JP2016153931A - Information processing method, information processing device, and information processing program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To derive a potential relationship between a first person and a second person.SOLUTION: An information processing method includes the steps of: determining a condition used for deriving a relationship between first data including the characteristic amount of information on a first person and second data including the characteristic amount of information on a second person, from existing data and relationship; and deriving the relationship between the first data and second data according to the condition.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an information processing method, an information processing apparatus, and an information processing program for deriving a relationship between a certain thing and another.

  In the case where one person is combined with another person to perform some action together, in order to perform the action well, the potential relationship between the two is important. For example, it is a case where a person in charge of a certain business is selected and the business is performed together with another person.

  Here, when selecting a person in charge of a certain business, generally, a business statement and a transfer candidate in a position that is assumed to be accepted by the candidate in charge for the population of candidates in charge registered in the database. This is done by matching the description contents of the business resume. Here, the business statement is a document in which a business name, business content, business performance requirements, and the like are described, and the business history is a document in which the business history is described.

  This matching is usually performed mainly by human work, but there is the following disclosure regarding a system that supports the work. Patent Document 1 discloses a project personnel selection support management system for maintaining a stable project team, and Patent Document 2 discloses a system that supports selection of members constituting a team in charge of a new project. Yes.

  Further, in relation to the present invention, Patent Document 3 discloses a technique related to latent semantic indexing (hereinafter referred to as “SSI”). In this technique, first, a vector of two input data belonging to a high-dimensional feature vector space is converted (feature extraction) into a low-dimensional concept space. The technique of the same technique then calculates the similarity between two input data by calculating the cosine similarity between these two vectors belonging to the same concept space.

  Further, in connection with the present invention, Non-Patent Document 1 discloses a method for automatically generating a feature vector. Further, in relation to the present invention, Non-Patent Document 2 describes a support vector machine library (A Library for Support Vector Machines).

JP 2011-008387 A JP 2007-122144 A US Pat. No. 8341095

Besparov, Dmitriy and Qi, Yanjun and Bai, Bing and Shokofuandeh, Ali, “Machine Learning and Knowledge Discovery in Database, 12”. 159-174 Chih-Chung Chang and Chih-JenLin, “LIBSVM--A Library for SupportVector Machines” [searched on January 7, 2015], Internet (http: //www.csie.nt. /)

  However, in any of the methods described in Patent Documents 1 and 2, information about the candidate needs to be quantified in advance by some means. For example, Patent Document 1 describes that human relations are taken into account when selecting project members. However, the human relationships to be considered when setting project members need to be prepared in advance by human evaluation and quantification.

  That is, these methods cannot derive a potential relationship, such as compatibility between one member and another member, from information based on facts that are not based on human evaluation.

  The objective of this invention is providing the information processing method etc. which can derive | lead-out the potential relationship with the other member which solves the said subject and one member.

  The information processing method according to the present invention determines a condition for deriving a relationship between first data including a feature amount of information relating to a first person and second data including a feature amount of information relating to a second person. The step of performing is included. The determination includes third data including a characteristic amount of information regarding a certain person, fourth data including a characteristic amount of information regarding the certain person or a person other than the certain person, and the certain person and the certain person or And a known relationship with a person other than the certain person. The information processing method of the present invention further includes a step of deriving a first data relationship, which is a relationship between the first data and the second data, based on the determined condition.

  With the information processing method, information processing apparatus, and information processing program of the present invention, a potential relationship between a person and another person can be derived.

It is a conceptual diagram showing the information processing method of 1st embodiment. It is a conceptual diagram showing the information processing method of 2nd embodiment. It is a conceptual diagram showing the structure of the information processing apparatus of 2nd embodiment. It is an image figure showing the 1st example of a 1st vector. It is an image figure showing the 2nd example of a 1st vector. It is a conceptual diagram showing the information processing method of 3rd embodiment. It is a conceptual diagram showing the structure of the information processing apparatus of 3rd embodiment. It is an image figure showing the structure of each group tree. It is a conceptual diagram showing the structure of the information processing apparatus of 4th embodiment. It is an image figure showing the example of human information. It is an image figure showing the example of the information about the relationship between a supervisor and a subordinate. It is an image figure showing the example of the information which a learning information storage part memorize | stores. It is an image figure showing the example of the information which a result memory | storage part memorize | stores. It is a conceptual diagram showing the processing flow in a prior learning step. It is a conceptual diagram showing the processing flow in a compatibility prediction step. It is a conceptual diagram showing the processing flow in an automatic selection step. It is a conceptual diagram showing the processing flow which calculates the gain of a 1st group. It is an image figure showing the group performance change by transfer.

<First embodiment>
This embodiment is an embodiment of the minimum information processing method of the present invention.
[Processing flow]
FIG. 1 is a conceptual diagram showing the information processing method of this embodiment.

First, the first data including the feature amount of information on the first person (hereinafter referred to as “first party”) and the second person (hereinafter referred to as “second party”). The condition for associating the second data including the feature amount of the information on the is determined. The determination includes third data including a characteristic amount of information regarding a certain person, fourth data including a characteristic amount of information regarding the certain person or a person other than the certain person, the certain person, and the certain person. Alternatively, it is performed from a known relationship with respect to a person other than the certain person. (S001)
Finally, a relationship between the first data and the second data is derived and output by the relationship based on the determined condition. (S002)
[effect]
With the information processing method of this embodiment, a potential relationship between a person and another person can be derived.
<Second embodiment>
This embodiment is an embodiment of an example of an information processing apparatus that generates vectors for the first party and the second party. The vector will be described later.
[Processing flow]
FIG. 2 is a conceptual diagram showing the information processing method of this embodiment.

