Recommendation device

Abstract

To construct a mechanism which makes a system recommend a sending destination considered to be appropriate on task regulation, and a mechanism which makes a system recommend a negotiation destination considered to be appropriate on request of a purchase applicant.SOLUTION: A mechanism makes a system recommend a sending destination considered to be appropriate on task regulation by performing machine learning with conversion of a paradigm which is a case of a transaction matter becoming a standard into data, expressing a relationship between a feature of the transaction matter and the sending destination through abstracted numerical processing, and automatically evaluating and calculating the relationship with respect to a new matter. Moreover, the mechanism makes the system recommend a negotiation destination considered to be appropriate on request of a purchase applicant by performing machine learning with conversion of a paradigm in which an appropriate negotiation destination is selected in performing negotiation of the transaction matter into data, expressing a relationship between a feature of dealt products and the request of the purchase applicant and the negotiation destination through abstracted numerical processing, and automatically evaluating and calculating the relationship with respect to the products and new request of the purchase applicant.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a recommendation device that recommends necessary and optimal coordinators and decision makers, or negotiating partners, when making decisions within a company or reaching an agreement in a transaction between companies. [Background technology]

[0002] When making decisions within a company, especially within a large company, information systems called electronic approval systems and workflows are commonly used (see Patent Document 1). The most important part of these systems is the selection of consultants and decision makers, known as destinations, who are determined based on various internal and external rules.

[0003] However, since it is difficult to completely automate the selection of the destination, there are many cases where human interpretation and correction of the rules are involved (see Patent Document 2). There are also proposals to provide and recommend templates for the body of electronic approval requests by replacing human judgment with machine learning techniques, but the crucial recommendation of the destination is not included (see Patent Document 3). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 11-232357 [Patent Document 2] JP 2002-82891 A [Patent Document 3] Patent No. 6985322 Summary of the Invention [Problem to be solved by the invention]

[0005] In order for a company to smoothly advance a business transaction, it is necessary to select an appropriate counterparty according to the contents of the transaction and then carry out the procedures and negotiations. The main situations in which the selection of an appropriate counterparty is required are (1) when deciding the organization and its role (position) that will discuss and approve the internal procedures for carrying out the business transaction (hereinafter referred to as "referral destination selection"), and (2) when deciding the department in the other company that will be responsible for negotiations when negotiating in a business-to-business transaction (hereinafter referred to as "negotiation destination selection").

[0006] <(1) Selection of destination> In many large companies, due to internal control requirements, orders and payments for commercial transactions are required to be submitted, confirmed, discussed, and approved through an information system known as an electronic approval system or workflow (hereafter collectively referred to as workflow system) before being carried out.

[0007] In a typical workflow system, the system registers categories according to the type of transaction based on the company's authority regulations and organizational chart, account items for budget management and financial accounting, and the names of the consultation and approval departments for transactions in that category.The system also registers the job titles (hereafter referred to as roles) of the decision makers or their representatives in each department, and the destinations for workflow (the order and destination of distribution to consultation and approval departments).The mainstream method is for applicants to select an appropriate application pattern from the workflow system's registration menu and submit an application each time they plan or execute a specific transaction (hereafter referred to as a transaction).

[0008] However, when formulating business rules that stipulate procedures for real-world transactions, various elements such as the organization in charge, account items, amount, type and purpose of the transaction, etc. are intricately intertwined. It is difficult to comprehensively organize such various elements, create patterns, and register and maintain them in a workflow system.

[0009] On the other hand, if business regulations and patterns are not properly organized, not only will the scope of internal control be limited, but the person in charge will be unable to select the appropriate circulation procedure, and will have difficulty in formulating the circulation procedure. Furthermore, the internal parties who carry out the checkpoints and checks, including legal regulations, are constantly changing due to organizational changes and personnel transfers, and the internal organizational hierarchy and the operation of delegation authority often change over time. For this reason, cases that cannot be handled by the pre-registered workflow menu are constantly occurring. In addition, in order to quickly detect such on-site conditions and reflect the registered contents in the system after internal discussion, numerous internal procedures and system revisions are required on an ongoing basis.

[0010] As mentioned above, it is difficult to realize workflow control that expresses the diversity and complexity of business transactions in a highly sophisticated manner with existing workflow systems. In addition, there is a drawback in that it is difficult to respond to timely changes in terms of both time and cost.

[0011] <(2) Selection of Negotiation Partner> In electronic commerce systems for business-to-business transactions provided on websites, etc., competitiveness in the market can be improved if buyers of merchandise can move away from providing them with a fixed product catalog and instead customize their options in consultation with suppliers. In real-life websites where electronic commerce systems are deployed, catalogs and templates of products that can be ordered based on combinations of product specifications, prices, delivery dates, etc. are almost always fixed. For this reason, buyers are often only able to freely specify the quantity and a rough delivery date.

