresponds to a set of a specific consumer group, the "first characteristic" corresponds to sales, and the "second characteristic" corresponds to purchasing volume.

[0161] [Seventh embodiment] According to the seventh embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and acquires the first dataset DS71 shown in Fig. 9 in S110. This first dataset DS71 represents the sales ranking of a group of products belonging to a specific product category in the market. Specifically, the first dataset DS71 has records representing the sales amount of the corresponding product for each sales ranking in the market.

[0162] The processor 11 further acquires, in S120, a second data set DS72 relating to purchases of a specific consumer group shown in Fig. 9. This second data set DS72 represents a ranking of purchase amounts of a group of products belonging to a specific product category in a specific consumer group. Specifically, the second data set DS72 represents purchase amounts by consumer attribute of the specific consumer group for the corresponding products for each purchase ranking. The purchase amounts by consumer attribute may be purchase amounts by age group.

[0163] Based on the acquired first dataset DS71 and second dataset DS72, the processor 11 generates and outputs a third dataset DS73 shown in FIG. 9 which explains the sales of corresponding products by consumer attributes for each sales ranking in the market for a group of products in a specific product category (S130, S140).

[0164] Here, the variable y r r is the sales rank index. That is, y r represents the sales of the product corresponding to rank r.

[0165] Furthermore, for the purchase amount by consumer attribute for each rank in the second data set DS72, the variable x mr That is, assign x mr represents the purchase amount of the product corresponding to purchase rank r of a consumer group having consumer attribute m among a specific consumer group.

[0166] Furthermore, for the sales by consumer attribute for each rank in the third dataset DS73, the variable u mr That is, u mr represents the market sales of the product corresponding to sales ranking r for the consumption group of consumer attribute m.

[0167] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> Sales u for the product with the sales ranking r by consumer attribute mr The sum of Σumr is the sales y of the product corresponding to the observed sales rank r. r The error is within the range based on .

[0168] <Second constraint> For each consumer attribute m, the sales of the product corresponding to the sales ranking r is mr is the sales of the product with the previous sales ranking r-1. mr-1 , the sales of the product next in sales ranking r+1 is u mr+1 , and the purchase amount x of the product corresponding to the observed purchase order r mr This second constraint is a linear relationship between the sales mr Includes constraints on differences.

[0169] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: r ,e mr respectively follow normal distribution.

[0170]

number

[0171] The processor 11 estimates the parameters by MCMC according to this model to calculate the sales u mr Calculate the posterior distribution of sales by market sales rank and consumer attributes, u mr A third dataset DS73 describing an estimated value of may be generated. The third dataset DS73 may be generated as a dataset explaining sales by segment that combines two or more consumer attributes for each sales rank.

[0172] According to this embodiment, the above-mentioned "first set of analysis targets" corresponds to the set of all consumers, the "second set" corresponds to a set of a specific consumer group, the "first characteristic" corresponds to sales, and the "second characteristic" corresponds to purchase amount.

[0173] The first dataset DS71 has sales y as the first feature for each sales rank as a series element in the sales ranking. r The second dataset DS72 describes the purchase amount x mr The third dataset DS73 has sales rank as a series element in the sales ranking, and sales u as a third feature. mr It describes:

[0174] [Eighth embodiment] According to the eighth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and in S110, acquires a first data set DS81 shown in Fig. 10. This first data set DS81 has records representing, for a specific product, sales per week in the market for a consumer group of the corresponding age for each consumer age. Hereinafter, sales per week are referred to as weekly sales.

[0175] The processor 11 further acquires, in S120, a second data set DS82 relating to the purchase amounts of the specific consumer group, as shown in Fig. 10. The second data set DS82 has records indicating the purchase amounts by age of the specific consumer group for each weekday and holiday in the above-mentioned one week.

[0176] Based on the acquired first dataset DS81 and second dataset DS82, the processor 11 generates and outputs a third dataset DS83 shown in FIG. 10 which explains sales by consumer age on weekdays and holidays across the entire market (S130, S140).

[0177] Here, we use the variable y r r is the age index. That is, y r represents the weekly sales of a particular product to a consumer group of age r.