First, information about the first party is set as the first vector. Here, the first vector is data including a feature amount of information about the first party. (S011)
Next, the information regarding the second party is set to the second vector. Here, the second vector is data including a feature amount of information about the second party. (S012)
Then, a condition for associating the first vector with the second vector is determined. . The determination includes a third vector in which information about a person is a vector, a fourth vector in which information about the person or a person other than the person is a vector, the person, the person, or the person We do it from a known relationship about a person other than a certain person. Here, the third vector is data including a feature amount of information regarding the certain person, and the fourth vector is data including a feature amount of information regarding the certain person or a person other than the certain person. (S013)
Finally, the relationship between the first vector and the second vector is derived according to the condition and output. (S014)
The order of the processes of S011 to S013 can be arbitrarily changed.

[Configuration and operation]
FIG. 3 is a conceptual diagram showing the configuration of the information processing apparatus of this embodiment.

  The information processing apparatus 100 of this embodiment includes a human information storage unit 001, a human information processing unit 003, a relationship derivation unit 004, a learning information storage unit 005, a learning processing unit 006, and a relationship output unit 007.

  The information processing apparatus 100 is typically a personal computer or a server. In the present embodiment, a server is assumed as the information processing apparatus 100. In addition, since the function and hardware configuration of the server are generally well known, description thereof is omitted here.

  The human information storage unit 001 stores information about each of the input first party and second party. The stored information is sent to the human information processing unit 003.

  The human information processing unit 003 generates a vector for the first party (hereinafter referred to as “first vector”) from the information about the first party sent from the human information storage unit 001. Here, the first vector is data including a feature amount of information on the first party. The human information processing unit 003 generates a vector for the second party (hereinafter referred to as “second vector”) from the information about the second party sent from the human information storage unit 001. Here, the second vector is data including a feature amount of information about the second party. The generated first vector and second vector are sent to the relationship deriving unit 004.

  In the present specification, a “vector” is an example of data including a feature amount of information related to a certain person, and is a numerical string in which numerical values are arranged. An item that defines what it is about. As shown in the example described later, in the numeric string, symbols other than numeric values, for example, commas, semicolons, colons, etc. may be inserted in addition to numeric values. Absent.

  The learning information storage unit 005 can perform a process for deriving a relationship between vectors generated by the relationship deriving unit 004 (hereinafter referred to as “vector relationship”) by performing learning and performing an appropriate process more suitable for the purpose. The inputted information is stored. The stored information is sent to the learning processing unit 006.

  The learning processing unit 006 uses the information sent from the learning information storage unit 005 to determine a condition for associating vectors in the relationship deriving unit 004. The determination is performed so that appropriate processing suitable for the purpose can be performed by learning.

  The relationship deriving unit 004 derives the vector relationship between the first vector and the second vector sent from the human information processing unit 003 according to the conditions determined by the learning processing unit 006. Information including the derived vector relationship is sent to the relationship output unit 007.

  The relationship output unit 007 outputs information including the vector relationship sent from the relationship deriving unit 004.

  There may be a plurality of human information storage units 001, and the human information storage units that store information about each of the first and second parties may be different in the plurality of human information storage units.

  The information about the first party and the second party input to the personal information storage unit 001 may be in a data format divided into items, a format described in text, or a combination thereof. I do not care.

  The generation of the first vector and the second vector from the information from the human information storage unit 001 in the human information processing unit 003 can be performed as follows, for example.

  Gender, age, final educational background, origin, qualification, etc. may be simply quantified. Alternatively, the age, final educational background, qualification and the like can be further classified, and the corresponding code item can be set to 1 and the non-applicable code item can be set to 0. For example, if it is age, it is subdivided into 0-20 years old, 21-40 years old, 41-60 years old, 61 years old or older, the corresponding code item is set to 1, and the non-applicable code item is set to 0.

  For example, text information included in a resume or the like decomposes text into words by a morphological analysis technique, counts frequencies, sets words as vector items, and sets frequencies as vector values. Since morphological analysis is a widely known technique, its explanation is omitted. The frequency can be counted from two to five words instead of every single word. In addition, it is possible to perform vectorization so that the features of the text content appear by limiting words that are frequently counted, such as by excluding words that frequently appear in all documents, for example, particles.

It should be noted that text posted by a first party or the like on a Web (World Wide Web) site or SNS (social networking service) can be converted into items and data constituting a vector as elements representing the characteristics of personnel. . The first party or the like may write a matter including the characteristics of the first party or the like by using such information because there are cases where the first party or the like has written in the SNS or blog. Similar to the above method of digitizing the text, the URL name or the like of the access destination is decomposed and the frequency or residence time is counted. And URL (Uniform Resource)
The access history can be vectorized by decomposing and counting the HTML (Hypertext Transfer Protocol) document of the Locator.

  Since various known methods can be used for acquiring these pieces of information, the description thereof is omitted here.

  FIG. 4 is an image diagram illustrating a first example of the first vector.

  The number sequence within the frame line in the figure is the first vector for the first party. This first vector has a gender of female, an age of 61 or more, a college degree, a European origin, the number of words “A” included in the text information is 6 to 8 and the number of the word “B” is 9 or It is about the first party such as 10. In this example, “,” is inserted at the boundary of each item.

  FIG. 5 is an image diagram illustrating a second example of the first vector.

  In this example, with respect to the number of words included in the age and text information, a number indicating the age and the number of words is written as it is, instead of providing an option and writing “1” in the corresponding option. Is different from the example shown in FIG.

  It is desirable that the first vector and the second vector are the same number of items, the item content, and the ordering sequence of the items, that is, the vectors expressed in the same vector space. This is because, since the vectors are represented in the same vector space, it is possible to apply the technique such as SSI described above without performing a process for converting into a vector in the same vector space. Of course, if a post-process that performs processing such as extracting a numerical value corresponding to a predetermined item from a vector and converting it to a vector represented in the same vector space is used, and the vector after the post-process is used, the same vector It does not have to be a vector belonging to space.

