Data analysis apparatus and data analysis program

Abstract

To provide a data analysis apparatus which can predict a promotion effect of various project conditions including those that have never been implemented.SOLUTION: A data analysis apparatus according to an embodiment of the present invention includes a first conversion unit, a second conversion unit, and a learning unit. The first conversion unit converts project conditions to latent variables according to similarity of contribution to KPI. The second conversion unit converts the latent variables to a prediction value for KPI. The learning unit learns at least parameters of the first conversion unit based on past conditions and past project results.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] Embodiments of the present invention relate to a data analysis apparatus and a data analysis program.

Background Art

[0002] An apparatus for optimizing marketing measures is known. For example, such an apparatus calculates a predicted value of a KPI (Key Performance Indicator), which is an evaluation index of marketing measures, according to a target product and basic attributes of customers. The KPI value includes sales, profit margin, trial rate, repeat rate, and the like. In the case of a target product for which no measures have been implemented, correction is made based on the "coefficient" and similarity of similar products.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] It is impossible to predict the KPI value under unknown measure conditions. That is, when measure conditions other than the target product (target customers, stores, regions, provided values, etc.) are different, the KPI value cannot be predicted. The correspondence between the similarity between products (for example, the similarity of list price, product classification, product name) and the KPI is unclear. Just because the similarity is high, the response (or coefficient) of the KPI value is not necessarily similar.

[0005] The problem to be solved by the present invention is to provide a data analysis apparatus and a data analysis program capable of predicting a promotion effect for various planning conditions including non-implementation.

Means for Solving the Problems

[0006] The data analysis apparatus according to the embodiment includes a first conversion unit, a second conversion unit, and a learning unit. The first conversion unit converts the planning conditions into latent variables according to the similarity of the contributions to the KPIs. The second conversion unit converts the latent variables into predicted values of the KPIs. The learning unit learns at least the parameters of the first conversion unit based on past conditions and past planning results.

[0007] The data analysis program according to the embodiment causes a computer to execute a first conversion function of converting planning conditions into latent variables according to the similarity of the contributions to the KPIs, a second conversion function of converting the latent variables into predicted values of the KPIs, and a learning function of learning at least the parameters of the first conversion function based on past conditions and past planning results.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Embodiments for Carrying Out the Invention

[0009] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0010] (Embodiment) First, referring to FIG. 1, the data analysis apparatus 10 according to the embodiment will be described. FIG. 1 is a diagram showing the functional blocks of the data analysis apparatus 10 according to the embodiment.

[0011] The data analysis apparatus 10 according to the embodiment includes a first conversion unit 11, a second conversion unit 12, and a learning unit 13.

[0012] The first conversion unit 11 takes the planning condition xi as an input. The first conversion unit 11 converts the planning condition xi into a latent variable zi according to the similarity of the contribution to the KPI, and outputs the latent variable zi. For example, the first conversion unit 11 performs the conversion using a learned model by machine learning.

[0013] The second conversion unit 12 takes the latent variable zi output by the first conversion unit 11 as an input. The second conversion unit 12 converts the latent variable zi into a result prediction value yi, and outputs the result prediction value yi. For example, the second conversion unit 12 performs the conversion using a learned model by machine learning.

[0014] The learning unit 13 learns the parameters of the first conversion unit 11 and the second conversion unit 12 based on the past condition X' and the past planning result Y'. For example, the learning unit 13 learns the parameters of the learned models used by the first conversion unit 11 and the second conversion unit 12.

[0015] For example, the data analysis apparatus 10 is configured by a computer. The computer executes a data analysis program to execute the functions of the first conversion unit 11, the second conversion unit 12, and the learning unit 13.

[0016] The planning conditions include, for example, characteristics of the target product, characteristics of the target customer, characteristics of the target store / region, and other characteristics.

[0017] The features of the target product include, for example, the product index (such as product code, etc.). Also, the features of the target product include, for example, the manufacturer, category, quantity. Also, the features of the target product include, for example, the tendency of co-sold (products that are likely to be purchased simultaneously) products, region / store and sales volume. Also, the features of the target product may include text data such as product descriptions, and other features.

