JP5116719B2 - Matching success calculation device and method - Google Patents

Description

  The present invention relates to a matching success calculation device and method.

  Conventionally, a good affinity member introduction device that extracts a combination of good affinity members and introduces them to the members has been proposed (for example, Patent Document 1). When a member inputs his / her own personal information, desired conditions, and an answer to a questionnaire, the good friendship member introducing apparatus proposed in Patent Document 1 determines the personality category of the member based on the answer to the questionnaire. Then, the affiliation member introduction device extracts a plurality of opposite sex members belonging to the personality category that is compatible with the personality category of the member with reference to the compatibility table stored in advance, and further, based on the desired condition of the member Thus, a candidate of the opposite sex is determined from the plurality of opposite sex members.

JP 2003-30306 A

  However, the affinity member introduction device proposed in Patent Document 1 only extracts members of the opposite sex based on the personality category of the opposite sex called the compatibility table, and selects candidates according to the desired condition from the members. Absent. For this reason, it is not shown that a past introduction result, that is, a matching result is learned by the apparatus and a candidate is selected based on the learning result.

  An object of this invention is to provide the apparatus or method which learns a matching result and can select a candidate based on a learning result.

  (1) A matching success degree calculation device for calculating a matching success degree of an arbitrary user, which is individual data of a user who performs matching, and quantifies profile information stored in a predetermined database in advance. As a vector element, a feature vector generating means for generating a user feature vector, a combined feature vector generating means for generating a combined feature vector based on each of the user feature vectors, and the combined feature vector, , Based on the matching storage unit that stores the past matching success history of each user of the synthetic feature vector in association with each other, the synthetic feature vector stored in the matching storage unit, and the past successful matching history Learning means to generate a learning model by generating a teacher vector, and any two people Matching success rate of the two arbitrary users based on the input receiving means for receiving the input of the user's synthetic feature vector, the teacher vector, and the input synthetic feature vector of the two arbitrary users A matching success degree calculating device comprising: a matching success degree calculating means for calculating

  According to the matching success degree calculation device described in (1), the feature vector generation means digitizes each piece of profile data stored in a predetermined database in advance for each piece of user data to be matched. By using vector elements, user feature vectors are generated, and a composite feature vector is generated based on individual user feature vectors by a composite feature vector generation means. Further, the matching storage means stores the combined feature vector and the past matching success history of each user of the combined feature vector in association with each other. Further, the learning unit generates a learning model by generating a teacher vector based on the synthetic feature vector stored in the matching storage unit and the past successful matching history, and the input receiving unit generates an arbitrary 2 Accepts input of synthetic feature vectors of human users. Further, the matching success degree calculating means calculates the matching success degree of any two users based on the teacher vector and the input composite feature vector of any two users.

  By doing in this way, the matching success degree calculation apparatus described in (1) can calculate the matching success degree of a combination of any two users based on the generated learning model. For this reason, in the present invention, by preparing a plurality of combinations of arbitrary two users and calculating the matching success level, it is possible to select matching candidates by comparing the matching success levels.

  (2) The synthetic feature vector generation means generates the synthetic feature vector by performing a predetermined operation on each element of the feature vectors of two users, and the matching success degree according to (1) Calculation device.

  According to the matching success rate calculation device described in (2), a composite feature vector is generated by performing a predetermined calculation for each element of the feature vectors of the two users. Similarities can be reflected in the composite feature vector.

  (3) The value of the inner product of the teacher vector and the combined feature vector of the two users who performed matching with which a relatively good result was obtained gives a relatively good result with the teacher vector. (2) The matching success degree calculation device according to (2), wherein the matching success degree calculation device has a higher value than an inner product value of a combined feature vector of two users who have not been matched.

  According to the matching success degree calculating device described in (3), the teacher vector has a higher inner product value with the combined feature vector of the two users who performed matching with relatively good results. Is generated. Therefore, when the inner product value of the teacher vector and the combined feature vector of the two users is relatively high, the combination of the two users corresponding to the relatively high inner product value is a relatively good result. Can be expected.

  (4) The input receiving unit receives two or more combined feature vectors generated from a feature vector of one user and each feature vector of a plurality of users who are candidates for the one user, and the matching success degree The calculating means calculates an inner product value of each of the two or more synthetic feature vectors and the teacher vector, and as a result of the calculation, a user of the synthetic feature vector having a high inner product value is successfully matched with the one user. The matching success degree calculating device according to (1), wherein the degree is high.

