Matching device and program

Abstract

To provide a matching apparatus and program for predicting, more accurately, a goodness of fit between a project and a partner that belongs to a different department.SOLUTION: A matching apparatus which predicts a goodness of fit between a partner and a project includes learning means and prediction means. Machine learning processing includes: first learning processing trained to receive, as input, a partner attribute value and a project attribute value and output a predicted goodness of fit which is closer to a ground-truth goodness of fit; and second learning processing trained to receive, as input, a partner attribute value and a project attribute value and to be used for predicting a goodness of fit in any department. The prediction means predicts a goodness of fit on the basis of information on the partner and information on the project using a trained model.SELECTED DRAWING: Figure 7

Description

[Technical Field]

[0001] This invention relates to a matching device and program. [Background technology]

[0002] Companies that conduct project-based business often temporarily bring in personnel from other companies as partners to address labor shortages or acquire technologies they do not possess themselves. In particular, highly skilled partners often play a central role in projects, and maintaining long-term collaboration with such partners, rather than just temporary partnerships, is a crucial management challenge.

[0003] To maintain long-term collaboration with partners, it's necessary to go beyond use within specific departments and collaborate across the entire company. However, cross-departmental collaboration with partners is currently not progressing. This is because the required skills and customer characteristics differ from department to department, making it difficult to determine whether a partner belonging to a different department is suitable for one's own department's project.

[0004] As a technique for predicting the degree of compatibility between partners and projects, Patent Document 1 discloses a technique for learning how to calculate the degree of compatibility from past partner-project matching results.

[0005] Specifically, a deep neural network (DNN) is used to extract features useful for calculating the degree of fit from the attribute values ​​of the partner and project, and the degree of fit between the partner and project is calculated based on these features. By training the parameters so that the degree of fit predicted by the DNN matches the actual degree of fit for partners and projects in past matching results, it becomes possible to accurately predict the degree of fit. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] Patent No. 6369053 [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] However, conventional technologies have presented a challenge in accurately predicting the degree of compatibility between partners belonging to one department and projects belonging to different departments.

[0008] To achieve this using the technology described in Patent Document 1, it is necessary to have matching records between partners belonging to different departments and projects, which are used as training data with similar characteristics to the target of prediction. However, given the current state of partner collaboration, there are few such matching records, and a sufficient amount of training data cannot be obtained. Therefore, it is difficult to accurately predict the degree of fit between partners belonging to different departments and projects using the prior art.

[0009] This invention was made to solve these problems and aims to provide a matching device and program that can more accurately predict the degree of compatibility between partners belonging to different departments and projects. [Means for solving the problem]

[0010] An example of a matching device according to the present invention is: A matching device that predicts the degree of compatibility between a partner and a project, The matching device comprises a learning means and a prediction means, The learning means generates a trained model by performing machine learning processing based on partner attribute values, project attribute values, partner correct answer category, project correct answer category, and correct answer fit. The aforementioned machine learning process is - A first learning process that takes the partner attribute value and the project attribute value as input and learns to output a predicted fit that is closer to the correct fit, - A second learning process that takes the aforementioned partner attribute values ​​and project attribute values ​​as input and learns a common prediction method that can be used to predict the degree of fit of any department, Includes, The prediction means uses the trained model to predict the degree of fit based on information about the partner and information about the project.

[0011] In one example, the first learning process is: A process to calculate partner features based on the aforementioned partner attribute values, A process for calculating project features based on the aforementioned project attribute values, A process for calculating the prediction fit based on the partner features and project features, A process to learn so that the accuracy error, which is the difference between the correct accuracy and the predicted accuracy, becomes smaller, Includes.

[0012] In one example, the second learning process is: Based on the aforementioned partner features, a process is performed to predict the partner prediction department to which the partner belongs. A process to calculate the partner division error as the error between the partner prediction division and the partner correct answer division, A process to learn so that the aforementioned partner division error becomes larger, Includes.

[0013] In one example, the second learning process is: Based on the aforementioned project features, a process is performed to predict the project forecasting department to which the project belongs. The process of calculating the project department error as the difference between the project prediction department and the project correct answer department, The process involves learning to increase the error in the aforementioned project division, Includes.

[0014] In one example, the partner feature and the project feature are represented by vectors. The aforementioned machine learning process is performed using a deep neural network.

[0015] In one example, the goodness of prediction is calculated using the squared error or cosine similarity.

[0016] In one example, the partner prediction sector is represented by a vector that indicates the likelihood of the partner belonging to each sector. The correct answer category for the partner is represented by a vector where the likelihood of the partner belonging to that category is 1, and the likelihood of the partner belonging to another category is 0. The aforementioned partner division error is calculated using the cross-entropy error function.

