JP2019095968A - Impression estimation model learning device, impression estimation device, impression estimation model learning method, impression estimation method, and program - Google Patents

Abstract

To provide a method or the like which stably estimates an average impression value.SOLUTION: A learning data storage unit 14 stores learning data, a correct answer value of an impression value, and an evaluator in association with each other. An evaluator tendency data storage unit 15 stores the evaluator and an evaluator tendency in association with each other. Impression estimation units 11-1 to 11-N estimate posterior probability of an impression value using an impression estimation model, which estimates posterior probability of an impression value from learning data, for each group obtained by classifying the evaluator tendency. A weight estimation unit 12 estimates weight using a weight estimation model which estimates weight from the evaluator tendency. A weight estimation model update unit 131 updates a weight estimation model from a difference between posterior probability of an impression value weighted by the weight and a correct answer value of an impression value. An impression estimation model update unit 132 updates an impression estimation model with the weight as learning weight from posterior probability of an impression value.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a technique for automatically estimating an impression of data.

  There is a need for a technique for automatically estimating the impression of data such as voice and images. For example, when the impression (brightness, lightness, etc.) of music or an image can be estimated, an impression value of the data can be automatically given and used for data search. In addition, if it is possible to estimate the impression (naturalness, affinity, etc.) of the call center operator, it will be possible to automatically evaluate the operator's response quality.

Automatic estimation of an impression of data using machine learning can be realized by manually creating data to which an impression value is given in advance and learning a machine learning model in which the data and the impression value are linked. At this time, for the impression value given manually, the evaluation criteria differ depending on the person, so when learning a machine learning model, for example, it is necessary to prepare learning data as described below (Non-Patent Document 1) reference).
1. Label the impression values (several evaluations) manually with multiple people.
2. If necessary, round several levels of impression values into two or three levels of labels.
3. In order to exclude data which is blurred by human beings from correct data, only labels which are matched by a plurality of persons are selected (if three or more persons are given, selection of labels by majority rule may be used).

  To learn a model that estimates stable impression values, select an assessor who can assign impression values stably (for example, an assessor who can assign the same impression value when listening to the same data on different days) Also, in order to apply the impression value while stabilizing the impression value, it is necessary to perform the impression value assignment work while taking a break (see Non-Patent Document 2).

Masaki Suzaka et al., "Construction of a natural speech database for emotion / emotion discrimination", Transactions of Information Processing Society of Japan, vol. 52, No. 3, pp. 1185-1194, 2011 Takamiya Takamiya et al., “Overview of Impression Evaluation Data”, [online], [Search on November 1, 2017], Internet <URL: http://pj.ninjal.ac.jp/corpus_center/csj/manu- f / impression.pdf>

  In order to learn a stable impression estimation model, it is necessary to select a stable evaluator, perform long-term evaluation with a break, and exclude blurry data, so the cost is extremely high. . Also, if an evaluator is selected to a certain extent, an impression estimation model specialized in the tendency of several evaluators is learned, and the estimated impression value is the impression value of the average sense of the user It may be different.

  The object of the present invention is to stably estimate an average impression value while taking into consideration the tendency of the evaluator's impression evaluation without selecting the evaluator and excluding the data in view of the above-described points. It is to realize the impression estimation technology that can be done.

  In order to solve the above problems, the impression estimation model learning device according to the first aspect of the present invention is an evaluation in which the learning data and the correct value of the impression value given to the learning data and the correct value of the impression value are given. Evaluator tendency data storage that stores an evaluator tendency database that associates a learning data storage unit that stores a learning database that associates a person with a person, and an evaluator tendency that represents an appearance probability of each evaluator with an evaluator An impression estimation model storage unit for storing an impression estimation model for estimating a posterior probability of an impression value from learning data; a weight estimation model storage unit for storing a weight estimation model for estimating weights from an evaluator tendency; An impression estimation unit that estimates a posteriori probability of an impression value from learning data using an impression estimation model for each of the groups into which the tendency has been classified, and is associated with the learning data using a weight estimation model Weight estimation model update that updates the weight estimation model from the difference between the weight estimation unit that estimates the weight from the evaluated evaluator tendency, and the posterior probability of the impression value weighted by the weight and the correct value of the impression value given to the learning data And an impression estimation model updating unit for updating an impression estimation model using learning weights from the posterior probability of impression values.