[0178] In addition, the variable x mr m is the index of the time attribute. The time attribute here means the attribute of a day classified as a weekday or a holiday. m takes the value 1 when the day corresponding to the purchase amount is a weekday, and takes the value 2 when it is a holiday. x mr represents the purchase amount of a specific product by a consumer group of age r on the day corresponding to time attribute m.

[0179] Furthermore, for the sales by consumer age on weekdays and holidays in the third dataset DS83, the variable u mr Assign u mr The interpretation of the index in x mr is the same as:

[0180] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> Sales u by time attribute of consumer age r mr The sum of Σu mr is the observed sales y r This is within the margin of error based on the standard.

[0181] <Second constraint> Sales u of consumer age r on the day corresponding to time attribute m mr is the sales u to the consumer group whose consumer age is r-1. mr-1 , sales to the consumer group whose consumer age is r+1, u mr+1 , and the purchase amount x of the observed consumer age r mr has a linear relationship with each of

[0182] <Third constraint> The sales amount is close between adjacent ages. Specifically, the sales amount of consumer age r is u mr and the sales of adjacent consumers of age r-1, u mr-1 Difference with |Σu mr -Σu mr-1 | is less than or equal to a threshold C.

[0183] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: r ,e mr respectively follow normal distribution.

[0184]

number

[0185] The processor 11 estimates the parameters by MCMC according to this model to calculate the sales u mr Calculate the posterior distribution of sales by consumer age in the market on weekdays and holidays, u mr A third data set DS83 describing estimates of

[0186] According to this embodiment, the above-mentioned "first set of analysis targets" corresponds to the set of purchasing events of all consumers, the "second set" corresponds to the set of purchasing events of a specific consumer group, the "first characteristic" corresponds to sales, and the "second characteristic" corresponds to purchase amount.

[0187] The first dataset, DS81, has weekly sales y as the first feature for each age group as a series element. r The second dataset DS82 describes the purchase amounts x for weekdays and holidays in a week as the second feature for each age as a series element. mr The third dataset DS83 has the following features: age as a series element, weekday and holiday sales u mr It describes:

[0188] [Ninth embodiment] According to the ninth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and in S110, acquires a first data set DS91 shown in Fig. 11. This first data set DS91 has records indicating the consumption amount of a specific commodity for each region (e.g., prefecture) throughout the country. The specific commodity may be, for example, a specific food such as tuna.

[0189] The processor 11 further acquires, in S120, a second dataset DS92 shown in FIG. 11 which explains the population by age group as a geographic characteristic of each town, which is a smaller area than the above-mentioned region, for the entire country.

[0190] Based on the acquired first dataset DS91 and second dataset DS92, the processor 11 generates and outputs a third dataset DS93 shown in FIG. 11 which explains the consumption of a specific product in each town throughout the country (S130, S140).

[0191] Here, the variable y r r is the index of the region, i.e., y r represents the consumption of a particular commodity in region r.

[0192] For the age-specific population of each town in the second data set, DS92, we use the variable x ma m is the town index and a is the age index. That is, x ma represents the population of age a in town m. For the consumption by town in the third dataset DS93, the variable u m Assign u m represents the consumption of a particular commodity in town m.

[0193] In this embodiment, the parameter estimation process is performed in S130 under the following constraints. <First constraint> Consumption by town in region r m The sum of Σu mis the observed consumption in region r, y r The error is within the range based on .

[0194] <Second constraint> Consumption by town m is the age-specific population of the corresponding town x ma is related to.

[0195] For parameter estimation subject to the above constraints, a simulation model can be put in place that follows the following equation: r ,e m respectively follow normal distribution.

[0196]

number

[0197] The processor 11 estimates the consumption amount u by parameter estimation using MCMC according to this model. m , and calculate the posterior distribution of the consumption of a particular product by town, u m A third dataset DS93 describing the estimates of consumption by segments of two or more towns may be generated.

[0198] According to this embodiment, the above-mentioned "set of analysis targets" corresponds to a set of towns, the "first characteristic" corresponds to consumption, and the "second characteristic" different from the first characteristic corresponds to the population by age, in other words, the population structure. According to this embodiment, the first dataset DS91 and the second dataset DS92 describe different first and second characteristics for the same set of towns, respectively.

[0199] According to this embodiment, a third dataset DS93 with finer information granularity regarding the first features can be generated based on a first dataset DS91 describing a first feature related to a set and a second dataset DS92 describing, with finer information granularity, a second feature related to the same set as the first dataset DS91.