  According to the matching success degree calculation device described in (4), two or more generated by the input receiving unit from the feature vector of one user and the feature vectors of a plurality of users who are candidates for the one user. And a matching success degree calculating means calculates inner product values of each of the two or more synthetic feature vectors and the teacher vector, and as a result of calculation, a user of the synthetic feature vector having a high inner product value is obtained. Assume that the matching success rate with one user is high. By doing in this way, about the some user who becomes a candidate of one user, the user whose matching success degree with this one user is relatively high can be confirmed.

  (5) A matching success degree calculation device for calculating a matching success degree of an arbitrary group, which is individual data of users belonging to a group to be matched, and quantifies profile information stored in a predetermined database in advance. By combining each numerical value as a vector element, a feature vector generating means for generating a user feature vector and an individual user feature vector belonging to the matching group are combined to generate a composite feature vector. A combined feature vector generating unit, a combined feature vector, a matching storage unit that stores past matching success history of each user of the combined feature vector in association with each other, and a combined feature vector stored in the matching storage unit And the past matching success history based on the teacher vector. Learning means for generating a learning model by generating a rule, selection receiving means for receiving selection of any two groups, the teacher vector, and a combined feature vector of a plurality of users belonging to the two groups for which selection has been received A matching success degree calculating device comprising: matching success degree calculating means for calculating a matching success degree between the two arbitrary groups based on the above.

  According to the matching success degree calculating device described in (5), the matching success degree of any two groups can be calculated by the matching success degree calculating means. By doing in this way, a group with a relatively high matching success degree can be confirmed by calculating the matching success degree of other several groups with respect to a certain group, and comparing the said matching success degree.

  (6) The matching success degree calculation means calculates the matching success degree of each of the plurality of users belonging to the second group with respect to each of the plurality of users belonging to the first group. And a composite feature vector of a plurality of users belonging to the first group and a plurality of users belonging to the second group, and calculating an average value of the matching success degrees in the first group and The matching success degree calculation device according to (5), wherein the matching success degree of the second group is calculated.

  According to the matching success degree calculating device described in (6), the matching success degree calculating unit calculates the matching success degree of the group based on the average value of the matching success degrees of the users belonging to the group. For this reason, it is possible to calculate the group matching success rate by equally evaluating each user belonging to the group.

  (7) A matching success degree calculation method for calculating a matching success degree of an arbitrary user, which is individual data of a user who performs matching, digitized in advance in profile information stored in a predetermined database, As a vector element, generating a user feature vector, generating a composite feature vector based on each individual user feature vector, each of the composite feature vector, and each of the composite feature vectors Generating a teacher vector based on the step of storing the past matching success history of the user in association with the synthetic feature vector stored in the storing step and the past matching success history The step of generating a learning model and the input of a composite feature vector of any two users And a step of calculating a matching success level of the two arbitrary users based on the teacher vector and a composite feature vector of the two arbitrary users that have been input. Degree calculation method.

  According to such a configuration, the same effect as in (1) can be expected by performing the method.

  ADVANTAGE OF THE INVENTION According to this invention, the matching result can be learned and the apparatus or method which can select a candidate based on a learning result can be provided.

It is a figure showing the whole matching success degree calculation device 1 composition concerning this embodiment. It is a figure which shows the profile table 21a which concerns on this embodiment. It is a figure which shows the feature vector table 21b which concerns on this embodiment. It is a figure which shows the performance table 21c which concerns on this embodiment. It is a figure which shows the input list table 22a which concerns on this embodiment. It is a figure which shows the model table 22b which concerns on this embodiment. It is a figure which shows the code | cord | chord which calculates the weight w which concerns on this embodiment. It is a figure which shows the hardware constitutions of the matching success calculation apparatus 1 which concerns on this embodiment. It is a flowchart which shows the flow of a process after the matching success calculation apparatus 1 which concerns on this embodiment produces | generates a user's feature vector until it produces | generates a learning model. It is a flowchart which shows the flow of a process in the learning part 13 of the matching success calculation apparatus 1 which concerns on this embodiment. It is a flowchart which shows the flow of a process in the matching success degree calculation part 15 of the matching success degree calculation apparatus 1 which concerns on this embodiment.

  Hereinafter, a preferred embodiment of the matching success degree calculation apparatus of the present invention will be described with reference to the drawings.