  In order to solve the above problems, an impression estimating apparatus according to a second aspect of the present invention includes an impression estimating model storage unit that stores an impression estimating model learned by the impression estimating model learning apparatus according to the first aspect; A weight estimation model storage unit for storing a weight estimation model learned by the impression estimation model learning device of the aspect, and an impression estimation model for each group into which an evaluator tendency indicating an appearance probability for each impression value in the evaluator is classified An impression estimation unit that estimates a posterior probability of an impression value from input data, a weight estimation unit that estimates a weight from an evaluator tendency input using a weight estimation model, and integration that is a posterior probability of an impression value weighted by a weight And an integrated value estimation unit for obtaining an impression value of input data based on the posterior probability.

  According to the impression estimation technique of the present invention, it is not necessary for all the evaluators to give an impression value to all the data of the learning data, so selection of evaluators and long-term restraint is not necessary, and selection of labels is unnecessary. Thus, the cost of creating learning data can be reduced. This makes it possible to stably estimate an average impression value for impression values that differ according to a person. That is, according to the impression estimation technology of the present invention, it is possible to stably estimate the average impression value while taking into consideration the tendency of the evaluator's impression evaluation without selecting the evaluator or excluding the data. it can.

FIG. 1 is a diagram illustrating a functional configuration of an impression estimation model learning device. FIG. 2 is a diagram illustrating the processing procedure of the impression estimation model learning method. FIG. 3 is a diagram illustrating a functional configuration of the impression estimating device. FIG. 4 is a diagram illustrating the processing procedure of the impression estimation method.

  The points of the present invention are as follows. First, the blur of the impression value by the evaluator is automatically classified into a group for each tendency by a plurality of impression estimation models. Next, weights are assigned according to the tendency of blur of the impression value by the evaluator to the result of automatically classifying the estimation result of the impression value, and the result is integrated to estimate the final impression value. In this way, an average blur of evaluation by the evaluator is given to the estimation result of the impression value to estimate the average impression value.

  Hereinafter, embodiments of the present invention will be described in detail. In the drawings, components having the same functions are denoted by the same reference numerals and redundant description will be omitted.

  In the embodiment of the present invention, an impression estimation model learning device for learning an impression estimation model and a weight estimation model using learning data and evaluator tendency data prepared in advance, a learned impression estimation model and a weight estimation model Can be roughly divided into an impression estimation device that estimates an impression value of input data using. Hereinafter, the functional configuration of the impression estimation model learning device and the impression estimation device of the embodiment and the details of the processing procedure will be described in order.

  The symbol “^” or the like used in the text should be written directly above the character immediately after it, but it is written immediately before the character due to the restriction of the text notation. In mathematical expressions, these symbols are described at their original positions, that is, directly above the characters.

<Impression Estimation Model Learning Device>
As shown in FIG. 1, the impression estimation model learning device 1 according to the embodiment includes N (≧ 2) impression estimation units 11-n (n = 1, 2,..., N), a weight estimation unit 12, A model learning unit 13, a learning data storage unit 14, an evaluator trend data storage unit 15, and a weight estimation model storage unit 16 are provided. Each impression estimation unit 11-n includes an impression estimation model storage unit 111-n. The model learning unit 13 includes a weight estimation model updating unit 131 and an impression estimation model updating unit 132. The impression estimation model learning method of the embodiment is realized by the impression estimation model learning device 1 performing the processing of each step illustrated in FIG. 2.

  The impression estimation model learning device 1 is configured by reading a special program into a known or dedicated computer having, for example, a central processing unit (CPU: Central Processing Unit), a main memory (RAM: Random Access Memory), etc. Special equipment. The impression estimation model learning device 1 executes each process, for example, under the control of the central processing unit. The data input to the impression estimation model learning device 1 and the data obtained by each process are stored, for example, in the main storage device, and the data stored in the main storage device is read out to the central processing unit as necessary. Used for other processing. At least a part of each processing unit of the impression estimation model learning device 1 may be configured by hardware such as an integrated circuit. Each storage unit included in the impression estimation model learning device 1 is, for example, a main storage device such as a random access memory (RAM), an auxiliary storage device including a semiconductor memory element such as a hard disk, an optical disk, or a flash memory. Or can be configured by middleware such as a relational database or key value store.