[0012] However, in business transactions, it is often preferable for both the buyer and the supplier to negotiate on a case-by-case basis to determine not only product specifications, but also price and delivery schedule specifications (for example, discounts and number of deliveries). This is because it allows the buyer to maximize his or her own utility, and it allows the supplier to build an advantage in the market by proposing a transaction that closely matches the buyer's needs. It also allows the provider of such an e-commerce system to increase their advantage in the market, as they can provide more value than e-commerce systems that cannot be customized.

[0013] However, the supplier organization with which a potential buyer should negotiate may differ depending on the merchandise handled in the transaction and the content of the potential buyer's request. Providers of electronic commerce systems need to comprehensively organize the merchandise handled and the requests of potential buyers, and link them to the appropriate organization to negotiate with. If the system cannot specify the appropriate negotiation partner, smooth commerce cannot be realized unless processes and tasks are prepared to assign a negotiation partner each time after understanding the content of the inquiry.

[0014] The aforementioned (1) destination selection and (2) negotiation partner selection have a common structure. That is, the workflow system in (1) corresponds to the electronic commerce system in (2), and the business management department aiming to strengthen internal control in (1) corresponds to the provider of the electronic commerce system aiming to build market dominance in (2). Also, it is possible to correspond the transaction item in (1) to the handled merchandise and the request of the potential buyer in (2), and the workflow destination in (1) corresponds to the negotiation partner in (2).

[0015] Regarding (1), the present invention aims to provide a recommendation device that constructs a mechanism for having a system recommend a routing destination that is considered appropriate in accordance with business regulations, without implementing workflow routing rules that comprehensively organize business regulations that define the procedural methods for business cases into the system.

[0016] In addition, based on a method similar to the solution of (1) above, with regard to (2) above, it is an object to provide a recommendation device that constructs a mechanism for having the system recommend negotiation partners that are considered appropriate based on the requirements of the prospective purchaser, without implementing into the system rules for designating negotiation partners that comprehensively organize combinations of products / requirements and candidate companies / departments for negotiation partners. [Means for solving the problem]

[0017] In the present invention, first, with regard to (1) above, instead of implementing workflow routing rules that comprehensively organize business regulations that define the procedure for business cases into a system, examples of business cases that are normative and in which destinations are selected in accordance with the business regulations or the ideas behind the business regulations for such routing (hereinafter referred to as "exemplary cases") are converted into data and machine-learned. The machine learning is used to express the relationship between the characteristics of the business case and the destinations through abstract numerical processing. Then, for new cases, a mechanism is constructed in which the system automatically evaluates and calculates the relationship to recommend destinations that are considered appropriate under the business regulations.

[0018] In addition, based on the same method as the solution for (1) above, for (2) above, instead of implementing a system with rules for designating a negotiation partner that comprehensively organizes combinations of products / requirements and candidate companies / departments, the system digitizes examples of selection of appropriate negotiation partners when negotiating business transactions and uses machine learning to learn them. Machine learning is used to express the relationship between the characteristics of the products handled and the requirements of the prospective buyer and the negotiation partners through abstract numerical processing. Then, for new products and requirements from the prospective buyer, a mechanism is constructed in which the system automatically evaluates and calculates the relationship to recommend a negotiation partner that is considered appropriate for the prospective buyer's requirements.

[0019] A recommendation device according to an embodiment for solving the above problem includes at least a master encoding unit, a business rule example data input unit, an encoding unit, a learning unit, a prediction unit, and a decoding unit.

[0020] In the preparation stage, The master encoding unit generates rules for converting a company's commercial transaction information and various master data for carrying out related procedures into numerical vectors that can be processed using machine learning, and registers the rules in the encoded master.

[0021] The business rule example data input unit receives instructions to register or delete case information for commercial transactions and examples of how to proceed with them, which have been created by the company's business regulation management department, as example data, and reflects the data registration and deletion operations in the business rule example data master.

[0022] The encoding unit receives information on a transaction and its normative procedures from the business rule example data master, and converts each item constituting the information into a numerical vector based on the conversion rule obtained from the encoding master.

[0023] The learning unit matches the configuration information of the transaction item that has been numerically vectorized by the encoding unit with the configuration information of its normative procedures through machine learning training, and stores the definition information of the computational model generated by the training in a business rule prediction master.

[0024] During the operational phase, The encoding unit receives information related to new commercial transaction cases for which appropriate procedural methods are to be recommended from the case information input unit of the existing system, and converts it into a numerical vector based on the conversion rules defined by the encoding master created in the preparation stage.

[0025] The prediction unit reads definition information of a prediction model of a transaction procedure from the business rule prediction master, and calculates a prediction model using as input the configuration information of the transaction that has been converted into a numerical vector by the encoding unit.