  The derivation of the vector relationship in the relationship derivation unit 004 can be performed, for example, by obtaining the similarity between vectors. As an index representing the similarity at this time, cosine similarity or any other index indicating the similarity (Pearson correlation coefficient, deviation pattern similarity, etc.) can be used. It can be evaluated that the larger these values are, the better the compatibility is. Alternatively, any index (L1 norm, L2 norm, etc.) indicating the distance between two vectors may be used. It can be evaluated that the smaller these values are, the better the compatibility is.

  The learning information stored in the learning information storage unit 005 typically includes a vector pair whose vector relationship has already been derived by means other than the information processing apparatus, such as evaluation by a person, and an already derived vector pair. This is a set of information representing the vector relationship being performed. For example, if the vector relationship between the first party and the second party to be derived in the relationship deriving unit 004 is such that the first party and the second party are compatible, the vector pair It is a degree that represents good or bad compatibility.

  Note that the known vector relationship used for learning is not necessarily binary information, and may be multivalued or continuous.

  In this case, for example, the learning processing unit 006 compares the vector relationship obtained by inputting the given vector pair to the relationship deriving unit 004 and the given known vector relationship. Then, the learning processing unit 006 determines conditions used for the vector relationship deriving process performed in the relationship deriving unit 004 so that the values are close to each other. More specifically, the learning processing unit 006 determines a value of a function parameter used in the process performed by the relationship deriving unit 004, for example.

  More specifically, the determination of the condition for deriving the vector relationship can use a machine learning parameter determination method in machine learning. As a machine learning means at this time, any supervised machine learning classifier including the SSI technique described in Patent Document 3 can be applied. For example, a support vector machine, a neural network, a Bayes classifier, or the like can be used.

  The determination of the conditions for deriving the vector relationship is usually performed by information representing a plurality of vector pairs and the already derived vector relationships for the vector pairs.

[effect]
The information processing method according to the present embodiment uses data including feature quantities of information that has been revealed with respect to a certain person and other persons, and derivation of the relationship between these data is possible by using separately prepared learning data. This is done according to the conditions determined so that information on the appropriate relationship is required.

For this reason, the information processing method of the present embodiment can derive a potential relationship between the two even though the information used for the processing itself is composed of information that has become apparent.
<Third embodiment>
The present embodiment is an embodiment of an information processing apparatus that generates a feature vector from a vector of a person or the like.
[Processing flow]
FIG. 6 is a conceptual diagram showing the information processing method of this embodiment.

  The description of S011 and S012 in the figure is the same as the description of the process indicated by the same number in FIG.

Following S011, a first feature vector is generated from the first vector generated in S011. (S021)
Following S012, a second feature vector is generated from the second vector generated in S012. (S022)
Next, a condition used when deriving a relationship between the first feature vector and the second feature vector is determined. The determination includes a third vector in which information about a person is a vector, a fourth vector in which information about the person or a person other than the person is a vector, the person, the person, or the person We do it from a known relationship about a person other than a certain person. (S023)
Finally, the relationship between the first feature vector and the second feature vector is derived and output according to the determined condition. (S024)
[Configuration and operation]
FIG. 7 is a conceptual diagram showing the configuration of the information processing apparatus of this embodiment.

  The information processing apparatus 101 includes a staff post-processing unit 008 in addition to the configuration of the information processing apparatus 100 according to the second embodiment whose conceptual diagram is shown in FIG.

  The description of the configuration other than the personnel post-processing unit 008 is the same as the description of the configuration indicated by the same number in the second embodiment, except for the description in the present embodiment.

  The first vector and the second vector generated by the human information processing unit 003 are sent to the personnel post-processing unit 008.

  The personnel post-processing unit 008 creates a first party feature vector from the first vector sent by the human information processing unit 003 and a second party feature vector from the second vector sent by the human information processing unit 003, respectively. . The first party feature vector and the second party feature vector are sent to the relationship deriving unit 004.

  The relationship deriving unit 004 derives the relationship between the first party feature vector and the second party feature vector sent from the human information processing unit 003 based on the conditions determined by the learning processing unit 006.

  The generation of the first party feature vector in the staff post-processing unit 008 is a process of selecting only a term that becomes a feature from the first vector and generating a compressed vector. In general, the first vector is data having a large vector length, and it may be difficult to apply the first vector to learning and prediction in the subsequent stage. For this reason, only a term that becomes a feature is selected and a compressed vector is generated. The same applies to the second vector.

  In this case, typical contents of the characteristic items are related to preferences, interests, hobbies, habits, habits, life, etc. in addition to sex, age, educational background, and origin. On the other hand, content that does not become a feature is content that is almost common for humans, and is typically content such as “eat”, “sleep”, and “walk”.

The method described in Non-Patent Document 1 can be applied to the generation of feature vectors. This method is a method of automatically generating a feature vector, but it is also possible to generate a feature vector by analyzing an important vector term by principal component analysis or the like and manually selecting the vector term.
[effect]
First, the information processing apparatus of this embodiment has the same effect as the information processing apparatus of the second embodiment.

In addition, the information processing apparatus of the present embodiment can derive and output a potential relationship between the first party and the second party even when the vector length of the first vector and the second vector is too long. it can. This is because only the portion corresponding to the feature item is extracted from the first vector and the second vector.
<Fourth embodiment>
The present embodiment is an embodiment related to an information processing apparatus in which the present invention is applied to a case of personnel change in an organization such as an actual company. Focus on the compatibility of the candidate with the supervisor who is expected to have a particularly deep business connection at the transfer destination, and evaluate the transfer from the viewpoint of the profit of the entire organization reflecting the compatibility.
[Premise in this embodiment]
The “candidate” is a transfer candidate belonging to the first group. “Supervisor” belongs to the second group, and becomes a candidate's supervisor when the candidate is changed. Here, the “second group” refers to a group that wishes to accept personnel. The “first group” refers to a group in which personnel are relinquished when personnel are transferred to the second group. Both the second group and the first group belong to the same upper group (hereinafter referred to as “organization”).

  The personnel composition of the first group and the second group can be expressed by a tree structure (hereinafter referred to as “group tree”).