  The learning data storage unit 14 stores a learning database in which learning data for learning an impression estimation model is accumulated. An example of a learning database is shown in Table 1. For the learning database, as shown in Table 1, learning data x (j), an impression value y (j), and an evaluator number i (for data number j (j = 0, 1, ..., J-1) j) are associated. Here, J is the total number of learning data.

  The learning data x (j) indicates data such as an image or voice. Since a plurality of evaluators assign impression values to one learning data x (j), a plurality of learning data x (j) having the same value exist in the learning database (for example, j = 0 in Table 1) And j = 1, j = 2, j = 3 and j = 4 etc.). The learning data x (j) is, for example, a known feature extracted from image data or voice data.

  The impression value y (j) is a manually applied impression value, that is, the correct value of the impression value. The impression value y (j) is a vector having a dimension of the number of stages to be evaluated, and is a One-hot vector in which the dimension of the evaluated value is 1 and the other dimensions are 0. For example, when "brightness" is rated in 5 steps of "very dark: 1", "dark: 2", "normal: 3", "bright: 4", "very bright: 5", "very dark: 1" When it is evaluated as "1, 0, 0, 0, 0", when it is evaluated as "dark: 2", when it is evaluated as [0, 1, 0, 0, 0], and when it is evaluated as "usual: 3" When [0, 0, 1, 0, 0] and “bright: 4” are evaluated, [0, 0, 0, 1, 0], and when “very bright: 5” is evaluated, [0, 0 , 0, 0, 1].

  The evaluator number i (j) is an identifier (ID) indicating the evaluator who assigned the impression value y (j) to the learning data x (j). The evaluator number i (j) is a number associated with the evaluator number i of the evaluator tendency database stored in the evaluator tendency data storage unit 15.

  In ordinary impression estimation, all evaluators assign impression values to certain data, and the average value is estimated as a label of learning data, and learning is performed by eliminating non-matching impression values. However, practically all evaluators label all data, and obtaining a sufficient amount of learning data requires enormous cost. Therefore, in the present embodiment, one or more evaluators assign an impression value to certain data, and learns by giving the evaluator's tendency as auxiliary information. This makes it possible to learn the impression value without requiring the evaluation of all the data.

  The evaluator tendency data storage unit 15 stores an evaluator tendency database in which evaluator tendencies, which are auxiliary information given when learning an impression estimation model, are accumulated. An example of the evaluator tendency database is shown in Table 2. In the evaluator tendency database, the evaluator tendency q (i, k) (k = 0, 1, ..., K-1) is associated with the evaluator number i (i = 0, 1, ..., I-1). It is done. Here, I is the total number of evaluators, and K is the number of types of evaluation values.

  The evaluator tendency q (i, k) is a vector representing to what stage the evaluator i gives the impression value (in other words, the appearance probability for each impression value k in the evaluator i). For example, when "brightness" is evaluated in 5 levels of "very dark: 1", "dark: 2", "normal: 3", "bright: 4", "very bright: 5", i = 0 "Very dark: 1%" rating 1%, "dark: 2" rating 2%, "normal: 3" rating 30%, "bright: 4" rating 58%, "very Bright: The evaluation of “5” is evaluated with a distribution of 9%, and the evaluator tendency q (i, k) at that time is a vector such as [1%, 2%, 30%, 58%, 9%] .

  Hereinafter, with reference to FIG. 2, the impression estimation model learning method which the impression estimation model learning apparatus 1 of embodiment performs is demonstrated.

  In step S11, the impression estimation unit 11-n uses the impression estimation model n stored in the impression estimation model storage unit 111-n to determine the posterior probability p (j, n) of the impression value for the learning data x (j). Calculate and output. The impression estimation model n is a model for estimating the posterior probability of an impression value from input data, and is a model in which the tendency of evaluation is learned. For example, if there is a group of evaluators who evaluate a certain impression with {4, 5}, the group of evaluators who rate with {3, 4, 5} or evaluation with {1, 2, 3, 4, 5} There is also a group of evaluators who At this time, the impression estimation unit 11-1 estimates an impression with {4, 5}, the impression estimation unit 11-2 estimates an impression with {3, 4, 5}, and the impression estimation unit 11-3 performs an impression. Each impression estimating unit 11-n performs impression estimation reflecting the tendency of the group of evaluators, such as estimation by {1, 2, 3, 4, 5}. The number N of impression estimation units 11-n may perform clustering for automatically determining the number of clusters using an evaluator tendency database, and set the calculated number of clusters as N. As clustering for automatically determining the number of clusters, X-means or the like described in reference 1 can be used.

