Order data inspection system, order data inspection method, and program

Abstract

Regarding the inspection target information included in order data, inspection based solely on judgment rules may not yield inspection results for information containing variations in notation or special notations. Widespread AI judgment would increase processing costs, and this approach aims to mitigate the decline in business continuity when AI becomes unavailable. [Solution] The order management system 1 includes an order data inspection function 10, which comprises: an acquisition unit 11 that acquires information to be inspected; a rule inspection unit 12 that inspects the consistency between predetermined items and reference information based on judgment rules; an AI judgment unit 13 that performs an AI judgment based on the information to be inspected if the rule inspection unit 12 cannot acquire an inspection result; an AI analysis unit 14 that performs an AI analysis if an inspection result indicating an abnormality is obtained; a fallback unit 15 that outputs standard suggestion information if the AI ​​analysis unit 14 is unavailable; and an output unit 16 that outputs the inspection result, judgment result, or interpretation result.

Description

Technical Field

[0001] The present invention relates to an order data inspection technique in an order management system, and more specifically, relates to an order data inspection system, an order data inspection method, and a program that perform inspection on inspection target information included in order data based on determination rules, and perform AI-based determination when an inspection result based on the determination rules cannot be obtained.

Background Art

[0002] Order data corresponding to orders received via EC sites or EC malls includes delivery destination addresses, telephone numbers, and various other types of information. If this information contains typographical errors, inconsistencies, or notation variations, it will cause delivery delays, returns, reconfirmation responses, and other issues. Therefore, there is a demand for a technique for inspecting the consistency of the information contained in order data in an order management system.

[0003] As this type of technique, a technique has been proposed that inspects whether a telephone number conforms to a predetermined telephone number system protocol, and if it does not conform, identifies the country corresponding to the telephone number using a machine learning system (see, for example, Patent Document 1).

[0004] Furthermore, a technique for maintaining data quality by applying verification rules to elements of a data set according to conditions has been proposed (see, for example, Patent Document 2).

[0005] Moreover, a technique for standardizing addresses by combining preprocessing, an address understanding model, verification using a dictionary or standard rules for input addresses has also been proposed (see, for example, Patent Document 3).

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

[0007] However, in the conventional technology described above, mechanical inspection based on judgment rules alone may not be sufficient to obtain inspection results for information to be inspected that includes variations in notation, special notations, or context-dependent information. On the other hand, configurations that widely implement AI-based judgment tend to lead to increased processing costs.

[0008] Therefore, the present invention aims to provide an order data inspection system, an order data inspection method, and a program that can output inspection results while ensuring processing efficiency by prioritizing inspection based on judgment rules for inspection target information included in order data, and performing AI-based judgment when inspection results based on said judgment rules cannot be obtained. [Means for solving the problem]

[0009] To solve the above problems, an order data inspection system according to one aspect of the present invention comprises: an acquisition unit that acquires inspection target information contained in order data; a rule inspection unit that inspects the consistency between predetermined items contained in the inspection target information and reference information based on predetermined judgment rules; an AI judgment unit that, if the inspection by the rule inspection unit does not result in an inspection result based on the judgment rules for the predetermined items, performs an AI judgment based on the inspection target information; and an output unit that outputs the inspection result by the rule inspection unit or the judgment result or interpretation result by the AI ​​judgment unit.

[0010] The order data inspection system may further include an AI analysis unit that performs AI-based analysis based on the inspection target information related to the inspection result indicating an anomaly when an inspection result indicating an anomaly is obtained as a result of the inspection by the rule inspection unit. The AI ​​analysis unit may generate explanatory information about the problem, generate confidence information, or generate correction candidate information. The AI ​​analysis unit may generate the correction candidate information based on candidate information narrowed down in advance by matching with the reference information, and the correction candidate information may include a plurality of prioritized candidates.

[0011] The order data inspection system may further include a fallback unit that outputs standardized suggestion information based on the inspection results indicating the anomaly when the AI ​​analysis unit is unavailable. The fallback unit may output the standardized suggestion information if the AI ​​analysis unit does not return output in a predetermined format, or if the action key included in the output of the AI ​​analysis unit does not conform to a predetermined permission list. The reference information may include master information, dictionary information, system information, or regional information. The rule inspection unit may perform at least one of consistency checks, range checks, format checks, or system checks.

[0012] The information subject to inspection may be information corresponding to individual orders, or it may be address-related information. The rule inspection unit may perform normalization processing on the address-related information, and the normalization processing may include at least one of the following: removal of spaces, conversion of numbers, unification of variant characters, and unification of hyphens. The rule inspection unit may confirm the consistency between predetermined items included in the address-related information and the reference information based on the postal code or regional information. The rule inspection unit may obtain information corresponding to the postal code included in the address-related information from the postal code master, and if there is no corresponding data in the postal code master, it may refer to the large business establishment master. The rule inspection unit may perform range confirmation regarding block numbers or house numbers. The AI ​​judgment unit may perform an AI judgment based on the address-related information if the rule inspection unit cannot obtain an inspection result based on the judgment rules due to numerical notation including Chinese numerals, notation including old characters or variant characters, or special notation.

[0013] The information subject to inspection may also be telephone number-related information. The rule inspection unit may perform a format check on the telephone number-related information, or perform a number system check, a regional correspondence check, or a special number range check on the telephone number-related information. The AI ​​judgment unit may perform an AI judgment based on the telephone number-related information if the rule inspection unit cannot obtain an inspection result based on the judgment rule due to ambiguous cases related to special notation, regional inconsistencies, or special number ranges.

[0014] The order data inspection system may further include a storage unit for holding the reference information, and the storage unit may periodically update the reference information. The order data inspection system may also further include a cache unit for caching the search results of the reference information. The fallback unit may output the standardized suggestion information based on the inspection results definitively obtained by the rule inspection unit. The fallback unit may also generate confidence information corresponding to the standardized suggestion information, and this confidence information may be set to indicate a high level of confidence among multiple levels of confidence.

[0015] Furthermore, an order data inspection method according to one aspect of the present invention includes: a) a step of acquiring inspection target information contained in order data; b) a step of inspecting the consistency between predetermined items contained in the inspection target information and reference information based on predetermined judgment rules; c) a step of performing an AI judgment based on the inspection target information if, as a result of the inspection in step b), an inspection result based on the judgment rules cannot be obtained for the predetermined items; and d) a step of outputting the inspection result in step b) or the AI ​​judgment result or interpretation result in step c). Furthermore, a program according to one aspect of the present invention is for causing a computer to perform each of the above steps. [Effects of the Invention]

[0016] According to the invention of claim 1, the inspection target information included in the order data is first inspected based on a judgment rule, and only if an inspection result based on the judgment rule cannot be obtained is an AI judgment performed. This makes it possible to output inspection results, judgment results, or interpretation results while ensuring processing efficiency.

[0017] According to the invention of claim 2, when an abnormal inspection result is obtained through rule inspection, it is possible to provide additional analysis results for abnormal cases by performing AI analysis.

[0018] According to the invention of claim 3, by generating explanatory information about the problem, it is possible to make it easier for the operator to understand the nature of the abnormality.

[0019] According to the invention of claim 4, by generating confidence information, it is possible to support appropriate decision-making based on the likelihood of the analysis results.

[0020] According to the invention of claim 5, it is possible to facilitate responses to abnormalities by generating candidate information for correction.

[0021] According to the invention of claim 6, by generating correction candidate information based on candidate information narrowed down in advance by collation with reference information, it is possible to improve the validity of candidate generation.

[0022] According to the invention of claim 7, by presenting a plurality of prioritized candidates, it is possible for the operator to easily select an adoption candidate.

[0023] According to the invention of claim 8, even when the AI analysis unit is unavailable, it is possible to ensure business continuity by outputting standard proposal information based on an inspection result indicating an abnormality.

[0024] According to the invention of claim 9, even when the AI analysis unit does not return an output in a predetermined format or returns an action key that does not conform to the permission list, the fallback unit can output the standard proposal information, so that it is possible to provide a stable support result even when the AI output is defective.

[0025] According to the invention of claim 10, by using master information, dictionary information, system information, or regional information as the reference information, it is possible to perform an appropriate inspection corresponding to the inspection target information.

[0026] According to the invention of claim 11, by performing at least one of consistency check, range check, format check, or system check, it is possible to realize various rule checks corresponding to the content of the inspection target information.

[0027] According to the invention of claim 12, by using information corresponding to an individual order as the inspection target information, it is possible to perform an inspection conforming to order receiving operations.

[0028] According to the invention of claim 13, by setting the inspection target information as address-related information, it is possible to appropriately inspect for typographical errors, inconsistencies, or notation variations related to delivery destination addresses.

[0029] According to the invention of claim 14, by performing normalization processing on address-related information, it is possible to suppress variations in notation and then perform rule checking.

[0030] According to the invention of claim 15, it is possible to standardize the notation of address-related information to a predetermined format by performing a normalization process that includes at least one of removing spaces, converting to numbers, standardizing variant characters, and standardizing hyphens.

[0031] According to the invention of claim 16, it is possible to efficiently detect regional inconsistencies in addresses by confirming the consistency between a predetermined item included in address-related information and reference information based on a postal code or regional information.

[0032] According to the invention of claim 17, by referring to the large business establishment master in addition to the postal code master, it is possible to appropriately verify the consistency of addresses related to large businesses.

[0033] According to the invention of claim 18, by performing a range check regarding the block number or house number, it is possible to detect an address where the block number or house number is inappropriate, even if the regional information is consistent.

[0034] According to the invention of claim 19, when the rule inspection unit cannot obtain an inspection result due to numerical notation including Chinese numerals, notation including old characters or variant characters, or special notation, AI-based determination is performed, thereby supplementing the determination even when the address notation is complex.