[0026] The decoding unit consults the coding master for an inverse transformation rule corresponding to each output value calculated by the prediction unit, and by inverse transformation, retrieves various masters corresponding to the recommended contents from the numerical vector, thereby forming the recommended contents. Effect of the Invention

[0027] According to the present invention, when an applicant for a transaction inputs an outline of the transaction on a general workflow system as a support function for internal procedures for commercial transactions in a company, the recommendation device of the present invention recommends an appropriate workflow destination (a conferee or an approver) based on the content of the transaction. If the applicant specifies the workflow destination based on the content and submits an application, the internal procedures can be realized through a process that complies with various internal business regulations established by the management department, without reflecting the various internal business regulations on the workflow system.

[0028] Furthermore, on the inter-company electronic commerce platform, if an outline of a transaction and a candidate business partner with whom one wishes to negotiate are input, the recommendation device of the present invention will recommend an appropriate negotiating partner for the requirements of the transaction, enabling negotiations to begin quickly and the transaction to proceed. [Brief description of the drawings]

[0029] [Figure 1] 1 is a configuration diagram showing a recommendation system to which an embodiment of a recommendation device according to the present invention is applied. [Diagram 2] FIG. 13 is an explanatory diagram of organization / personnel master (organization master). [Diagram 3] FIG. 13 is an explanatory diagram of transaction item master data (account item master data). [Figure 4] FIG. 13 is an explanatory diagram of transaction item master data (amount level master data). [Diagram 5] FIG. 13 is an explanatory diagram of transaction item master data (item keyword master data). [Figure 6] FIG. 13 is an explanatory diagram of transaction item master data (target keyword master data). [Figure 7] FIG. 13 is an explanatory diagram of transaction item master data (consultation / approval destination master data). [Figure 8] FIG. 13 is an explanatory diagram of transaction item master data (negotiation partner master data). [Figure 9]FIG. 13 is an explanatory diagram of organization / personnel master data (employee master data). [Figure 10] FIG. 1 is an explanatory diagram of One Hot coding. [Figure 11] FIG. 1 is an illustration of the encoding of a master of nominal concepts having a hierarchical structure. [Figure 12] FIG. 13 is an illustration of the encoding of a master of quantitative concepts. [Figure 13] FIG. 11 is an explanatory diagram showing an encoding process of text information. [Figure 14] This is an explanatory diagram of the business rule learning data master used to build a predictive model of consultation organizations and roles in the selection of destinations. [Figure 15] This is an explanatory diagram of the business rule learning data master used to build a prediction model of the decision-making organization and role in the selection of the destination. [Figure 16] This is an explanatory diagram of the business rule learning data master used to build a predictive model of a negotiating partner company / organization when selecting a negotiating partner. [Figure 17] FIG. 13 is an explanatory diagram of the output from the prediction unit in the selection of a destination. [Figure 18] FIG. 13 is an explanatory diagram of the output of a prediction unit in selecting a negotiation partner. [Figure 19] 13 is a flowchart showing an encoding procedure for master data excluding a keyword master data among transaction item master data. [Figure 20] 13 is a flowchart showing a processing procedure for creating a keyword master for a transaction master from a business rule example data master. [Figure 21] 13 is a flowchart showing a processing procedure of a business rule example data input unit. [Figure 22] 13 is a flowchart showing a procedure for creating a business rule learning data master. [Diagram 23] 13 is a flowchart showing a processing procedure in a prediction unit. [Figure 24] This is an example of a screen for registering example data in the business rule example data input section for selecting a destination. [Diagram 25]13 is an example of a case information input screen for selecting a destination. [Figure 26] 13 is an example of a recommended screen in the recommended result display section for destination selection. [Figure 27] 13 is an example of a screen for entering the reason for change in the recommended result display section of the destination selection. [Figure 28] 13 is an example of a screen for registering example data in a business rule example data input section for selecting a negotiation partner. [Figure 29] 13 is an example of a screen for inputting project information in selecting a negotiation partner. [Diagram 30] 13 is an example of a recommendation screen in a recommendation result display section for selecting a negotiation partner. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0030] Configuration of the embodiment FIG. 1 is a block diagram showing the configuration of a system to which an embodiment of a recommendation device according to the present invention is applied.

[0031] The recommendation device 100 shown in FIG. 1 is a server that performs input and output to an existing system 200, and the recommendation device 100 and the existing system 200 are connected via a network.

[0032] The recommendation device 100 includes a master encoding unit 101, an encoding master 102, a business rule example data input unit 103, a business rule example data master 104, an encoding unit 105, a business rule learning data master 106, a learning unit 107, a business rule prediction master 108, a prediction unit 109, a decoding unit 110, a selection result collection unit 111, and a recommendation result history storage unit 112.

[0033] The existing system 200 is composed of approval rules 201, transaction master data 202, organization and personnel master data 203, a workflow system or commercial transaction platform 204, a transaction information input unit 205, and a recommendation result display unit 206. The transaction information input unit 205 and the recommendation result display unit 206 in the existing system 200 can also be configured to be included in the recommendation device 100.

[0034] The following is a description of the preparation stage of the recommendation device 100.