[0017] In one example, the machine learning process is: - A first deep neural network that takes the aforementioned partner attribute values ​​as input and outputs the aforementioned partner features, - A second deep neural network that takes the aforementioned project attribute values ​​as input and outputs the aforementioned project features, - A third deep neural network that takes the aforementioned partner features as input and outputs the partner prediction department to which the partner belongs, - A fourth deep neural network that takes the aforementioned project features as input and outputs the project prediction department to which the project belongs, It is done using In the first learning process, the first deep neural network and the second deep neural network are trained. In the second learning process, the first deep neural network, the second deep neural network, the third deep neural network, and the fourth deep neural network are trained.

[0018] In one example, in the first learning process, - Update the parameters of the first deep neural network to reduce the aforementioned goodness-of-fit error, - Update the parameters of the second deep neural network to reduce the aforementioned goodness-of-fit error, In the second learning process described above, - Update the parameters of the first deep neural network to increase the partner division error, - Update the parameters of the second deep neural network to increase the project division error, - Update the parameters of the third deep neural network to reduce the partner division error, - Update the parameters of the fourth deep neural network to reduce the project division error.

[0019] An example of a program according to the present invention causes a computer to function as the matching device described above. [Effects of the Invention]

[0020] According to the present invention, the degree of compatibility between partners belonging to different departments and a project can be predicted more accurately. [Brief explanation of the drawing]

[0021] [Figure 1] A functional block diagram of a matching device 100 according to Embodiment 1 of the present invention. [Figure 2] Hardware configuration of matching device 100. [Figure 3] A schematic flowchart illustrating the operation of the matching device 100. [Figure 4] Example configuration for partner table T1. [Figure 5] Example configuration for project table T2. [Figure 6] Example configuration of matching results table T3. [Figure 7] Overview of the learning process using learning method 101. [Figure 8] An overview of the effects of common feature learning processing. [Figure 9]A mathematical formula that explains the details of the calculation process performed by learning method 101. [Figure 10] A diagram illustrating the details of the computation process performed by learning method 101. [Figure 11] Flowchart of the operation of learning method 101. [Figure 12] Example structure of the learning result table T4. [Figure 13] Example configuration of the prediction target table T5. [Figure 14] Flowchart of the operation of the prediction means 102. [Figure 15] Example structure of prediction result table T6. [Modes for carrying out the invention]

[0022] First, the outline of the present invention will be described. This specification discloses a matching device that can accurately predict the degree of fit between partners and projects belonging to different departments by learning a common prediction method that can be used to predict the degree of fit for any department, by excluding department-specific characteristics from the matching results of partners and projects belonging to the same department, for which a sufficient amount of training data is available.

[0023] Specifically, the model is trained so that it can no longer predict which department a partner or project belongs to based on its features, and then the goodness of fit is predicted based on those features. The inability to predict the department means that partners and projects from different departments also possess the same features, which can be said to be common to all departments.

[0024] Embodiments of the present invention will be described in detail below with reference to the drawings. Below, an example is shown in which a partner procurement officer for a project selects partners to be used in that project from a list of partners used within the company, using one embodiment of the present invention.

[0025] [Embodiment 1] Figure 1 shows a functional block diagram of a matching device 100 according to Embodiment 1 of the present invention. The matching device 100 is a device that predicts the degree of compatibility between a partner and a project. A partner refers to, for example, an individual, and a project refers to, for example, a group of individuals organized within a company for a certain purpose. However, a partner does not necessarily have to be an individual within a company, and a project does not necessarily have to be a group within a company.

[0026] The matching device 100 includes a learning means 101 and a prediction means 102.

[0027] Figure 2 shows the hardware configuration of the matching device 100. The matching device 100 has a hardware configuration similar to that of a known computer and includes an input device 103, an output device 104, an arithmetic unit 105, and a storage device 106.

[0028] The input device 103 includes a keyboard, mouse, network interface, etc. The output device 104 includes a display device (such as a display), printing device (such as a printer), network interface, etc. The arithmetic unit 105 includes, for example, a processor. The storage device 106 includes, for example, storage media such as semiconductor memory devices and magnetic disk devices. The storage device 106 may include main memory and auxiliary storage.

[0029] The storage device 106 stores the partner table T1, project table T2, matching performance table T3, learning results table T4, prediction target table T5, and prediction results table T6. The contents of these tables will be described later.

[0030] The storage device 106 may store a program (not shown). The arithmetic unit 105 may execute this program, causing the computer to function as a matching device 100. In other words, this program may cause the computer to function as a matching device 100.

[0031] Figure 3 shows a schematic flowchart illustrating the operation of the matching device 100. In step S1, the matching device 100 starts processing. In step S2, the matching device 100 decides whether to operate in learning mode or prediction mode. This decision is made, for example, in response to instructions from the user, such as a procurement officer.

[0032] If the matching device 100 is operating in learning mode, in step S3, the learning means 101 of the matching device 100 learns how to calculate the degree of fit between the partner and the project based on past matching results.

