Pattern determination device, cross tabulation device, pattern determination method, and program

The pattern determination device efficiently determines cross-tabulation patterns to produce useful results by iteratively sampling and noise addition, addressing the inefficiencies of conventional methods in cross-tabulation while maintaining confidentiality.

WO2024247028A1PCT designated stage expired Publication Date: 2024-12-05NIPPON TELEGRAPH & TELEPHONE CORP
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
PCT/JP2023/019874
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-29
Publication Date
2024-12-05

AI Technical Summary

Technical Problem

Conventional methods for cross-tabulation in statistical calculations, while maintaining confidentiality, often result in increased calculation and processing time and may not produce useful results due to the need to repeatedly change cross-tabulation patterns and add differential privacy noise, potentially leading to undesirable outcomes like negative values in aggregate results.

Method used

A pattern determination device and method that pre-create multiple cross-tabulation patterns, sample records, perform cross-tabulation, add differential privacy noise, and iteratively change patterns until useful results are obtained, then apply these patterns to all records, thereby reducing computational burden and maintaining confidentiality.

Benefits of technology

This approach allows for the attainment of useful cross-tabulation results while minimizing the increase in calculation and processing time, ensuring confidentiality by effectively managing noise addition and pattern selection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2023019874_05122024_PF_FP_ABST
    Figure JP2023019874_05122024_PF_FP_ABST
Patent Text Reader

Abstract

Provided are a pattern determination device and the like for determining a cross tabulation pattern in order to obtain a useful cross tabulation result while keeping confidentiality and inhibiting increase in calculation amount and processing time as compared to conventional devices. The pattern determination device comprises: a sampling unit that samples N records from among N' records; a pattern storage unit that stores a plurality of pre-created cross tabulation patterns; a first cross tabulation unit that selects a cross tabulation pattern that has not yet been used, among the plurality of cross tabulation patterns stored in the pattern storage unit, and performs cross tabulation for the N records on the basis of the selected cross tabulation pattern; and a first noise addition unit that adds noise of differential privacy to the cross tabulation result of the first cross tabulation unit. The processes by the first cross tabulation unit and the first noise addition unit are repeated by changing the cross tabulation pattern.
Need to check novelty before this filing date? Find Prior Art

Description

Pattern determination device, cross tabulation device, pattern determination method, and program

[0001] The present invention relates to a technique for performing statistical calculations while maintaining confidentiality.

[0002] When a company combines its own customer information with that of other companies using a common ID to obtain statistics on attribute information that the company does not possess, it needs to be able to output useful statistics without identifying the customer.

[0003] Traditionally, companies have held information on customers who purchase their products or use their services, and have converted this information into anonymously processed information or pseudonymized information as necessary to use for marketing new products, etc. Anonymously processed information is information that has been processed so that a specific individual cannot be identified and that personal information cannot be restored, while pseudonymized information is information about an individual that has been processed so that a specific individual cannot be identified unless it is compared with other information.

[0004] In addition, when marketing to new targets that could not be identified using only the data held by the company, the company used publicly available statistical information and anonymously processed information provided by other companies. However, this statistical information and anonymously processed information cannot be matched with the customer information held by the company.

[0005] To uncover new attribute information about their customers, companies can match their own customer information with that of other companies using a common ID while keeping it encrypted, and obtain statistical information that meets their needs. New attribute information could be, for example, purchase history with other companies, such as what other products a customer has purchased in addition to their own company's products. For example, consider a situation where two companies, Company α and Company β, sell products using the same e-commerce site. When using the e-commerce site, users register and create an ID. This ID is used in common when using the e-commerce site and is shared by both Companies α and β. Figure 1 shows an example of customer information held by Companies α and β, and the e-commerce site holds both sets of information.

[0006] However, when matching customer information between companies, from the perspective of protecting personal information, it is necessary to ensure that individuals cannot be identified from the output results.

[0007] For example, matching is performed based on the IDs of tables T1 and T2 in Figure 1, tables T1 and T2 are joined, and information that can identify individuals (such as IDs) is deleted to obtain a new table T(1+2) as shown in Figure 2.