[0035] The master encoding unit 101 generates reversible conversion rules for converting transaction item master data 202 such as organizations and account items shown in Fig. 3 to Fig. 8 and organization and personnel master data 203 constituting the destination and negotiation destination shown in Fig. 2 and Fig. 9 into numerical vectors for calculating them by machine learning, and stores the generated conversion rules in the encoding master data 102. The master data for generating the conversion rules includes master data referenced in the business regulations held by the approval regulations 201.

[0036] Here, the forward transformation of reversible transformations, i.e., the conversion of codes and names of account items, organizations, etc. in master data into numerical vectors, is called encoding, and the reverse conversion from numerical vectors to names and codes is called decoding.

[0037] The encoding master 102 holds a conversion rule that is a correspondence relationship between the numerical vector encoded by the master encoding unit 101 and the information before encoding. Since this conversion is a reversible conversion, the conversion rule held by the encoding master 102 is not only a conversion rule for encoding, but also a conversion rule for decoding.

[0038] The business rule example data input unit 103 receives instructions for new data to be registered and data to be deleted for example data that has been associated with transaction item information such as organizations and account items and with destinations and negotiation parties, which are the contents of the transaction item procedures. In the example of Fig. 1, the business rule example data input unit 103 is configured to be used on the recommendation device 100. However, it may be configured to be used from an input device on the existing system 200, such as the case information input unit 205.

[0039] The business rule example data master 104 is a storage area (database) that holds the example data registered in the business rule example data input unit 103 in its original form, and registers new example data or deletes specified existing example data.

[0040] The encoding unit 105 compares the encoding master 102 with the business rule example data master 104, and converts the example data into a numerical vector. The converted example data is stored in the business rule learning data master 106.

[0041] The learning unit 107 reads out the vectorized example data from the business rule learning data master 106, and inputs information on the transaction item, such as the organization and account item, which are controlling in the selection of the destination and negotiation partner. Then, for each component of the destination and negotiation partner information in the transaction item, such as the organization and its role that will be the consultation, decision-making, or negotiation partner, machine learning is performed on the input / output relationship with the transaction item information as input. The definition information of the calculation model of the input / output relationship obtained by learning is stored in the business rule prediction master 108. The business rule prediction master 108 is dictionary-format information that holds parameters for machine learning calculations and their values.

[0042] The following is a description of the operational stages of the recommendation device 100.

[0043] When case information is input to the case information input unit 205, the encoding unit 105 obtains the encoding rule for each item from the encoding master 102 and encodes each item. The encoded data is combined into one numerical vector and input to the prediction unit 109.

[0044] The prediction unit 109 extracts a prediction model for each element (organization, role, etc.) of the destination / negotiation destination from the business rule prediction master 108, calculates the output using the numerical vector as input, and returns the output result to the decoding unit 110. The output of the prediction unit 109 in the selection of the destination is shown in FIG.

[0045] The decoding unit 110 refers to the encoding master 102 for each output result for the destination / negotiation element, acquires the code and name that are the source of encoding of the master data, forms recommended content, and displays the recommended content in the recommended result display unit 206.

[0046] The selection result collection unit 111 collects, from the recommendation result display unit 206, the recommended contents concerning the procedures for the transaction item, information on the destination / negotiation party actually selected by the user, and the reason for the change when the user modifies the recommended contents and specifies the actual destination / negotiation party, and stores them in the recommendation result history storage unit 112.

[0047] The recommendation result history storage unit 112 records and stores information on a business transaction entered by a user of a system incorporating the present invention, the recommended consultation and negotiation partners for that transaction, and the actually designated destinations and negotiation partners, in association with each other.

[0048] Processing Procedure of the Embodiment Next, the processing procedures of the recommendation device 100 according to an embodiment of the present invention will be described in the following order: (a) preparation processing of the encoded master (Figures 19 and 20), (b) preparation processing of the business rule example data (Figure 21), (c) preparation processing of the business rule learning data master (Figure 22), (d) preparation processing of the predictive model, and (e) processing during operation (Figure 23).

[0049] <(a) Preparation of the encoding master> FIG. 19 shows a processing procedure in the master encoding unit 101.

[0050] First, when various masters are input from the existing system 200 (step S1), for the master data defining the list of distribution destinations as shown in Fig. 7 and the master data defining the list of negotiation destinations as shown in Fig. 8 (step S2 True), each master value is converted into a numerical vector by One Hot Encoding (step S5). Fig. 10 shows an example of One Hot Encoding.

[0051] In cases other than the destination master data and the negotiation destination master data, the encoding process is assigned based on whether the input master data is a nominal concept (a concept intended for distinction) or a quantitative concept (a concept having an order or a magnitude relationship on the same axis or dimension) (step S3) and, if it is a nominal concept, whether the primary key of the master data has a hierarchical structure or not (step S4).

[0052] As components of a transaction, the contract type such as "basic contract" and "individual contract", whether the project is planned or not, and other classifications such as project type, classification, and flag are generally nominal concepts. Among the master data related to nominal concepts, those whose primary key does not have a hierarchical structure (step S4) are converted into a numeric vector by One Hot encoding (step S5).