[0018] The features of the target customer include, for example, demographics, questionnaires, hobbies, and others. Also, the features of the target customer include, for example, the features of the purchased products. Also, the features of the target customer may include other features.

[0019] The features of the target store / region include, for example, the region (one-hot vector, latitude and longitude), climate (temperature, rainfall,...), store brand, scale. Also, the features of the target store / region may include the features of the sold products, text data such as store descriptions, and other features.

[0020] Other features include, for example, the implementation date and time (season, month, day of the week, xx's day, morning / noon / night), weather. Also, other features may include other features.

[0021] The extraction of the planning conditions is performed, for example, as follows.

[0022] For category features (such as product index, customer ID, target store, etc.), one-hot vectorization may be used. Alternatively, pre-learned feature extraction may be used.

[0023] Feature extraction can be learned based on purchase data. For example, for each customer or each target store, a matrix showing the sales / purchase quantity (or transformed value such as logarithm, etc.) of each product / category can be prepared, and feature embedding can be learned by applying various dimensionality reduction methods. As the dimensionality reduction method, for example, matrix decomposition can be used. As matrix decomposition, for example, PCA (Principal Component Analysis), ICA (Independent Component Analysis), NMF (Non-Negative Matrix Factorization), AE (Autoencoder), etc. can be used.

[0024] For text data (WEB, product / store descriptions, etc.), it is performed using feature extraction (embedding) and feature word search (score value for a given search word). Feature extraction may be extracted using a pre-trained large language model (LLM: Large language Models).

[0025] Continuous value data such as numerical features or time features may be subjected to a predetermined input transformation. The input transformation performs, for example, a basis transformation. For example, the basis transformation includes RBF (Radial Basis Function), polynomial, sine wave, wavelet, etc.

[0026] The learning unit 13 performs learning using a loss function and a link function according to the KPI value. The KPI value is, for example, any one value selected from sales, profit margin, trial rate, repeat rate, etc. For learning, MSE (Mean Squared Error), logloss (logarithmic loss), poisson (Poisson loss), etc. can be used.

[0027] The learning range is the parameters (coefficients) of the first conversion unit 11 and the parameters (coefficients) of the second conversion unit 12. When the planning condition xi includes category features and text data, the coefficients of feature extraction may also be learned.

[0028] In addition to parameter learning, the learning unit 13 may also optimize the hyperparameters of the learning unit 13. Hyperparameters include regularization strength, latent feature dimension, conversion unit function shape, and the like. For example, the hyperparameters are optimized based on the CV (Cross-Validation Loss) loss criterion.

[0029] The first conversion unit 11 and the second conversion unit 12 perform conversion using a linear function, a single-layer perceptron, or a multi-layer perceptron. In addition, the first conversion unit 11 and the second conversion unit 12 perform regularization. Regularization includes, for example, L1 regularization, L2 regularization, and dropout. When performing regularization, the number of latent feature dimensions may be utilized.

[0030] According to the data analysis apparatus 10 according to the embodiment, for various planning conditions, for example, even for planning conditions that have never been implemented, the promotion effect can be predicted by referring to the results of similar conditions. In addition, since modeling is performed including the results of similar conditions, the prediction accuracy is improved particularly when the planning conditions are diverse, when there is a lot of noise, or when the data is insufficient.

[0031] (First Modification Example) Next, with reference to FIG. 2, the data analysis apparatus 10 according to the first modification example will be described. FIG. 2 is a diagram showing the functional blocks of the data analysis apparatus 10 according to the first modification example.

[0032] The configuration of the data analysis apparatus 10 according to the first modification example is the same as that of the data analysis apparatus 10 according to the embodiment. However, the input and output of the first conversion unit 11, the input of the second conversion unit 12, and the learning target of the learning unit 13 are different.