  At this time, the root represents the group length, the node and the leaf represent the members of the group, and the edge represents the relationship between the superior and the subordinate. Here, the branch has a weight, and the weight is a value that quantitatively expresses whether the boss and the subordinate are connected to the branch.

  The structure of the group tree for each of the first group and the second group is shown in FIG.

In this expression, the performance of the entire organization in the first group and the second group (hereinafter referred to as “group performance”) is defined as the sum of the weights of the branches in the group tree. In other words, it is assumed that the performance of the organization as a whole is higher as the organization of the compatibility of the superior and subordinates is larger.
[Configuration and operation]
FIG. 9 is a conceptual diagram showing the configuration of the information processing apparatus of this embodiment.

  The information processing apparatus 102 includes a human information storage unit 1, a structure information storage unit 2, a candidate information processing unit 3, a boss information processing unit 4, a first post-processing unit 5, a second post-processing unit 6, A relationship learning unit 7 is provided. The information processing apparatus 102 further includes a learning information storage unit 8, a learning result storage unit 9, a relationship derivation unit 10, a result storage unit 11, and a result display unit 12.

  The human information storage unit 1 stores input information regarding candidates and superiors. Of the stored information, information related to the candidates is sent to the candidate information processing unit 3, and information related to the boss is sent to the boss information processing unit 4.

  The structure information storage unit 2 stores the input information relating to the personnel composition of the first group and the second group. Of the stored information, information on the first group is sent to the candidate information processing unit 3, and information on the second group is sent to the boss information processing unit 4.

  The candidate information processing unit 3 vectorizes information related to the candidates based on the information sent from the human information storage unit 1 and the structure information storage unit 2 to generate a candidate vector. Here, the candidate vector is data including a feature amount of information related to the candidate. Information including the candidate vector is sent to the first post-processing unit 5.

  The boss information processing unit 4 vectorizes information related to the boss based on the information sent from the human information storage unit 1 and the structure information storage unit 2 to generate a boss vector. Here, the boss vector is data including a feature amount of information related to the boss. Information including the boss vector is sent to the second post-processing unit 6.

  The first post-processing unit 5 extracts a portion corresponding to an item that is a candidate feature from the candidate vector sent from the candidate information processing unit 3, and generates a candidate feature vector. The generated candidate feature vector is sent to the relationship deriving unit 10.

  The second post-processing unit 6 extracts a portion corresponding to an item that is a feature of the boss from the boss vector sent from the boss information processing unit 4 and generates a boss feature vector. The generated boss feature vector is sent to the relationship deriving unit 10.

  The learning information storage unit 8 stores the input learning information and sends the learning information to the relationship learning unit 7.

  The relationship learning unit 7 uses the learning information sent from the learning information storage unit 8 to determine conditions used for relationship derivation performed by the relationship derivation unit 10. The determined condition is sent to the learning result storage unit 9.

  The learning result storage unit 9 stores the condition sent from the relationship learning unit 7. The stored condition is sent to the relationship deriving unit 10.

  The relationship deriving unit 10 obtains a vector relationship between the candidate feature vector and the boss vector based on the condition sent from the learning result storage unit 9. The obtained vector relationship is sent to the result storage unit 11.

  The result storage unit 11 stores the vector relationship sent from the relationship deriving unit 10 and sends the vector relationship to the result display unit 12.

  The result display unit 12 displays information including the vector relationship sent from the result storage unit 11.

  FIG. 10 is an image diagram illustrating an example of personnel attribute information of the first group and the second group. The person information storage unit 1 stores personnel attribute information of the first group and the second group as shown in FIG.

  Personnel attribute information includes, for example, an ID (refers to a symbol for identifying a person), basic information (name, age, gender, educational background, work experience, transfer history, qualification, etc.), work description, performance evaluation (personnel appraisal) , Bonuses, annual income, etc.), health checkup results, and other evaluation results. Note that the personnel attribute information of the second group job offer frame (R6 in the figure) is N / A, which is a symbol indicating that there is no corresponding information for all items.

  FIG. 11 is an image diagram illustrating an example of structure information included in the group tree of the first group and the second group. The table on the left side of the figure describes the first boss and the second boss of each person having the personnel ID of the leftmost term. N / A is described when there is no corresponding supervisor. The tree structure on the right side of the figure is an image of the corresponding table on the left side.

  The first boss is a boss of a member belonging to the group in the current group tree, and the second boss is a boss of a member belonging to the group in the group tree when the first boss is transferred. The first boss ID and the second boss ID are the IDs of the first boss and the second boss, respectively.

  The structure information storage unit 2 stores information as shown in FIG. For example, the candidate information processing unit 3 refers to the structure information storage unit 2 in accordance with an instruction from the first post-processing unit 5 and identifies a candidate for the target personnel, and the personnel attribute for the candidate from the human information storage unit 1 Information is read and candidate vectors are generated using the personnel attribute information.

  The boss information processing unit 4 reads, for example, personnel attribute information about the boss from the human information storage unit 1 in accordance with an instruction from the second post-processing unit 6 and generates a boss vector using the personnel attribute information.

  The relationship learning unit 7 uses a data set including three pieces of data to machine-learn relationships so that compatibility can be appropriately obtained. The three data are the candidate for learning read from the candidate feature vector, the supervisor feature vector, and the learning information storage unit 8 for the learning candidate and the learning boss for which compatibility has already been obtained. And the learning boss.

  FIG. 12 is an image diagram showing information stored in the learning information storage unit 8.

  The learning information storage unit 8 stores a data set of a candidate ID, a boss ID, and a compatibility score for a corresponding candidate-boss pair. The compatibility score is a compatibility score between the personnel and the boss, and is expressed by a binary value of 1 when good, 0 when bad, for example. The compatibility score can be obtained from a candidate's performance evaluation (personnel appraisal, bonus amount, annual income, etc.), a subjective evaluation value of the candidate for the boss, and the like. For example, if the candidate's performance evaluation is equal to or higher than the threshold, the compatibility is good, if the evaluation is lower than the threshold, the compatibility is bad, or if the evaluation from the subordinates is equal to or higher than the threshold, the compatibility is good, and if the evaluation is lower than the threshold, the compatibility is good It's bad.