[0200] [Tenth embodiment] According to the tenth embodiment, the processor 11 starts the data generation process shown in Fig. 2 based on an execution command, and in S110, acquires a first data set DS101 shown in Fig. 12. This first data set DS101 has records indicating the annual consumption amount of a specific commodity in a specific region for each year.

[0201] The processor 11 further obtains, in S120, a second data set DS102 shown in Fig. 12, which describes, for each combination of year and month, the average temperature for that month as a geographic characteristic of the corresponding month in the corresponding year, by area in a specific region. The term "month" hereinafter should be understood to mean a month that is distinguished by the combination of year and month.

[0202] The processor 11 generates and outputs a third dataset DS103 shown in FIG. 12 that explains the monthly consumption of a specific commodity in a specific region based on the acquired first dataset DS101 and second dataset DS102 (S130, S140).

[0203] Here, the variable y r where r is the year index, i.e., y r represents the annual consumption for the corresponding year r.

[0204] For the monthly mean temperatures in the second data set, DS102, we use the variable x ma m is the month index and a is the area index. That is, x ma represents the average temperature in area a for month m. For the monthly consumption of a particular commodity in the third dataset DS103, the variable u m Assign u m represents the monthly consumption of a particular commodity in month m.

[0205] In this embodiment, the parameter estimation process is performed under the following constraints. <First constraint> Monthly consumption u in a given year r m The sum of Σu m is the observed annual consumption in year r, y r This is within the margin of error based on the standard.

[0206] <Second constraint> Monthly consumption m is the average temperature by area for the corresponding month x ma is related to.

[0207] For parameter estimation according to the above constraints, a simulation model according to the same formula as in the ninth embodiment can be set. The processor 11 calculates the monthly consumption u by parameter estimation using MCMC according to this model. m , and calculate the monthly consumption in a specific area u m A third dataset DS103 describing an estimate of consumption by segments may be generated. The third dataset DS103 may be generated as a dataset describing consumption by segments taken together for two or more months.

[0208] According to this embodiment, the above-mentioned "set of analysis targets" corresponds to a set of time on a monthly basis, the "first characteristic" corresponds to the consumption amount, and the "second characteristic" different from the first characteristic corresponds to the average temperature by area.

[0209] Depending on the type of product, the type of geographical characteristics that affect the behavioral characteristics of consumers may differ. For example, precipitation or sunshine hours may affect the behavioral characteristics of consumers. Therefore, as a modification of this embodiment, an example in which precipitation or sunshine hours by area are used instead of the average temperature by area may be used.

[0210] Geographical characteristics other than temperature, precipitation, and sunshine hours may be used in the analysis. Examples of geographical characteristics include meteorological characteristics such as atmospheric pressure in the corresponding area. Examples of geographical characteristics include the demographic composition of the corresponding area, as described in the ninth embodiment. Examples of geographical characteristics may further include the number of specific facilities or stores in the corresponding area, characteristics of television broadcasting content in the corresponding area, characteristics such as interests, personalities, and temperaments of residents, and characteristics related to the classification and distribution of land in the corresponding area.

[0211] Furthermore, the consumption amount is one of the feature quantities representing the behavioral characteristics of a consumer. The technical idea of ​​this embodiment may be used to estimate another feature quantity representing the behavioral characteristics of a consumer instead of the consumption amount. Examples of the behavioral characteristics may include, for example, characteristics related to one or more of purchasing behavior, media contact behavior, facility utilization behavior, entertainment activities, indoor and outdoor movement behavior, and web browsing behavior. Media contact behavior may include contact behavior with content provided by information transmission media such as television broadcasts, radio broadcasts, newspapers, magazines, outdoor advertisements, and the Internet.

[0212] According to this embodiment, it is possible to estimate the feature amount related to the behavioral characteristics of a person in a fine time unit based on the data set that describes the geographical characteristics by area and time. A similar effect can be obtained by the ninth embodiment. According to the ninth embodiment, it is possible to estimate the feature amount related to the behavioral characteristics of a person in a fine area unit.

[0213] [Other embodiments] The present disclosure is not limited to the above-described embodiments, and may take various forms. In the first to tenth embodiments, a data set related to human behavior is handled, but the technology of the present disclosure may be applied to generate a fine-grained data set of information related to non-human organisms, non-biological objects, or physical phenomena.