[0033] In the data analysis device 10 according to the first modification example, the first conversion unit 11 takes the condition xi regarding the promotion target as an input. Hereinafter, for the sake of convenience, the condition xi regarding the promotion target is referred to as the promotion target condition xi. The promotion target condition xi is a product feature or the like. For example, in the above-described example, the promotion target condition xi is the feature of the target product of the planning condition. The first conversion unit 11 converts the promotion target condition xi into a coefficient θi used for conversion by the second conversion unit 12.

[0034] The second conversion unit 12 takes the coefficient θi converted by the first conversion unit 11 and the detailed promotion condition vi as inputs. Hereinafter, for the sake of convenience, the detailed promotion condition vi is referred to as the promotion detailed condition vi. The promotion detailed condition vi is a planning condition excluding the promotion target condition xi. The promotion detailed condition vi includes the offered value or the like. For example, in the above-described example, the promotion detailed condition vi includes the features of the target customer, the features of the target store / region, and other features excluding the features of the target product from the planning condition. The second conversion unit 12 converts the promotion detailed condition vi into a result prediction value yi based on the coefficient θi.

[0035] The second conversion unit 12 converts the promotion detailed condition vi into a result prediction value yi by a given function based on the coefficient θi. The given function is, for example, a linear function. In this case, the second conversion unit 12 calculates the result prediction value yi based on the following formula.

[0036] yi = Σk (θik × xik) However, k corresponds to each dimension of the promotion target condition x.

[0037] The second conversion unit 12 may apply a predetermined basis conversion to xi and then input it to a linear function. In addition, the given function may be a single-layer or multi-layer perceptron. In this case, the coefficient θi corresponds to the weight of each layer.

[0038] The learning unit 13 learns the parameters of the first conversion unit 11 based on the past conditions (past promotion target conditions X' and past promotion detailed conditions V') and the past planning results Y'. For example, the learning unit 13 learns the parameters (coefficients) of the machine learning model used by the first conversion unit 11.

[0039] The data analysis device 10 according to the first modification example has the same advantages as the data analysis device 10 according to the embodiment. In addition, according to the data analysis device 10 according to the first modification example, the effect according to the promotion detailed conditions corresponding to the promotion target can be evaluated. For example, the base response rate, value dependence, etc. corresponding to the target product can be evaluated. Thereby, it becomes easy to estimate more appropriate detailed conditions for each promotion target.

[0040] (Second Modification Example) Next, with reference to FIG. 3, the data analysis device 10 according to the second modification example will be described. FIG. 3 is a diagram showing the functional blocks of the data analysis device 10 according to the second modification example.

[0041] The configuration of the data analysis device 10 according to the second modification example has a sampling unit 14 in addition to the configuration of the data analysis device 10 according to the embodiment.

[0042] In the data analysis device 10 according to the second modification example, the first conversion unit 11 takes the promotion target condition xi as an input, similar to the data analysis device 10 according to the first modification example. The first conversion unit 11 converts the promotion target condition xi into parameters of a probability distribution, that is, a Bayesian posterior distribution, according to the likelihood of the coefficient θi used for the conversion by the second conversion unit 12. The probability distribution may be approximated by, for example, a multivariate normal distribution. In this case, the approximate posterior distribution parameters include the coefficient expected value μi and the coefficient variance Σi.

[0043] The sampling unit 14 takes the coefficient expected value μi and the coefficient variance Σi converted by the first conversion unit 11 as inputs. The sampling unit 14 samples the coefficient θi based on the multivariate normal distribution N(μi,Σi) from the coefficient expected value μi and the coefficient variance Σi. For example, the sampling unit 14 samples the coefficient θij (j = 1...n) at a ratio of n samples per 1 data sample. n is an integer of 1 or more. The sampling unit 14 outputs the sampled coefficient θij to the second conversion unit 12.

[0044] The second conversion unit 12 takes as input the coefficient θi output by the sampling unit 14 and the promotion detail condition vi. Based on the coefficient θij, the second conversion unit 12 converts the promotion detail condition vi into the result prediction value yi.