[0033] When the matching device 100 operates in prediction mode, in step S4, the prediction means 102 of the matching device 100 predicts the degree of fit between the user-specified partner and the project based on the degree of fit calculation method learned in the learning process. The process in step S4 is performed using the learning result table T4, but since the learning result table T4 does not contain any data in its initial state, the matching device 100 must operate in learning mode (step S3) at least once.

[0034] After step S3 and step S4, the process ends in step S5.

[0035] <Learning process in Embodiment 1> The following provides a more detailed explanation of step S3. In the learning process of step S3, the system learns how to calculate the degree of fit between partners and projects based on past matching results.

[0036] Before executing the learning process, the user stores the necessary data in the partner table T1, project table T2, and matching performance table T3 using the input device 103.

[0037] Figure 4 shows an example of the structure of Partner Table T1. Partner Table T1 stores teacher-partner information related to a partner. Teacher-partner information is information related to a partner, and in the example in Figure 4, the key is information that identifies the partner (Partner ID), and it associates information such as the department to which the partner belongs, whether the partner has skills (C language skills and Java language skills), the partner's personality (whether they are cheerful or not, and whether they are calm or not), the partner's employment rate, and the partner's years of experience.

[0038] In this embodiment, the teacher partner information includes, as mandatory items, a partner ID and information representing the department. Other items can be modified as appropriate by those skilled in the art, and it is preferable to include various items other than those shown in the illustration.

[0039] Teacher partner information is preferably in a format suitable for learning. In this embodiment, since a DNN is used, teacher partner information is preferably numerical data that can be input into the DNN. In the example in Figure 4, for the sake of explanation, "department" is represented as text data, but in reality, departments are also represented as numerical data.

[0040] Of the information shown in Figure 4, the presence or absence of C language skills, the presence or absence of Java language skills, whether or not the person has a cheerful personality, whether or not they have a calm personality, the employment rate, and the number of years of experience represent attribute values ​​related to the partner.

[0041] Figure 5 shows an example of the structure of project table T2. Project table T2 stores teacher project information related to the project. Teacher project information is information related to the project, and in the example in Figure 5, the key is information that identifies the project (project ID), and it associates information such as the department to which the project belongs, whether the partner's skills are required for the project (C language skills and Java language skills), the characteristics of the customer related to the project (new customer or returning customer), and the personality of the employee related to the project (whether they are cheerful or not, and whether they are calm or not). Note that "employee" may include partners, or it may be defined regardless of whether they are partners or not.

[0042] In this embodiment, the teacher project information includes, as mandatory items, a project ID and information representing the department. Other items can be modified as appropriate by those skilled in the art, and it is preferable to include various items other than those shown in the illustration.

[0043] Of the information shown in Figure 5, the following represent attribute values ​​related to the project: whether or not C language skills are required, whether or not Java language skills are required, whether or not it is a new customer, whether or not it is a continuing customer, whether or not the employees have cheerful personalities, and whether or not the employees have calm personalities.

[0044] Teacher project information is preferably in a format suitable for learning. In this embodiment, since a DNN is used, teacher project information is preferably numerical data that can be input into the DNN. In the example in Figure 5, for the sake of explanation, "department" is represented as text data, but in reality, departments are also represented as numerical data.

[0045] Figure 6 shows an example of the structure of the matching performance table T3. The matching performance table T3 uses pairs of current or past projects and the partners that belonged to those projects as keys, and stores the degree of fit (correct fit) between the project and the partner for each pair. The correct fit is represented by a real number between 0 and 1, for example, with a higher value indicating a better fit between the project and the partner.

[0046] For example, the first row of Figure 6 means that the partner with partner ID "PTN0001" has belonged to the project with project ID "PRJ0001," and that the compatibility score entered by the person in charge (for example, the user of matching device 100) for these pairs was 0.9.

[0047] Figure 7 shows an overview of the learning process by the learning means 101. In this embodiment, the learning means 101 uses a DNN to learn how to calculate the goodness of fit between partners and projects from past matching results. The processing related to the DNN includes feature calculation processing, goodness of fit calculation processing, goodness of fit error calculation processing, departmental calculation processing, and departmental error calculation processing.

[0048] In Figure 7, the white squares represent scalar values, and the figure formed by three connected white squares represents a vector value.

[0049] The input to the DNN is information (attribute values) stored in partner table T1 and project table T2. Since it is difficult to predict the degree of fit by directly comparing attribute values, the feature extraction calculation process converts the information (features) into a format suitable for predicting the degree of fit based on multiple attribute values.

[0050] In the goodness-of-fit calculation process, the goodness-of-fit of a pair is calculated by comparing the features of the partner and the project. In the goodness-of-fit error calculation process, the error (goodness-of-fit error) between the goodness-of-fit predicted in the goodness-of-fit calculation process and the correct goodness-of-fit stored in the matching performance table T3 is calculated. By updating the parameters of the feature extraction calculation process to minimize this error, the learning means 101 learns how to calculate the goodness-of-fit.