[0035] According to the invention of claim 20, by using telephone number-related information as the information to be inspected, it is possible to appropriately inspect for errors, inconsistencies, or special notations related to telephone numbers.

[0036] According to the invention of claim 21, by performing normalization processing on telephone number-related information, it is possible to appropriately verify the format or structure of telephone numbers.

[0037] According to the invention of claim 22, it is possible to suppress variations in the notation of telephone numbers by performing a normalization process that includes at least one of converting full-width numbers to half-width numbers, converting full-width hyphens to half-width hyphens, removing parentheses, processing country codes, and removing spaces.

[0038] According to the invention of claim 23, by performing a format check on telephone number-related information, it is possible to detect format errors such as insufficient or excessive digits or the inclusion of prohibited characters.

[0039] According to the invention of claim 24, by performing a number system check, a regional correspondence check, or a special number range check, it is possible to detect telephone numbers that do not exist in a predetermined number system, telephone numbers that do not conform to the delivery destination region, or telephone numbers that belong to a special number range.

[0040] According to the invention of claim 25, when the rule inspection unit is unable to obtain inspection results due to ambiguous cases related to special notation, regional inconsistencies, or special number ranges, AI-based determination is performed, making it possible to make determinations that can handle complex situations, including ambiguous cases, regarding telephone numbers as well.

[0041] According to the invention of claim 26, by providing a storage unit that holds reference information, it is possible to stably utilize the reference information necessary for rule checking.

[0042] According to the invention of claim 27, it is possible to maintain the freshness of the reference information by periodically updating the reference information stored in the memory unit.

[0043] According to the invention of claim 28, it is possible to streamline repeated inspection processes by caching the search results of the reference information.

[0044] According to the invention of claim 29, by outputting standardized suggestion information based on the inspection results definitively obtained by the rule inspection unit, it is possible to provide highly reliable support results even during fallback.

[0045] According to the invention of claim 30, by setting the reliability information corresponding to the standardized suggestion information as high reliability, it is possible to make it easier for the operator to appropriately evaluate the fallback result.

[0046] According to the invention of claim 31, it is possible to realize an order data inspection method in which inspection target information included in order data is prioritized based on a judgment rule, and when an inspection result based on the judgment rule cannot be obtained, an AI-based judgment is performed.

[0047] According to the invention of claim 32, the method can be implemented in an information processing device by providing a program for causing a computer to execute the method.

[0048] Furthermore, the specific configurations of the present invention, such as the configuration in which the consistency between predetermined items and reference information in the order data is checked based on a judgment rule, and an AI judgment is performed only when an inspection result based on the judgment rule cannot be obtained, the configuration in which AI analysis or fallback is performed when an inspection result indicating an abnormality is obtained, and address consistency verification by referring to the postal code master and the large business establishment master, are not described at all in Patent Documents 1 to 3. [Brief explanation of the drawing]

[0049] [Figure 1] This figure shows an example of the overall configuration of an order management system. [Figure 2] This figure shows an example of the basic processing flow in the order data inspection function. [Figure 3] This diagram shows an example of a processing flow that executes AI analysis or a fallback when an abnormal test result is obtained through rule testing. [Figure 4] This figure shows an example of the processing flow for rule checking and AI judgment on address-related information. [Figure 5] This table shows a step-by-step inspection process for address-related information, along with an example of a specific case. [Figure 6]This table shows an example of the address normalization process. [Figure 7] This table shows an example of specific examples of the pre- and post-conversion states in address normalization processing. [Figure 8] This figure shows an example of a processing flow for verifying the consistency of address-related information by referring to the postal code master and the large business establishment master. [Figure 9A] This figure shows an example of the processing flow for rule checking of telephone number-related information. [Figure 9B] This diagram shows an example of a processing flow where AI judgment is performed when rule-based inspection is insufficient to determine the phone number-related information. [Figure 10A] This table shows an example of specific cases of telephone number normalization, both before and after conversion. [Figure 10B] This table shows an example of a specific case for format verification of telephone number-related information. [Figure 10C] This table shows an example of a specific case for verifying regional coverage of telephone number-related information. [Figure 10D] This table shows an example of a specific case where AI analysis is performed on telephone number-related information. [Figure 11] This table shows an example of the process involved in normalizing telephone numbers. [Figure 12] This table shows an example of specific cases of telephone number normalization, both before and after conversion. [Figure 13] This figure shows an example of a functional configuration that verifies the consistency of address-related information by referencing the postal code master and the large business establishment master. [Figure 14] This table shows the criteria for setting reliability information and provides an example of a specific example. [Figure 15] This table shows an example of information regarding suggested corrections. [Figure 16A] This table shows an example of the structure of information to be input into the AI ​​analysis department. [Figure 16B] This table shows an example of the structure of information output from the AI ​​analysis unit. [Figure 16C] This table shows an example of implementation-dependent information when performing AI analysis. [Figure 17] This figure shows a concrete example of AI analysis results in JSON format. [Figure 18] This table shows specific examples of conditions under which the system will switch to a fallback mode when the AI ​​analysis department becomes unavailable. [Figure 19] This diagram shows an example of a processing flow that performs AI analysis or a fallback when an abnormal test result is obtained, and outputs the result to the operator terminal. [Figure 20] This table shows an example of standardized suggestion information associated with anomaly types. [Figure 21] This figure shows an example of a processing flow related to retaining and updating reference information and caching search results. [Figure 22] This figure shows an example of a step-by-step analysis process flow for information in the remarks column. [Figure 23] This table shows a step-by-step analysis of the information in the remarks column and provides an example of a specific case. [Figure 24] This figure shows an example of a step-by-step inconsistency detection process flow for name-related information. [Figure 25] This table shows a step-by-step examination process for name-related information, along with an example of a specific case. [Figure 26] This diagram shows a modified processing flow that also performs AI analysis on cases that have been identified as abnormal by the AI. [Figure 27] This diagram shows the processing flow of a modified example that automatically applies highly reliable correction candidates. [Figure 28A] This table shows the first example of a functional enhancement configuration. [Figure 28B] This table shows a second example of a functional enhancement configuration. [Figure 28C] This table shows a third example of a functional enhancement configuration. [Figure 29] This table shows examples of response methods based on the level of confidence. [Figure 30] This figure shows an example of the display of the order data inspection results screen. [Modes for carrying out the invention]

[0050] The order data inspection system, order data inspection method, and program according to the present invention will be described with reference to Figures 1 to 30. Hereinafter, the order management system 1 shown in Figure 1 will be used as an example of the order data inspection system according to the present invention. Furthermore, the program executed in the order management system 1 will be used as an example of the program according to the present invention. The order data inspection method according to the present invention is realized by the execution of the program in the order management system 1.

[0051] First, the system configuration and basic processing flow of the order management system 1 will be explained with reference to Figures 1 and 2. Figure 1 shows the overall configuration of the order management system 1, and Figure 2 shows the basic processing flow of the order data inspection function 10. The order management system 1 is a system that performs rule inspections on the inspection target information 32 contained in the order data 31 based on reference information 51 and judgment rules 34, and performs an AI judgment if an inspection result cannot be obtained through the rule inspection.

[0052] As shown in Figure 1, the order management system 1 processes order data 31 corresponding to orders received through e-commerce sites, e-commerce malls, and other order acceptance means. The order management system 1 includes an order data inspection function 10 for checking the consistency of various information contained in the order data 31. The order data inspection function 10 consists of an acquisition unit 11, a rule inspection unit 12, an AI judgment unit 13, an AI analysis unit 14, a fallback unit 15, an output unit 16, a storage unit 17, a cache unit 18, and a periodic data acquisition processing unit 19. For example, an operator terminal 2 is used as the output destination of the output unit 16.

[0053] The order management system 1 may be composed of one or more information processing devices equipped with a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), storage device (HDD, SSD, etc.), and a communication interface. Each functional unit in the order management system 1, such as the acquisition unit 11, rule inspection unit 12, AI judgment unit 13, AI analysis unit 14, fallback unit 15, output unit 16, storage unit 17, cache unit 18, and periodic acquisition processing unit 19, may be realized by the CPU reading a predetermined program stored in ROM or storage device into RAM and executing it, and cooperating with various hardware.

[0054] Order data 31 is data corresponding to an individual order and includes, for example, the shipping address, telephone number, remarks field, customer name, recipient name, and other information used in order processing. In this embodiment, the information within the order data 31 that is subject to inspection is referred to as inspection target information 32. Examples of inspection target information 32 include address-related information 32A, telephone number-related information 32B, remarks field information 32C, and name-related information 32D. However, inspection target information 32 is not limited to these and may include other items included in the order data 31.

[0055] Among the information included in the information to be inspected 32, items that are subject to consistency verification in relation to the reference information 51 are called designated items 33. Examples of designated items 33 include the postal code 61, regional information 62, block number 63, and house number 64 included in the address-related information 32A. In addition, the number of digits, leading digit, area code, mobile prefix, regional correspondence information, and special number range information included in the telephone number-related information 32B may also be treated as designated items 33. The reference information 51 is information including master information, dictionary information, system information, or regional information, and is used as a standard when the rule inspection unit 12 performs mechanical inspections.

[0056] The judgment rule 34 is a rule used by the rule inspection unit 12 to inspect the information to be inspected 32. The judgment rule 34 may include rules relating to consistency checks, range checks, format checks, and system checks. Consistency checks are processes that, for example, verify whether the regional information 62 corresponding to the postal code 61 matches the regional information in the input address. Range checks are processes that, for example, verify whether the block number 63 or street number 64 is within a reasonable range for a particular area. Format checks are processes that, for example, verify the number of digits in a telephone number or whether there are any prohibited characters, and system checks are processes that, for example, verify whether the area code or mobile prefix is ​​included in a predetermined numbering system.