[0008] However, simply deleting information that can identify an individual (such as ID) may make it possible to identify the individual from the joined tables. Performing a cross-tabulation based on table T(1+2) in Figure 2 yields the tabulated result C1 shown in Figure 3. In this case, company α, which has information on the attributes "age" and "BMI," may be able to identify the individual from "age" and "BMI" in the tabulated result C1, and obtain the "place of employment" and "annual income" of the identified individual.

[0009] To avoid such a situation, there is a method of performing cross-tabulation after removing information that can identify individuals (such as IDs), and then adding differential privacy noise to the cross-tabulation results. Figure 4 shows an example of a cross-tabulation result C2 with added noise. Non-Patent Document 1 is known as a conventional technique for adding differential privacy noise.

[0010] Reo Erikuchi, Atsutake Ichikawa, and Noboru Kunihiro, "Efficient Noise Generation Achieving Differential Privacy for Applications to Secure Computing," SCIS2020, 2020.

[0011] However, depending on the level of noise, useful cross-tabulation results may not be obtained. For example, in the example shown in Figure 4, the number of items that should be an integer value greater than or equal to 0 is "-1," making the cross-tabulation results unhelpful. In this case, to obtain useful cross-tabulation results, the pattern of data items to be multiplied (hereinafter also referred to as the "cross-tabulation pattern") must be changed. The cross-tabulation process must be repeated with different cross-tabulation patterns until useful cross-tabulation results are obtained, which increases the amount of calculations and processing time. For example, if "BMI" and "place of employment" are removed from the data items that produce cross-tabulation result C1 in Figure 3, cross-tabulation result C3 in Figure 5 is obtained. However, adding noise to cross-tabulation result C3 may not produce the same useful cross-tabulation results as adding noise to cross-tabulation result C1. Furthermore, by changing the range of the data item "age" from "10-20 years old," "20-30 years old," "30-40 years old," and "40-50 years old" to "10-30 years old" and "30-50 years old," and by changing the range of the data item "annual income" from "5 million to 6 million yen," "6 million to 7 million yen," "7 million to 8 million yen," "8 million to 10 million yen," and "10 million to 15 million yen" to "5 million to 7 million yen" and "7 million to 15 million yen," the cross-tabulation result C4 in Figure 6 can be obtained. In this example, if noise is added to the cross-tabulation result C4, the useful cross-tabulation result C5 in Figure 7 can be obtained, in which the "number of cases" does not become "-1."

[0012] The present invention aims to provide a pattern determination device that determines patterns of data items to be multiplied in cross-tabulation, a cross-tabulation device that performs cross-tabulation using the determined patterns, a pattern determination method, and a program, in order to obtain useful cross-tabulation results while maintaining confidentiality and suppressing increases in the amount of calculations and processing time compared to conventional methods.

[0013] In order to solve the above problem, according to one aspect of the present invention, a pattern determination device includes: a sampling unit that samples N records L(1,n), L(2,n), ..., L(S,n) from N' records L(1,n'), L(2,n'), ..., L(S,n'), where n'=1, 2, ..., N', n=1, 2, ..., N, N'>N, and S is an integer equal to or greater than 2; a pattern storage unit that stores a plurality of pre-created cross-tabulation patterns; a first cross-tabulation unit that selects an unused cross-tabulation pattern from the plurality of cross-tabulation patterns stored in the pattern storage unit and performs cross-tabulation on N records L(1,n), L(2,n), ..., L(S,n) based on the selected cross-tabulation pattern; and a first noise addition unit that adds noise for differential privacy to the cross-tabulation results of the first cross-tabulation unit; and the cross-tabulation pattern is changed and the processing by the first cross-tabulation unit and the first noise addition unit is repeated.

[0014] The present invention has the effect of suppressing increases in the amount of calculation and processing time compared to conventional methods, while maintaining confidentiality and obtaining useful cross-tabulation results.

[0015] A diagram showing an example of customer information held by a company. A diagram showing an example of a new table obtained by joining two tables. A diagram showing an example of a cross-tabulation result. A diagram showing an example of a cross-tabulation result to which noise has been added. A diagram showing an example of a cross-tabulation result. A diagram showing an example of a cross-tabulation result to which noise has been added. A functional block diagram of a cross-tabulation device according to a first embodiment. A diagram showing an example of the processing flow of a cross-tabulation device according to the first embodiment. A diagram showing examples of cross-tabulation patterns stored in a pattern storage unit. A diagram showing an example of the configuration of a computer to which the present method is applied.