[0051] The departmental calculation process and departmental error calculation process are designed to learn a common prediction method that can be used to predict the degree of fit for any department, by excluding departmental-specific characteristics from the matching history of partners and projects belonging to the same department.

[0052] The departmental calculation process and the departmental error calculation process will be collectively referred to as the "common feature learning process." In the departmental calculation process, the department to which each partner belongs is predicted by extracting departmental-specific features from the partner and project features. In the departmental error calculation process, the error (departmental error) between the department predicted in the departmental calculation process and the correct department stored in the partner table T1 or project table T2 is calculated. By updating the parameters of the feature calculation process to increase this error (so that departments cannot be predicted from the features), features that exclude departmental-specific features are learned. As a result, common features that can be used to calculate the goodness of fit of any department are learned, enabling accurate prediction of the goodness of fit of partners and projects in different departments. In addition, the parameters of the departmental calculation process are also updated sequentially to extract departmental-specific features from the features at each learning stage. Specifically, the parameters are updated to reduce the departmental error.

[0053] Figure 8 illustrates the general effect of the common feature learning process. In Figure 8, the white figures and the hatched figures represent partners belonging to different departments.

[0054] Figure 8(a) is a conceptual diagram showing the case where common feature learning is not applied. Generally, most of the paired data included in the matching performance table T3 are pairs of partners and projects belonging to the same department. As a result, features with different distributions for each department are learned, which allows for accurate prediction of the goodness of fit only for pairs within each department.

[0055] Figure 8(b) is a conceptual diagram of the case when common feature learning processing is applied. By learning the features in such a way that it is no longer possible to predict the department from the features, department-specific features are removed from the features, and the distribution of features for each department overlaps (they become similar values). In this case, in the feature space, features of partners in different departments are placed near the features of a partner in one department, so it can be said that features common to multiple departments are extracted.

[0056] The learning process by the learning means 101 is described below. The learning means 101 uses four DNNs and generates a trained model by performing machine learning processing based on partner attribute values, project attribute values, partner correct department, project correct department, and correct fit. In this embodiment, four DNNs are used, but the number of DNNs can be changed as appropriate by those skilled in the art. For example, if the partner or project attribute values ​​obtained differ for each department, a DNN for feature calculation processing may be created for each department.

[0057] The details of the computation process by the learning method 101 will be explained using the mathematical formulas shown in Figure 9 and the diagram shown in Figure 10. Below, the partner is defined as i∈S ptn Represented as j∈S, the project is j∈S prj It is represented as S. ptn S is the set of partners. prj is a set of projects. t∈{ptn,prj} are indices representing either a partner (ptn) or a project (prj).

[0058] The attribute value of partner i is x i (Represented by a vector with dimensions equal to the number of elements in the partner's attribute values), and the attribute values ​​of project j are x j It is represented as a vector with dimensions equal to the number of elements in the project's attribute values.

[0059] The learning method 101 comprises a partner feature calculation DNN201 (first deep neural network), a project feature calculation DNN202 (second deep neural network), a partner division calculation DNN203 (third deep neural network), and a project division calculation DNN204 (fourth deep neural network). Partner feature calculation DNN201 and project feature calculation DNN202 are fully connected DNNs, where n∈{1,…,N} represents the layer number and N is the number of layers. Partner division calculation DNN203 and project division calculation DNN204 are fully connected DNNs, where m∈{1,…,M} represents the layer number and M is the number of layers.

[0060] In the feature quantity calculation process, the partner feature quantity calculation DNN201 uses equations (1) and (2) to calculate the partner feature quantity y i based on the attribute value x of partner i i N . y i N is a vector with a dimension equal to the number of elements of the partner feature quantity. A ptn is the weight coefficient of the neuron connection in the partner feature quantity calculation DNN201, and A ptn 0 is a matrix of [number of elements of partner feature quantity] dimension × [number of elements of partner attribute value] dimension, and A ptn n is a matrix of [number of elements of partner feature quantity] dimension × [number of elements of partner feature quantity] dimension. a ptn (that is, a ptn 0 and a ptn n ) are the biases of each layer and are vectors with a dimension equal to the number of elements of the partner feature quantity. LeakyReLU is a leaky rectified linear unit.

[0061] By calculating equation (2) for each layer, the partner feature quantity y used for predicting the fitness as the output is calculated i N . Thus, the partner feature quantity calculation DNN201 takes the attribute value x of partner i as the input and outputs the partner feature quantity y i . i N

[0062] Similarly, using the mathematical formula obtained by changing the subscript i in equations (1) and (2) to j, the project feature quantity calculation DNN202 calculates the project feature quantity y j based on the attribute value x of project j j N . y j N is a vector with a dimension equal to the number of elements of the project feature quantity. A prj is the weight coefficient of the neuron connection in the project feature quantity calculation DNN202, and A prj 0This is a matrix of the dimensions [number of elements in project features] × [number of elements in project attribute values], and A prj n This is a matrix of dimensions [number of elements in project features] x [number of elements in project features]. prj (that is, a prj 0 and a prj n ) represents the bias of each layer and is a vector with dimensions equal to the number of elements in the project features.