[0057] As shown in Figure 2, the acquisition unit 11 first acquires order data 31. Next, the acquisition unit 11 acquires inspection target information 32 from the order data 31. The rule inspection unit 12 performs an inspection on the acquired inspection target information 32 based on reference information 51 and judgment rules 34. If the rule inspection unit 12 is able to acquire an inspection result 35, the inspection result 35 is sent to the output unit 16. On the other hand, if the rule inspection unit 12 is unable to acquire an inspection result 35 based on judgment rules 34 for a predetermined item 33, the AI ​​judgment unit 13 performs an AI judgment based on the inspection target information 32 and sends the judgment result 36 or interpretation result 37 to the output unit 16.

[0058] Here, inspection result 35 refers to the result mechanically obtained by the rule inspection unit 12. For example, this includes results corresponding to normal, mismatch, out of range, format invalid, unassigned number range, or non-existent postal code. Inspection result 38 indicating an abnormality refers to the result of inspection result 35 that requires confirmation, correction, or additional action by an operator. Judgment result 36 is the result output by the AI ​​judgment unit 13 based on the inspection target information 32. Interpretation result 37 refers to the result of the AI ​​judgment unit 13 interpreting free text, special notation, or ambiguous notation, and is information used for subsequent judgments or displays.

[0059] The AI ​​judgment unit 13 is not executed continuously for all order data 31. The AI ​​judgment unit 13 is executed when the rule inspection unit 12 cannot obtain an inspection result 35 based on the judgment rule 34. Therefore, in this embodiment, rule-based inspection is prioritized first, and supplementary judgment by AI is performed only when the rule-based inspection is insufficient to make a judgment. For this reason, it is easier to suppress the increase in processing load and cost compared to a configuration in which the AI ​​is activated for all items.

[0060] If the rule inspection unit 12 obtains an inspection result 38 indicating an anomaly, the AI ​​analysis unit 14 may perform an additional analysis on the anomaly case. The AI ​​analysis unit 14 generates, for example, explanatory information 41, confidence information 42, and correction candidate information 43. The explanatory information 41 is information indicating the content of the anomaly, the confidence information 42 is information indicating the likelihood of the analysis result, and the correction candidate information 43 is information indicating possible countermeasures.

[0061] As shown in Figure 3, if the rule inspection unit 12 obtains an inspection result 38 indicating an anomaly, subsequent processing for the anomaly case is performed. First, it is determined whether the AI ​​analysis unit 14 is available for the anomaly case. If the AI ​​analysis unit 14 is available, the AI ​​analysis unit 14 performs an AI analysis based on the inspection target information 32 related to the anomaly case and generates at least one of the explanatory information 41, confidence information 42, and correction candidate information 43. On the other hand, if the AI ​​analysis unit 14 is unavailable, the fallback unit 15 generates standard suggestion information 44 based on the inspection result 38 indicating an anomaly.

[0062] As shown in Figure 19, the information generated by the AI ​​analysis unit 14 or the fallback unit 15 is output to the operator terminal 2 by the output unit 16. If the AI ​​analysis unit 14 is available, at least one of the explanatory information 41, confidence information 42, and correction candidate information 43 may be output. On the other hand, if the AI ​​analysis unit 14 is unavailable, standardized suggestion information 44 generated by the fallback unit 15 may be output. This allows the operator terminal 2 to be presented with the support results for abnormal cases.

[0063] If the AI ​​analysis unit 14 does not return output in a predetermined format, if the returned output cannot be parsed as JSON, if the output structure does not conform to a predetermined format, or if the action keys included in the output do not conform to a predetermined allow list, the system may not adopt the AI ​​analysis result and switch processing to the fallback unit 15. The fallback unit 15 may generate standardized suggestion information 44 based on the inspection result 38 indicating an anomaly and output it in the same display format as when the AI ​​analysis unit 14 was operating normally. This allows support processing for the operator to continue even if there are output defects or insufficient validity in the AI ​​analysis unit 14.

[0064] The suggested correction information 43 is not limited to simply the corrected value. In one aspect, the suggested correction information 43 may include action information that specifies the processing content. For example, the suggested correction information 43 may include information indicating a process to correct the postal code, a process to correct the address, a process to correct the telephone number, a process to contact the purchaser for confirmation, or a process to ignore the problem and continue processing. Alternatively, an action key may be used as part of the output items included in the AI ​​analysis results. In this case, the action key is information that identifies the type of recommended action.

[0065] Figure 3 shows how the AI ​​analysis unit 14 generates explanatory information 41, confidence information 42, and correction candidate information 43, and sends them to the output unit 16. Figure 19 shows how this information, or the standard suggestion information 44 generated by the fallback unit 15, is ultimately output to the operator terminal 2. Therefore, the order management system 1 is implemented as a complete system, including operator support after anomaly detection.

[0066] The following describes an embodiment of the address-related information 32A. The address-related information 32A is information about the delivery address included in the order data 31. The address-related information 32A may include, for example, the prefecture name, city / ward / town / village name, town / district name, block number 63, street number 64, and building name. In this embodiment, the postal code 61 and regional information 62 included in the address-related information 32A are used as the primary matching targets. Regional information 62 refers to address elements that include at least the prefecture name, city / ward / town / village name, and town / district name.

[0067] As shown in Figure 4, the inspection of address-related information 32A is performed in the following order: address normalization processing 65, first-stage regional matching confirmation 66, second-stage range confirmation 67, and third-stage AI judgment as needed. In this embodiment, the rule inspection unit 12 prioritizes processing matters that can be determined mechanically, and the AI ​​judgment unit 13 performs a complementary judgment only when the rule inspection unit 12 cannot obtain an inspection result 35 based on the judgment rule 34. This realizes a rule-prioritizing, AI-supplementary configuration even in address inspection.

[0068] As shown in Figure 4, the rule inspection unit 12 first performs address normalization processing 65 on the address-related information 32A. Next, the rule inspection unit 12 performs a regional match check 66 to confirm whether the regional information 62 corresponding to the postal code 61 matches the regional information 62 included in the input address-related information 32A. If the result of the regional match check 66 is a mismatch, the rule inspection unit 12 outputs an inspection result 38 indicating an abnormality. On the other hand, if the regional match check 66 is passed, the rule inspection unit 12 performs a range check 67 to confirm whether the block number 63 or house number 64 is within a predetermined range. If the result of the range check 67 is outside the range, the rule inspection unit 12 outputs an inspection result 38 indicating an abnormality. Furthermore, if an inspection result 35 based on the judgment rule 34 cannot be obtained in the regional match check 66 or range check 67, as a third step, the AI ​​judgment unit 13 performs an AI judgment based on at least one of the matching results between the address-related information 32A, the predetermined items 33 extracted from the address-related information 32A, and the reference information 51. The AI ​​determination unit 13 outputs the determination result 36 or interpretation result 37, which may be displayed by the output unit 16 or used in subsequent processing.

[0069] Figure 5 schematically illustrates the step-by-step inspection process and specific examples for address-related information 32A. In the first stage of Figure 5, a mismatch between the postal code 61 and the regional information 62 of the input address is shown. In the second stage of Figure 5, an example is shown where the town area matches, but the block number 63 is outside the range. In the third stage of Figure 5, since the rule inspection unit 12 has difficulty uniquely extracting block number 63 or house number 64 for notations such as "Rokujo," an example is shown in which the AI ​​judgment unit 13 is used. Thus, Figure 5 is a table of specific examples to explain what kind of events each stage of address inspection targets.

[0070] The address normalization processing 65 is pre-processing for suppressing notation variations in the address-related information 32A and stably executing subsequent rule checks. As shown in Figures 6 and 7, the address normalization processing 65 may perform blank removal, line break removal, wave dash unification, large-area character / small-area character removal, small-sized katakana unification, place name particle unification, conversion of full-width numerals to half-width numerals, conversion of Chinese numerals to Arabic numerals, hyphen unification, and old character / variant character unification. These processes may be applied not only to the input address character string but also to the master address character string that serves as the reference destination.

[0071] Figure 6 is a table listing the processing contents of the address normalization processing 65. Figure 7 is a table showing specific examples before and after conversion corresponding to each normalization process. For example, in blank removal, "Tokyo-to Shibuya-ku" is converted to "Tokyo-to Shibuya-ku". In converting full-width numerals to half-width numerals, "Minami 6-jo Higashi" is converted to "Minami 6-jo Higashi". In Chinese numeral conversion, "Roku-jo", "Nijuuichi-chome" and "Juuroku-banchi" are converted to "6-jo", "21-chome" and "16-banchi", respectively. In hyphen unification, "2ー2ー1" is unified to "2-2-1". In old character / variant character unification, the notations corresponding to "Shiogama City", "Kawasaki City" and "Takasaki City" may each be unified to a predetermined standard notation.

[0072] The Chinese numeral conversion in the address normalization processing 65 may be configured so as not to erroneously convert Chinese numerals included in place names. For example, only when a subsequent character such as "chome", "ban", "go", "jo", "sen" or "chi" exists, the immediately preceding Chinese numeral may be converted to an Arabic numeral. By doing so, it is possible to prevent Chinese numerals included in place names such as "Hachioji" or "Mitaka" from being mistakenly recognized as the chome number 63 or the house number 64. Further, the application order of normalization may be set in consideration of dependencies. For example, by performing the conversion of full-width numerals to half-width numerals first, the subsequent regular expression processing may be simplified.