[0053] Among the masters related to nominal concepts, if the primary key has a hierarchical structure or can be organized hierarchically (step S4 True), one-hot coding is performed for each piece of information indicating a hierarchy, and the results are linked (step S6). Information on transaction items that can be organized hierarchically is generally a classification system such as a corporate organization as shown in Figure 2, account items as shown in Figure 3, and item categories.

[0054] For example, as shown in Figure 11, if there are three organizations in the hierarchy immediately below organization 1, they can be expressed as 11, 12, and 13, and if there are two organizations in the hierarchy immediately below organization 11, they can be expressed as 111 and 112, so One Hot Coding is applied to this representation for each hierarchy. Even for different hierarchies, by using One Hot Coding to make the bits corresponding to common hierarchies the same value, machine learning can be used to output that reflects the characteristics shown by example data for similar hierarchies with a common structure, even for hierarchies that were not given as example data.

[0055] On the other hand, for quantitative concepts such as the amount and duration of a transaction, the master data corresponds to the level or standard for treating them as ordered nominal concepts. Figure 4 shows an example of a master data that determines the level of the settlement amount, and has the function of uniquely determining the level of the amount by judging each standard for the amount. The master data related to quantitative concepts is coded so that the continuous nature of the level can be expressed (step S7).

[0056] Specifically, as shown in Figure 12, a bit string whose length is the number of criteria (a number one less than the number of levels) is prepared, and the value is encoded into a numeric vector by setting it to 1 if it is greater than the criteria value, and 0 if it is smaller. In the case of settlement amounts in internal procedures, generally, a criteria value is set for each case or account item according to which the authority of the settler and procedures are changed, and various judgments are made by converting the settlement amount into a monetary level that indicates which criteria it meets, rather than the actual settlement amount which is broken down finely for each case.

[0057] For example, if the rule is "Purchases of equipment and facilities under the asset expense item require approval from the superior if the amount is less than 100,000 yen, approval from the division head if the amount is between 100,000 yen and 1 million yen, approval from the CFO if the amount is between 1 million yen and 10 million yen, and approval from the CEO if the amount is over 10 million yen," the approval method is determined not by the actual amount of approval, but by which of the four levels the amount falls into: "less than 100,000 yen," "100,000 yen and 1 million yen," "1 million yen and 10 million yen," and "10 million yen or more." In this example, the amount of approval is determined by four levels, so "less than 100,000 yen" is coded as 000, "100,000 yen and 1 million yen" is coded as 100, "1 million yen and 10 million yen" is coded as 110, and "10 million yen or more" is coded as 111.

[0058] The same applies to other quantitative concepts such as period and quantity; for period, the master levels are defined as "less than 3 business days," "3 to 5 business days," "5 to 10 business days," and "10 or more business days," while for quantity, the master levels are defined as "less than 10 items," "10 to 100 items," and "100 items or more," and are coded in the same manner as above.

[0059] In addition to the information in which important elements such as the organization and settlement amount are organized as items, the case information may also include text information such as the item name of the transaction subject and a description of its purpose. For the text information, a list of keywords or tags that control the transaction procedure is made into a hash table and stored in the encoding master 102.

[0060] Figure 5 shows an example of a hash table of text information related to case items, and Figure 6 shows an example of a hash table of text information related to case objectives.

[0061] If the keyword list is not prepared as a master in the transaction master data 202 or the existing system 200, the master can be generated from the example data registered in the business rule example data master 104 by following the procedure in FIG.

[0062] In the flowchart of FIG. 20, first, the necessary parts of speech, such as nouns and verbs, are specified (step S11), and text information from all example data registered in the business rule example data master 104 is obtained and input (step S12), and a keyword list for storing a list of keywords is initialized (step S13).

[0063] Next, for each example data, morphological analysis of the text information is performed to extract a list of parts of speech (step S14). If the extracted parts of speech are not included in the current keyword list (step S15), the parts of speech are added to the keyword list (step S16).

[0064] Finally, the keyword list is rearranged in ascending order to create a hash table (step S17), and the hash table is registered in the encoding master 102 (step S18).

[0065] As shown in Figure 13, to encode text information, a bit string whose length is the number of hashes in the hash table is prepared, and the text information is encoded into a vector in which the value corresponding to the keyword that appears in the text information is 1 and the other values ​​are 0.

[0066] When controlling the procedures of a transaction in detail, it is necessary to detail the corresponding business regulations and corresponding master data. If the business regulations are not defined in detail, it is highly likely that the controlling elements of the procedure are contained in the text information such as the item and purpose of the transaction, so by creating example data with text information and using machine learning, it is expected that the recommendation of the destination and negotiation partner that reflects the intention of the management department in detail can be made.

[0067] <(b) Preparation of business rule example data> 21 shows the processing procedure of the business rule example data input unit 103. First, a request is made to the business regulation management department of the company to instruct the business rule example data master 104 to perform operations such as "register", "delete", and "end" (step S21), and the instruction is acquired (step S22).