[Entire system configuration]
FIG. 1 is a diagram illustrating an overall configuration of a matching success degree calculation apparatus 1 according to the present embodiment.

  The matching success level calculation device 1 is a device that calculates the matching success level of any two users who use the matching service. The matching success degree calculation device 1 includes a control unit 10 and a storage unit 20.

  The storage unit 20 includes various programs (not shown) for functioning as the matching success degree calculation device 1, a program (not shown) for executing the functions of the present invention, a user database (hereinafter referred to as DB) 21 and a learning DB 22. Remember.

  The user DB 21 stores information related to users who use the matching service. The user DB 21 includes a profile table 21a, a feature vector table 21b, and a performance table 21c.

  FIG. 2 is a diagram showing a profile table 21a according to the present embodiment. The profile table 21a stores profile information of users who use the matching service, and indicates a “user ID” field that uniquely identifies the user, a “user name” field that indicates the first and last name of the user, and the age of the user. An “age” field, a “gender” field indicating the user's gender, and a “marital calendar” field indicating the user's marriage calendar are included. The profile table 21a is referred to when a user feature vector is generated by a later-described feature vector generating unit 11 and when a matching success degree of any two users is calculated by a later-described matching success degree calculating unit 15. .

  FIG. 3 is a diagram showing the feature vector table 21b according to the present embodiment. A feature vector is stored in the feature vector table 21b. The feature vector is a vector element obtained by digitizing user profile information stored in the profile table 21a. The feature vector table 21b includes a “user ID” field indicating a user ID and a “feature vector” field indicating a user feature vector. The feature vectors generated by the feature vector generator 11 are stored in the feature vector table 21b. The feature vector table 21b is referred to when an input list is generated by the input list generation unit 12 described later.

  FIG. 4 is a diagram showing a result table 21c according to the present embodiment. The result table 21c stores information indicating the results of users who have used the matching service, that is, past matching success histories. The actual result table 21c includes a “user ID 1” field indicating the user ID of the first user, a “user ID 2” field indicating the user ID of the second user, A matching success history by the combination of the user and the second user, that is, a “success level” field indicating the success level is included. The performance table 21c is referred to when the input list generation unit 12 generates an input list.

  Here, in the present embodiment, the success level is set to one of three values depending on the success level. If the matching is successful, the success level is set to 3. If the matching fails midway, the success level is set to 2. If the matching fails first, the success level is set to 1. Is done. That is, when the success is more successful, the success value is set higher. In the present embodiment, one of the three types of values is set depending on the degree of success. However, the present invention is not limited to this. For example, it may be set differently from three kinds of values such as five kinds of values and ten kinds of values.

  For example, it is assumed that the matching is a so-called match, and the users “B”, “C”, and “D” are selected as candidates for the user “A”. When the user “B” accepts the matching with the user “A” and the match is established by engagement or the like, the success degree of the users “A” and “B” is set to 3. In addition, when the user “C” has accepted the match with the user “A” once, but the match is not established due to the withdrawal or the like, the success rate of the users “A” and “C” is 2 Is set. Further, when the user “D” declines without accepting the match with the user “A”, 1 is set to the success degrees of the users “A” and “D”.

  Returning to FIG. 1, the learning DB 22 stores input / output information of the learning unit 13 described later. The learning DB 22 includes an input list table 22a and a model table 22b.

  FIG. 5 is a diagram showing the input list table 22a according to the present embodiment. In the input list table 22a, as input information to the learning unit 13, a composite feature vector and past matching success history of each user of the composite feature vector are stored in association with each other. The input list table 22a includes a “serial ID” field for uniquely identifying a synthesized feature vector obtained by synthesizing the feature vectors of two users, a “synthesized feature vector” field indicating a synthesized feature vector, A success history of the user, that is, a “success level” field indicating the success level is included. The input list table 22a stores the input list generated by the input list generation unit 12. Further, the input list table 22a is referred to when the learning unit 13 generates a learning model.

  FIG. 6 is a diagram showing a model table 22b according to the present embodiment. The model table 22b stores the learning model generated by the learning unit 13, and includes a “learning model” field indicating the learning model. The model table 22b stores a model generated by the learning unit 13. The model table 22b is referred to when the matching success level calculation unit 15 calculates the matching success level.

  Returning to FIG. 1, the control unit 10 is a part that comprehensively controls each function related to the matching success degree calculating device 1. As each function according to the present invention, a feature vector generation unit 11 and an input list generation unit 12. A learning unit 13, an input receiving unit 14, and a matching success degree calculating unit 15.