[0073] In address verification, before proceeding to stages 1 through 3 in Figure 4, a preliminary check for the existence of postal code 61 may be performed. That is, the rule verification unit 12 may check whether the entered postal code 61 exists in either Japan Post's postal code master 56 or large business establishment master 57. If it does not exist in either master, an inspection result 38 indicating an anomaly as a non-existent postal code may be output immediately. For example, if the non-existent postal code "999-9999" is entered and there are 0 hits in both masters, an anomaly report as a non-existent postal code may be made.

[0074] In the first stage, the regional match check 66 verifies whether the regional information 62 corresponding to the postal code 61 matches the regional information 62 included in the normalized address-related information 32A. In Figure 4, the regional match check 66 is performed by referring to the postal code 61 and the regional information 62. More specifically, the master record corresponding to the postal code 61 may be obtained, and it may be checked whether the address string contained in the master record matches the beginning of the normalized input address. This check may also be performed as a prefix match check.

[0075] If the result of the regional match check 66 indicates that the regional information 62 of the input address does not match the regional information 62 corresponding to the postal code 61, the rule check unit 12 can output an inspection result 38 indicating an anomaly. For example, if the postal code "100-0001" has master-side regional information of "Chiyoda, Chiyoda-ku, Tokyo" but the input address is "Motomachi, Naka-ku, Hiroshima-shi, Hiroshima-ken", the rule check unit 12 may output an inspection result 38 indicating an anomaly due to a regional mismatch. For this type of anomaly, standard suggestion information 44 that prompts confirmation of the input content of the address or postal code may be associated with it.

[0076] The second stage, range confirmation 67, is performed if the first stage, area matching confirmation 66, is passed. In range confirmation 67, it is checked whether the block number 63 or house number 64 is included in the range information set for the area in question. Here, range information refers to information indicating the block range, house number range, or other area range included in the postal code master 56 or large business master 57, for example. If block number 63 or house number 64 can be mechanically extracted from the input address and the extracted result is within the range information, the rule inspection unit 12 can output a normal inspection result 35.

[0077] For example, consider the case where the town area corresponding to the postal code "060-0042" is "Odori Nishi," and its block range is "1 to 19 chome." When the input address is "Odori Nishi 15 chome 1-1," "15" can be extracted as block number 63, and since this value is within the range, it may be judged as normal. On the other hand, when the input address is "Odori Nishi 25 chome 1-1," "25" can be extracted as block number 63, but since this value is outside the range, an inspection result 38 indicating an anomaly may be output. For this type of anomaly, standard suggestion information 44 prompting confirmation of block number 63 or house number 64 of the address may be associated.

[0078] If block number 63 or street number 64 cannot be mechanically extracted from the input address, the rule inspection unit 12 may treat it as unable to obtain an inspection result 35 based on the judgment rule 34. For example, in numerical notation including Chinese numerals, notation including old characters or variant characters, or special notation, such as "Odori Nishi 15-chome" or "Odori Nishi 15-chome", extraction processing using regular expressions alone may fail to extract block number 63. In such cases, as a third step, the AI ​​judgment unit 13 performs an AI judgment based on the address-related information 32A and outputs a judgment result 36 or an interpretation result 37.

[0079] Note that "the case where the inspection result based on the determination rule cannot be obtained" is not limited to the case where the character string extraction itself fails due to Chinese numerals, old character forms, variant characters, or special notations as described above. For example, this also includes cases where the character string extracted by the rule inspection unit 12 does not conform to a predetermined data format, the extraction result branches into a plurality of candidates and cannot be uniquely determined, or no valid candidate is obtained as a result of collation with master data. In these cases, the rule inspection unit 12 may hand over the processing to the AI determination unit 13, assuming that the inspection result based on the determination rule 34 cannot be obtained.

[0080] The AI determination unit 13 may determine whether the address is within the range information by, for example, interpreting "十五" as "15" or "壱伍" as "15". For example, in the case of "大通西十五丁目", the AI determination unit 13 may interpret it as "15-chome" and determine that it is within the range. Further, in the case of "大通西二十五丁目", it may be interpreted as "25-chome" and determined to be outside the range. As such, the AI determination unit 13 plays a role of complementarily interpreting special notations that are difficult for the rule inspection unit 12 to process.

[0081] As shown in FIG. 8 and FIG. 13, in the inspection of the address-related information 32A, the postal code master 56 and the large-scale business establishment master 57 of Japan Post can be referred to. The rule inspection unit 12 may first extract the postal code 61, and refer to the postal code master 56 to confirm the presence or absence of corresponding data. If there is corresponding data in the postal code master 56, the area information 62 and the range information are acquired from the data and used for consistency check with the address-related information 32A. On the other hand, if there is no corresponding data in the postal code master 56, the large-scale business establishment master 57 may be referred to.

[0082] As shown in Figure 8, the rule inspection unit 12 first obtains the postal code 61 from the address-related information 32A and checks whether the information corresponding to the postal code 61 exists in the postal code master 56. If the corresponding data exists in the postal code master 56, it obtains the regional information 62 and range information contained in that data and uses them for regional matching confirmation 66 and range confirmation 67. On the other hand, if the corresponding data does not exist in the postal code master 56, or if it is not possible to determine whether it is normal or abnormal based solely on the confirmation based on the postal code master 56, the large business establishment master 57 may be referred to. If the corresponding data exists in the large business establishment master 57, regional matching confirmation 66 and range confirmation 67 as necessary can be performed based on that data. If the corresponding data does not exist in either the postal code master 56 or the large business establishment master 57, the rule inspection unit 12 may output an inspection result 38 indicating an abnormality, which indicates that the postal code does not exist.

[0083] Furthermore, in this embodiment, even if the relevant data is obtained in Japan Post's postal code master 56, if a mismatch occurs in the first or second stage, the same flow may be performed on the large business establishment master 57. That is, even if an abnormality is detected in the regional match check 66 or range check 67 in the postal code master 56, the regional match check 66, range check 67, and the third stage AI judgment as needed may be performed on the same postal code 61 and input address in the large business establishment master 57. If a match is found in the large business establishment master 57, it may be judged as normal. On the other hand, if there is a mismatch in both masters, the judgment result from the postal code master 56 may be prioritized and reported as abnormal.

[0084] The Large Business Establishment Master 57 is a master containing business establishment information corresponding to postal codes individually assigned to companies or organizations. Therefore, even if there is no match in the Postal Code Master 56, a match may be found in the Large Business Establishment Master 57. For example, the postal code "100-8111" may not exist in the general Postal Code Master 56, but it may be associated with "1-1 Chiyoda, Chiyoda-ku, Imperial Household Agency" in the Large Business Establishment Master 57. In this case, a regional matching check 66 and, if necessary, a range check 67 may be performed based on the Large Business Establishment Master 57.

[0085] If the relevant data does not exist in either the postal code master 56 or the large business establishment master 57, the rule inspection unit 12 may output an inspection result 38 indicating an anomaly as a non-existent postal code. For this anomaly, standard suggestion information 44 may be associated with prompting the user to confirm the entered postal code. This allows for the detection of cases where the postal code itself is incorrect before proceeding to the regional matching check 66 or range check 67.

[0086] Figure 13 shows an example of a functional configuration in which, after obtaining address-related information 32A, the rule inspection unit 12 performs a postal code search and a large business establishment search, and then verifies consistency with reference information. From Figure 13, it can be understood that both the postal code master 56 and the large business establishment master 57 are positioned as reference information sources for address inspection. Therefore, Figure 13 is a diagram that supplements the processing flow of Figure 8 from the perspective of functional blocks.

[0087] The AI ​​analysis unit 14 may generate additional explanatory information 41, confidence information 42, and correction candidate information 43 for the anomalous case if it obtains an inspection result 38 indicating an anomaly in the address-related information 32A. Figure 14 shows an example of the setting criteria and specific examples for the confidence information 42. For example, if it can be clearly determined that there is an error from the matching result of the postal code master 56 and a single correct value can be identified, the confidence information 42 may be set to high. If there is a high possibility of an error but there remains a possibility of variation in notation or a legitimate reason, it may be set to medium. If there is some inconsistency but there is still a good chance that it is a normal case, it may be set to low.

[0088] Figure 15 shows a specific example of the correction candidate information 43. For example, for a mismatch between a postal code and an address, the correction candidate may include options to correct the postal code, options to correct the address, options to contact the buyer for confirmation, and options to ignore the problem. For an anomaly outside the district range, the correction candidate information 43 may include multiple prioritized candidates, and the most recommended candidate may be marked with a recommendation flag.

[0089] The AI ​​judgment unit 13 may accept at least one of the following as input: the information to be inspected 32, predetermined items 33 extracted from the information to be inspected 32, the intermediate inspection results by the rule inspection unit 12, and the results of comparison with the reference information 51. Based on these inputs, the AI ​​judgment unit 13 may output a judgment result 36 corresponding to normal, abnormal, or judgment pending, or an interpretation result 37 indicating the interpretation of the input notation. For example, for address-related information 32A, the AI ​​judgment unit 13 may accept the postal code 61, regional information 62, block number 63, street number 64, and reference candidate information as inputs, and for telephone number-related information 32B, the AI ​​judgment unit 13 may accept the telephone number, delivery address, numbering system information, and special number range information as inputs. This allows the AI ​​judgment unit 13 to make a complementary judgment based on the context of the entire order data 31 for cases where the rule inspection unit 12 has difficulty obtaining a unique conclusion.

[0090] Furthermore, the AI ​​analysis unit 14 may accept at least one of the following as input: an inspection result 38 indicating an anomaly, inspection target information 32 related to the anomaly case, the result of comparison with reference information 51, and the judgment result 36 or interpretation result 37 of the AI ​​judgment unit 13. Based on these inputs, the AI ​​analysis unit 14 may output at least one of the following: explanatory information 41 indicating the reason for the anomaly, confidence information 42 indicating the likelihood of the explanation or candidate, and correction candidate information 43 indicating a candidate response. Thus, the AI ​​analysis unit 14 may be positioned not as a function to determine the presence or absence of an anomaly itself, but as a function to generate information that supports subsequent response decisions.