[0214] Furthermore, the above-mentioned datasets, such as the first dataset, may not be based on a census of the corresponding analysis target population, but may be based on a sample survey. For example, the above-mentioned dataset may be generated by estimating characteristics of the entire population from a sample. Data obtained from a large-scale panel survey with a large number of samples may have substantially the same value as census data for analytical purposes.

[0215] In addition, the function of one component in the above embodiment may be distributed among multiple components. The functions of multiple components may be integrated into one component. Some of the configurations of the above embodiment may be omitted. All aspects included in the technical idea specified by the wording of the claims are embodiments of the present disclosure. [Explanation of symbols]

[0216] 1...data generation system, 11...processor, 12...memory, 13...storage, 15...display, 17...input device, 18...media reader / writer, 19...communication device, DS11, DS21, DS31, DS41, DS51, DS61, DS71, DS81, DS91, DS101...first dataset, DS12, DS22, DS32, DS42, DS52, DS62, DS72, DS82, DS92, DS102...second dataset, DS13, DS23, DS33, DS43, DS53, DS63, DS73, DS83, DS93, DS103...third dataset.

Claims

1. An acquisition unit configured to acquire a first data set and a second data set; a generator configured to generate a third dataset based on the first dataset and the second dataset; Equipped with the first dataset is a dataset expressing a first feature related to a first set at a first granularity, describing a first feature amount related to the first feature for each element group with respect to one or more element groups in the first set, the first set includes a plurality of elements, each of the element groups is an element group corresponding to the first granularity in the first set, The second dataset is a dataset in which a second feature different from the first feature for a second set that is the same as the first set, or a second feature that is the same as or different from the first feature for a second set that is smaller than the first set, is expressed at a second granularity finer than the first granularity, and describes a second feature amount for the second feature for each of a plurality of elements in the second set, the second set includes a plurality of elements, and each of the elements is one or more elements corresponding to the second granularity in the second set, the third dataset is a dataset expressing the first feature for the first set at a third granularity that is finer than the first granularity and coarser than the second granularity or equal to the second granularity, and describes a third feature amount for the first feature for each of a plurality of segments in the first set, each of the segments being one or more elements corresponding to the third granularity in the first set; The generation unit is generating the third dataset by estimating the third feature amount for each of the segments based on the first dataset and the second dataset under a first constraint condition regarding the estimation of the third feature amount based on a relationship between the first feature amount and the third feature amount and a second constraint condition regarding the estimation of the third feature amount based on a relationship between the second feature amount and the third feature amount; When generating the third data set, an optimum value of a design parameter conforming to the first constraint condition and the second constraint condition of an estimation model for estimating the third feature quantity from the first feature quantity and the second feature quantity is searched for based on the first data set and the second data set by statistical analysis taking into account a probability distribution regarding the design parameter, and the third feature quantity is estimated according to the estimation model corresponding to the optimum value of the design parameter. Data generation system.

2. The data generation system according to claim 1, A data generation system, wherein the second dataset is a dataset in which the second feature, which is the same or different from the first feature for the second set which is smaller than the first set, is expressed at the second granularity which is finer than the first granularity.

3. 3. The data generation system according to claim 1, the first dataset is a dataset describing the first feature amount in a time series, and describes, for each of a plurality of time points, the first feature amount for each of the element groups at a corresponding time point; The second data set is a data set describing the second feature amount in a time series, and describes, for each of the plurality of time points, the second feature amount for each of the entities at a corresponding time point; The generation unit is a data generation system that estimates the third feature amount for each segment at at least one of the multiple time points based on the first dataset and the second dataset.

4. 4. The data generating system according to claim 3, The generation unit estimates the third feature amount for each of the segments at the multiple time points; At least one of the first constraint condition and the second constraint condition includes a constraint condition regarding a time change of the third feature amount.

5. 4. The data generating system according to claim 3, The generation unit estimates the third feature amount for each of the segments at the multiple time points; The second constraint condition includes a constraint condition that, for each of the multiple time points, the third feature at a corresponding time point depends in a predetermined relationship on the third feature at a time point prior to the corresponding time point and on the second feature at the corresponding time point.