[0045] In the second modification, the first conversion unit 11 converts to the parameters of the posterior distribution of the coefficient θi. The learning unit 13 learns the parameters of the first conversion unit 11 so as to approximate the posterior distribution of the coefficient θi with respect to a given prior distribution based on the past conditions (past promotion target conditions X' and past promotion detail conditions V') and the past planning results Y'.

[0046] When the past planning result Y' is given, the posterior distribution of the coefficient θ is shown by the following formula according to Bayes' theorem with respect to the given prior distribution p(θ).

[0047] p(θ|Y') ∝ p(Y'|Y)p(θ)

[0048] However, Y is the result prediction value, which is calculated by the second conversion unit 12 from the coefficient θ and the promotion detail condition V. p(Y'|Y) is the likelihood function, and for example, a normal distribution, a binomial distribution, a Poisson distribution, etc. may be used according to the KPI value. The prior distribution p(θ) may be arbitrarily set, for example, as a multi-dimensional normal distribution, etc.

[0049] The learning unit 13 may learn the parameters of the first conversion unit 11 so as to obtain the coefficient expectation value μi and the coefficient variance Σi that approximate the posterior distribution using the variational Bayes method. In this case, the probability distribution (for example, N(μi, Σi)) obtained by the output of the first conversion unit 11 is used as the approximate posterior distribution q(θ), and learning is performed so as to minimize the KL divergence between the posterior distribution p(θ|Y') and q(θ). In this case, the coefficient samples generated by the sampling unit 14 may be used to approximately evaluate the KL divergence based on the sample expectation value of the log-likelihood.

[0050] When the first conversion unit 11 evaluates the probability distribution of the coefficient θi as a multivariate normal distribution, either a diagonal matrix or a Cholesky lower triangular matrix can be used as the parameter of the covariance matrix Σi. When using the Cholesky lower triangular matrix, the first conversion unit 11 may calculate and output the covariance matrix by the product of the matrix and its transposed matrix.

[0051] The second conversion unit 12 calculates the result prediction value yi based on the coefficient θi sampled by the sampling unit 14. For each of the n samples of the coefficient θij (j = 1...n) of 1 data sample sampled by the sampling unit 14, the second conversion unit 12 may calculate and output a plurality of result prediction values yij in the same manner as in the first modification example. Alternatively, the result prediction value yi may be calculated and output by taking the average of the n samples for yij.

[0052] The data analysis device 10 according to the second modification example has the same advantages as the data analysis device 10 according to the embodiment. In addition, according to the data analysis device 10 according to the second modification example, since the accuracy in an unknown promotion target can be evaluated from the variation of a plurality of result prediction values, a planning design assuming a prediction error can be implemented.

[0053] (Third Modification Example) Next, with reference to FIG. 4, the data analysis device 10 according to the third modification example will be described. The third modification example is a further modification example of the second modification example. FIG. 4 is a diagram showing the functional blocks of the data analysis device 10 according to the third modification example.

[0054] The configuration of the data analysis device 10 according to the third modification example has a third conversion unit 15 in addition to the configuration of the data analysis device 10 according to the second modification example.

[0055] In the data analysis apparatus 10 according to the third modification example, similar to the data analysis apparatus 10 according to the second modification example, the first conversion unit 11 takes the promotion target condition xi as an input. The first conversion unit 11 converts the promotion target condition xi into a probability distribution according to the likelihood of the coefficient θi used for the conversion by the second conversion unit 12, that is, into a parameter of the Bayesian posterior distribution. The probability distribution may be approximated by, for example, a multivariate normal distribution. In this case, the approximate posterior distribution parameters include the coefficient expected value μi and the coefficient variance Σi.