  The feature vector generation unit 11 refers to the user profile information stored in the profile table 21a, generates a user feature vector, and stores the generated feature vector in the feature vector table 21b. The feature vector uses each field constituting the profile table 21a as a vector component. The value of each vector component is a value between 0 and 1 for normalization. For this reason, the feature vector is represented by a sequence of numerical values between 0 and 1.

  The input list generation unit 12 generates an input list for input to the learning unit 13 and stores the input list in the input list table 22a. More specifically, referring to the result table 21c, a combination of two users who have matched in the past is extracted. Then, the feature vectors of the first user and the second user constituting the extracted combination are combined, that is, a composite feature vector is generated based on the feature vectors of the first user and the second user. Then, a serial ID is assigned to the synthetic feature vector in order to uniquely identify the synthetic feature vector, and an input list is generated by associating the success levels of the two users stored in the result table 21c. . The generated input list is stored in the input list table 22a. Here, as a method for calculating the composite feature vector, a difference between the feature vectors of the first user and the second user is taken. By doing in this way, the difference and similarity of two users can be reflected in a synthetic feature vector.

  The learning unit 13 generates a learning model by generating a teacher vector based on the input list stored in the input list table 22a, that is, the synthetic feature vector and the past matching success history (success). To do. Subsequently, the learning unit 13 stores the generated learning model in the model table 22b.

  More specifically, the learning unit 13 refers to the input list stored in the input list table 22a and generates a set of synthetic feature vectors for each success level. The learning unit 13 then defines two sets R and N from the set of synthetic feature vectors for each success level. The definitions of the two sets R and N are performed for all combinations of success levels. That is, in this embodiment, since the success level is set to one of 1, 2, and 3, the combination of the set R and the set N is a success level 3 and a success level 2, a success level 3 and a success level 1, and a success. Three are defined: degree 2 and degree of success 1. Further, it is assumed that the success level corresponding to the set R is not smaller than the success level corresponding to the set N.

Subsequently, the learning unit 13 extracts one synthetic feature vector from each of the sets R and N, and generates (d _nq , d _rq ) as a combination S of two synthetic feature vectors. For example, it is assumed that a set of synthetic feature vectors having a success level of 3 is R, and the set R is composed of synthetic feature vectors d1a and d1b. Assume that a set of synthetic feature vectors having a success level of 2 is N, and the set N is composed of synthetic feature vectors d2a and d2b. In this case, four combinations (d2a, d1a), (d2b, d1a), (d2a, d1b), and (d2b, d1b) are generated as the combination S of the two composite feature vectors.

Subsequently, the learning unit 13 generates a weight w that is a teacher vector as a learning model. The weight w is a weight w and matching in which the inner product value of the combined feature vectors of the two users who performed matching with which a relatively good result was obtained did not obtain the weight w and a relatively good result. This is a vector having a higher value than the value of the inner product of the combined feature vectors of the two users who performed the above. The weight w is generated by the learning unit 13 performing an operation using the code shown in FIG. In the code shown in FIG. 7, T is the number of combinations of the sets R and N, and Score (d _nq , w ⁱ ) is the value of the inner product of d _nq and w ⁱ . In the code shown in FIG. 7, the value of the weight w is adjusted so that the higher the value of the inner product of the weight w that is the learning model and the comparison target vector, the higher the degree of success. This adjustment is performed when the value of the inner product of the vectors included in the set with the weight w and low success is higher than the value of the inner product of the vectors included in the set with the weight w and high success. For this reason, in the vector component of the weight w, the vector component that tends to affect the high success level is weighted with respect to the other components.

That is, the learning unit 13 compares the inner product value of the combined feature vector _dnq and the weight w constituting the set N with the inner product value of the combined feature vector d _rq and the weight w forming the set R. When the value of the inner product of the composite feature vector and the weight w constituting the set N is equal to or greater than the value of the inner product of the composite feature vector and the weight w constituting the set R, the learning unit 13 The weight w is updated by adding the number obtained by multiplying the difference between d _rq and d _nq by η _q which is the reciprocal of the number of all combinations. Then, the learning unit 13 calculates the weight w by performing this calculation for all combinations. Then, the learning unit 13 stores the calculated weight w as a learning model in the model table 22b.