[0063] By calculating equation (2) for each layer, the project features y used for predicting the goodness of fit are obtained as the output. j N This is calculated. In this way, the project feature calculation DNN202 calculates the attribute value x of project j. j The input is the project feature y j N This will be the output.

[0064] It is preferable that the number of elements in each feature and the number of layers in each DNN be configured to be changeable from the outside.

[0065] In the goodness-of-fit calculation process, the learning means 101 uses equation (3) to calculate the pair of features (y) of partner i and project j. i N ,y j N ) From this, the predicted goodness of fit for the pair is the predicted goodness of fit p ij Calculate the goodness of fit p. ij is a real number (scalar value). MSE is the squared error function, and tanh is the hyperbolic function. Prediction fit p ij The value is calculated as a value between 0 and 1, where 1 means a high goodness of fit and 0 means a low goodness of fit.

[0066] In this embodiment, the predictive fit p ij This is calculated using the squared error, but (y i N ,y j NThe calculation may be performed using the cosine similarity between ) or by using other calculation methods.

[0067] In the goodness-of-fit error calculation process, the learning means 101 uses equation (4) to calculate the predicted goodness-of-fit p ij The correct fit score^p is stored in the matching performance table T3. ij As an error, the degree of fit error e ij evl Calculate the result. Note that for notational purposes, the hat symbol ^ is placed to the left of p, but the correct placement of the hat symbol ^ is above p, as shown in equation (4). The same applies hereafter.

[0068] goodness of fit error e ij evl The degree of fit error e is a real number (scalar value). ij evl By updating the parameters of partner feature computation DNN201 and project feature computation DNN202 to reduce the predictive fit p ij The calculation method for determining this will be learned as a parameter for each DNN.

[0069] Thus, the machine learning process by the learning means 101 includes a process (first learning process) that takes partner attribute values ​​and project attribute values ​​as input and learns to output a predicted degree of fit that is closer to the correct degree of fit. More specifically, the first learning process includes the following: - Attribute value x of partner i i Based on the partner feature y i N The process of calculating - Attribute value x of project j j Based on the project features y j N The process of calculating - Partner feature y i N and project features y j N Based on this, the predictive goodness of fit p ij The process of calculating - The correct fit score^p is stored in the matching performance table T3.ij and the prediction fitness p ij The fitness error e, which is the error from ij evl A process of training the partner feature amount calculation DNN 201 and the project feature amount calculation DNN 202 so that

[0070] In this embodiment, in this first training process, the partner feature amount calculation DNN 201 and the project feature amount calculation DNN 202 are trained, but the partner department calculation DNN 203 and the project department calculation DNN 204 are not trained. (Note that in a modification example, the partner department calculation DNN 203 and the project department calculation DNN 204 may also be trained.)

[0071] In the department calculation process, the partner department calculation DNN 203 uses expressions (5) to (8) to calculate the partner prediction department q i N based on the partner feature amount y i q i is a vector of dimensions equal to the number of departments, and each element represents the probability or likelihood that partner i belongs to each department. B ptn is the weight coefficient of the neuron indirect connection of the partner department calculation DNN 203, and B ptn m is a matrix of [number of elements of partner feature amount] dimensions × [number of elements of partner feature amount] dimensions, and B ptn M+1 is a matrix of [number of departments] dimensions × [number of elements of partner feature amount] dimensions. b ptn is the bias of each layer, and in particular, b ptn m is a vector of dimensions equal to the number of elements of the partner feature amount, and b ptn M+1 is a vector of dimensions equal to the number of departments. Softmax represents the softmax function.

[0072] Thus, the partner department calculation DNN 203 takes the partner feature amount y i N as input and outputs the partner prediction department q i as output.

[0073] Similarly, using the formulas obtained by changing the subscript i to j in equations (5) to (8), the project department calculation DNN204 calculates the project feature y j N Based on the project forecast department q j Calculate q. j B is a vector with dimensions equal to the number of departments, where each element represents the probability or likelihood that project j belongs to each department. prj is the weighting coefficient for interneuronal connections in the project division calculation DNN204, and B prj m This is a matrix of dimensions [number of elements in project features] x [number of elements in project features], and B prj M+1 This is a matrix of dimensions [number of departments] x [number of elements in project features]. prj This represents the bias of each layer, especially b prj m is a vector with dimensions equal to the number of elements in the project features, and b prj M+1 This is a vector with dimensions equal to the number of departments.

[0074] Thus, the project division calculation DNN204 calculates the project features y j N The input is the project forecast department q. j This will be the output.