[0016] Hereinafter, an embodiment of the present invention will be described. In the drawings used in the following description, components having the same functions and steps performing the same processing are denoted by the same reference numerals, and duplicated explanations will be omitted. In the following description, processing performed on each element of a vector or matrix is ​​assumed to apply to all elements of that vector or matrix, unless otherwise specified.

[0017] <Key Points of the First Embodiment> (1) In this embodiment, multiple cross-tabulation patterns are created before cross-tabulation is performed using all records to be cross-tabulated. A record is information having a specific data structure, and in this example, it is information having a data structure made up of attribute information linked to an ID indicating an object (an ID indicating a customer in the example of FIG. 1). For example, in table T1 in FIG. 1, information having a data structure made up of "age" and "BMI" linked to a certain "ID" constitutes one record, and in table T2 in FIG. 1, information having a data structure made up of "place of employment" and "annual income" linked to a certain "ID" constitutes one record.

[0018] (2) A predetermined number of records are sampled from all records to be cross-tabulated, and cross-tabulation is performed on the sampled records using the cross-tabulation pattern created in (1) above, and differential privacy noise is added to the cross-tabulation results. Differential privacy is a technology that makes it impossible to infer input data from output data; adding noise to the output data ensures indistinguishability, protecting privacy data by preventing the input data from being inferred from the output data. Privacy protection can be achieved without changing the meaning of the output data by making the noise have a small effect on the output data and a large effect on the input data.

[0019] (3) Repeat the cross-tabulation and noise addition while changing the cross-tabulation pattern until the cross-tabulation results after noise addition in (2) above become useful.

[0020] (4) Using the cross-tabulation pattern when useful cross-tabulation results are obtained, cross-tabulation is performed on all records that are the subject of cross-tabulation, and differential privacy noise is added to the cross-tabulation results.

[0021] By adopting this configuration, it is possible to obtain useful cross-tabulation results while minimizing increases in the amount of calculations and processing time compared to when all records to be cross-tabulated are used to determine whether a useful cross-tabulation pattern can be obtained.

[0022] First Embodiment FIG. 8 is a functional block diagram of a cross tabulation device 100 according to a first embodiment, and FIG. 9 shows the processing flow thereof.

[0023] The cross-tabulation device 100 includes a first table memory unit 101, a second table memory unit 103, a third table memory unit 105, a sampling unit 110, a first combining unit 120, a first cross-tabulation unit 130, a pattern memory unit 131, a first noise addition unit 140, a determination unit 150, a second combining unit 160, a second cross-tabulation unit 170, and a second noise addition unit 180.

[0024] The cross tabulation device 100 receives P records L(1,p) and Q records L(2,q), performs cross tabulation, and outputs a cross tabulation result D.

[0025] The cross tabulation device 100 is a special device configured by loading a special program into a publicly available or dedicated computer having, for example, a central processing unit (CPU) and a main memory (RAM). The cross tabulation device 100 executes various processes under the control of the central processing unit. Data input to the cross tabulation device 100 and data obtained from each process are stored in the main memory, for example. The data stored in the main memory is read by the central processing unit as needed and used for other processes. At least a portion of each processing unit of the cross tabulation device 100 may be configured with hardware such as an integrated circuit. Each memory unit included in the cross tabulation device 100 may be configured with, for example, a main memory such as RAM (Random Access Memory) or middleware such as a relational database or key-value store. However, each memory unit does not necessarily have to be included within the cross tabulation device 100 itself; it may be configured with an auxiliary memory device implemented by a hard disk, optical disk, or semiconductor memory device such as flash memory, and be external to the cross tabulation device 100.

[0026] Each part will be explained below.

[0027] <First table storage unit 101, second table storage unit 103, third table storage unit 105, sampling unit 110, and first combining unit 120> A first table is stored in the first table storage unit 101, and the first table includes P input records L(1, p). Here, p = 1, 2,..., P.

[0028] A second table is stored in the second table storage unit 103, and the second table includes Q input records L(2, q). Here, q = 1, 2,..., Q.