[0091] The confidence level information 42 may be determined based on the degree of agreement with the reference information 51, the number of candidates extracted, the clarity of the reason for the anomaly, and whether or not there is competition among multiple candidates. For example, if a candidate is uniquely narrowed down by matching with the reference information 51 and the reason for the anomaly is clear, it may be set as high confidence. On the other hand, if multiple candidates coexist and it is not possible to uniquely determine which one to adopt, it may be set as medium confidence or low confidence. This makes it easier for the operator to decide whether to adopt the candidate as is or to perform additional verification based on the confidence level information 42.

[0092] The correction candidate information 43 may be generated based on candidate information narrowed down in advance by matching with reference information 51, anomaly types, and contextual information contained in order data 31. For example, for address-related information 32A, correction candidates may be generated using regional information 62, block ranges, or candidate addresses obtained by matching with postal code master 56 or large business master 57. Also, for telephone number-related information 32B, candidates related to correction, confirmation contact, or continuation of processing may be generated using numbering system information, regional correspondence information, or special number range information. In this way, the correction candidate information 43 can be presented as a candidate with objective grounds based on reference information 51.

[0093] As shown in Figure 16A, the information input to the AI ​​analysis unit 14 may include the type of anomaly, the specific reason for the anomaly, the detection stage, a snapshot of the order data, and the result of comparison with reference information 51. Examples of anomaly types may include mismatch between postal code and address, non-existent postal code, failure to extract block number, incorrect telephone number format, regional mismatch, or ambiguous cases related to special number ranges. Examples of detection stages may include values ​​indicating whether the anomaly was detected in the first stage regional matching check 66, the second stage range check 67, or whether the case was extracted in the format check 73, number system check 74, regional correspondence check 75, or special number range check 76 for telephone number related information 32B. This makes it easier for the AI ​​analysis unit 14 to generate explanatory information 41 and correction candidate information 43 according to the nature of the anomaly.

[0094] The snapshot of the order data may include, for example, the delivery address, postal code, name, remarks, and other order information. The results of the comparison with reference information 51 may also include information such as the expected town area, information source, block range, extracted block number, and other information obtained as a result of the comparison with the postal code master 56 or the large business master 57. Of this information, the expected town area or block range, etc., are candidate information that has been narrowed down in advance by the comparison with reference information 51. The AI ​​analysis unit 14 can reduce the risk of freely generating values ​​that do not exist in the master by referring to such candidate information and generating the corrected candidate information 43.

[0095] As shown in Figures 16B and 17, the information output from the AI ​​analysis unit 14 may include, for example, an explanatory text corresponding to explanatory information 41, a confidence level corresponding to confidence level information 42, and a group of candidates corresponding to the correction candidate information 43. Each element of the candidate group may include a label, a description, a recommendation flag, and parameters necessary at runtime. Here, the recommendation flag is information that identifies the most recommended candidate from among multiple candidates. Parameters may include, for example, a recommended postal code, a recommended address, a name of the item to be checked, or a name of the field to be reflected.

[0096] The AI ​​analysis unit 14 may return output in JSON format. When using JSON format, the explanatory information 41, confidence information 42, and correction candidate information 43 may be output, each associated with a predetermined key. Each element of the candidate group may include an action key, button label, description, recommendation flag, and parameters. For example, action keys corresponding to postal code correction, address correction, purchaser verification, or ignore may be used. Figure 17 shows a specific example of such JSON format.

[0097] Figure 16C shows an example of implementation-dependent information when performing AI analysis. For example, the AI ​​judgment unit 13 or the AI ​​analysis unit 14 may access an external large-scale language model via the AWS Bedrock Runtime API. As a model, for example, an Anthropic Claude model may be used. In addition, HTTPS communication, authentication method, timeout value, API version, temperature parameter, or maximum number of tokens may be set as appropriate depending on the operating environment. However, the information shown in Figure 16C is merely one implementation example, and the present invention is not limited to these.

[0098] The output returned by the AI ​​analysis unit 14 does not need to be adopted unconditionally. For example, the system may verify whether the output of the AI ​​analysis unit 14 conforms to a predetermined format. It may also verify whether the action key included in the correction candidate information 43 or action information is included in a predetermined permission list. Candidates with action keys not included in the permission list may be automatically excluded. This helps to prevent the AI ​​analysis unit 14 from returning unexpected processing content.

[0099] Figure 18 shows specific examples of conditions under which the system transitions to a fallback when the AI ​​analysis unit 14 becomes unavailable. The AI ​​analysis unit 14 may be unavailable in cases of API communication timeout, server-side error, client-side error, missing text content, JSON parsing failure, invalid JSON structure, or zero valid actions. In such cases, the system may discontinue using the AI ​​analysis results and automatically switch to rule-based standardized suggestion information 44.

[0100] Furthermore, in this embodiment, the use of AI may be a stateless design on a per-request basis. Therefore, after the AI ​​has recovered successfully, AI analysis or AI judgment may automatically resume from the next request without any explicit recovery processing. In the current implementation example, a configuration without a dedicated circuit breaker may be adopted. Such a configuration may simplify the processing in the event of an AI failure and the processing after the AI ​​has recovered.

[0101] Furthermore, if an AI failure occurs, this fact may be recorded in a log. For example, the failure type, exception message, abnormality type, identification information of the target order data, and the fact that fallback was executed may be recorded in the application log at a warning level. This makes it easier to track the frequency or cause of AI failures later on. It also makes it easier to audit whether the switch to the fallback unit 15 was executed appropriately.

[0102] Figure 20 shows an example of standardized suggestion information 44 associated with the type of anomaly. For example, in the case of a mismatch between the postal code and address, suggested postal code corrections, suggested address corrections, and suggested buyer confirmations may be stored as standardized suggestion information 44. In the case of a non-existent postal code, a suggestion prompting confirmation of the postal code may be stored. In the case of being outside the district range, a suggestion prompting confirmation of district number 63 or street number 64 may be stored. In the case of an invalid telephone number format or invalid numbering system, a suggestion prompting confirmation of the entered telephone number may be stored.

[0103] The confidence information 42 generated by the fallback unit 15 may be set to high confidence. This is because the fallback unit 15 generates standard suggestion information 44 based on the inspection result 38 indicating an anomaly definitively obtained by the rule inspection unit 12. For example, if a regional mismatch is confirmed by matching with the postal code master 56, the explanation may be set to "mismatch between postal code and address," the confidence level may be set to high, and recommended candidates may be clearly indicated. This allows the operator to refer to the confidence information 42 and determine the appropriate course of action, regardless of whether it is an AI analysis result or a fallback result.

[0104] The following describes an embodiment of the telephone number-related information 32B. The telephone number-related information 32B is information about the telephone number included in the order data 31. The telephone number-related information 32B may include the delivery address telephone number, the orderer's telephone number, the contact telephone number, and other numbers used for order processing or delivery communication. In this embodiment, the telephone number used for contacting the delivery address will be described as the main subject of inspection. The predetermined items 33 included in the telephone number-related information 32B include the number of digits, the leading digit, the area code, the mobile prefix, regional information, and special number range information.

[0105] As shown in Figures 9A and 9B, the inspection of telephone number-related information 32B is performed in the following order: telephone number normalization processing 72, first-stage format verification 73, second-stage number system verification 74, regional correspondence verification 75 or special number range verification 76, and third-stage AI judgment as needed. Therefore, for telephone number-related information 32B as well as address-related information 32A, rule-based inspection is prioritized, and the AI ​​judgment unit 13 operates supplementarily only when a judgment cannot be made by the rule-based inspection. This makes it easier to ensure the accuracy of anomaly detection while suppressing processing load and operational costs in telephone number inspection as well.

[0106] As shown in Figure 9A, the rule inspection unit 12 first performs telephone number normalization processing 72 on the telephone number-related information 32B. Next, it performs a format check 73 to confirm whether the telephone number satisfies a predetermined format. If the format check 73 indicates an invalid format, the rule inspection unit 12 outputs an inspection result 38 indicating an abnormality. On the other hand, if the format check 73 is passed, the second stage involves a number system check 74, a regional correspondence check 75, or a special number range check 76. If the number system check 74, regional correspondence check 75, or special number range check 76 can determine whether it is normal or abnormal, the rule inspection unit 12 outputs the inspection result 35.

[0107] As shown in Figure 9B, if the telephone number-related information 32B that has passed the format check 73 cannot be determined to be normal or abnormal even after performing the number system check 74, regional correspondence check 75, or special number range check 76, the AI ​​determination unit 13 performs an AI-based determination as the third stage. The AI ​​determination unit 13 makes a determination based on the telephone number-related information 32B, the delivery address, and other related information contained in the order data 31, and outputs a determination result 36 or an interpretation result 37. This allows the rule inspection unit 12 to make a complementary determination even in ambiguous cases where a unique conclusion cannot be reached.

[0108] Figures 10A, 11, and 12 show the contents and specific examples of the telephone number normalization process 72. The telephone number normalization process 72 may include processes for converting full-width numbers to half-width numbers, converting full-width hyphens to half-width hyphens, removing leading country code notation to match the domestic format, removing parentheses, and removing spaces. For example, "03-1234-5678" may be normalized to "03-1234-5678", and "+81-90-1234-5678" may be normalized to "090-1234-5678". Also, "(03)1234-5678" and "03 1234 5678" may be unified to the same domestic format.