[0075] In the division error calculation process, the learning means 101 uses equation (9) to calculate the predicted division for partner i (partner predicted division q). i ) and the correct answer category (Partner correct answer category^q i ) The error with the partner division error e i dpt Calculate the correct answer for partner category ^q i is a one-hot vector representing the department to which partner i belongs (a vector where the probability or likelihood of partner i belonging to that department is 1, and the probability or likelihood of partner i belonging to another department is 0). CrossEntropy represents the cross-entropy error function.

[0076] Similarly, the learning tool 101 uses the formula obtained by changing the subscript i in formula (9) to j to determine the predicted department for project j (project predicted department q). j ) and the correct answer category (Project correct answer category^q j ) The error is project department error e j dpt Calculate the correct answer for the project. j This is a one-hot vector representing the department to which project j belongs (a vector where the probability or likelihood of project j belonging to that department is 1, and the probability or likelihood of project j belonging to another department is 0).

[0077] Learning method 101 is partner division error e i dpt and project department error e j dpt The parameters of Partner Feature Calculation DNN201 and Project Feature Calculation DNN202 are updated to increase the value. This learning process allows for efficient learning of only the features common to each department. Furthermore, the learning method 101 is the partner division error e i dpt and project department error e j dpt The parameters of Partner Department Computation DNN203 and Project Department Computation DNN204 are updated to reduce the value. Through this learning process, Partner Department Computation DNN203 and Project Department Computation DNN204 will be able to extract department-specific features from the features at each stage of learning.

[0078] As mentioned above, the partner feature calculation DNN201 uses a goodness-of-fit error e ij evl Learning that reduces the partner division error e i dpt It is updated in two types of learning processes: one that increases the learning process and another that increases the goodness-of-fit error e. Similarly, the project feature calculation DNN202 uses the goodness-of-fit error e. ijevl Learning that reduces the project division error e j dpt The value is updated in two types of learning processes: one that increases the value and another that increases the value. Therefore, in order to adjust the influence of learning on goodness of fit and learning on departments, the calculation of the department error is multiplied by a weight w (real number), as shown in equation (9).

[0079] Thus, the machine learning process by the learning means 101 takes partner attribute values ​​and project attribute values ​​as input and includes a process (second learning process) that makes predictions between the department to which the partner belongs and the department to which the project belongs more difficult. More specifically, the second learning process includes the following: - Partner feature y i N Based on this, Partner Forecast Department q to which Partner i belongs i Prediction process, - Partner Forecast Division q i And, Partner Correct Answer Category ^q i As an error, the partner division error e i dpt The process of calculating - Partner division error e i dpt The process of training the partner feature computation DNN201 to make it larger, - Partner division error e i dpt The process of training the partner division computation DNN203 to make it smaller, - Project features y j N Based on this, Project J belongs to Project Forecasting Department Q. j Prediction process, - Project forecasting department q j And, Project Correct Answer Category ^q j As an error, project department error e j dpt The process of calculating - Project department error e j dptThe process of training the DNN202 project feature computation is performed so that it becomes larger. - Project department error e j dpt This process trains the project-departmental computation DNN204 to make the value smaller.

[0080] In this embodiment, in this second learning process, the partner feature y i N and project features y j N Regarding the calculation method, it is possible to exclude department-specific features and learn a common prediction method that can be used to predict the degree of fit for any department. This allows for learning that enables more accurate prediction of the degree of fit of partners and projects belonging to different departments.

[0081] Figure 11 shows a flowchart of the operation of the learning means 101. This flowchart shows the details of step S3 in Figure 3.

[0082] In step S11, the learning means 101 starts processing. Next, in step S12, the learning means 101 extracts partner and project pairs and the degree of fit (correct degree of fit^p) of those pairs from the matching performance table T3. ij ) obtain.

[0083] Next, in step S13, the learning means 101 obtains the partner attribute value x from the partner table T1 and the project table T2. i and project attribute value x j Obtain the following: Also, the department to which each belongs (correct answer department) is the Partner Correct Answer Department^q i and Project Correct Answer Category ^q j These are obtained. These correct answers are then processed into one-hot vectors.

[0084] Next, in step S14, the learning means 101 initializes the parameters of each DNN. For example, the initial values ​​of all parameters may be determined using uniformly distributed random numbers between -1 and 1, but this method is not limited to this.

[0085] Next, the learning means 101 repeats the process in steps S15 to S23 a predetermined number of times. Preferably, the number of repetitions is configured to be input or changed by the user of the matching device 100. In step S15, the learning means 101 starts the repetition unit.

[0086] In step S16, the learning means 101 uses the attribute value x of the partner obtained in step S13. i and project attribute value x j Based on this, partner feature y i N and project features y j N Calculate.

[0087] Next, in step S17, the learning means 101 calculates the partner feature y calculated in step S16. i N and project features y j N Based on this, the predictive fit p for that pair (i,j) ij Calculate.

