JP2014048801A - Statistic model generation device, statistic model generation method, and computer program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To efficiently create a statistic model capable of estimating what a degree of association he/she has toward others using a life log data of another person.SOLUTION: A statistic model used when estimating a degree of association is created by using teacher data representing the information of a person or the attribute of an object and a degree of association representing the strength of association with the person or the object. The degree of association is calculated by using verification data representing the information of the preset attribute and the degree of association, and the statistic model created in the creation step. The estimation accuracy which is a value representing a matched ratio of the degrees of association is calculated by comparing the degree of association set in the verification data and the degree of association calculated in the calculation step, and the accuracy of the statistic model is determined based on the estimation accuracy calculated in the estimation accuracy calculation step.

Description

  The present invention relates to a technique for estimating the strength of connection between people or things.

When providing communication support to people, technology to understand the relationship between people is important.
2. Description of the Related Art Conventionally, as a technique for grasping the relationship between people, there is a technology for finding a relationship between a person and another person from externally observable information such as sensors and GPS (Global Positioning System) and a reference relationship on the web. As a first technique, there is a technique in which a communication frequency is measured using a tag and a good relationship is determined when a lot of communication is performed (for example, see Non-Patent Document 1). As a second technique, there is a technique for evaluating intimacy by combining a function in the frequency of mail transmission / reception (see, for example, Non-Patent Document 2).

In addition, in web services, there is Spysee that uses semantic technology to find out the relationship between people on the web (see, for example, Non-Patent Document 3). Spysee is a service that can find connections between people by automatically collecting and organizing information published on the web, such as blogs and paper co-authorship relationships.
However, the person-to-person relationship that can be specified by the above technique is limited to a superficial relationship such as the relationship between the person and the person that can be specified from the reference relationship and the closeness of communication. For this reason, it has been difficult to apply the above technique to elderly people and children who are not good at improving human relations and communication in the workplace.

  Therefore, in order to solve the above problem, a technique for grasping the inner information of a person who is difficult to understand from the superficial observation information is required. Inner information has long been studied in the field of psychology. However, in order to use it in engineering, technology that connects information on the web, externally observable data that can be collected with sensors, GPS, etc., and internal information is required. In the present invention, internal information on how a person thinks of a person is called “degree of connection”. Estimate (estimate) the degree of connection using information on the web, information collected from sensors or GPS, and personal information such as the user's gender and age, as described in blogs and Twitter (registered trademark) The technique for this is demonstrated.

The degree of connection that expresses what a person thinks of a person depends largely on the subjectivity of the person. Therefore, in order to estimate the connection degree as accurately as possible, customization for each person is required to minimize individual differences.
Since the present invention is premised on customization by a statistical method, there are two methods for customizing the teacher data used for estimating the connection degree. The first method is a customization method using only one's own life log data. The second method uses another person's data up to a certain level and customizes the differences individually.

  First, the first method will be described. The life log data is information obtained by accumulating personal information such as an individual's behavior history, persons involved, and preferences. Since life log data is stored information for each individual, it is possible to determine how the individual is concerned about the other party. In the first method, when life log data is accumulated for a long time, analysis data is enormous and a lot of miscellaneous information is included. Therefore, it takes a lot of work for pre-processing and information organization for analyzing information of necessary connection degree. On the other hand, when life log data is accumulated for a short period of time, since the partner with whom the individual is involved is limited, sufficient teacher data cannot be obtained. For this reason, it is impossible to accurately estimate the degree of connection for an unknown partner who has never met an individual. Thus, it takes time and labor to use the life log data for each individual as the teacher data for customizing. Therefore, although the first method has existed as an idea for a long time, there are few people using it because there is a problem in convenience.

  Next, the second method will be described. The second method is not a customization using one's own life log data, but a method using other person's life log data having similar circumstances and preferences as teacher data. The second method assumes that even if it is impossible to generalize completely, it is possible to estimate the degree of connection by grouping individuals to some extent using attributes related to human characteristics and characteristics. It is a method that is based on. Therefore, unlike the life log, there is an advantage that it is not necessary to arrange a large amount of personal life log data. However, the second method uses other person's life log data as teacher data for customizing his / her life log data. For this reason, it is very difficult to classify what type of person he / she has the same degree of connection. Further, if the typing is failed, the estimation accuracy of the connection degree is only a low value.