[0056] Also, the third conversion unit 15 takes at least a part of the promotion target condition xi as an input. (For example, the promotion target condition xi includes a product index and a product category, and only the product category among them is input to the third conversion unit 15). The third conversion unit 15 converts a part of the promotion target condition xi into a probability distribution of the coefficient θpi used for the conversion by the second conversion unit 12. The probability distribution may be approximated by, for example, a multivariate normal distribution. In this case, the approximate distribution parameters include the coefficient expected value μpi and the coefficient variance Σpi. The third conversion unit 15 may use, similar to the first conversion unit 11, a linear function, a single-layer neural network, a multi-layer neural network, etc., and may be learned by the learning unit 13.

[0057] The sampling unit 14 takes the coefficient expected value μi and the coefficient variance Σi converted by the first conversion unit 11, and the coefficient expected value μpi and the coefficient variance Σpi converted by the third conversion unit 15 as inputs. The sampling unit 14 samples a coefficient θi based on the multivariate normal distribution N(μi,Σi) from the coefficient expected value μi and the coefficient variance Σi. Also, the sampling unit 14 samples a coefficient θpi based on the multivariate normal distribution N(μpi,Σpi) from the coefficient expected value μpi and the coefficient variance Σpi. The sampling unit 14 outputs the sampled coefficients θi and θpi to the second conversion unit 12.

[0058] The second conversion unit 12 takes the coefficient θi and the coefficient θpi output by the sampling unit 14, and the promotion detail condition vi as inputs. The second conversion unit 12 converts the promotion detail condition vi into a result prediction value yi based on the coefficients θi and θpi.

[0059] In the third modification example, instead of using a given distribution for the prior distribution p(θ), a probability distribution obtained from the output of the third conversion unit 15 (for example, when approximated by a multivariate normal distribution, N(μpi, Σpi)) is used. The learning unit 13 may learn the parameters of the first conversion unit 11 so that the coefficient expected value μi and the coefficient variance Σi that approximate the posterior distribution are obtained, using the variational Bayes method in the same manner as in the second modification example. In addition, the learning unit 13 may learn the parameters of the third conversion unit 15 by minimizing the same KL divergence as in the second modification example. Alternatively, after setting an appropriate prior distribution, the posterior distributions of μpi and Σpi may be further estimated (for example, approximate estimation based on variational inference).

[0060] The third conversion unit 15 may calculate μpi and Σpi, for example, by variational Gaussian process regression. In GPR, the expected value / variance is determined according to the distance on the kernel. That is, a prior distribution with a similar shape is obtained for similar products.

[0061] The sampling unit 14 may sample the coefficient θpi based on the probability distribution (N(μpi, Σpi)) obtained from the output of the third conversion unit 15. The probability distribution corresponds to the prior distribution of the coefficient θi and is calculated based on a part of the promotion target condition xi (for example, product category, manufacturer, brand, etc.). Thereby, even for an unknown promotion target product, it is possible to make a prediction based on the learning results of similar products.

[0062] The data analysis device 10 according to the third modification example has the same advantages as the data analysis device 10 according to the second modification example. Furthermore, the data analysis device 10 according to the third modification example can evaluate the effects and prediction accuracy in an unknown promotion target by learning the prior distribution. In addition, if there are coefficients with high commonality among the promotion targets, they can be learned and the prediction accuracy can be improved. (Other Modification Examples)

[0063] In Modification 2, an appropriate promotion detail condition vi may be output to the outside (e.g., a user, etc.) using the probability distribution of the promotion result prediction yi for a given promotion target. For example, for any measure condition v, v that maximizes any index value such as the probability of improvement, the expected improvement, or the upper confidence bound may be output. Thereby, the measure condition to be tried next can be determined from the existing data.

[0064] Also, the learning unit 13 re-learns based on the new promotion measure results. That is, the learning unit 13 additionally receives xi (or xi and vi) and yi and re-learns.

[0065] (Operation example) Next, with reference to FIG. 5, an operation example of the data analysis apparatus 10 according to the embodiment and the modification examples (the first to third modification examples) will be described. FIG. 5 is a flowchart showing an operation example of the data analysis apparatus 10 according to the embodiment and the modification examples.