  By doing in this way, the weight w which is a learning model is generated so that the value of the inner product with the matching which has obtained a relatively good result based on the past result becomes high. Therefore, when the inner product value of the combined feature vector of the weight w and the combination of the two new users is relatively high, the combination of the new two users can obtain a relatively good result. Can be expected.

  The input receiving unit 14 receives the input of the composite feature vector of any two users. More specifically, the input reception unit 14 displays a combination of any two users on a display unit described later, and accepts the selection of the combination of any two users. When the selection is accepted, the input accepting unit 14 generates a synthetic feature vector based on the profile information of the users constituting the selected combination, and delivers it to the matching success degree calculating unit 15.

  In the present embodiment, the input receiving unit 14 receives the input of the composite feature vector of any two users, but is not limited thereto. For example, the input reception unit 14 receives arbitrary two user names and user IDs. And the input reception part 14 produces | generates the synthetic | combination feature vector of arbitrary 2 persons according to having received arbitrary 2 user names and user ID, and passes them to the matching success degree calculation part 15. Also good.

  Based on the learning model (teacher vector) stored in the model table 22b and the two combined feature vectors input by the input receiving unit 14, the matching success degree calculation unit 15 The matching success level of the user is calculated. Specifically, when the matching success level calculation unit 15 receives an arbitrary two person's composite feature vector from the input receiving unit 14, the arbitrary two person's composite feature vector input by the input receiving unit 14, and The value of the inner product with the learning model (teacher vector) stored in the model table 22b is calculated. Then, the calculated result is output as a matching success level to the display unit. As described above, the matching success degree calculation unit 15 calculates the matching success degree of any two users. For this reason, when the composite feature vector of any two users is received a plurality of times by the input receiving unit 14, a plurality of matching successes are calculated by the matching success level calculating unit 15. The degree can be compared.

  In the present embodiment, the matching success level calculation unit 15 receives the composite feature vector of any two people input by the input reception unit 14 and calculates the matching success level of the two arbitrary people. However, it is not limited to this. For example, the input receiving unit 14 receives two or more combined feature vectors generated from a feature vector of one user and each feature vector of a plurality of users who are candidates for the one user. Then, the matching success level calculation unit 15 calculates an inner product value between each of the two or more combined feature vectors received by the input reception unit 14 and the learning model (teacher vector) stored in the model table 22b. Then, as a result of the calculation, the matching success degree calculation unit 15 may cause the display unit to display a user with a synthetic feature vector having a high inner product value as a high matching success degree with the one user. By doing in this way, about the some user who becomes a candidate of one user, the user whose matching success degree with this one user is relatively high can be confirmed.

  In addition, the matching success level calculation unit 15 sorts the calculated multiple matching success levels when there are a plurality of arbitrary composite feature vectors delivered from the input reception unit 14 within a predetermined time. It is good also as displaying on a display part in order with a high matching success degree. By doing in this way, a relatively favorable combination can be selected from any two combinations corresponding to the combined feature vector delivered from the input receiving unit 14 within a predetermined time. Note that the predetermined time in this case is the time from the input start operation of the composite feature vector to the input end operation performed by the operation unit described later.

[Hardware configuration of matching success calculation device]
FIG. 8 is a diagram illustrating a hardware configuration of the matching success degree calculating device 1 according to the present embodiment. The matching success degree calculation apparatus 1 in which the present invention is implemented may be a standard one, and an example of the configuration is shown below.

  The matching success degree calculating device 1 includes a bus line 1000, a CPU (Central Processing Unit) 1010 constituting a control unit 10 (in a multiprocessor configuration, a plurality of CPUs such as a CPU 1012 may be added), a communication I / F (I / F F: Interface) 1020, main memory 1030, BIOS (Basic Input Output System) 1040, display device 1050, input device 1060, I / O (Input / Output) controller 1070, hard disk 1074, optical disk drive 1076, optical disk 1077, and semiconductor memory 1078. The hard disk 1074, the optical disk drive 1076, and the semiconductor memory 1078 are collectively referred to as the storage unit 20.

  The CPU 1010 is a part that comprehensively controls various functions related to the matching success degree calculation device 1, and cooperates with the above-described hardware by appropriately reading and executing various programs stored in the hard disk 1074 or the like, Various functions according to the present invention are realized. The communication I / F 1020 is a network adapter when the matching success calculation device 1 transmits / receives information to / from an external server or the like via a communication network. The communication I / F 1020 may include a modem, a cable modem, and an Ethernet (registered trademark) adapter.