That is, in the second method, the problem is how to create a group of the same type as that of the teacher data (clustering) and create a statistical model that can estimate the degree of connection.
Although it is possible to use an attribute that is said to be effective for clustering in the psychological field such as gender and age group, clustering from only that attribute can only estimate with low accuracy.

"" NK3-Human Relations Improvement Plan-Android OS-equipped smartphone app ", [online], [Search August 9, 2012], Internet (URL: http://nk3.nii.ac.jp/content/ intro.html) Kim Nare and two others, “Examination of Presence Service System to Support Informal Communication Between Acquaintances”, IPSJ Research Report, March 18, 2009, Vol. 33, p. 1-6 “SPYSEE”, [online], [Search August 13, 2012], Internet (URL: http://spysee.jp/help/index/)

  In view of the above circumstances, an object of the present invention is to provide a technique for efficiently creating a statistical model that can estimate what degree of connection a person has with another person using other person's life log data.

  One aspect of the present invention is to create a statistical model used for estimating a connection degree using teacher data representing attribute information of a person or a thing and a connection degree representing the strength of the connection of the person or the thing. Set in the verification data, a calculation unit for calculating the connection degree using the preset attribute information and the verification data representing the connection degree, and the statistical model generated by the preparation part An estimated accuracy calculation unit that compares the connection degree calculated with the connection degree calculated by the calculation unit and calculates an estimation accuracy that is a value representing a proportion of connection degrees, and the estimation accuracy calculation unit calculates And a determination unit that determines the accuracy of the statistical model created by the creation unit based on the estimation accuracy.

  One aspect of the present invention is the statistical model generation device described above, wherein the creation unit selects and selects an attribute having a high correlation with the connection degree from among a plurality of types of attributes set in advance in the teacher data. The statistical model is generated based on the attribute information.

  One aspect of the present invention is to create a statistical model used for estimating a connection degree using teacher data representing attribute information of a person or a thing and a connection degree representing the strength of the connection of the person or the thing. A creation step for calculating the connection degree using the attribute information set in advance and the verification data representing the connection degree, and the statistical model created in the preparation step, and the verification data In the estimation accuracy calculation step that compares the set connection degree and the connection degree calculated in the calculation step to calculate an estimation accuracy that is a value representing a proportion of the connection degree, and in the estimation accuracy calculation step And a determination step of determining the accuracy of the statistical model based on the calculated estimation accuracy.

  One aspect of the present invention is to create a statistical model used for estimating a connection degree using teacher data representing attribute information of a person or a thing and a connection degree representing the strength of the connection of the person or the thing. A creation step for calculating the connection degree using the attribute information set in advance and the verification data representing the connection degree, and the statistical model created in the preparation step, and the verification data In the estimation accuracy calculation step that compares the set connection degree and the connection degree calculated in the calculation step to calculate an estimation accuracy that is a value representing a proportion of the connection degree, and in the estimation accuracy calculation step A determination step of determining the accuracy of the statistical model based on the calculated estimation accuracy; Is a computer program.

  According to the present invention, it is possible to efficiently create a statistical model that can estimate what degree of connection a person has with a partner using life log data of another person.

It is a figure which shows the system configuration | structure of the connection degree estimation system in this invention. 3 is a schematic block diagram illustrating a functional configuration of a connection degree estimation device 30. FIG. It is a block diagram of attribute information DB. It is a flowchart which shows the flow of a process of statistical model preparation. It is a flowchart which shows the flow of a connection degree calculation process. It is a flowchart which shows the flow of AIC calculation processing. It is a flowchart which shows the flow of the process which calculates a correlation coefficient. It is a flowchart which shows the flow of a statistical model creation process. It is a block diagram of attribute information DB regarding a person and a thing.