[0068] If there are any changes in the current business transaction regulations or concepts regarding the example data already registered in the business rule example data master 104, an instruction to delete the inappropriate example data is received (step S23), the target data is selected (step S24), and the data is deleted from the business rule example data master 104 (step S25).

[0069] When the management department registers new example data (step S26), input of the example data is requested (step S27), and the input example data is registered in the business rule example data master 104 (step S28).

[0070] <(c) Preparation process for business rule learning data master> FIG. 22 shows the procedure for generating data to be stored in the business rule learning data master 106. For each record input obtained from the business rule example data master 104 (step S31), it is determined whether or not the data item is text information (indicated as "text data" in the figure) (step S32). If the item is not text information, the encoding master 102 is consulted to obtain a conversion rule corresponding to the item, and the data is converted into a numerical vector (step S33). If the item is text information, morphological analysis is performed (step S34), and the conversion rule is referred to the encoding master 102 for the extracted keyword, and the data is converted into a numerical vector (step S35). Finally, the results of the conversion into numerical vectors for each item are integrated and stored in the business rule learning data master 106 (step S36).

[0071] An example of the business rule learning data master 106 for the prediction model of the negotiator or intermediate approver in the selection of the destination to which the document is to be forwarded is shown in Fig. 14. Also, an example of the business rule learning data master 106 for the prediction of the decision maker is shown in Fig. 15. Furthermore, an example of the business rule learning data master 106 for the prediction model of the negotiating organization in the selection of the negotiating partner is shown in Fig. 16.

[0072] <(d) Preparing the prediction model> The learning unit 107 trains a dedicated machine learning model for each component of the routing destination / negotiation destination recommended by the recommendation device 100.

[0073] The machine learning model uses a method capable of multi-dimensional classification and regression, such as a neural network. The term "multi-dimensional" refers to the requirement that arises from the fact that the selection of a consultation or negotiation partner is generally not a binary choice, but a selection is made from a large number of candidates.

[0074] Regarding the selection of the destination, the method of forwarding in the workflow system differs depending on the company, but in most cases, it is configured as confirmation, consultation, and approval by multiple consultants or intermediate approvers, and final approval by one final approver. Figure 17 shows an example of the machine learning output when recommending an arbitrary number of consultants and one approver as the destination. Since the specific consultants or intermediate approvers and the approver are specified by the combination of organizations and roles, the recommended candidates for the destination of the workflow are prepared as follows: a consultant prediction model that recommends any number of combinations of organizations and roles as consultants or intermediate approvers, and an approver prediction model that recommends only one combination of organization and role as the approver. For the approver prediction model, machine learning is trained as a classification model (selecting only one from multiple candidates), and for elements such as consultants and intermediate approvers that can specify any number of destinations, machine learning is trained as a regression model (predicting continuous values), and it is determined that the list of candidates whose machine learning output exceeds an appropriate judgment criterion (e.g., 0.5) will be the recommended destination. However, if it is necessary to determine the order in which the workflow is routed to intermediate approvers, etc., the order can be determined based on the recommended organizational hierarchy and role hierarchy.

[0075] Regarding the selection of a negotiation partner, the recommendation of a negotiation partner in an electronic commerce system is realized by a negotiation partner prediction model that recommends one or more pairs of candidate companies related to the case and organizations within each company that will conduct direct negotiations. Figure 18 shows an example of recommending only one from a pair of multiple companies and their organizations. When recommending multiple negotiation partners, the output value is determined in the same way as the recommendation of the negotiator in the case of the aforementioned selection of a forwarding party.

[0076] <(e) Processing during operation> FIG. 23 shows the processing procedure when the recommendation device is in operation.

[0077] First, when a user inputs information about a case for which a routing / negotiation destination is recommended through the case information input unit 205 (step S41), it is determined for each item of the case information whether the input case information is the text information described above (referred to as "text data" in the figure) (step S42). In the case of text information, the encoding master 102 is referenced to obtain a list of keywords to be extracted and a conversion rule, the text information is subjected to morphological analysis (step S43), and is converted into a numerical vector according to the conversion rule (step S44).

[0078] For items that are not text information, the corresponding conversion rule is referenced in the encoding master 102, and each item data is encoded (step S45). After each item is converted using the corresponding conversion rule, the data are concatenated into one numerical vector (step S46).

[0079] The numerical vector has the same form as the input portion of the data held by the business rule learning data master 106 (Figs. 14, 15, and 16). Using the numerical vector as a common input, definition information of the corresponding prediction model is obtained from the business rule prediction master 108 for each element of the prediction task of the destination / negotiation partner, such as organization or role, and the prediction unit 109 calculates the output value of the prediction model (step S47). For the output value, a conversion rule corresponding to the encoding master 102 is referenced, and the master value recommended by the prediction model is obtained by decoding (step S48), and output as the recommended content (step S49).