  The main memory 1030 temporarily stores information generated when the CPU 1010 executes various programs. The BIOS 1040 records a boot program executed by the CPU 1010 when the matching success degree calculating device 1 is started, a program depending on the hardware of the matching success degree calculating device 1, and the like. The display device 1050 includes a display device such as a cathode ray tube display device (CRT) or a liquid crystal display device (LCD), and functions as a display unit. The display device 1050 displays the information output by the matching success level calculation device 1. The input device 1060 includes a keyboard, a mouse, and the like, and receives an input from the administrator of the matching success calculation device 1 and functions as an operation unit. The I / O controller 1070 is connected to devices such as a hard disk 1074, an optical disk drive 1076, and a semiconductor memory 1078, and performs input / output control of information.

  The hard disk 1074 stores various programs for causing the hardware to function as the matching success calculation device 1, a program for executing the functions of the present invention, and the above-described DB. The matching success calculation device 1 can also use a hard disk (not shown) separately provided as an external storage device.

  As the optical disk drive 1076, for example, a DVD-ROM drive, a CD-ROM drive, a DVD-RAM drive, and a Blu-ray Disc (registered trademark) drive can be used. When the optical disk drive 1076 is used, the optical disk 1077 corresponding to the optical disk drive 1076 is used. A program or data may be read from the optical disk 1077 by the optical disk drive 1076 and provided to the main memory 1030 or the hard disk 1074 via the I / O controller 1070.

  Note that the computer in the present invention refers to an information processing device including a storage device, a control unit, and the like, and the matching success calculation device 1 is information including the control unit 10 and the storage unit 20 as described above. The information processing apparatus is constituted by a processing apparatus and is included in the concept of the computer of the present invention.

  In addition, the matching success calculation device 1 is not limited in the number of hardware, and may be configured by one or a plurality of hardware as necessary. In the case of a plurality of hardware, each hardware may be connected via a communication network. For example, the functions of the present embodiment may be realized by using separate servers (devices) for each function and linking the servers by transmitting and receiving signals between the servers.

[Processing flow]
FIG. 9 is a flowchart showing a flow of processing from when the matching success degree calculation apparatus 1 according to the present embodiment generates a user feature vector to when a learning model is generated.

  In step S11, the control unit 10 (feature vector generation unit 11) refers to the user profile information stored in the profile table 21a and generates a user feature vector. Then, the control unit 10 (feature vector generation unit 11) stores the generated feature vector in the feature vector table 21b.

  In step S12, the control unit 10 (input list generation unit 12) refers to the result table 21c and extracts a combination of users that have been matched in the past. In step S13, the control unit 10 (input list generation unit 12) takes the difference between the elements of the feature vectors of the first user and the second user that constitute the combination of users extracted in step S12, A composite feature vector for the second user is generated.

  In step S14, the control unit 10 (input list generation unit 12) generates an input list and stores it in the input list table 22a. More specifically, the control unit 10 (input list generation unit 12) assigns a serial ID to the composite feature vector generated in step S13 and determines the success of the combination of users stored in the result table 21c. Create an input list by associating. The generated input list is stored in the input list table 22a.

  In step S15, the control unit 10 (learning unit 13) generates a learning model based on the input list. Details will be described with reference to FIG.

  FIG. 10 is a flowchart showing the flow of processing in the learning unit 13 of the matching success degree calculating apparatus 1 according to the present embodiment.

  In step S151, the control unit 10 (learning unit 13) refers to the input list table 22a and generates a set of synthetic feature vectors for each success level. In step S152, the control unit 10 (learning unit 13) defines two sets R and N from the set of synthetic feature vectors for each success degree generated in step S151.

In step S153, the control unit 10 (learning unit 13) extracts one synthetic feature vector from each of the sets R and N defined in step S152, and sets (d _nq , d) as a combination of two synthetic feature vectors. _rq ). d _nq is a composite feature vector of the set N, and d _rq is a composite feature vector of the set R.

In step S154, the control unit 10 (learning unit 13) determines that the inner product value of the combined feature vector _dnq and the weight w ⁱ constituting the set N is the inner product of the combined feature vector d _rq and the weight w ⁱ configuring the set R. It is determined whether or not the value is greater than or equal to If the determination result is YES, the process proceeds to step S155, and if the determination result is NO, the process proceeds to step S156. Here, the weight w ⁱ indicates the weight w updated for the i-th time.