Hereinafter, an embodiment of the present invention will be described with reference to the drawings. In the following description, it is assumed that the user itself is a cognitive person and a person other than the user is a recognized person. In addition, the attribute generally represents the nature and characteristics of what exists in the world. For example, attributes are the color and shape of an object, human ability, human identity, social relationships, and the like.
FIG. 1 is a diagram showing a system configuration of a connection degree estimation system according to the present invention.
The connection degree estimation system of the present invention includes an information input terminal 10, a sensing device 20, a connection degree estimation apparatus 30, and a display terminal 40.

The information input terminal 10 is configured using an information processing apparatus such as a personal computer. The information input terminal 10 is operated by a user (recognition person), and transmits a recognizer attribute, a recognized person attribute, and a connection degree to the connection degree estimation device 30. The cognitive attribute is information about the cognitive person and represents, for example, the gender and occupation of the cognitive person. The perceived person attribute is information about the perceived person, and represents, for example, the gender of the perceived person or the relationship with the perceived person. The degree of connection is information indicating the strength of connection between a person (recognition person) and a person (recognized person).
The sensing device 20 detects the behavior of the cognitive person or the cognitive person using the sensor, and measures information related to the emotions of the cognitive person or the cognitive person. The sensing device 20 may be, for example, an RFID (Radio Frequency IDentification) tag, a human sensor, a video or audio storage device, a physiological sensor, or the like. The sensing device 20 measures information regarding the behavior that has occurred between the cognitive person and the cognitive person as a life log attribute. Specifically, for example, the life log attribute is information such as experiences shared by the cognitive person and the cognitive person, emotions held by the cognitive person when experiencing the experience, busyness of the cognitive person, and poor physical condition.

The connection degree estimation device 30 estimates a connection degree representing the strength of connection between people. The connection degree estimation device 30 accumulates and manages the cognitive person attribute, the recognized person attribute, and the connection degree transmitted from the information input terminal 10 or the sensing device 20.
The display terminal 40 is configured using an information processing device such as a personal computer. The display terminal 40 displays the connection degree estimation result output from the connection degree estimation device 30.

FIG. 2 is a schematic block diagram illustrating a functional configuration of the connection degree estimation device 30.
The connection degree estimation device 30 includes a CPU (Central Processing Unit), a memory, an auxiliary storage device, and the like connected by a bus, and executes a connection degree estimation program. By executing the connection degree estimation program, the connection degree estimation device 30 is configured to include the information management unit 100, the information storage unit 110, the extraction unit 120, the creation unit 130, the connection degree calculation unit 140, the determination coefficient calculation unit 150, and the estimation accuracy calculation unit 160. , Functioning as a device including a determination unit 170 and an output unit 180. Further, the creation unit 130 functions as an analysis processing unit 131, an AIC selection unit 132, a correlation coefficient calculation unit 133, and a generation processing unit 134. Note that all or part of each function of the connection degree estimation device 30 may be realized using hardware such as an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA). good. Further, the connection degree estimation program may be recorded on a computer-readable recording medium. The computer-readable recording medium is, for example, a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, or a storage device such as a hard disk built in the computer system. The connection degree estimation program may be transmitted / received via a telecommunication line.

The information management unit 100 records the information on the cognitive person attribute and the person to be recognized attribute transmitted from the information input terminal 10 and the information on the life log attribute transmitted from the sensing device 20 in the information storage unit 110.
The information storage unit 110 is configured using a storage device such as a magnetic hard disk device or a semiconductor storage device. The information storage unit 110 stores an attribute information DB having information on a cognitive person attribute, a recognized person attribute, a life log attribute, and a connection degree (hereinafter referred to as “attribute information”) output from the information management unit 100. In addition, a cognitive person attribute, a to-be-recognized person attribute, and a life log attribute are each one aspect | mode of an attribute.