  The learning result storage unit 9 stores the learning result learned by the relationship learning unit 7.

  The relationship deriving unit 10 reads the information stored in the learning result storage unit 9 and derives a predicted value of the compatibility score from the unknown candidate feature vector corresponding to the candidate ID and the boss ID and the boss feature vector. Specifically, there are the following two patterns of unknown candidate ID and boss ID pairs.

<Persons in the first group, bosses in the second group>
<Each member of the first group, the second boss of the first group> * However, the number of personnel assigned to the second boss is limited.

  FIG. 13 is an image diagram illustrating an example of information stored in the result storage unit 11. As shown in the figure, for example, the result storage unit 11 holds a compatibility score for a set of unknown candidate ID and boss ID calculated by the relationship deriving unit 10.

For example, the result display unit 12 reads the information stored in the result storage unit 11 and takes into account the performance improvement of the second group and the performance degradation of the first group accompanying the transfer from the first group to the second group. To do. And the result display part 12 calculates | requires and displays the aptitude to the recruitment frame of the 2nd group of the staff of a 1st group.
[Processing flow]
Hereinafter, an example of a processing flow in the information processing apparatus 102 will be described. This processing flow includes three steps: a pre-learning step, a compatibility prediction step, and an automatic selection step.

  In the pre-learning step, machine learning is performed from information on learning candidates and compatibility managers for which compatibility data has already been obtained. That is, a feature vector is generated for the learning candidate and the learning boss, machine learning is performed based on the already obtained result of the compatibility, machine learning parameters are determined, and the learning result is recorded. Here, the machine learning parameter is an example of a condition used for the above-described relationship derivation.

  In the compatibility prediction step, the compatibility score of the candidate and the boss whose compatibility score has not yet been obtained is calculated by the process of setting the machine learning parameter obtained in the prior learning step.

  In the automatic selection step, the profit of the entire organization is obtained using the compatibility score obtained between each of the plurality of candidates and the boss, and the transfer suitability of each candidate from that viewpoint can be compared.

  FIG. 14 is a conceptual diagram illustrating an example of a processing flow in the pre-learning step. The description of each process shown in the figure is as follows.

S101: Reading of compatibility score First, the relationship learning unit 7 reads all data sets of <candidate ID, boss ID, compatibility score> from the learning information storage unit 8 for pairs of learning candidates and learning bosses. Here, the ID is an ID uniquely assigned to all members of the second group and the first group. The supervisor ID is the supervisor ID of the person. The compatibility score is already obtained for the pair of the learning candidate and the learning boss.

S102: Repeat START for the number of compatibility scores
The relationship learning unit 7 repeats the following processes from S103 to S108 for the number of data sets read in S101.

S103: Generation of Boss Vector The relationship learning unit 7 drives the boss information processing unit 4 and instructs reading of the boss information corresponding to the ID acquired in S101. The route for performing the instruction is arbitrary, but is performed via the learning result storage unit 9, the relationship derivation unit 10, and the second post-processing unit 6, for example. The boss information processing unit 4 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a boss vector.

  As a method for vectorizing the boss information, for example, the method described for the human information processing unit 003 of the second embodiment can be used.

S104: Generation of Boss Feature Vector The relationship learning unit 7 drives the first post-processing unit 5 and instructs the reading of the boss vector generated in S103. The route for giving the instruction is arbitrary, but for example, it is made via the learning result storage unit 9 and the relationship deriving unit 10.

  The first post-processing unit 5 performs feature extraction of the boss vector and generates a boss feature vector. The description about the generation of the feature vector is the same as the description in the third embodiment.

S105: Generation of Candidate Vector The relation learning unit 7 drives the candidate information processing unit 3 and instructs reading of candidate information corresponding to the candidate ID acquired in S101. The route for giving the instruction is arbitrary, but for example, it is made via the learning result storage unit 9, the relationship derivation unit 10, and the first post-processing unit 5. The candidate information processing unit 3 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a candidate vector.

  The explanation of vectorization of candidate information is the same as that for S103.

S106: Generation of Candidate Feature Vector The relation learning unit 7 drives the second post-processing unit 6 and instructs reading of the candidate vector generated in S105. The route for giving the instruction is arbitrary, but for example, it is made via the learning result storage unit 9 and the relationship deriving unit 10. The second post-processing unit 6 performs feature extraction of candidate vectors and generates candidate feature vectors.

  The description about the feature extraction of the concession vector is the same as the description about S104.

S107: Calculation of similarity between feature vectors The relationship learning unit 7 calculates the similarity between the boss feature vector generated in S104 and the candidate feature vector generated in S106. The description about the calculation of the similarity is the same as the description in the second embodiment.

S108: Determination of Machine Learning Parameter The relationship learning unit 7 determines a machine learning parameter using the similarity calculated in S107 and the affinity score (value of 0 or 1) acquired in S101. The description about the determination of the machine learning parameter is the same as the description in the second embodiment.

S109: Repeat END for the number of compatibility scores
The relationship learning unit 7 repeats the processes of S103 to S108 for the number of data sets read in S101.

S110: Writing Learning Result The relationship learning unit 7 records the machine learning parameter determined in S108 in the learning result storage unit 9.

  FIG. 15 is a conceptual diagram illustrating an example of a processing flow in the compatibility prediction step. The description of each process shown in the figure is as follows.

S201: Reading of Learning Result The relationship deriving unit 10 reads the machine learning parameter recorded in S110 from the learning result storage unit 9.