In step S155, the control unit 10 (learning unit 13) adds the number obtained by multiplying the weight w ⁱ by the difference between d _rq and d _nq by η _q which is the reciprocal of the total value of the combination S. Update w ⁱ .

In step S156, the control unit 10 (learning unit 13) determines whether or not another combination (d _nq , d _rq ) of synthetic feature vectors exists. If the determination result is YES, the process proceeds to step S154, and if the determination result is NO, the process proceeds to step S157.

In step S157, the control unit 10 (learning unit 13) determines whether another set can be defined. If the determination result is YES, the process proceeds to S152, and if the determination result is NO, the process proceeds to S158. In step S158, the control unit 10 (the learning unit 13) stores the weights ^{w i} in the model table 22b. When this process ends, the process in the learning unit 13 ends. By doing in this way, a learning result will be memorize | stored in the model table 22b.

  FIG. 11 is a flowchart showing a flow of processing in the matching success level calculation unit 15 of the matching success level calculation device 1 according to the present embodiment.

  In step S21, the control unit 10 (matching success level calculation unit 15) receives an input of a combination of two users. In step S22, the control unit 10 (matching success level calculation unit 15) generates a combined feature vector of two users constituting the combination received in step S21. More specifically, the control unit 10 (matching success calculation unit 15) refers to the profile table 21a for the users constituting the combination, and generates a feature vector of the user. Then, a synthetic feature vector is generated based on the generated feature vector.

  In step S23, the control unit 10 (matching success level calculation unit 15) calculates the inner product of the combined feature vector generated in step S22 and the learning model stored in the model table 22b.

  Thus, in this embodiment, the learning unit 13 learns past matching results, generates a learning model as a learning result, and stores the learning model in the model table 22b. And the matching success degree calculation part 15 can calculate the matching success degree of a new combination of two users based on the learning model of the model table 22b. For this reason, in the present invention, a plurality of new combinations of two users are prepared, and the matching success level is calculated by causing the matching success level calculation unit 15 to calculate the matching success level, thereby selecting candidates suitable for matching. Can be selected.

  Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments, and modifications, improvements, and the like within the scope that can achieve the object of the present invention are included in the present invention.

  The matching success level calculation device 1 according to the present embodiment is assumed to calculate the matching success level of two users, such as a match or a business partner, but is not limited thereto. For example, the matching success level of two groups may be calculated based on past matching results. More specifically, in addition to the functions shown in FIG. 1, a group DB that stores information on the group to which the user belongs and a group matching success rate calculation unit are provided. And the matching success degree calculation part 15 specifies each user who belongs to the 1st group with reference to group DB, and each user who belongs to the 2nd group about each user who belongs to the 1st group The matching success level with the user is calculated by the matching success level calculation unit 15. Then, the group matching success level calculation unit calculates an average value of the calculated matching success levels, and uses the average value as the matching success level of the first group and the second group. By doing in this way, with respect to a certain group, it is possible to determine the ranks of matching success levels of other groups by calculating the matching success levels of other groups and comparing the matching success levels. Moreover, since the matching success level of the group is calculated based on the average value of the matching success levels of the users belonging to the group, it is possible to calculate the matching success level of the group by equally evaluating each user belonging to the group.

  Moreover, in this embodiment, although the synthetic | combination feature vector was decided by taking the difference of each element of the feature vector of a 1st user and a 2nd user, it does not restrict to this. That is, it is good also as producing | generating a synthetic | combination feature vector by performing a predetermined calculation based on the feature vector of a 1st user and a 2nd user. For example, a method of generating a combined feature vector by simply combining feature vectors of a first user and a second user, or generating a vector by combining two user feature vectors of success calculation targets, A method of generating a combined feature vector by combining the difference between feature vectors of two users whose success levels are to be calculated with respect to the generated vector, and specifying a plurality of elements in the elements constituting the feature vector, The synthetic feature vector may be generated using other methods such as a method of generating a synthetic feature vector by performing an operation on a plurality of designated elements.

  Further, in the present embodiment, a learning model (Jonathan Elsas, Fast learning of Document Ranking Functions with the Commitee Perceptron (WSDMit 2008ro Pent) is used by the processing shown in FIG. 10, that is, the page rank learning method in the search system. However, the present invention is not limited to this, and the learning model may be generated using another learning method, that is, in the present embodiment, for example, Online Passive-Aggressive Algorithm proposed by Koby Crammer et al. JMLR 2006, Passive Aggressive) and T. Joachims et al. Optimized search engines using clicking through data (KDD 2002, “RankingSVM”), and Zhe Cao et al.'S “Learning to Rank: ICP” which is proposed by Zhe Cao et al. A learning model may be generated using a learning method.