[0029] Each table is composed of columns and tuples (rows), and the tuples correspond to records. In a predetermined column of each table, an identifier of an object to be cross-tabulated is included. It is assumed that the column including the identifier of the object to be cross-tabulated is predetermined in each table.

[0030] For example, the sampling unit 110 extracts P records L(1, p) from the first table storage unit 101, extracts Q records L(2, q) from the second table storage unit 103, and stores N' records L(1, n'), L(2, n') associated with the same identifier in the third table storage unit 105 so that the correspondence relationship is known. Here, N' < P, N' < Q, and n' = 1, 2,..., N'. The records L(1, n'), L(2, n') are records included in the P records L(1, p) and the Q records L(2, q) respectively, and the identifiers included in the record L(1, n') and the record L(2, n') are the same. For example, the sampling unit 110 searches for a record having the same identifier as that of a certain record L(1, p) from among the Q records L(2, q), and if there is a record having the same identifier, the record L(1, p) and the record L(2, q) are associated and stored in the third table storage unit 105 as records L(1, n'), L(2, n'). This process is performed for all p.

[0031] Furthermore, the sampling unit 110 samples N records L(1,n) and L(2,n) from among N' records L(1,n') and L(2,n') respectively (S110), and outputs the records L(1,n) and L(2,n). Here, N < N', and n = 1, 2, …, N. The records L(1,n) and L(2,n) are the n-th records sampled from among N' records L(1,n') and L(2,n') respectively, and the IDs included in the record L(1,n) and the record L(2,n) are the same. Note that the sampling number N may be appropriately set according to the population N', the allowable error, the cross-tabulation pattern, etc. Here, N records may be randomly sampled from among N' records, or N' records may be narrowed down to records that satisfy a predetermined condition, and N records may be randomly sampled from among the narrowed-down records. The predetermined condition may be set as needed. For example, "the value of the attribute 'BMI' is 10 to 30" etc. may be set as the predetermined condition.

[0032] For example, assume a case where a certain company α and another company β sell products using the same EC site, and the users of the EC site perform a registration operation and create an ID when using the EC site. This ID corresponds to the ID of the object and is commonly used when using the EC site, and is common to companies α and β. Company α holds information on users who have purchased products of company α, company β holds information on users who have purchased products of company β, and the EC site holds information on users who have purchased products via the EC site. The record L(1,p) held by company α is stored in the first table storage unit 101, and the record L(2,q) held by company β is stored in the second table storage unit 103. Company α conceals the record L(1,p) from company β, and company β conceals the record L(2,q) from company α. In short, management is performed so that only those with the right to access can access the records.

[0033] 1, it has been explained that companies α and β hold information corresponding to the same customer, but companies α and β do not need to hold information corresponding to all the same customers, and they may hold information corresponding to different customers. Also, in FIG. 1, companies α and β hold information with different attributes for the same customer, but they may hold information with some of the same attributes.

[0034] The first combining unit 120 receives N records L(1,n) and L(2,n), combines records L(1,n) and L(2,n) (S120), deletes information that can identify an individual (such as an ID) from the combined record, and outputs the combined and deleted record R(n).

[0035] <First Cross Tabulation Unit 130 and Pattern Storage Unit 131> The pattern storage unit 131 stores T cross tabulation patterns that have been created in advance, where T is an integer of 2 or greater.

[0036] The first cross tabulation unit 130 receives N records R(n), selects an unused cross tabulation pattern from among the multiple cross tabulation patterns stored in the pattern storage unit 131, performs cross tabulation based on the selected cross tabulation pattern (S130), and outputs a cross tabulation pattern identifier t and a cross tabulation result C(t), where t=1, 2, ..., T.

[0037] 10 shows an example of a cross tabulation pattern stored in the pattern storage unit 131. In this example, the pattern storage unit 131 stores a cross tabulation pattern identifier t and the cross tabulation pattern (type and range of attribute).