The extraction unit 120 extracts necessary information from the attribute information DB stored in the information storage unit 110, and generates teacher data and verification data.
The creating unit 130 creates a statistical model used when estimating the connection degree. Hereinafter, a specific configuration of the creation unit 130 will be described.
The analysis processing unit 131 performs analysis of variance for each combination of one or more attributes (hereinafter referred to as “attribute group”) using the teacher data extracted by the extraction unit 120. The analysis processing unit 131 extracts an attribute group exceeding a predetermined significance level using the F test table.

  The AIC selection unit 132 calculates an AIC value using AIC (Akaike's Information Criterion) for each piece of attribute information stored in the information storage unit 110 or attribute information of teacher data. The AIC value is a value calculated using the AIC. The AIC selection unit 132 selects one AIC value effective for estimating the connection degree from the plurality of calculated AIC values. For example, the AIC selection unit 132 selects a minimum AIC value from a plurality of AIC values.

The correlation coefficient calculation unit 133 clusters data with the attribute group extracted by the analysis processing unit 131, and the degree of connection between the clustered data and the attribute group with the minimum AIC value selected by the AIC selection unit 132 for each clustered data The correlation coefficient is calculated.
The generation processing unit 134 creates a statistical model by performing regression analysis on an attribute group in which the correlation coefficient calculated by the correlation coefficient calculation unit 133 is greater than or equal to a threshold value.

The connection degree calculation unit 140 calculates a connection degree value based on the statistical model created by the generation processing unit 134 using the verification data extracted by the extraction unit 120.
The determination coefficient calculation unit 150 calculates the determination coefficient R ^ 2 of the statistical model created by the generation processing unit 134 ("^" represents a power). The determination coefficient is an index indicating the high accuracy of the statistical model generated by the regression analysis. For example, the determination coefficient R ^ 2 may be calculated in the range of “0” to “1” and defined as follows. The determination coefficient R ^ 2 is “0.8” or more and represents that the accuracy is very good. The determination coefficient R ^ 2 is “0.5” or more, indicating that the accuracy is good. The determination coefficient R ^ 2 indicates that the accuracy is normal at "0.25" or more. The determination coefficient R ^ 2 is less than “0.25” and indicates that the accuracy is not good.

The estimation accuracy calculation unit 160 calculates an average value EAav of estimation accuracy that represents a ratio between the connection degree value calculated by the connection degree calculation unit 140 and the connection degree value of the verification data. For example, the average value EAav of the estimation accuracy may be defined as follows. The average value EAav of the estimation accuracy indicates that the rate of coincidence at “0.8” or higher is very good. The average value EAav of the estimation accuracy indicates that the ratio of matching is good at “0.5” or more.
The determination unit 170 determines whether the determination coefficient R ^ 2 calculated by the determination coefficient calculation unit 150 and the average value EAav of the estimation accuracy calculated by the estimation accuracy calculation unit 160 are equal to or greater than a threshold value.
The output unit 180 outputs the connection degree estimation result to the display terminal 40.

FIG. 3 is a configuration diagram of the attribute information DB.
The attribute information DB has a plurality of records 50. One attribute information is defined in one record 50. The attribute information defined in the record 50 has values of data number, cognitive person attribute, cognitive person attribute, other attributes, and connection degree.
The data number represents an identification number assigned to each record 50.

  Specific examples of the cognitive attributes defined in the record 50 include attributes such as a principal ID, a principal sex, and a principal occupation. The principal ID represents the identification number of the cognitive person represented by the record 50. The identity gender represents the gender of the cognitive person represented by the record 50. The principal occupation represents the occupation of the cognitive person represented by the record 50.

Specific examples of the recognized person attributes defined in the record 50 include attributes such as a partner ID, a partner sex, and a relationship with the recognizer.
The partner ID represents the identification number of the recognized person represented by the record 50. The opponent gender represents the gender of the recognized person represented by the record 50. The value of the relationship with the cognitive person represents the relation between the cognitive person represented by the record 50 and the recognized person.
The other attributes represent attribute information other than the recognizer attribute and the recognized person attribute. The other attribute may be a life log attribute, for example.