6. The data generation system according to claim 3 or 5, The generation unit estimates the third feature amount for each of the segments at the multiple time points; A data generation system in which the first constraint condition includes a constraint condition that, for each of the multiple time points, the sum of the third feature for each of the segments at a corresponding time point corresponds to the first feature of the entire segment at the corresponding time point.

7. 3. The data generation system according to claim 1, the first data set describes, for each location, the first feature for each of the element groups for a corresponding location, for a plurality of locations; the second data set describes, for each of the plurality of locations, the second feature amount for each of the entities related to the corresponding location; The generation unit is a data generation system that estimates the third feature amount for each segment at at least one location among the multiple locations based on the first dataset and the second dataset.

8. 3. The data generation system according to claim 1, The first set is a set of a plurality of areas including the area as an element, The element group is a region including multiple geographically adjacent areas, The element is one of the plurality of regions, The first data set represents, for each of the regions, a feature amount related to a behavioral characteristic of a person in the corresponding region as the first feature amount; The second dataset represents, for each of the sections smaller than the region, a feature amount related to a geographic characteristic of the corresponding section as the second feature amount; The generation unit is a data generation system that estimates, for each area as the segment, a feature related to the behavioral characteristics of people within the corresponding area as the third feature based on the first dataset and the second dataset.

9. 3. The data generation system according to claim 1, the first set is a set of a plurality of unit periods including a unit period as an element, the element group is a first period that is longer than a plurality of unit periods and is composed of two or more unit periods among the plurality of unit periods; The element is a second period that is composed of one or more unit periods among the plurality of unit periods and has the same time length as the unit period or is longer than the unit period, the second period is shorter than the first period, The first data set represents, for each of the first periods, a feature amount related to a behavioral characteristic of a person within a corresponding first period as the first feature amount; The second dataset represents, for each of the second time periods, a feature amount related to a geographic characteristic of the corresponding second time period as the second feature amount; The generation unit is a data generation system that estimates, for each of the second periods as the segments, a feature related to the behavioral characteristics within the corresponding second periods as the third feature based on the first dataset and the second dataset.

10. The data generation system according to any one of claims 1 to 9, At least one of the first constraint condition and the second constraint condition includes an error term related to a variance of the third feature amount.

11. A computer program 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 10.

12. 1. A computer-implemented method for generating data, comprising: Obtaining a first data set and a second data set; generating a third data set based on the first data set and the second data set; Including, the first dataset is a dataset expressing a first feature related to a first set at a first granularity, describing a first feature amount related to the first feature for each element group with respect to one or more element groups in the first set, the first set includes a plurality of elements, each of the element groups is an element group corresponding to the first granularity in the first set, The second dataset is a dataset in which a second feature different from the first feature for a second set that is the same as the first set, or a second feature that is the same as or different from the first feature for a second set that is smaller than the first set, is expressed at a second granularity finer than the first granularity, and describes a second feature amount for the second feature for each of a plurality of elements in the second set, the second set includes a plurality of elements, and each of the elements is one or more elements corresponding to the second granularity in the second set, the third dataset is a dataset expressing the first feature for the first set at a third granularity that is finer than the first granularity and coarser than the second granularity or equal to the second granularity, and describes a third feature amount for the first feature for each of a plurality of segments in the first set, each of the segments being one or more elements corresponding to the third granularity in the first set; the generating step includes estimating the third feature quantity for each of the segments based on the first dataset and the second dataset under a first constraint condition for estimating the third feature quantity based on a relationship between the first feature quantity and the third feature quantity and a second constraint condition for estimating the third feature quantity based on a relationship between the second feature quantity and the third feature quantity, and generating the third dataset, wherein the data generating method includes searching for an optimum value of a design parameter that complies with the first constraint condition and the second constraint condition of an estimation model for estimating the third feature quantity from the first feature quantity and the second feature quantity by statistical analysis based on the first dataset and the second dataset, taking into account a probability distribution for the design parameter, estimating the third feature quantity according to the estimation model corresponding to the optimum value of the design parameter, and generating the third dataset.

13. The data generation method according to claim 12, comprising: A data generation method, wherein the second dataset is a dataset in which the second feature, which is the same as or different from the first feature for the second set which is smaller than the first set, is expressed at the second granularity which is finer than the first granularity.

Images