[0109] Figure 10B shows a specific example of format check 73. Format check 73 verifies whether the normalized telephone number is formally valid as a telephone number within Japan. Format check 73 can include, for example, a digit count check to confirm that the number of digits after hyphen removal is 10 or 11, a leading number check to confirm that the leading digit is 0, a prohibited character check to confirm that no characters other than numbers and hyphens remain, and a pattern check to confirm that it matches a predetermined valid pattern. For example, "03-1234-5678" and "090-1234-5678" may pass, while "03-1234-567", "090-1234-56789", "90-1234-5678", "03-12AB-5678", or "00-1234-5678" may output an abnormal inspection result 38 indicating a format error.

[0110] In the second stage, telephone numbers that have passed format verification 73 undergo number system verification 74, regional correspondence verification 75, or special number range verification 76. Number system verification 74 is a process to confirm whether the telephone number belongs to a valid number range in light of the Japanese telephone number system. For example, it may be checked whether the first two to five digits exist in a designated area code master, or for 11-digit numbers, whether the first three digits are a valid mobile phone prefix or IP phone prefix. This allows for the mechanical detection of unassigned area codes or unassigned mobile phone prefixes. Regional correspondence verification 75 is a process to confirm whether the region indicated by the area code of a landline telephone matches the region indicated by the delivery address geographically. Special number range verification 76 is a process to confirm whether toll-free numbers, navigation numbers, or other service numbers are entered as contact numbers for personal delivery addresses.

[0111] For example, information such as "03-1234-5678" corresponding to Tokyo, "06-1234-5678" corresponding to Osaka Prefecture, and "099-123-4567" corresponding to Kagoshima Prefecture may be stored as reference information 51. If the region corresponding to the area code matches the prefecture of the delivery address, a normal inspection result 35 may be output. On the other hand, since mobile phone numbers or IP phone numbers are not tied to a region, they may be treated as excluded from the regional correspondence check 75.

[0112] Figure 10C shows a specific example of regional correspondence check 75. For example, if the delivery address is Shinjuku-ku, Tokyo for "03-1234-5678", or if the delivery address is Osaka-shi, Osaka Prefecture for "06-1234-5678", the regional correspondence check 75 may determine that it is normal. Also, for "090-1234-5678" or "050-1234-5678", since they are mobile phones or IP phones, the regional check may be omitted and they may be determined to be normal. On the other hand, "039-123-4567" is an unassigned area code, so it may be determined to be abnormal by number system check 74, and "060-1234-5678" is an unassigned mobile prefix, so it may be similarly determined to be abnormal.

[0113] Special number range verification 76 may also be performed in the second stage. As mentioned above, special number range verification 76 is a process to check whether a service number such as a toll-free number has been entered as the personal delivery address. For example, if a telephone number starting with "0120-123-456" or "0570" has been entered, that telephone number may be detected as a special number range. However, just because a number is in the special number range does not mean it is always abnormal; if the delivery address is a corporation, it may be a legitimate use. For this reason, in this embodiment, special number range verification 76 may function as a trigger to extract ambiguous cases and pass them on to the third stage.

[0114] Figure 10D shows a specific example of performing an AI judgment on telephone number-related information 32B. For example, if the area code is "06-1234-5678" indicating Osaka, but the delivery address is Sapporo, Hokkaido, the AI ​​judgment unit 13 may generate explanatory information 41 and reliability information 42 that take into account gift delivery, maintaining the number after moving, and other circumstances. Also, even if "0120-123-456" is entered as the delivery address telephone number, if the recipient name includes a company name, the AI ​​judgment unit 13 may output an interpretation result 37 indicating that it may be acceptable as a corporate representative number. On the other hand, even if it is the same special number range, if the delivery address is in the name of an individual, the AI ​​judgment unit 13 may output a judgment result 36 indicating that there is a high possibility that the delivery company will not be able to contact the recipient.

[0115] The AI ​​judgment unit 13 can also complementarily handle similar numbers with a single digit difference or mixed natural language notations. For example, if "03-1234-5678" exists for another order to the same address, but "03-1234-5679" is entered in the target order, the AI ​​judgment unit 13 may output a judgment result 36 indicating the possibility of a mix-up with a fax number or an input error. Also, if the remarks column states "The phone number is zero nine zero one two three four five six seven eight," the AI ​​judgment unit 13 may interpret this as "090-1234-5678" and determine its consistency with the value in the phone number field. In this way, the AI ​​judgment unit 13 can complement natural language or context-dependent issues that are difficult to handle with format verification 73 and number system verification 74 alone.

[0116] Furthermore, the area code master, mobile prefix table, or special number range information used for telephone number verification may not be implemented or may be implemented only to a limited extent in the current implementation. On the other hand, as an example of future expansion, these may be stored in the storage unit 17 as reference information 51 and subject to periodic import or batch update. In this way, the same management method as the master management configuration for address-related information 32A can be extended to telephone number-related information 32B.

[0117] As shown in Figure 21, the reference information 51 is stored in the storage unit 17. The storage unit 17 is a functional unit that stores various masters, dictionaries, or system information used by the rule inspection unit 12 for inspection. In this embodiment, at least the postal code master 56 and the large business master 57 are included in the reference information 51 stored in the storage unit 17. When the rule inspection unit 12 receives an inspection request based on the order data 31, it retrieves the necessary reference information 51 from the storage unit 17 and performs the inspection based on the reference information 51.

[0118] As shown in Figure 21, when the rule inspection unit 12 receives a search request for reference information 51 based on the information to be inspected 32, it first refers to the cache unit 18 to check whether a search result cache 59 corresponding to the search request exists. If a search result cache 59 exists, the rule inspection unit 12 retrieves the search result cache 59 and performs an inspection based on its contents. On the other hand, if a search result cache 59 does not exist, the rule inspection unit 12 refers to the storage unit 17 to obtain the necessary reference information 51. The rule inspection unit 12 may then save the search results obtained from the storage unit 17 to the cache unit 18 and then perform an inspection based on those search results. In addition, the periodic data acquisition processing unit 19 may obtain externally updated master data and update the reference information 51 held in the storage unit 17, and invalidate or delete the search result cache 59 in accordance with the update.

[0119] Reference information 51 is not obtained by directly querying an external public site or external API each time an inspection is performed, but may be stored within the system. For example, the postal code master 56 may be generated based on a CSV file of postal code data published by Japan Post Co., Ltd. Similarly, the large business establishment master 57 may be generated based on a CSV file of large business establishment data published by Japan Post Co., Ltd. This makes the system less susceptible to external communication failures during inspection execution and facilitates high-speed rule inspection.

[0120] The periodic import processing unit 19 is a functional unit that acquires master data updated externally at predetermined intervals and updates the reference information 51 held in the storage unit 17. Figure 21 shows an example in which the periodic import processing unit 19 acquires external master data and updates the postal code master 56 and the large business establishment master 57. In this embodiment, a CSV file updated monthly by Japan Post Co., Ltd. may be downloaded and reflected in the system through a dedicated import process. This allows for continuous inspection based on the latest postal code or large business establishment information while maintaining the freshness of the reference information 51.

[0121] The update method for the reference information 51 may be, for example, a wash-and-replace method. In the wash-and-replace method, existing master data may be deleted, and then newly acquired data may be registered in bulk. The import process may also be performed within a transaction, or registration may be performed in units of a certain number of records. For example, a configuration that registers in units of 1000 records may be adopted. By using such an update method, it is easier to keep the contents of the reference information 51 consistent and to suppress the mixing of old and new data. However, the update method is not limited to this, and differential update methods or other methods may be adopted.

[0122] The reference information 51 stored in the memory unit 17 may be accessed by all tenants in common. That is, instead of each e-commerce business maintaining its own individual master copy, multiple tenants may access a master stored in a common shared database. This allows for centralized management of master data and simplifies update operations. Furthermore, high-speed searching may be achieved by setting indexes for postal codes and other search keys for each master.

[0123] The cache unit 18 does not have the function of permanently storing the reference information 51 itself, but rather it is a function unit that temporarily stores the results of the rule inspection unit 12's search of the reference information 51 as a search result cache 59. As shown in Figure 21, when the rule inspection unit 12 receives an inspection request based on the order data 31, it first refers to the cache unit 18 to check whether a search result cache 59 corresponding to the inspection request exists. If a search result cache 59 exists, the rule inspection unit 12 retrieves the search result cache 59 and performs the inspection based on the result.

[0124] On the other hand, if no corresponding search result cache 59 exists in the cache unit 18, the rule inspection unit 12 refers to the storage unit 17 to obtain the necessary reference information 51. For example, it may obtain regional information 62 and range information corresponding to the entered postal code 61 from the storage unit 17. Then, the rule inspection unit 12 may store the obtained search results in the cache unit 18 and then perform an inspection based on those search results. This reduces the burden of performing the same search again when the same or similar inspection requests occur repeatedly.

[0125] The unit of the search result cache 59 may be determined according to the type of information to be inspected 32. For example, for address-related information 32A, the postal code 61 may be used as the key to store the corresponding regional information 62, block range, and reference master type together. This makes it easier to avoid duplicate search processing for multiple order data 31 that contain the same postal code 61. In addition, for telephone number-related information 32B, the search result cache 59 may be stored using the area code, mobile prefix, or number range information as the key.

[0126] The search result cache 59 stored in the cache unit 18 may be managed to be consistent with updates to the reference information 51 by the periodic import processing unit 19. For example, if the periodic import processing unit 19 updates the postal code master 56 or the large business master 57, the search result cache 59 corresponding to the update may be invalidated or deleted. This prevents the continued use of the search result cache 59 based on the old version of the reference information 51. Alternatively, a retention period may be set for the search result cache 59, and a re-search may be performed after a predetermined time has elapsed.