The connection degree value represents the strength of the connection between the cognitive person and the person to be recognized represented by the record 50. The connection degree value defined in the record 50 may be arbitrarily determined by the user. The value of the connection degree indicates that the larger the positive value, the higher the connection degree between the cognitive person and the recognized person.
For example, in FIG. 3, the value of the connection degree is represented in seven stages from “−3” to “+3”. That is, the value of the degree of connection from “−3” to “−1” represents that the larger the negative value having the larger absolute value, the worse the relationship between the cognitive person and the recognized person. A connection degree value of “0” indicates that the cognitive person and the cognitive person have a relationship like others. The value of the degree of connection from “1” to “3” indicates that the larger the positive value, the better the relationship between the cognitive person and the recognized person.

FIG. 4 is a flowchart showing the flow of processing for preparing a statistical model.
First, the information management unit 100 inputs attribute information transmitted by the user (step S101). The information management unit 100 stores the input attribute information in the attribute information DB (step S102). The determination unit 170 determines whether there is a predetermined number or more of attribute information in the attribute information DB stored in the information storage unit 110 (step S103).

  When attribute information more than a predetermined number exists in attribute information DB (step S103-YES), the extraction part 120 extracts attribute information at a predetermined ratio from attribute information DB at random, and supervises teacher data and verification data. Generate (step S104). The determination unit 170 determines whether or not the extraction process has been repeated a predetermined number of times (step S105). When the process of step S104 is repeated a predetermined number of times (step S105-YES), the process ends.

On the other hand, in the process of step S103, when attribute information more than a predetermined number does not exist in the attribute information DB (step S103-NO), the connection degree estimation device 30 stands by until the attribute information is input. When the attribute information is input, the connection degree estimation device 30 executes the processes after step S101 (step S106).
In the process of step S105, when the process of step S104 is less than the predetermined number of times (step S105-NO), the extraction unit 120 repeatedly executes the process of step S104.

FIG. 5 is a flowchart showing the flow of the connection degree calculation process.
First, in the teacher data extracted by the extraction unit 120, the value of the connection degree is set as the objective variable, and the value of each attribute is set as the explanatory variable. The analysis processing unit 131 performs an analysis of variance for each attribute group of explanatory variables (step S201). The analysis processing unit 131 uses the F test table to extract an attribute group that exceeds a predetermined significance level. (Step S202).

FIG. 6 is a flowchart showing the flow of AIC calculation processing.
The AIC selection unit 132 calculates an AIC value using the AIC for each attribute group of the attribute information DB stored in the information storage unit 110. (Step S301). The AIC selection unit 132 selects the minimum AIC value from the calculated plurality of AIC values (step S302).

FIG. 7 is a flowchart showing a flow of processing for calculating a correlation coefficient.
The correlation coefficient calculation unit 133 clusters data with the attribute group extracted by the analysis processing unit 131, and the degree of connection between the clustered data and the attribute group with the minimum AIC value selected by the AIC selection unit 132 for each clustered data The correlation coefficient is calculated. (Step S401). The determination unit 170 determines whether the calculated correlation coefficient is greater than or equal to a threshold value (step S402). If the correlation coefficient is equal to or greater than the threshold (step S402—YES), the determination unit 170 determines that there is a correlation, and stores an attribute group having the correlation in the attribute information DB (step S403).

The determination unit 170 determines whether correlation coefficients have been calculated for all attribute groups extracted by the analysis processing unit 131 (step S404). When the correlation coefficient is calculated for all the attribute groups (step S404—YES), the process ends.
In the process of step S402, when the correlation coefficient is less than the threshold (step S402-NO), the determination unit 170 determines that there is no correlation, and proceeds to the process of step S404.
When the correlation coefficient is not calculated for all the attribute groups in the process of step S404 (step S404-NO), the correlation coefficient calculation unit 133 returns to step S401, and the attribute group for which the correlation coefficient has not been calculated. A correlation coefficient is calculated for.