[0080] Regarding the selection of the destination, when an actual workflow destination is generated in the existing system 200, the employee master shown in FIG. 9 is referenced with the information on the recommended content, and a specific destination is determined, including the employee.

[0081] Example 1: Selection of destination As an embodiment of the present invention for a workflow system in a company, (1) an example of preparation of a recommendation device in selecting a destination for delivery, and (2) an example of operation of the recommendation device in selecting a destination for delivery will be described in that order.

[0082] <(1) Example of preparation of recommendation device for selecting a destination> First, various master data related to the procedures of a transaction are obtained from an existing system 200 and input to a master encoding unit 101 to prepare an encoded master 102.

[0083] The set of master data required for encoding is specified from the contents of the approval rules 201, etc. For example, if the approval method is determined in the approval rules document with the item category of the transaction as a condition, the item category is controlling in the workflow distribution, so the master data of the item category is the target of encoding.

[0084] Next, a plurality of example data are created and input to the business rule example data input unit 103. Specifically, contents such as "When the general affairs department purchases a headset for 3,000 yen for use in telephone conferences under the expense item of OA equipment, the general affairs department manager shall approve the purchase after consultation with the materials procurement department" (FIG. 24) and "When the information systems department purchases a memory chip for 100,000 yen to improve the performance of development computers, the information systems department manager shall approve the purchase after consultation with the technology department" are entered for each item in the input field determined by the business rule example data input unit 103, and registration is proceeded. The example data may be created by referring to the encoding master 102 or the transaction matter masters 202 and organization / personnel masters 203 that are the source of the data, and the specific implementation method of the creation and management of the data may be determined according to the requirements of the application destination of the present invention. For example, when preparing an interface for creating and registering example data as shown in Fig. 24, the interface may be made to execute the processing procedure of the business rule example data input unit (Fig. 21), and an interface for issuing instructions for creating and registering example data may be provided in an input device on the existing system 200. Alternatively, a set of example data to be registered may be created on the existing system 200, and the processing procedure of the business rule example data input unit 103 (Fig. 21) may be called to transmit the data file, thereby registering the example data all at once.

[0085] After a plurality of examples are registered, each data of the business rule example data master 104 is converted into a numerical vector by the encoding unit 105 and stored in the business rule learning data master 106 .

[0086] The learning unit 107 acquires data from the business rule learning data master, generates prediction models for each of the coordinators and approvers, which are elements to which the workflow is to be routed, by machine learning, and stores the models in the business rule prediction master 108.

[0087] <(2) Example of the operation of the recommendation device in selecting a destination> As an example of a transaction for which a workflow destination is to be recommended, a case where "an employee of the general affairs department purchases a display for a PC to be used for clerical work for 25,000 yen" will be described.

[0088] First, when logging in to a workflow system to which an embodiment of the present invention is applied, the application organization "General Affairs Department" is automatically entered from the organization to which the logged-in user belongs, as shown in the screen configuration of Fig. 25. The user selects "OA office supplies" as the account title from the list and enters "25,000 yen" as the amount. In addition, the user enters on the screen as case information the item "24-inch LCD display" and the purpose "to improve the efficiency of clerical work within the department" and executes the recommendation.

[0089] As shown in Figure 26, the recommended results are presented by machine learning, which indicates appropriate organizations and roles of decision makers and organizations and roles that should participate as conferees. However, system users can modify the recommended results and select the destination to which the transaction will actually proceed.

[0090] If the user makes a selection different from the recommended content, the user inputs the reason for the selection into the system. The information on the destination before and after the correction and the reason for the selection are acquired by the selection result collection unit 111 from the recommendation result display unit 206 and sent to the recommendation result history storage unit 112, where they are stored as reference data for checking the compliance with rules and correcting the example.

[0091] Fig. 27 shows an example of entering a reason for modifying the recommended content on the input screen. The example in Fig. 27 is an example of entering a reason for changing the recommended workflow destination.

[0092] Example 2: Selection of a negotiating partner As an embodiment of the present invention for a negotiation partner selection function in an electronic commerce system between companies deployed on a website or the like, we will explain (1) an example of preparation of a recommendation device in selecting a negotiation partner, and (2) an example of operation of the recommendation device in selecting a negotiation partner.

[0093] <(1) Example of preparation of recommendation device for selecting a negotiation partner> Preparation of the encoded master 102, creation of example data, and input to the business rule example data input unit 103 are the same as in the first embodiment. A specific example of the example data is "If you want to negotiate with Company A about a replacement hard disk drive for the purpose of increasing the capacity of your PC at a low cost, the product planning department of Company A is the place to negotiate" (FIG. 28). Subsequent processing is also the same as in the first embodiment.

[0094] <(2) Example of the operation of the recommendation device in selecting a negotiation partner> As an example of a business transaction for which you would like to have a negotiating partner recommended, let us explain the case where you say, "We urgently need to procure computer memory to upgrade the performance of the PCs used in our company, and we would like to negotiate a delivery date with Company A or Company B."