S202: Generation of Compatibility Prediction Target List The relationship deriving unit 10 generates a list of candidate ID and boss ID pairs that are targets of compatibility score prediction. The list is automatically generated according to the following two rules.
[Rule 1]
・ Candidate ID shall be the ID of each person in the first group.

-The boss ID is the ID of the boss of the second group job offer.
[Rule 2]
The candidate ID is the ID of the person who is set by the second boss in the human information storage unit 1.

  The boss ID is the second boss ID of the person who is set as the second boss in the human information storage unit 1.

  The list generation according to rule 1 is performed in order to simulate the compatibility score between the personnel and the manager of the job offer frame when each member of the first group is transferred to the job offer frame of the second group in the automatic selection step.

  The list generation according to rule 2 is to simulate the compatibility score between the candidate of the member and the second boss of the candidate when each member of the first group is transferred to the recruitment frame of the second group in the automatic selection step. To do.

S203: Repeat START for the number of items in the list.
The relationship deriving unit 10 repeats the processes of S204 to S209 for the number of lists generated in S202.

S204: Generating Candidate Vector The relationship deriving unit 10 drives the candidate information processing unit 3 to instruct reading of candidate information corresponding to the candidate ID acquired in S202. The route for performing the instruction is arbitrary, but is performed via the first post-processing unit 5, for example. The candidate information processing unit 3 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a candidate vector.

  The description about vectorization of candidate information is the same as the description about S103.

S205: Generation of Candidate Feature Vector The relationship deriving unit 10 drives the first post-processing unit 5 and instructs reading of the candidate vector generated in S204. The first post-processing unit 5 performs feature extraction of candidate vectors and generates candidate feature vectors.

  The description about the feature extraction of the candidate vector is the same as the description about S104.

S206: Generation of Boss Vector The relationship deriving unit 10 drives the boss information processing unit 4 and instructs reading of the boss information corresponding to the ID acquired in S202. The route for performing the instruction is arbitrary, but is performed via the first post-processing unit 5, for example. The boss information processing unit 4 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a boss vector.

  The explanation about vectorization of the boss information is the same as the explanation about S103.

S207: Generation of superior feature vector The relationship deriving unit 10 drives the second post-processing unit 6 and instructs the reading of the superior vector generated in S206. The second post-processing unit 6 performs feature extraction of the boss vector and generates a boss feature vector.

  The description of feature vector feature extraction is the same as that for S104.

S208: Calculation of compatibility score The relationship deriving unit 10 uses the candidate feature vector generated in S205 and the boss feature vector generated in S207, and based on the machine learning parameter read in S201, the boss and candidate compatibility score. Is calculated.

  The compatibility score is, for example, a numerical value called supportability (confidence, reliability) of the support vector machine (see Non-Patent Document 2).

S209: Writing Analysis Result The relationship deriving unit 10 records the compatibility score calculated in S208 in the result storage unit 11 in the format of <candidate ID, boss ID, compatibility score>.

S210: Repeat END for the number of items in the list
The relationship deriving unit 10 repeats the processes of S204 to S209 for the number of lists generated in S202.

  FIG. 16 is a conceptual diagram showing processing in the automatic selection step. The description of each process shown in the figure is as follows.

S 301: The analysis result reading result display unit 12 reads the compatibility score from the result storage unit 11.

S302: Generation of job offer frame boss vector The result display unit 12 drives the boss information processing unit 3 and instructs reading of the boss information corresponding to the boss ID of the job offer frame. The boss information processing unit 3 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a boss vector.

  The explanation about vectorization of the boss information is the same as the explanation about S103.

S303: Generation of Boss Feature Vector The result display unit 12 drives the second post-processing unit 5 and instructs the reading of the boss vector generated in S302. The route for performing the instruction is arbitrary, but is performed via the result storage unit 11, the relationship derivation unit 10, and the first post-processing unit 5, for example. The second post-processing unit 5 performs feature extraction of the boss vector and generates a boss feature vector.
The description of feature vector feature extraction is the same as that for S104.

S304: Generation of Transfer Target List The result display unit 12 generates a personnel list as a transfer candidate target. The list is automatically generated by sequentially applying the following two rules.

[Rule 1]
In the human information storage unit 1, the ID of the person whose group is the first group is extracted.

[Rule 2]
In the structure information storage unit 2, the personnel (group head) whose first boss is N / A is excluded from the ID extracted in the rule 1.

  As a rule, rule 1 is a rule that lists all personnel in the first group as candidates for transfer.

  Rule 2 is a rule for excluding the group length from the transfer candidate list extracted in rule 1. When the group head is selected as a transfer candidate, it is provided so that the group tree after the group head transfer is not fixed.

  In addition to the above two rules, it is also possible to narrow down the transfer target list by applying any additional rules to the personnel list extracted in Rule 2, such as the score of a predetermined test for age and English. is there.

S305: Repeat START for the number of items in the list
The result display unit 12 repeats the processes from S306 to S310 for the number of lists generated in S304.

S306: Candidate Vector Generation The result display unit 12 drives the candidate information processing unit 4 to instruct reading of candidate information corresponding to the candidate ID acquired in S304. The route for performing the instruction is arbitrary, but is performed via the result storage unit 11, the relationship derivation unit 10, and the second post-processing unit 6, for example. The candidate information processing unit 4 reads the attribute information of the corresponding ID from the human information storage unit 1, converts the information into a vector format, and generates a candidate vector.

  The description about vectorization of candidate information is the same as the description about S103.

S307: Generation of Candidate Feature Vector The result display unit 12 drives the first post-processing unit 6 and instructs to read the candidate vector generated in S306. The route for performing the instruction is arbitrary, but is performed via the result storage unit 11 and the relationship deriving unit 10, for example. The first post-processing unit 6 performs feature extraction of candidate vectors and generates candidate feature vectors.

  The description about the feature extraction of the candidate vector is the same as the description about S104.

S308: First Group Gain Calculation The result display unit 12 calculates the gain of the first group when each person acquired in S304 is a transfer candidate. The gain is calculated by virtually constructing a group tree of the first group after the personnel change and adding up the weights of all branches of the group tree. A specific processing flow will be described later.