FIG. 8 is a flowchart showing the flow of statistical model creation processing.
The generation processing unit 134 reads all the attribute groups determined to have a correlation in the determination unit 170 from the attribute information DB (step S501). The generation processing unit 134 performs a linear multiple regression analysis for each attribute group read from the attribute information DB, and creates a statistical model (step S502). The connection degree calculation unit 140 calculates the connection degree for each statistical model created by the generation processing unit 134 using each verification data extracted by the extraction unit 120 (step S503).
The estimation accuracy calculation unit 160 calculates the connection degree value calculated by the connection degree calculation unit 140 and the connection degree value defined in advance in the verification data used by the connection degree calculation unit 140 to calculate the connection degree. An estimation accuracy EA representing the matching ratio is calculated (step S504). The estimation accuracy calculation unit 160 performs the process of step S504 using each verification data extracted by the extraction unit 120. The estimation accuracy calculation unit 160 calculates an average value EAav of estimation accuracy based on Equation 1 (step S505). In Equation 1, x represents the number of verification data. Further, each value (EA1, EA2,..., EAx) represented in Expression 1 represents a value of the estimated accuracy EA calculated based on each verification data. For example, the estimation accuracy calculated based on the nth (n = 1 to x) integer verification data is represented as “EAn”.

Further, the determination coefficient calculation unit 150 calculates the determination coefficient R ^ 2 for each statistical model based on Expression 2 (step S506).

S1 is calculated based on Equation 3. S2 is calculated based on Equation 4.

Y1 of Formula 3 is a value of the connection degree of verification data. Y2 in Equation 4 is a connection degree value calculated by the statistical model using the verification data. yav is the average value of the connection degrees of the verification data extracted by the extraction unit 120.
The determination unit 170 determines whether or not both of the statistical model determination coefficient R ^ 2 calculated by the determination coefficient calculation unit 150 and the average value EAav of the estimation accuracy calculated by the estimation accuracy calculation unit 160 are equal to or greater than a threshold value. Is determined (step S507).

  When both of the determination coefficient R ^ 2 and the average value EAav of the estimation accuracy are equal to or greater than the threshold (step S507-YES), the determination unit 170 can represent the statistical model created by the generation processing unit 134 by a linear equation. (Appropriate statistical model). The determination unit 170 stores the statistical model determined to be appropriate in step S507, the statistical model determination coefficient R ^ 2, and the statistical model estimation accuracy average value EAav in the attribute information DB (step S508).

  On the other hand, in the process of step S507, when one or both of the determination coefficient R ^ 2 and the average value EAav of the estimation accuracy are less than the threshold (step S507-NO), the determination unit 170 generates The statistical model created by the processing unit 134 is determined as a statistical model that cannot be expressed by a linear equation (an inappropriate statistical model). Note that different thresholds are set for the determination coefficient R 2 and the average value EAav of the estimation accuracy.

  The connection degree estimation device 30 configured as described above creates a statistical model that can calculate the subjective connection degree by three-stage processing with different calculation amount weights. The connection degree estimation device 30 performs processing in order from processing with a lighter weight of calculation amount, and reduces the number of attribute candidates applied to the statistical model each time. The connection degree estimation device 30 selects attribute information correlated with the connection degree by performing such processing. Therefore, the connection degree estimation device 30 can efficiently create a statistical model that can calculate the subjective connection degree.

  As specific processing, for example, the following three-stage processing is performed. As a first step, the connection degree estimation device 30 extracts an attribute group at a significant level using the F test table. As a second stage, the connection degree estimation device 30 selects an attribute group having a correlation with the connection degree using the AIC. As a third stage, the connection degree estimation device 30 selects an appropriate statistical model using the determination coefficient R 2 and the average value EAav of the estimation accuracy. Through the above-described three-stage processing, the connection degree estimation device 30 can efficiently create a statistical model used for estimating the subjective connection degree.

  The connection degree estimation device 30 uses the determination coefficient R ^ 2 and the average value EAav of the estimation accuracy for the created statistical model, and the connection degree estimation device 30 does not require customization for each individual when creating the statistical model. Therefore, the connection degree estimation device 30 can create a statistical model that can efficiently calculate the subjective connection degree.