  In the present embodiment, the matching success level is calculated using an index called the success level, but the present invention is not limited to this. For example, an index that can objectively evaluate a plurality of users, such as a rank and a score, may be used.

DESCRIPTION OF SYMBOLS 1 Matching success degree calculation apparatus 10 Control part 11 Feature vector generation part 12 Input list generation part 13 Learning part 14 Input reception part 15 Matching success degree calculation part 20 Storage part 21 User DB
22 Learning DB

Claims (7)

A matching success calculation device for calculating a matching success of an arbitrary user,
Individual data of the user to be matched, profile information stored in advance in a predetermined database is digitized, and each numeric value is used as a vector element, thereby generating a feature vector generation means for generating a user feature vector;
Synthetic feature vector generation means for generating a composite feature vector based on individual user feature vectors;
Matching storage means for storing the synthetic feature vector and past matching success history of each user of the synthetic feature vector in association with each other;
Learning means for generating a learning model by generating a teacher vector based on the synthetic feature vector stored in the matching storage means and the past matching success history;
Input accepting means for accepting input of a composite feature vector of any two users;
Matching success calculation means comprising: a matching success degree calculating means for calculating a matching success degree of the arbitrary two users based on the teacher vector and the composite feature vector of the arbitrary two users inputted. Degree calculation device.   2. The matching success degree calculating apparatus according to claim 1, wherein the synthetic feature vector generation unit generates the synthetic feature vector by performing a predetermined operation on each element of the feature vectors of two users.   The inner product value of the teacher vector and the combined feature vector of the two users who performed matching with which a relatively good result was obtained could not obtain a relatively good result with the teacher vector. 3. The matching success degree calculation device according to claim 2, wherein the matching success degree calculation apparatus has a higher value than a value of an inner product with a combined feature vector of two users who have performed matching. The input receiving unit receives two or more combined feature vectors generated from a feature vector of one user and each feature vector of a plurality of users who are candidates for the one user,
The matching success degree calculation means calculates an inner product value of each of the two or more combined feature vectors and the teacher vector, and as a result of the calculation, a user of the combined feature vector having a higher inner product value is determined as the one user. The matching success degree calculation device according to claim 1, wherein the matching success degree is high. A matching success calculation device for calculating the matching success of an arbitrary group,
Individual vector data of users belonging to the group to be matched, profile information stored in advance in a predetermined database is digitized, and each numeric value is used as a vector element to generate a feature vector of the user Means,
A combined feature vector generating means for generating a combined feature vector by combining the feature vectors of individual users belonging to the group performing the matching;
Matching storage means for storing the synthetic feature vector and the past successful matching history of each user of the synthetic feature vector in association with each other;
Learning means for generating a learning model by generating a teacher vector based on the synthetic feature vector stored in the matching storage means and the past matching success history;
Selection accepting means for accepting selection of any two groups ;
Successful matching comprising: the teacher vector; and a matching success degree calculating means for calculating a matching success degree of the two arbitrary groups based on a combined feature vector of a plurality of users belonging to the two groups for which selection has been accepted. Degree calculation device.   The matching success level calculating means calculates the matching success level of each of the plurality of users belonging to the second group with respect to each of the plurality of users belonging to the second group. Based on a combined feature vector of a plurality of users belonging to the first group and a plurality of users belonging to the second group, an average value of the matching success degrees is calculated based on the first group and the second group 6. The matching success degree calculating device according to claim 5, wherein the matching success degree is calculated as a matching success degree of the group. A matching success calculation method executed by a matching success calculation device for calculating a matching success of an arbitrary user,
Individual data of users to be matched, the profile information stored in advance in a predetermined database is digitized, and each numeric value is used as a vector element to generate a user feature vector;
Generating a composite feature vector based on individual user feature vectors;
Storing the synthetic feature vector and the past matching success history of each user of the synthetic feature vector in association with each other;
Generating a learning model by generating a teacher vector based on the combined feature vector stored in the storing step and the past matching success history;
Receiving input of composite feature vectors of any two users;
A matching success degree calculation method comprising: calculating a matching success degree of the arbitrary two users based on the teacher vector and the input synthetic feature vector of the arbitrary two users.

Images