[0127] The embodiments of the remarks column information 32C and name-related information 32D will be described below. These embodiments reinforce the central configuration related to address-related information 32A and telephone number-related information 32B and serve as the basis for future expansion or division. However, in these embodiments, the step-by-step processing of rule priority and AI supplementation can also be applied to these.

[0128] The remarks column information 32C is free-form text information from the buyer or customer included in the order data 31. The remarks column information 32C may include standard requests such as "Please provide a receipt," "Please include a gift wrapping," "Please deliver on the morning of March 15th," or "Please leave it at the door." In addition, the remarks column information 32C may include free-form text containing multiple requests, or ambiguous descriptions that do not easily fit into standard patterns.

[0129] As shown in Figures 22 and 23, the analysis of the remarks column information 32C is performed in the following order: first stage keyword detection, second stage pattern classification, and third stage free text interpretation. In the first stage, the rule inspection unit 12 performs a comparison with a predefined keyword list. In the second stage, the rule inspection unit 12 extracts structured information using regular expressions and other pattern matching rules. In the third stage, if sufficient results cannot be obtained in each of the above stages, the AI ​​judgment unit 13 performs free text interpretation.

[0130] For example, if the message includes "Please provide a receipt," the AI ​​may generate a keyword detection result 81 indicating that a receipt is required. Similarly, if the message includes "Please deliver it on the morning of March 15th," the AI ​​may generate structured information indicating a specified delivery date and time as a pattern classification result 82. Furthermore, if the message is free-form text such as "Thank you for your continued support; I'll leave the packaging to you, preferably in blue," the AI ​​judgment unit 13 may generate a free-form text interpretation result 83, extracting the request for gift wrapping and color specification.

[0131] As a result of the third stage for the remarks column information 32C, in addition to the free text interpretation result 83, explanatory information 41, confidence information 42, and corresponding candidate information may be output as needed. Here, the corresponding candidate information is information indicating candidate processing content that can be taken for the remarks column correspondence. In this embodiment, the corresponding candidate information may be positioned as one form of the correction candidate information 43. That is, the correction candidate information 43 is not limited to correction candidates for addresses or telephone numbers, but may also include action candidates related to the remarks column correspondence. In this case, the corresponding candidate information may also include action information or action keys.

[0132] Name-related information 32D is information about names included in the order data 31. Name-related information 32D may include the name of the orderer, the name of the delivery address, the name of the recipient, and other name items. In this embodiment, we will describe inconsistency detection focusing particularly on the relationship between the orderer's name and the delivery address. However, the use of name-related information 32D is not limited to this, and it may also be used as auxiliary information to determine the consistency between the orderer's name and the remarks column information 32C, or between the delivery address and the delivery address.

[0133] As shown in Figures 24 and 25, the detection of mismatches in name-related information 32D is performed in the following order: first stage: exact match determination, second stage: normalized comparison, and third stage: context determination. In the first stage, the rule inspection unit 12 checks whether the strings of the orderer's name and the recipient's name are an exact match. In the second stage, the rule inspection unit 12 performs a surname-only match determination, a re-comparison after standardizing to old characters, and other normalized comparisons. In the third stage, if it is difficult to mechanically determine whether there is a match or a mismatch in any of the previous stages, the AI ​​determination unit 13 performs a context determination.

[0134] As shown in Figure 24, the rule inspection unit 12 first determines, in the first stage, whether the strings of the orderer's name and the recipient's name are exactly the same. If it is determined in the first stage that they are exactly the same, the name-related information 32D may be considered a match and the process may be terminated. On the other hand, if they are not exactly the same in the first stage, the second stage involves checking for a match only in the surname, re-comparison after unifying to old characters, or other normalized comparisons. If it is determined that they are a match in the second stage, the name-related information 32D may be considered a match and the process may be terminated. Furthermore, if it is difficult to mechanically determine whether they are a match or a mismatch even after the second stage, the AI ​​determination unit 13 may perform a contextual determination in the third stage and output a contextual determination result 85, a determination result 36, or an interpretation result 37.

[0135] For example, if the orderer's name is "Taro Tanaka" and the recipient's name is also "Taro Tanaka," the system may output a normal check result 35, indicating an exact match. Alternatively, if the orderer's name is "Taro Watanabe" and the recipient's name is "Taro Watanabe," the system may determine a match after comparing the variant characters. On the other hand, if the orderer's name and the recipient's name are completely different, and there are no indications of gift wrapping, gift wrapping, gift-related notes in the remarks section, or other circumstances, the system may output a result 36 indicating a possible input error or misplacement of the input field.

[0136] As shown in Figure 26, in a modified example, for cases where the rule inspection unit 12 cannot obtain an inspection result 35 based on the judgment rule 34, the AI ​​judgment unit 13 performs an AI judgment. If the AI ​​judgment unit 13 determines that the case is abnormal, the case may be handed over to the AI ​​analysis unit 14. The AI ​​analysis unit 14 performs a subsequent analysis based on the inspection target information 32, judgment result 36, or interpretation result 37 related to the case, and may generate at least one of the explanatory information 41, confidence information 42, and correction candidate information 43. Thus, in addition to the mode in which the AI ​​analysis unit 14 operates mainly for cases where the rule inspection unit 12 obtains an inspection result 38 indicating an abnormality, as in the basic embodiment, subsequent support can also be provided for cases where the rule inspection unit 12 cannot make a judgment and the AI ​​judgment unit 13 has determined that the case is abnormal.

[0137] As shown in Figure 27, in another modified example, the AI ​​analysis unit 14 may first acquire an abnormal case and generate at least one of the following for the abnormal case: explanatory information 41, confidence information 42, and correction candidate information 43. Next, based on the generated confidence information 42, it is determined whether the confidence level of the abnormal case is above a predetermined value. If the confidence level is above the predetermined value, the recommended correction candidate may be automatically applied without operator confirmation. On the other hand, if the confidence level is below the predetermined value, automatic application may not be performed, and the candidate may be presented to the operator terminal 2 for confirmation, and processing may be carried out according to the confirmation result. This allows for the automation of processing for cases with high confidence levels, while requiring operator confirmation for cases with insufficient confidence levels.

[0138] The correction candidates subject to automatic application may, for example, be candidates with a recommendation flag among the correction candidate information 43. In addition, the conditions for automatic application may require that the confidence information 42 be high, the candidate be unique, the candidate's action key be included in a predetermined list of automatically applicable allowed lists, or the validity of the candidate be confirmed by consistency checks with reference information 51. This can enhance the security of automatic application.

[0139] Figures 28A to 28C show examples of functional enhancements. In the first example shown in Figure 28A, the detection targets include mismatches between postal codes and addresses, non-existent postal codes, and addresses outside the specified block range. Candidate generation involves the generation of explanatory text, confidence levels, and correction candidates by AI. The notification method uses periodic screen updates, and the AI ​​utilizes a cloud AI service. In the second example shown in Figure 28B, the detection targets are further expanded to include fraudulent phone numbers, analysis of requests in the remarks field, suspected duplicate orders, and mismatches between names and delivery addresses. In the third example shown in Figure 28C, automatic correction methods include automatically correcting high-confidence issues without operator confirmation, controlling the level of automation according to confidence levels as a step-by-step automation method, and visualizing the trend of detections and automatic correction rates by tenant or rule as an analysis function.

[0140] As shown in Figure 29, the response method may be changed according to the confidence level. For example, if the confidence level is high, automatic correction may be performed; if the confidence level is medium, operator confirmation may be requested with a recommended action; and if the confidence level is low, operator confirmation may be requested without a recommendation. On the other hand, in earlier operational modes, a method may be adopted in which candidates are displayed on the screen for high, medium, and low confidence levels. In this way, confidence level information 42 may be used as an indicator to control the level of automation.

[0141] Figure 30 shows an example of the display of the order data inspection results screen. The operator terminal 2 may be provided with an error list display field 92, an explanatory information display field 93, a reliability display field 94, a correction candidate display field 95, a recommended candidate display field 96, and a group of operation buttons 97. The operator can select a candidate from the correction candidate display field 95 or the recommended candidate display field 96 while referring to the explanatory information display field 93 and the reliability display field 94. In addition, in operational modes where automatic correction is permitted, the group of operation buttons 97 may be provided with buttons for approval after automatic application, rejection, or log reference.

[0142] It is not necessary to implement all of the above-described modifications and functional enhancements simultaneously. For example, during initial implementation, rule checking, AI analysis, and fallback may be implemented for address-related information 32A, and then sequentially extended to telephone number-related information 32B, remarks column information 32C, and name-related information 32D. Furthermore, automatic correction may be initially disabled and enabled only for highly reliable candidates after sufficient operational experience has been accumulated. Thus, the present invention can be flexibly configured according to the implementation, operation, and enhancement stages.

[0143] The functional divisions of the AI ​​judgment unit 13, the AI ​​analysis unit 14, and the fallback unit 15 may be modified in various ways. For example, in one embodiment, the AI ​​judgment unit 13 may only perform the abnormality determination, while the AI ​​analysis unit 14 may perform the generation of explanatory information 41, confidence information 42, and correction candidate information 43. In another embodiment, the activation of the AI ​​analysis unit 14 may be switched depending on the output content of the AI ​​judgment unit 13. Furthermore, the fallback unit 15 may be used not only when the AI ​​analysis unit 14 is unavailable, but also as an auxiliary when the automatic application conditions are not met. None of these modifications deviate from the spirit of the present invention.