S309: Second Group Gain Calculation The result display unit 12 calculates the gain of the second group when each person acquired in S304 is a transfer candidate. The gain is calculated by searching a record that matches the ID of the transfer candidate in the ID string of the result storage unit 11 and acquiring the compatibility score of the record.

S310: Gain calculation of entire organization The result display unit 12 calculates a gain (A + B) obtained by adding the gain (A) of the first group calculated in S308 and the gain (B) of the second group calculated in S309. And the gain of the whole organization.

S311: Repeat END for the number of items in the list
The result display unit 12 repeats the processes from S306 to S310 for the number of lists generated in S304.

S312: Candidate narrowing down The result display unit 12 narrows down the personnel having the maximum gain of the entire organization calculated in S310 from among the personnel included in the transfer target list generated in S304 as effective transfer candidates.
The candidates may be sorted in descending order according to the gain of the entire organization calculated in S310, and the top N people may be narrowed down as transfer candidates.

  FIG. 17 is a conceptual diagram showing the process of calculating the gain of the first group in S308.

S30801: Read the group tree structure information of the first group The result display unit 12 reads the group tree structure information of the first group held in the structure information storage unit 2.

S30802: Gain ← 0 (gain is initialized to 0)
The result display unit 12 initializes the gain of the first group (hereinafter referred to as “gain”) to 0 when the personnel acquired in S304 is a transfer candidate.

S30803: Repeat START for the number of structural information
The result display unit 12 repeats the processes of S30804 to S308141 for the number of records of the group tree structure information of the first group acquired in S30801.

S30804: ID is a transfer candidate?
The result display unit 12 checks whether or not the ID string of the group tree structure information of the first group acquired in S30801 matches the transfer candidate ID.

  (YES) If they match, the result display unit 12 proceeds to S30805.

  (NO) If not matched, the result display unit 12 proceeds to S30808.

S30805: ID ← Change candidate, superior ID ← first superior The result display unit 12 uses the ID column value as the ID and the first superior ID column value as the superior in the group tree structure information of the first group acquired in S30801. Set each ID.

S30806: Search for compatibility score from learning information storage unit 8 The result display unit 12 reads from the learning information storage unit 8 a compatibility score that matches both the ID set in S30805 and the supervisor ID. Here, it is assumed that the compatibility between the candidate and the candidate's superior in the first group has already been evaluated and provided as learning information. If the compatibility is not required, for example, it is necessary to obtain a compatibility score from others or to determine the compatibility score by the compatibility prediction step.

S30807: Gain ← Gain 0-compatibility score (subtract the compatibility score from gain)
The result display unit 12 acquires the value (0 or 1 value) of the affinity score string of the information searched in S30806, and subtracts the value from the gain. This is a decrease in group performance due to the absence of transfer candidates from the first group.

S30808: Is the first boss ID a transfer candidate?
The result display unit 12 checks whether or not the first boss ID column of the group tree structure information of the first group acquired in S30801 matches the transfer candidate ID.

  (YES) If they match, the result display unit 12 proceeds to S30809.

  (NO) If not matched, the result display unit 12 proceeds to S30812.

S30809: ID ← Transfer candidate, superior ID ← second superior The result display unit 12 uses the ID column value as the ID and the second superior ID column value as the superior in the group tree structure information of the first group acquired in S30801. Set each ID.

S30810: Search for compatibility score from analysis result The result display unit 12 searches the result storage unit 11 for information that matches both the ID set in S30809 and the supervisor ID.

S30811: Gain <-gain + compatibility score (add compatibility score to gain)
The result display unit 12 acquires the value of the compatibility score string (the real value of 0.0 to 1.0) of the information searched in S30810, and adds the value to the gain. This is an increase in group performance due to the absence of transfer candidates from the first group.

S30812: ID ← candidate for change, superior ID ← change candidate The result display unit 12 uses the ID column value in which the ID column value is set as the superior ID column in the group tree structure information of the first group acquired in S30801. (That is, the candidate ID of the transfer candidate) is set as the ID, and the value of the ID string (that is, the ID of the transfer candidate) is set as the supervisor ID.

  If there are a plurality of transfer candidates, S30813 to S30814 described below are repeated for the number of candidates.

S30813: Search for compatibility score from learning information storage unit 8. The result display unit 12 searches the learning information storage unit 8 for a record that matches both the ID set in S30812 and the supervisor ID. Here, it is assumed that the compatibility between the candidate and the candidate's superior in the first group has already been evaluated and provided as learning information. If the compatibility is not required, for example, it is necessary to obtain a compatibility score from others or to determine the compatibility score by the compatibility prediction step.

S30814: gain <-gain-compatibility score (subtract the compatibility score from gain)
The result display unit 12 acquires the value (0 or 1) of the compatibility score string of the record searched in S30813, and subtracts the value from the gain.

  This is a decrease in the group performance due to the transfer of candidates from the first group.

S308152: Repeat END for the number of structural information
The result display unit 12 repeats the processes of S30804 to S308141 for the number of records of the group tree structure information of the first group acquired in S30801.
FIG. 18 is a conceptual diagram showing a change in group performance of the entire organization when the personnel whose IDs are P1 to P6 are changed. In this example, when a person whose ID is P5 is moved, the highest profit can be obtained as an organization person. Therefore, the person with ID P5 is the most promising candidate for the change.
[effect]
First, the information processing apparatus of this embodiment has the same effect as the information processing apparatus of the third embodiment.

  In addition to this, the information processing apparatus according to the present embodiment can obtain the profit of the entire organization including both groups related to the transfer of the first party from the compatibility between the superior and the subordinate.

  In addition to this, the information processing apparatus of the present embodiment can evaluate the transfer suitability of the first party from the benefits of the entire organization.