[0038] There are no limitations on the patterns of data items to be multiplied as cross-tabulation patterns. All patterns may be stored in the storage unit 131 as cross-tabulation patterns, or cross-tabulation patterns created in advance manually or automatically according to predetermined rules may be stored. However, if all patterns are used as cross-tabulation patterns, the number of patterns may become enormous depending on the number of original attribute types and the values ​​of the attribute values. In such cases, it is possible that the process will not be completed within a realistic time frame or that the system memory will become enormous. Furthermore, narrowing the range of attribute values ​​increases the likelihood that each aggregated value included in the cross-tabulation results will take on a small value, and adding noise may result in a negative value, making the cross-tabulation results unusable. Therefore, to avoid an enormous number of patterns and ensure useful cross-tabulation results, it is desirable to prepare cross-tabulation patterns in which the attribute types and attribute value ranges used in the cross-tabulation are appropriately set, either manually or according to predetermined rules. Furthermore, among cross-tabulation patterns that can produce useful cross-tabulation results, those that use a large number of attribute types in the cross-tabulation and have a narrow range of attribute values ​​enable more detailed analysis. Therefore, when the first cross-tabulation unit 130 selects a cross-tabulation pattern, it is desirable to select the pattern in order of the number of attribute types used in the cross-tabulation and the narrowest range of attribute values. <First Noise Addition Unit 140> The first noise addition unit 140 receives the cross-tabulation pattern identifier t and the cross-tabulation result C(t), adds differential privacy noise to the cross-tabulation result C(t) (S140), and outputs the cross-tabulation pattern identifier t and the cross-tabulation result D(t) after the addition. Various conventional techniques can be used to add differential privacy noise. For example, this can be achieved by non-patent document 1.

[0039] <Determination Unit 150> The determination unit 150 receives the cross-tabulation pattern identifier t and the cross-tabulation result D(t) as input and determines whether the cross-tabulation result D(t) is useful (S150). If the cross-tabulation result D(t) is useful (YES in S150), it outputs the useful cross-tabulation pattern identifier t' to the second cross-tabulation unit 170. If the cross-tabulation result D(t) is not useful (NO in S150), it outputs a control signal to the first cross-tabulation unit 130 and the first noise addition unit 140 to instruct them to change the cross-tabulation pattern and repeat the process. Note that in this example, if the cross-tabulation pattern is not useful, the sampling unit 110 and the combining unit 120 are controlled not to repeat the process. However, it may also be controlled to repeat the process together with the first cross-tabulation unit 130 and the first noise addition unit 140. The determination of usefulness is made based on whether the cross-tabulation result includes only possible values. For example, if the tabulation value is smaller than the minimum possible value (e.g., "0") or larger than the maximum possible value (e.g., the total number of records, "N"), it is determined to be unuseful. If it is determined that none of the T cross-tabulation patterns stored in the pattern storage unit 131 are useful, the system may be configured to control the system so that a message indicating that there are no useful cross-tabulation patterns is presented to the user via a display or the like. <Second Combining Unit 160> The second combining unit 160 retrieves N' records L(1,n') and L(2,n') from the third table storage unit 105, combines record L(1,n') with record L(2,n') (S160), deletes information that can identify individuals (such as IDs) from the combined record, and outputs the combined and deleted record R(n'). This may be achieved, for example, by using the same technology as the first combining unit 120.

[0040] <Second Cross Tabulation Unit 170 and Second Noise Addition Unit 180> The second cross tabulation unit 170 receives record R(n') and the identifier t' of a useful cross tabulation pattern, extracts a cross tabulation pattern corresponding to the identifier t' of the useful cross tabulation pattern from the pattern storage unit 131, performs cross tabulation based on the extracted cross tabulation pattern (S170), and outputs the cross tabulation result C. The cross tabulation can be performed in the same manner as in the first cross tabulation unit 130.

[0041] The second noise addition unit 180 receives the cross tabulation result C as input, adds differential privacy noise (S180), and outputs the cross tabulation result D after addition. The differential privacy noise can be added in a similar manner to that used by the first noise addition unit 140.

[0042] <Effects> With the above configuration, it is possible to obtain useful cross-tabulation results while maintaining confidentiality and reducing the amount of calculation and processing time compared to conventional methods.

[0043] <Modification> Each piece of data in this embodiment may be encrypted, and the processing in each unit may be performed by secure computation. For example, this can be realized by the method disclosed in Reference 1.