[0088] Next, in step S18, the learning means 101 uses the goodness-of-fit error calculation process to calculate the predicted goodness-of-fit p calculated in step S17. ij And the correct fit^p obtained in step S12. ij Therefore, the degree of fit error e ij evl Calculate.

[0089] Next, in step S19, the learning means 101 uses the departmental prediction process to calculate the partner feature y calculated in step S16. i N and project features y j NBased on this, Partner Forecast Department q i and project forecasting department q j Calculate.

[0090] Next, in step S20, the learning means 101 uses the departmental error calculation process to calculate the partner predicted department q calculated in step S19. i and project forecasting department q j And, the correct partner category obtained in step S13 ^q i and Project Correct Answer Category ^q j Based on this, the partner department error e i dpt and project department error e j dpt Calculate.

[0091] Next, in step S21, the learning means 101 calculates the goodness-of-fit error e calculated in step S18. ij evl And the partner division error e calculated in step S20 i dpt and project department error e j dpt Based on this, the gradients of the parameters of each DNN are calculated. The gradient calculation can be performed using, but is not limited to, the backpropagation method.

[0092] Next, in step S22, the learning means 101 updates the parameters of each DNN as follows. - Partner feature calculation DNN201: Goodness of fit error e ij evl Minimize the partner department error e i dpt Maximize - Project feature calculation DNN202: goodness-of-fit error e ij evl Minimize project department error e j dpt Maximize - Partner Division Calculation DNN203: Partner Division Error e i dpt Minimize - Project Division Calculation DNN204: Project Division Error e j dpt Minimize

[0093] Furthermore, Adam's rules can be used to calculate the specific update difference of the parameters in S22.

[0094] Next, in step S23, the learning means 101 terminates the iteration unit. The iteration unit may be such that the parameters are updated each time a partner-project pair is processed, or that the parameters are updated each time a predetermined number of partner-project pairs are processed, or that the timing of parameter updates is determined based on other criteria.

[0095] When the iterative processing of steps S15 to S23 is completed, the machine learning process is finished and each DNN is generated as a trained model. Next, in step S24, the learning means 101 stores the parameters resulting from the training performed in steps S15 to S23 in the learning result table T4. Next, in step S25, the learning means 101 terminates the process.

[0096] Figure 12 shows an example of the structure of the learning result table T4. This example shows some of the parameters that define the partner feature calculation DNN201. Parameter name "A ptn 0 For the matrix identified by , a pair of row and column numbers is defined to identify each element of the matrix, and a numerical value is stored for each pair. Other parameters of the Partner Feature Calculation DNN201, as well as other DNNs, can be expressed in a similar format.

[0097] <Prediction processing in Embodiment 1> The following describes step S4 in Figure 3 in more detail. In the prediction process of step S4, the degree of fit between the user-specified partner and the project is predicted based on the degree of fit calculation method learned in the learning process of step S3.

[0098] The user of the matching device 100 generates the prediction target table T5 using the input device 103 before step S4 is executed.

[0099] Figure 13 shows an example of the structure of the prediction target table T5. The prediction target table T5 stores pairs of partners and projects for which the degree of fit to be predicted is to be determined. Specifically, the partner ID and project ID are used as keys.

[0100] Here, if you are looking for a partner to use in a specific project, you simply enter pairs of all partners and that specific project. Similarly, if you want to find recommended projects for a specific partner, you simply enter pairs of all projects and that specific partner. Figure 13 shows a scenario where you are looking for a partner to use in project PRJ1001, and pairs of project PRJ1001 and all partners have been entered.

[0101] Figure 14 shows a flowchart of the operation of the prediction means 102. This flowchart shows the details of step S4 in Figure 3.

[0102] In step S31, the prediction means 102 starts processing. Next, in step S32, the prediction means 102 retrieves the partner and project pairs to be predicted from the prediction target table T5.

[0103] Next, in step S33, the prediction means 102 obtains attribute values ​​for the partners and projects acquired in step S32. The attribute values ​​can be obtained, for example, from the partner table T1 and the project table T2, but may also be obtained from other sources (i.e., they may be different from the teacher partner information and teacher project information).

[0104] Next, in step S34, the prediction means 102 retrieves the parameters of each DNN from the learning result table T4 and sets them in each DNN, thereby reproducing each DNN as a trained model.

[0105] Next, in step S35, the prediction means 102 uses a feature calculation process to calculate partner features and project features from the partner attribute values ​​and project attribute values ​​obtained in step S33. This calculation is performed by the partner feature calculation DNN201 and the project feature calculation DNN202.

[0106] Next, in step S36, the prediction means 102 uses a goodness-of-fit calculation process to calculate the goodness-of-fit of the pair from the partner features and project features calculated in step S35.