The connection degree estimation device 30 selects an attribute having a high correlation with the connection degree. Therefore, the connection degree estimation device 30 can create a statistical model with a smaller number of attributes. Therefore, while maintaining the estimation accuracy of the connection degree, it is possible to limit the number of attributes provided to the cognitive person and realize privacy protection.
The statistical model generated by the connection degree estimation device 30 configured as described above is expressed by a linear function. Therefore, it is possible to shorten the time for estimating the connection degree as compared with a method using state transition probabilities such as Bayes and Markov models.

<Modification>
The information input terminal 10 may be configured to display the connection degree estimation result.
A cognitive person attribute does not need to be limited to the attribute of this embodiment, and may be any information as long as it is information about the cognitive person. The perceived person attribute need not be limited to the attribute of the present embodiment, and may be any information as long as it is information about the perceived person.
As long as the process of step S202 is a method which can extract the attribute group more than a significance level, methods other than F test may be used. For example, an attribute group having a significance level or higher may be extracted using a T test. Other methods may also be adopted.

  In the record 50 registered in the attribute information DB, it is not always necessary to register values for all attributes. For example, there may be a record 50 in which values are not registered for some attributes. In such a situation, the analysis processing unit 131 may be configured not to select an attribute group including attributes whose values are not registered in the attribute information DB by a predetermined number or more, such as analysis of variance. good. If a statistical model is created using an attribute with a small number of registered values, the accuracy may be reduced. Therefore, the accuracy of the statistical model can be maintained by adopting the above-described configuration. Furthermore, since the number of attribute groups to be subjected to selection processing such as analysis of variance is reduced, a statistical model can be generated more efficiently.

  The degree of connection estimated by the connection degree estimation device 30 need not be limited to the strength of the connection between people. The connection degree estimation device 30 may be configured to estimate the strength of connection between a person and a thing or the strength of connection between a thing and a thing.

  In the present embodiment, the perceived person attribute may be limited to “a person related to the workplace”. In this case, the value of the degree of connection represents the feeling of “comfort-irritated” by the cognitive person to the perceived person in seven levels from “−3” to “+3”. The value of the degree of connection from “−3” to “−1” indicates that the greater the absolute value, the more angry the cognitive person is with respect to the person being recognized. A connection degree value of “0” indicates that the cognitive person and the cognitive person have a relationship like others. The value of the degree of connection from “1” to “3” represents that the larger the value, the easier the cognitive person is with respect to the perceived person.

  Such a degree of connection may be estimated based on, for example, a statistical model generated using only the following four attributes. Emotions (regression coefficient: -0.41 to 0.49) held by past experiences with the recognized person, type of experience (regression coefficient: 0 to 0.78), frequency of e-mail (regression coefficient: 0.22), attributes of the cognitive age group (regression coefficient: -0.29). As a result, it is possible to estimate about 60% of the degree of connection between the people related to the workplace as “smooth-irritated” by using only four attributes. The regression coefficient is an index that represents the relationship between the objective variable (degree of connection) and the explanatory variable in numerical form.

  In the present embodiment, the perceived person attribute may be limited to “a person related to the workplace”. In this case, the value of the degree of connection represents the feeling of “respect-contempt” for the cognitive person in seven levels from “−3” to “+3”. The value of the degree of connection from “−3” to “−1” represents that the cognitive person despises the recognized person as the negative value having a large absolute value increases. A connection degree value of “0” indicates that the cognitive person and the cognitive person have a relationship like others. The value of the connection degree from “1” to “3” represents that the greater the positive value, the more the recognizer respects the person to be recognized.

  Such a degree of connection may be estimated based on, for example, a statistical model generated using only the following five attributes. Emotions held by past experiences with the perceived person (regression coefficient: -0.34 to 0.38), type of experience (regression coefficient: 0 to 0.68), and recognition person's experience when experiencing Attributes of status (regression coefficient: -0.37 to 0.07), e-mail frequency (regression coefficient: 0.14), and telephone frequency (regression coefficient: -0.18). As a result, it is possible to estimate the degree of connection of “respect-contempt” for people in the workplace by about 70% by using only five attributes.