[0044] (Reference 1) "Secure Computation System and Its Principles," [online], 2018, Nippon Telegraph and Telephone Corporation, [Retrieved May 19, 2023], Internet<URL:https: / / www.rd.ntt / sil / project / sc / NTT-himitsu-keisan.pdf> In this embodiment, the cross-tabulation process (S130, S170) is performed after the data merging process (S120, S160). However, the cross-tabulation process (S130, S170) may be performed based on two tables without the merging process. In this case, the cross-tabulation device 100 does not need to include the first merging unit 120 and the second merging unit 160. The first cross-tabulation unit 130 receives the sampled N records L(1,n) and L(2,n), selects an unused cross-tabulation pattern from the multiple cross-tabulation patterns stored in the pattern storage unit 131, and performs cross-tabulation based on the selected cross-tabulation pattern. The second cross-tabulation unit 170 receives the identifier t' of a useful cross-tabulation pattern, retrieves N' records L(1,n') and L(2,n') from the third table storage unit 105, retrieves the cross-tabulation pattern corresponding to the identifier t' of the useful cross-tabulation pattern from the pattern storage unit 131, and performs cross-tabulation based on the retrieved cross-tabulation pattern.

[0045] In this embodiment, a case has been described in which the identifiers of the objects contained in the first table and the second table do not necessarily all match, but when the identifiers of the objects all match and the correspondence is known, as in Figure 1, the sampling unit 110 may omit the process of storing N' records L(1,n') and L(2,n') linked to the same identifier in the third table storage unit 105 so that the correspondence is known, and may instead perform sampling directly from the N' records contained in the first table and the second table. In this case, P = Q = N', and the cross tabulation device 100 may not include the third table storage unit 105.

[0046] The cross tabulation device 100 may function as a pattern determination device without including the first table storage unit 101, the second table storage unit 103, the second combining unit 160, the second cross tabulation unit 170, and the second noise adding unit 180. The pattern determination device acquires N' records L(1,n') and L(2,n') from an external database based on the identifiers of the objects to be cross tabulated. Subsequently, processing is performed by the sampling unit 110, the first combining unit 120, the first cross tabulation unit 130, the first noise adding unit 140, and the determination unit 150, as in the first embodiment.

[0047] If the cross tabulation result D(t) is useful (YES in S150), the determination unit 150 outputs the cross tabulation pattern corresponding to the identifier t' of the cross tabulation pattern that is useful to the second cross tabulation unit 170 to the external cross tabulation device. In this case, the external cross tabulation device performs processing in the second combining unit 160, the second cross tabulation unit 170, and the second noise adding unit 180. In this case, by using the cross tabulation pattern determined by the pattern determination device, the external cross tabulation device can obtain useful cross tabulation results while maintaining confidentiality and reducing the amount of calculation and processing time compared to conventional methods.

[0048] Although this embodiment shows an example in which there are two tables, cross-tabulation may be performed using records contained in S tables, where S is any integer equal to or greater than 2, and N' records L(1,n'), L(2,n'), ..., L(S,n') linked to the same identifier are stored in the third table storage unit 105 so that the correspondence can be determined, and similar processing may be performed.

[0049] In this embodiment, the determining unit 150 determines whether the cross tabulation result D(t) is useful, but the determination process may also be performed by a user of the cross tabulation device 100. For example, a cross tabulation pattern and the cross tabulation results obtained when the cross tabulation pattern is used may be displayed on an output unit (not shown) such as a display, and the user may view the cross tabulation results to determine whether they are useful, and then specify the cross tabulation pattern determined to be most useful via an input unit (not shown) such as a keyboard or mouse.

[0050] In this embodiment, when the determining unit 150 determines that the cross tabulation result D(t) is useful, it outputs the identifier t' of the cross tabulation pattern that is useful at that time. However, it is also possible to perform cross tabulation for all cross tabulation patterns and select one from among the cross tabulation patterns that correspond to useful cross tabulation results. For example, it is possible to select a cross tabulation pattern that uses a large number of attribute types and a narrow range of attribute values ​​from among the cross tabulation patterns that correspond to useful cross tabulation results, or to select a cross tabulation pattern that is least likely to be determined to be unuseful.

[0051] The above-mentioned variations may be combined to form a pattern determination device or a cross-tabulation device.