[0107] Next, in step S37, the prediction means 102 writes the goodness of fit calculated in step S36 to the prediction result table T6. Then, in step S38, the prediction means 102 terminates the process.

[0108] Figure 15 shows an example of the structure of the prediction results table T6. The prediction results table T6 stores all pairs obtained in step S32 as keys, and for each pair, it stores the goodness of fit calculated in step S36 in association with them. For example, the first row of Figure 15 shows that the goodness of fit between partner PTN0001 and project PRJ1001 is 0.234.

[0109] A user of the matching device 100 (for example, a procurement officer) can select a partner to use for a project based on the degree of compatibility between the partner and the project stored in the prediction results table T6. For example, for each project, the partner with the highest degree of compatibility can be selected.

[0110] As mentioned above, the parameters of each DNN learn a common prediction method that is independent of departments. Therefore, even when partners and projects belonging to different departments are input, their degree of fit can be predicted more accurately. [Explanation of Symbols]

[0111] 100... Matching device 101...Learning methods 102… Prediction methods 103...Input device 104…Output device 105...Arithmetic device 106...Storage device T1…Partner Table T2...Project Table T3... Matching Performance Table T4...Learning result table T5…Prediction target table T6…Prediction result table 201... Partner Feature Calculation DNN (First Deep Neural Network) 202...Project Feature Calculation DNN (Second Deep Neural Network) 203... Partner Division Computational DNN (Third Deep Neural Network) 204...Project Division Computational DNN (Fourth Deep Neural Network)

Claims

1. A matching device that predicts the degree of compatibility between a partner and a project, The matching device comprises a learning means and a prediction means, The learning means generates a trained model by performing machine learning processing based on partner attribute values, project attribute values, partner correct answer category, project correct answer category, and correct answer fit. The aforementioned machine learning process is - A first learning process that takes the partner attribute value and the project attribute value as input and learns to output a predicted fit that is closer to the correct fit, - A second learning process that takes the aforementioned partner attribute values ​​and project attribute values ​​as input and learns a common prediction method that can be used to predict the degree of fit of any department, Includes, The prediction means uses the trained model to predict the degree of fit based on information about the partner and information about the project. The first learning process described above is: A process to calculate partner features based on the aforementioned partner attribute values, A process for calculating project features based on the aforementioned project attribute values, A process for calculating the prediction fit based on the partner features and project features, A process to learn so that the accuracy error, which is the difference between the correct accuracy and the predicted accuracy, becomes smaller, Includes, The second learning process described above is: Based on the aforementioned partner features, a process is performed to predict the partner prediction department to which the partner belongs. A process to calculate the partner division error as the error between the partner prediction division and the partner correct answer division, A process to learn so that the aforementioned partner division error becomes larger, Includes, The second learning process described above is: Based on the aforementioned project features, a process is performed to predict the project forecasting department to which the project belongs. The process of calculating the project department error as the difference between the project prediction department and the project correct answer department, The process involves learning to increase the error in the aforementioned project division, including, Matching device.

2. A matching device according to claim 1, The aforementioned partner features and project features are represented by vectors. The aforementioned machine learning process is performed using a deep neural network. Matching device.

3. A matching device according to claim 1, The aforementioned goodness of prediction is calculated using the squared error or cosine similarity. Matching device.

4. A matching device according to claim 1, The aforementioned partner prediction sector is represented by a vector that indicates the likelihood of the partner belonging to each sector. The correct answer category for the partner is represented by a vector where the likelihood of the partner belonging to that category is 1, and the likelihood of the partner belonging to another category is 0. The aforementioned partner division error is calculated using the cross-entropy error function. Matching device.

5. A matching device according to claim 2, The aforementioned machine learning process is - A first deep neural network that takes the aforementioned partner attribute values ​​as input and outputs the aforementioned partner features, - A second deep neural network that takes the aforementioned project attribute values ​​as input and outputs the aforementioned project features, - A third deep neural network that takes the aforementioned partner features as input and outputs the partner prediction department to which the partner belongs, - A fourth deep neural network that takes the aforementioned project features as input and outputs the project prediction department to which the project belongs, It is done using In the first learning process, the first deep neural network and the second deep neural network are trained. In the second learning process, the first deep neural network, the second deep neural network, the third deep neural network, and the fourth deep neural network are trained. Matching device.

6. A matching device according to claim 5, In the first learning process described above, - Update the parameters of the first deep neural network to reduce the aforementioned goodness-of-fit error, - The parameters of the second deep neural network are updated to reduce the aforementioned goodness-of-fit error, In the second learning process described above, - Update the parameters of the first deep neural network to increase the partner division error, - Update the parameters of the second deep neural network to increase the error in the aforementioned project division, - Update the parameters of the third deep neural network to reduce the aforementioned partner division error, - Update the parameters of the fourth deep neural network to reduce the project division error. Matching device.

7. A program that causes a computer to function as a matching device according to any one of claims 1 to 6.

Images