[0144] Furthermore, although this embodiment describes an example targeting order data 31 in the order management system 1, the present invention is not limited to e-commerce orders and may be applied to the inspection of various types of data, including delivery address information, contact information, or free-text information. For example, the rule-first / AI-completion configuration of the present invention may also be applied to address, telephone number, remarks field, or name-related information included in application data, member registration data, delivery instruction data, or inquiry data. Therefore, the information to be inspected 32 in the present invention is not limited to information included in order data 31, but may be extended to various types of information for which consistency verification between predetermined items 33 and reference information 51 is required.

[0145] As described above, according to this embodiment, the rule inspection unit 12 prioritizes determining items that can be mechanically processed by the inspection target information 32 included in the order data 31, and only uses the AI ​​determination unit 13 or the AI ​​analysis unit 14 when such determination is difficult. Furthermore, if the AI ​​is unavailable, the fallback unit 15 can output standard suggestion information 44. Therefore, according to this embodiment, it is possible to realize order data inspection that easily balances processing efficiency, consistency, business continuity, and operator support.

[0146] Furthermore, the steps of each process described herein may be executed in any order, as long as there is no inconsistency in the processing content, and some may be executed in parallel. In addition, some or all of the functional units constituting the order management system 1 may be implemented by dedicated hardware circuits such as ASICs (Application Specific Integrated Circuits) or FPGAs (Field Programmable Gate Arrays). Moreover, the programs for implementing each functional unit may be provided by recording them on a computer-readable non-temporary recording medium, or by providing them via a telecommunications line. [Explanation of symbols]

[0147] 1. Order Management System 2 Operator terminal 10. Order Data Inspection Function 11 Acquisition Department 12. Rules Inspection Department 13 AI judgment section 14 AI Analysis Department 15 Fallback section 16 Output section 17 Memory section 18. Cache Section 19 Periodic Data Acquisition Processing Unit 31. Order Data 32. Information to be tested 32A Address-related information 32B Telephone number related information 32C Remarks column information 32D Name-related information 33 Prescribed items 34 Judgment rules 35 Test Results 36 Judgment result 37 Interpretation Results 38 Abnormal test results 41. Explanatory Information 42. Reliability Information 43. Information on proposed revisions 44 Standard proposal information 51 Reference Information 56 Postal Code Master 57 Large Business Establishment Master 59 Search Results Cache 61 Postal Code 62 Local Information 63rd Street Number 64 65 Address Normalization Processing 66. Confirmation of regional match 67. Range confirmation 72. Telephone number normalization process 73 Format confirmation 74 Numbering System Confirmation 75. Regional coverage confirmation 76 Special number range confirmation 81 Keyword Detection Results 82 Pattern Classification Results 83. Results of Free Text Interpretation 84 Name Matching Result 85 Context determination result 91 Order Data Inspection Results Screen 92 Anomaly List Display Field 93. Explanation Information Display Field 94. Confidence level display field 95 Correction candidate display field 96 Recommended Candidates Display Area 97 Control Buttons

Claims

1. An acquisition unit that acquires inspection target information included in order data, A rule inspection unit that checks the consistency between predetermined items included in the information to be inspected and reference information based on predetermined judgment rules, If, as a result of the inspection by the rule inspection unit, an inspection result based on the judgment rule cannot be obtained for the predetermined item, the AI ​​judgment unit will perform an AI judgment based on the inspection target information. If the inspection by the aforementioned rule inspection unit yields an inspection result indicating an anomaly, the AI ​​analysis unit performs an AI analysis based on the inspection target information related to the inspection result indicating the anomaly, and generates at least one of the following: problem explanation information, reliability information, or correction candidate information. Output section, Equipped with, The output unit is, If the rule inspection unit is able to obtain an inspection result for the predetermined item based on the judgment rule, it outputs the inspection result. If the test result indicates an abnormality and the AI ​​analysis unit performs the analysis, then at least one of the problem description information, the reliability information, or the correction candidate information generated by the AI ​​analysis unit is output. If the rule inspection unit is unable to obtain an inspection result for the predetermined item based on the judgment rule, the AI ​​judgment unit outputs a judgment result or interpretation result. A system for checking order data, characterized by the following features.

2. The order data inspection system according to claim 1, characterized in that the AI ​​analysis unit generates the correction candidate information based on candidate information narrowed down in advance by matching with the reference information.

3. The order data inspection system according to claim 2, characterized in that the aforementioned correction candidate information includes a plurality of prioritized candidates.

4. The aforementioned order data inspection system is: The order data inspection system according to claim 3, further comprising a fallback unit that outputs standardized suggestion information based on the inspection results indicating the abnormality when the AI ​​analysis unit is unavailable.

5. The order data inspection system according to claim 4, characterized in that the fallback unit outputs the standardized suggestion information if the AI ​​analysis unit does not return an output in a predetermined format, or if the action key included in the output of the AI ​​analysis unit does not conform to a predetermined permission list.

6. The order data inspection system according to claim 5, characterized in that the aforementioned reference information includes master information, dictionary information, system information, or regional information.

7. The order data inspection system according to claim 6, characterized in that the rule inspection unit performs at least one of consistency check, range check, format check, or system check.

8. The order data inspection system according to claim 7, characterized in that the information subject to inspection corresponds to information for individual orders.

9. The order data inspection system according to claim 8, characterized in that the information subject to inspection is address-related information.

10. The order data inspection system according to claim 9, characterized in that the rule inspection unit performs normalization processing on the address-related information.

11. The order data inspection system according to claim 10, characterized in that the normalization process includes at least one of removing whitespace, converting to numbers, unifying variant characters, and unifying hyphens.

12. The order data inspection system according to claim 11, characterized in that the rule inspection unit confirms the consistency between a predetermined item included in the address-related information and the reference information based on the postal code or regional information.

13. The order data inspection system according to claim 12, characterized in that the rule inspection unit obtains information corresponding to the postal code included in the address-related information from the postal code master, and if there is no corresponding data in the postal code master, it refers to the large business establishment master.

14. The order data inspection system according to claim 13, characterized in that the rule inspection unit performs range verification regarding block numbers or street addresses.

15. The order data inspection system according to claim 14, characterized in that the AI ​​determination unit performs an AI determination based on the address-related information when the rule inspection unit cannot obtain an inspection result based on the determination rule due to numerical notation including Chinese numerals, notation including old characters or variant characters, or special notation.

16. The order data inspection system according to claim 8, characterized in that the information subject to inspection is telephone number-related information.

17. The order data inspection system according to claim 16, characterized in that the rule inspection unit performs normalization processing on the telephone number-related information.

18. The order data inspection system according to claim 17, characterized in that the normalization process includes at least one of converting full-width digits to half-width digits, converting full-width hyphens to half-width hyphens, removing parentheses, processing country codes, and removing spaces.

19. The order data inspection system according to claim 18, characterized in that the rule inspection unit performs a format check on the telephone number-related information.

20. The order data inspection system according to claim 19, characterized in that the rule inspection unit performs number system verification, regional coverage verification, or special number range verification on the telephone number-related information.

21. The order data inspection system according to claim 20, characterized in that the AI ​​determination unit performs an AI determination based on the telephone number-related information when the rule inspection unit cannot obtain an inspection result based on the determination rule due to ambiguous cases related to special notation, regional inconsistencies, or special number ranges.

22. The aforementioned order data inspection system is: The order data inspection system according to claim 6, further comprising a storage unit for holding the aforementioned reference information.

23. The order data inspection system according to claim 22, characterized in that the storage unit periodically updates the reference information.

24. The aforementioned order data inspection system is: The order data inspection system according to claim 23, further comprising a cache unit for caching the search results of the aforementioned reference information.

25. The order data inspection system according to claim 4, characterized in that the fallback unit outputs the standardized suggestion information based on the inspection results definitively obtained by the rule inspection unit.

26. The aforementioned fallback section is Reliability information corresponding to the aforementioned standardized proposal information is generated, The order data inspection system according to claim 25, characterized in that the reliability information is set to indicate a high level of reliability among multiple levels of reliability.

27. A computer-based order data inspection method, a) A step of obtaining inspection target information included in the order data, b) A step of checking the consistency between the specified items included in the information to be inspected and the reference information based on predetermined judgment rules, c) If, as a result of the inspection in step b), an inspection result based on the judgment rule cannot be obtained for the predetermined item, the AI ​​will perform a judgment based on the inspection target information; d) If the inspection in step b) yields an abnormal result, the AI ​​performs an analysis based on the inspection target information related to the abnormal inspection result and generates at least one of the problem description information, confidence information, or correction candidate information. e) If an inspection result based on the judgment rule can be obtained for the predetermined item in step b), the inspection result is output. If the test result indicates an abnormality, and the AI ​​analysis in step d) is performed, then at least one of the problem description information, confidence information, or correction candidate information generated in step d) is output. If, in step b), an inspection result based on the judgment rule cannot be obtained for the predetermined item, the AI ​​will output the judgment result or interpretation result in step c), A method for inspecting order data, characterized by including the following:

28. a) A step of obtaining inspection target information included in the order data, b) A step of checking the consistency between the specified items included in the information to be inspected and the reference information based on predetermined judgment rules, c) If, as a result of the inspection in step b), an inspection result based on the judgment rule cannot be obtained for the predetermined item, the AI ​​will perform a judgment based on the inspection target information; d) If the inspection in step b) yields an abnormal result, the AI ​​performs an analysis based on the inspection target information related to the abnormal inspection result and generates at least one of the problem description information, confidence information, or correction candidate information. e) If an inspection result based on the judgment rule can be obtained for the predetermined item in step b), the inspection result is output. If the test result indicates an abnormality, and the AI ​​analysis in step d) is performed, then at least one of the problem description information, confidence information, or correction candidate information generated in step d) is output. If, in step b), an inspection result based on the judgment rule cannot be obtained for the predetermined item, the AI ​​will output the judgment result or interpretation result in step c), A program that causes a computer to execute something.

Images