[0066] In ACT11, the learning unit 13 learns the parameters of the conversion unit (the first conversion unit 11 (and the second conversion unit 12)) based on past promotion data (past planning conditions X’ (or past conditions X’, V’) and past planning results Y’).

[0067] In ACT12, the first conversion unit 11 converts the planning condition xi (or the promotion target condition xi) into a latent variable zi according to the similarity of the contribution to the KPI.

[0068] In ACT13, the second conversion unit 12 converts the latent feature zi into a result prediction value yi of the KPI value.

[0069] In ACT14, using the result prediction value yi, an appropriate planning condition xi (or the promotion target condition xi and the promotion detail condition vi) is estimated and proposed.

[0070] In ACT15, the user refers to the proposed content, determines the planning conditions, and conducts the promotion.

[0071] In ACT15, the promotion data (planning condition X (or promotion target condition X and promotion detail condition V) and planning result Y) is saved.

[0072] (Hardware Configuration) Next, the hardware configuration of the data analysis device 10 according to the embodiment and the modification example will be described. Here, an example in which the data analysis device 10 is configured by a computer 20 will be described.

[0073] The hardware configuration of the computer 20 that constitutes the data analysis device 10 according to the embodiment and the modification example is shown in FIG. 6. The computer 20 includes a processor 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, an auxiliary storage device 24, and an input / output interface 25.

[0074] The processor 21, the ROM 22, the RAM 23, the auxiliary storage device 24, and the input / output interface 25 are electrically connected to each other via a bus 26, and data is transmitted and received via the bus 26.

[0075] The processor 21 is configured by a general-purpose hardware processor including, for example, a CPU (Central Processing Unit), a GPU (Graphical Processing Unit), or the like. The processor 21 controls the entirety of the ROM 22, the RAM 23, the auxiliary storage device 24, and the input / output interface 25.

[0076] The ROM 22 is a non-volatile memory that forms part of the main memory device. The ROM 22 non-temporarily stores the startup program necessary when the processor 21 starts up. The processor 21 starts up by executing the program in the ROM 22. The ROM 22 is composed of, for example, an EPROM (Erasable Programmable Read Only Memory), and in addition to the startup program, stores various settings at startup.

[0077] The RAM 23 is a volatile memory that forms part of the main memory device. The RAM 23 temporarily stores the programs necessary for the processing of the processor 21 and the data necessary for the execution of the programs. The processor 21 executes the programs in the RAM 23 to calculate the data in the RAM 23 and stores the calculation results in the RAM 23.

[0078] The auxiliary storage device 24 is composed of a non-volatile memory such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The auxiliary storage device 24 non-temporarily stores the programs executed by the processor 21 and the data necessary for the execution of the programs. The processor 21 reads the programs and data in the auxiliary storage device 24 into the RAM 23 and executes the programs to execute various functions.

[0079] The input / output interface 25 is connected to an external input device 31, an output device 32, etc., and enables the input of information from the input device 31 and the output of information to the output device 32. For example, the input / output interface 25 may be a wired interface or a wireless interface. The wired interface includes ports to which devices are connected. The wireless interface includes Bluetooth (registered trademark), WiFi (registered trademark), etc.

[0080] The input device 31 may include a keyboard, a mouse, a touch panel, a receiving device, a disk drive, etc. The input device 31 is not limited thereto and may include any other input device. The output device 32 may include a display, a transmitting device, a disk drive, etc. The output device 32 is not limited thereto and may include any other output device. The input device 31 and the output device 32 may be configured by an input / output device 33 having both functions.

[0081] The program non-temporarily stored in the auxiliary storage device 24 is provided to the computer 20, for example, via a recording medium 34 readable by the computer 20 that non-temporarily records the program. Such a recording medium 34 is called a non-temporary computer-readable recording medium. The non-temporary computer-readable recording medium includes disks such as flexible disks, optical disks (CD-ROM, CD-R, DVD-ROM, DVD-R, etc.), magneto-optical disks (MO, etc.), and semiconductor memories.