[0052] <Other Modifications> The present invention is not limited to the above-described embodiments and modifications. For example, the various processes described above may not only be executed in chronological order as described, but may also be executed in parallel or individually depending on the processing capacity of the device that executes the processes or as needed. Other modifications are possible as long as they do not deviate from the spirit of the present invention.

[0053] <Program and Recording Medium> The various processes described above can be implemented by loading a program that executes each step of the above method into the recording unit 2020 of the computer 2000 shown in FIG. 11 and operating the control unit 2010, input unit 2030, output unit 2040, display unit 2050, etc.

[0054] The program describing the processing contents can be recorded on a computer-readable recording medium, which may be, for example, a magnetic recording device, an optical disk, a magneto-optical recording medium, a semiconductor memory, or any other suitable recording medium.

[0055] The program may be distributed by, for example, selling, transferring, lending, etc. portable recording media such as DVDs and CD-ROMs on which the program is recorded. Furthermore, the program may be stored in a storage device of a server computer, and then transferred from the server computer to other computers via a network, thereby distributing the program.

[0056] A computer that executes such a program may first temporarily store the program recorded on a portable recording medium or transferred from a server computer in its own storage device. Then, when executing a process, the computer reads the program stored on its own recording medium and executes the process in accordance with the read program. Alternatively, the computer may read the program directly from a portable recording medium and execute the process in accordance with the program. Furthermore, the computer may execute the process in accordance with each program transferred from the server computer. Alternatively, the server computer may not transfer the program to the computer, but may instead execute the process through a so-called ASP (Application Service Provider) service, which realizes the processing function by issuing an execution instruction and obtaining the results. In this embodiment, the program includes information used for processing by a computer that is equivalent to a program (e.g., data that is not a direct instruction to the computer but has properties that define computer processing).

[0057] Furthermore, in this embodiment, the device is configured by executing a predetermined program on a computer, but at least a part of the processing contents may be realized by hardware.

Claims

1. A pattern determination device comprising: a sampling unit which samples N records L(1,n), L(2,n), ..., L(S,n) from N' records L(1,n'), L(2,n), ..., L(S,n'), respectively, where n'=1,2, ...,N', n=1,2, ...,N, N'>N, and S is an integer equal to or greater than 2; a pattern storage unit which stores a plurality of cross tabulation patterns created in advance; a first cross tabulation unit which selects a cross tabulation pattern that has not yet been used from the plurality of cross tabulation patterns stored in the pattern storage unit, and performs cross tabulation on N records L(1,n), L(2,n), ..., L(S,n) based on the selected cross tabulation pattern; and a first noise addition unit which adds noise for differential privacy to a cross tabulation result of the first cross tabulation unit, and which changes the cross tabulation pattern and repeats processing in the first cross tabulation unit and the first noise addition unit.

2. A cross-tabulation device including the pattern determination device of claim 1, comprising: a second cross-tabulation unit that performs cross-tabulation on N' records L(1, n'), L(2, n'), ..., L(S, n') based on a cross-tabulation pattern corresponding to the useful cross-tabulation result when the cross-tabulation result to which noise has been added is useful; and a second noise addition unit that adds noise for differential privacy to the cross-tabulation result of the second cross-tabulation unit.

3. A pattern determination method in which a pattern storage unit stores a plurality of cross tabulation patterns created in advance, the method comprising: a sampling step of sampling N records L(1, n), L(2, n), ..., L(S, n) from N' records L(1, n'), L(2, n), ..., L(S, n'), respectively, where n'=1, 2, ..., N, N'>N, and S is an integer equal to or greater than 2; a first cross tabulation step of selecting a cross tabulation pattern that has not yet been used from the plurality of cross tabulation patterns stored in the pattern storage unit, and performing cross tabulation on N records L(1, n), L(2, n), ..., L(S, n) based on the selected cross tabulation pattern; and a first noise addition step of adding noise for differential privacy to the cross tabulation result of the first cross tabulation step, wherein the cross tabulation pattern is changed and the processes in the first cross tabulation step and the first noise addition step are repeated.

4. A program for causing a computer to function as the pattern determination device of claim 1.

Citation Information

Patent Citations

  • Privacy protection device

    JP2017203904A

  • Information processing system and anonymizing method

    WO2012067213A1