FIG. 9 is a configuration diagram of an attribute information DB related to people and things.
The attribute information DB has a plurality of records 60. The attribute information defined in the record 60 has values of data number, cognitive person attribute, cognitive person attribute, and connection degree.
The data number represents the identification number of the attribute information represented by the record 60.
Specific examples of the cognitive attributes defined in the record 60 include attributes such as a principal ID, a principal sex, and a principal occupation.

The principal ID represents the identification number of the cognitive person represented by the record 60. The identity gender represents the gender of the cognitive person represented by the record 60. The principal occupation represents the occupation of the cognitive person represented by the record 60.
Specific examples of the recognized person attributes defined in the record 60 include each attribute such as an object ID, an object genre (name), and a relationship with the recognizer.

The object ID represents the identification number of the recognized person represented by the record 60. The genre (name) of the object represents the name of the object represented by the record 60. The value of the relationship with the cognitive person represents the relation between the cognitive person represented by the record 60 and the recognized person.
The connection degree value represents the strength of the connection between the cognitive person and the person to be recognized represented by the record 60. The connection degree value may be arbitrarily determined by the user. The connection degree value indicates that the larger the positive value, the better the relationship between the cognitive person and the cognitive person.

  The embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to this embodiment, and includes designs and the like that do not depart from the gist of the present invention.

DESCRIPTION OF SYMBOLS 10 ... Information input terminal, 20 ... Sensing device, 30 ... Connection degree estimation apparatus, 40 ... Display terminal, 100 ... Information management part, 110 ... Information storage part, 120 ... Extraction part, 130 ... Creation part, 131 ... Analysis processing part , 132 ... AIC selection unit, 133 ... correlation coefficient calculation unit, 134 ... generation processing unit, 140 ... connection degree calculation unit, 150 ... determination coefficient calculation unit (calculation unit), 160 ... estimation accuracy calculation unit, 170 ... determination unit , 180 ... output section

Claims (4)

A creation unit that creates a statistical model to be used when estimating the connection degree using teacher data representing the attribute information of the person or the object and the connection degree indicating the strength of the connection of the person or the object,
A calculation unit for calculating a connection degree using the preset attribute information and the verification data representing the connection degree, and the statistical model created by the creation unit;
An estimation accuracy calculation unit that calculates an estimation accuracy that is a value representing a proportion of matching degrees by comparing the connection degree set in the verification data and the connection degree calculated by the calculation unit;
A determination unit that determines the accuracy of the statistical model created by the creation unit based on the estimation accuracy calculated by the estimation accuracy calculation unit;
A statistical model generation apparatus comprising:   The creation unit selects an attribute having a high correlation with a connection degree among a plurality of types of attributes preset in the teacher data, and generates the statistical model based on information on the selected attribute. 1. The statistical model generation device according to 1. A creation step of creating a statistical model for use in estimating the connection degree using teacher data representing the attribute information of the person or the object and the connection degree representing the strength of the connection of the person or the object,
A calculation step for calculating a connection degree using the preset attribute information and the verification data indicating the connection degree, and the statistical model created in the creation step;
An estimation accuracy calculation step of calculating an estimation accuracy which is a value representing a proportion of connection degrees by comparing the connection degree set in the verification data and the connection degree calculated in the calculation step;
A determination step of determining the accuracy of the statistical model based on the estimation accuracy calculated in the estimation accuracy calculation step;
A statistical model generation method comprising: A creation step of creating a statistical model for use in estimating the connection degree using teacher data representing the attribute information of the person or the object and the connection degree representing the strength of the connection of the person or the object,
A calculation step for calculating a connection degree using the preset attribute information and the verification data indicating the connection degree, and the statistical model created in the creation step;
An estimation accuracy calculation step of calculating an estimation accuracy which is a value representing a proportion of connection degrees by comparing the connection degree set in the verification data and the connection degree calculated in the calculation step;
A determination step of determining the accuracy of the statistical model based on the estimation accuracy calculated in the estimation accuracy calculation step;
A computer program for causing a computer to execute.

Images