[0095] First, an interface for inputting case information is started from a menu on an electronic commerce system to which an embodiment of the present invention is applied, and the user inputs items such as "Company Z's 16GB computer memory" and purpose "We would like to urgently improve the performance of the company's PC terminals in conjunction with the launch of a research organization." In addition, "Company A" and "Company B" are input as negotiation partners, and "delivery date" is input as a negotiation item, and a recommendation is executed (FIG. 29). However, since the companies that are candidates for negotiation partners may be limited based on the user's company's customer list and business regulations, the system automatically inputs the candidate companies without the user specifying them, or the candidate companies do not need to be specified at all.

[0096] In the recommendation result, the company and department to be the negotiation partner are presented by machine learning (FIG. 30). In addition, in the example of FIG. 30, only one department of the candidate companies A and B is recommended, but it is also possible to recommend a negotiation partner for each company. The rest is the same as in the first embodiment.

[0097] As described above, according to an embodiment of the present invention, it is possible to provide a recommendation device that constructs a mechanism for having a system recommend a routing destination that is considered appropriate in accordance with business regulations, without having to implement workflow routing rules that comprehensively organize business regulations that define the procedural methods for business transactions into the system.

[0098] Furthermore, according to an embodiment of the present invention, it is possible to provide a recommendation device that constructs a mechanism for having the system recommend negotiation partners that are considered appropriate based on the requirements of a prospective purchaser, without having to implement into the system rules for designating negotiation partners that comprehensively organize combinations of products / requirements and candidate companies / departments for negotiation partners.

[0099] Although the embodiment of the present invention has been described above, this is merely an example and does not limit the scope of the invention. Various omissions, substitutions, and modifications can be made to the embodiment without departing from the gist of the invention. [Explanation of symbols]

[0100] 100... recommendation device, 101... master encoding unit, 102... encoding master, 103... business rule example data input unit, 104... business rule example data master, 105... encoding unit, 106... business rule learning data master, 107... learning unit, 108... business rule prediction master, 109... prediction unit, 110... decoding unit, 111... selection result collection unit, 112... recommendation result history storage unit, 200... existing system, 201... approval regulations, 202... transaction case master, 203... organization / personnel master, 204... workflow system or business transaction platform, 205... case information input unit, 206... recommendation result display unit.

Claims

1. A master encoding unit that generates conversion rules for converting master data that defines various elements of a transaction, including organizations and account items, and master data that defines destinations and negotiation parties for the transaction procedures, into numerical vectors used in machine learning, and registers the conversion rules in an encoding master; A business rule example data input unit adds or deletes example data to or from a business rule example data master that holds normative cases as example data for the procedures of the transaction, such as the selection of a negotiator or an approver according to conditions including the account title and settlement amount of the transaction, and the selection of a department of a company with which to negotiate; an encoding unit that converts input case information into a numerical vector by referring to a conversion rule stored in the encoding master, and that stores the numerical vector in a business rule learning data master when the example data is input; A learning unit that reads data from the business rule learning data master, inputs case information, generates a prediction model of a destination and a negotiation destination by machine learning, and saves definition information of the prediction model in the business rule prediction master; A prediction unit that reads out definition information of a prediction model from the business rule prediction master, and predicts a destination or a negotiation partner by inputting a numerical vector converted by an encoding unit for a new transaction; a decoding unit that queries the output of the prediction unit against a coding master, obtains a master value before conversion, and forms a recommendation for a destination or a negotiation destination; A recommendation device comprising:

2. 2. The recommendation device according to claim 1, wherein the decoding unit recommends to a workflow system that advances internal procedures for a business transaction within a company, based on information about the business transaction, a conferee or intermediate approver who will be responsible for confirmation or consultation, a decision maker, and, if necessary, an order of consultation or approval, to which the workflow will be forwarded.

3. The recommendation device according to claim 1 , wherein the decoding unit recommends one or more companies or companies and their divisions as candidates for negotiation partners on an electronic commerce platform, based on information on a transaction item.

4. The recommendation device of claim 1, wherein the learning unit uses examples that can quickly express the appropriateness of current business operations as training data for machine learning, thereby enabling appropriate procedures for business transactions without creating and operating corresponding business regulation documents.

5. The recommendation device of claim 1, wherein the learning unit creates learning data by inputting text information including elements that have not been mastered for a transaction together with existing master information, thereby making it possible to recommend a method of procedurally processing a transaction that reflects requirements regarding the transaction procedure that cannot be fully expressed by existing business regulations.

6. The recommendation device according to claim 1, wherein the encoding unit performs machine learning to generate a numerical vector representing a hierarchical structure for master data having a hierarchical structure relating to corporate transaction cases, thereby enabling predictions based on data of a similar hierarchy for input relating to an unknown hierarchy.

7. The recommendation device according to claim 1, further comprising a selection result collection unit that collects the selection results actually made by the user when selecting a destination or negotiation partner, and, if the recommendation content or implementation content by the decoding unit is changed, collects the reason for this and stores it in a recommendation result history storage unit.

Images