  Although the present invention has been described with reference to the preferred embodiments, the present invention is not necessarily limited to the above embodiments, and various modifications can be made within the scope of the technical idea.

Moreover, although a part or all of said embodiment can be described also as the following additional remarks, it is not restricted to the following.
(Appendix 1)
The determination of the conditions used when deriving the relationship between the first data including the feature amount of the information related to the first person and the second data including the feature amount of the information related to the second person Third data including the characteristic amount of information, fourth data including the characteristic amount of information relating to the certain person or a person other than the certain person, and the certain person and the certain person or a person other than the certain person A known relationship about and the steps to take from
Deriving and outputting a first data relationship, which is a relationship between the first data and the second data, according to the condition for which the determination has been performed;
An information processing method including:
(Appendix 2)
The information processing method according to appendix 1, wherein the first data relationship represents a potential relationship between the first person and the second person.
(Appendix 3)
The information processing method according to appendix 1 or 2, wherein the first data relationship represents compatibility between the first person and the second person.
(Appendix 4)
The information processing method according to any one of supplementary notes 1 to 3, wherein all of the first to fourth data are data of the same number of items, item contents, and order of items.
(Appendix 5)
The information described in any one of appendices 1 to 4, wherein the second person is a person who is scheduled to become the boss of the first person when the first person makes the change. Processing method.
(Appendix 6)
Converting the information related to the third person belonging to the certain group into third data including the characteristic amount of the information related to the third person;
Deriving a second data relationship, which is a relationship between the third data and the second data, according to the determined condition;
The information processing method according to any one of appendices 1 to 5, further including:
(Appendix 7)
The information processing method according to any one of supplementary notes 1 to 6, further comprising a step of evaluating a profit of the certain group from the first data relationship.
(Appendix 8)
The information processing method according to any one of supplementary notes 1 to 7, further comprising a step of evaluating a profit of the other group from the first data relationship.
(Appendix 9)
The information processing method according to any one of appendices 1 to 8, including a step of evaluating the first person from the first data relationship.
(Appendix 10)
The first person is a candidate for transfer from one group to another group, and the second person is a person belonging to the other group.
The information processing method described in any one of appendices 1 to 9.
(Appendix 11)
The information processing method according to appendixes 10 to 10, wherein both the certain group and the other group belong to the same upper group.
(Appendix 12)
The information processing method according to claim 11, further comprising a step of calculating a profit of the upper group from the first data relationship.
(Appendix 13)
The information processing method according to attachment 12, wherein the first person is evaluated from the profit of the upper group.
(Appendix 14)
The information described in any one of Supplementary notes 1 to 13, wherein the first data is data obtained by extracting the characteristics of the first person, and the second data is data obtained by extracting the characteristics of the second person. Processing method.
(Appendix 15)
An information processing method according to any one of supplementary notes 1 to 14, wherein the relationship is similarity.
(Appendix 16)
The information processing method according to any one of supplementary notes 1 to 15, wherein the data relationship is a relationship representing compatibility between the first person and the second person.
(Appendix 17)
The determination of the conditions used when deriving the relationship between the first data including the feature amount of the information related to the first person and the second data including the feature amount of the information related to the second person Third data including the characteristic amount of information, fourth data including the characteristic amount of information relating to the certain person or a person other than the certain person, and the certain person and the certain person or a person other than the certain person And a known processing relationship between
A relationship deriving unit for deriving a first data relationship, which is a relationship between the first data and the second data, according to the determined condition;
An information processing apparatus comprising:
(Appendix 18)
The determination of the conditions used when deriving the relationship between the first data including the feature amount of the information related to the first person and the second data including the feature amount of the information related to the second person Third data including the characteristic amount of information, fourth data including the characteristic amount of information relating to the certain person or a person other than the certain person, and the certain person and the certain person or a person other than the certain person A known relationship with respect to and processing to be performed from
A process of deriving a first data relationship, which is a relationship between the first data and the second data, according to the condition in which the determination has been performed;
Processing program for causing a computer to execute a process including:

001, 1 person information storage part 003 person information processing part 004, 10 relationship derivation part 005, 8 learning information storage part 006 learning processing part 007 relation output part 008 personnel post-processing part 1 person information storage part 2 structure information storage part 3 candidates Information processing unit 4 Boss information processing unit 5 First post-processing unit 6 Second post-processing unit 7 Relationship learning unit 8 Learning information storage unit 9 Learning result storage unit 10 Relationship deriving unit 11 Result storage unit 12 Result display unit 100, 101, 102 Information processing apparatus

Claims (10)

The determination of the conditions used when deriving the relationship between the first data including the feature amount of the information related to the first person and the second data including the feature amount of the information related to the second person Third data including the characteristic amount of information, fourth data including the characteristic amount of information relating to the certain person or a person other than the certain person, and the certain person and the certain person or a person other than the certain person Steps to take from a known relationship with
Deriving and outputting a first data relationship, which is a relationship between the first data and the second data, according to the condition for which the determination has been performed;
An information processing method including:   The information processing method according to claim 1, wherein the first data relationship represents a potential relationship between the first person and the second person.   The information processing method according to claim 1 or 2, wherein the first data relationship represents compatibility between the first person and the second person.   The information processing method according to claim 1, wherein all of the first to fourth data are data of the same number of items, item contents, and order of items. Converting the information related to the third party belonging to the certain group into third data including the characteristic amount of the third party;
Deriving a second data relationship, which is a relationship between the third data and the second data, according to the determined condition;
The information processing method according to any one of claims 1 to 3, further comprising:   The first person is a candidate for transfer from one group to another group, and the second person is a person belonging to the other group. Information processing method.   The information processing method according to claim 6, wherein the certain group and the other group belong to the same upper group.   The information processing method according to claim 7, further comprising a step of calculating a profit of the upper group from the first data relationship.   The information processing method according to claim 8, wherein the first person is evaluated from the profit of the upper group.   The information processing method according to claim 1, wherein the relationship is similarity.

Images