[0082] The program non-temporarily stored in the auxiliary storage device 24 includes a data analysis program. The data analysis program is a program that causes the computer 20 to execute at least some of the functions of the components of the data analysis device 10. For example, the data analysis program causes the computer 20 to execute the functions of the first conversion unit 11, the second conversion unit 12, and the learning unit 13 (and the sampling unit 14) of the data analysis device 10.

[0083] When the recording medium 34 is a disk, the program non-temporarily stored in the auxiliary storage device 24 is read into and non-temporarily stored in the auxiliary storage device 24 via the disk drive that is the input device 31 and the input / output interface 25. When the recording medium 34 is a semiconductor memory, the program is read into and non-temporarily stored in the auxiliary storage device 24 via the port that is the input / output interface 25. Also, the program may be stored in a server on a network, downloaded from the server via the input / output interface 25, and non-temporarily stored in the auxiliary storage device 24.

[0084] When the computer 20 is powered on, the processor 21 executes the program in the ROM 22, reads and boots the OS into the RAM 23. Under the control of the OS, the processor 21 monitors instruction inputs and connections to external devices. Also under the control of the OS, the processor 21 sets a program area and a data area in the RAM 23. In response to an instruction input to start the data analysis device 10, the processor 21 reads the data analysis program from the auxiliary storage device 24 into the program area of the RAM 23 and reads the data necessary for the execution of the data analysis program from the auxiliary storage device 24 into the data area of the RAM 23. The processor 21 calculates the data in the data area according to the data analysis program and writes the calculation result into the data area. Through such operations, the processor 21, the RAM 23, the auxiliary storage device 24, the input / output interface 25, and the bus 26 cooperate to execute at least some of the functions of the components of the data analysis device 10. For example, the processor 21, the RAM 23, the auxiliary storage device 24, the input / output interface 25, and the bus 26 cooperate to execute the functions of the first conversion unit 11, the second conversion unit 12, and the learning unit 13 (and the sampling unit 14) of the data analysis device 10.

[0085] Although some embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention and are included in the invention described in the claims and its equivalent scope.

Description of Reference Numerals

[0086] 10…Data analysis device, 11…Conversion unit, 12…Conversion unit, 13…Learning unit, 14…Sampling unit, 20…Computer, 21…Processor, 22…ROM, 23…RAM, 24…Auxiliary storage device, 25…Input / output interface, 26…Bus, 31…Input device, 32…Output device, 33…Input / output device, 34…Recording medium.

Claims

1. A first conversion unit that converts planning conditions into latent variables according to the similarity of contribution to KPI, a second conversion unit that converts the latent variables into predicted values of the KPI, and a learning unit that learns at least the parameters of the first conversion unit based on past conditions and past planning results. A data analysis device.

2. The planning conditions are conditions related to promotion targets, the first conversion unit converts the conditions into coefficients used for the conversion by the second conversion unit, and the second conversion unit converts the detailed conditions of the promotion into predicted values of the KPI based on the coefficients. The data analysis device according to Claim 1.

3. The first conversion unit converts the conditions into posterior distribution parameters of the coefficients, the second conversion unit converts the coefficients sampled from the posterior distribution of the coefficients and the detailed conditions into predicted values of the KPI, and the learning unit further learns the prior distribution parameters of the coefficients. The data analysis device according to Claim 2.

4. The second conversion unit converts the coefficients sampled from the posterior distribution, or in the case of unknown products, the coefficients sampled from the prior distribution, and the detailed conditions into predicted values of the KPI. The data analysis device according to Claim 3.

5. The past conditions include the conditions and the detailed conditions, and the learning unit learns at least the parameters of the first conversion unit based on the conditions, the detailed conditions, and the past planning results. The data analysis device according to Claim 4.

6. A data analysis program that causes a computer to execute a first conversion function that converts planning conditions into latent variables according to the similarity of contribution to KPI, a second conversion function that converts the latent variables into predicted values of the KPI, and a learning function that learns at least the parameters of the first conversion function based on past conditions and past planning results. A data analysis program.

Images