  [Reference 1] Dan Pelleg, Andrew W. Moore, "X-means: Extending K-means with Efficient Estimation of the Number of Clusters", ICML 2000, pp. 727-734, 2000.

  In addition, the evaluator's tendency is

Since q (i, k) is under the same conditions as the Direlet distribution (the sum of each dimension is always 1). Reference 1 calculates the Bayesian Information Criterion (BIC), which determines clusters by the k-means algorithm and determines whether to increase clusters or not, assuming that the clusters follow a normal distribution. However, the number of groups of evaluator tendencies can be determined more strictly by calculating the number of clusters using the EM algorithm, and also calculating the BIC calculation based on the mixed Dirichlet distribution, instead of using the k-means instead of k-means. be able to.

  The posterior probability indicates the probability that the learning data x (j) is estimated at each stage. For example, if the posterior probability when evaluating “brightness” in the above five steps is p (j, n) = [0.10, 0.20, 0.30, 0.25, 0.15], the learning data x (j) Dark to 1: probability of 0.1, Dark: 2 probability of 0.2, Normal: 3 probability of 0.30, Bright: 4 probability of 0.25, Very bright: 5 probability of It will be 0.15.

  The impression estimation unit 11-n can be estimated, for example, by a general machine learning method such as a neural network or a support vector machine (SVM). The impression estimation model n used by the impression estimation unit 11-n randomly sets parameters at the learning initial value.

  In step S12, using the weight estimation model stored in the weight estimation model storage unit 16, the weight estimation unit 12 weights from the evaluator tendency q (i, k) stored in the evaluator tendency data storage unit 15. Find α (j, n). The weight estimation model is a model that estimates weights from the input evaluator tendency. The weight α (j, n) is a value representing how much an impression estimation unit 11-n is considered as a final estimation result among the impression estimation units 11-1 to 11 -N. The weight estimation unit 12 can also obtain the weight α (j, n) as the posterior probability, similarly to the impression estimation unit 11-n. For example, when the weights α (j, 1) = 0.1, α (j, 2) = 0.6 and α (j, 3) = 0.3 in the impression estimation units 11-1 to 11-3, the impression estimation unit 11-1 Is 10%, the impression estimation unit 11-2 is 60%, and the impression estimation unit 11-3 is 30%, to be considered as the final estimation result.

  The weight estimation unit 12 can be estimated by a general machine learning method such as a neural network or a support vector machine. The weight estimation model used by the weight estimation unit 12 randomly sets parameters at the learning initial value.

  In step S13, the model learning unit 13 stores the weight α (j, n) output by the weight estimation unit 12, the posterior probability p (j, n) output by the impression estimation unit 11-n, and the learning data storage unit 14. The weight estimation model stored in the weight estimation model storage unit 16 and the impression stored in the impression estimation model storage unit 111-n using the learning data x (j) and the impression value y (j) Update the estimation model n. The model learning unit 13 includes both the weight estimation model updating unit 131 and the impression estimation model updating unit 132, and alternately updates the weight estimation model and the impression estimation model n when repeating learning.

  In step S131, the weight estimation model updating unit 131 updates the weight estimation model when the posterior probability p (j, n) is given. The weight estimation model is modeled by, for example, a neural network or a support vector machine. Weighted sum of impression value y (j) of training data x (j) and posterior probability p (j, n)

Update the weight estimation model from the difference between When the weight estimation model is a neural network, the weight estimation model can be updated from a general learning method (error propagation method) of the neural network. If the weight estimation model is a support vector machine,

The impression estimation model number ^ n which is is determined, and the support vector machine is learned so that the impression estimation model number ^ n can be estimated with the evaluator tendency q (i (j), k) as an input.

  In step S132, the impression estimation model updating unit 132 updates the impression estimation model n when the weight α (j, n) is given. The impression estimation model n is updated by a general method for each model, but the learning weights of the samples x [j] and y [j] used for learning each model are given by α (j, n) and learning is performed Do.

  By alternately updating the weight estimation model updating unit 131 and the impression estimation model updating unit 132, the groups of the evaluator tendencies of impressions are automatically clustered into the impression estimation models 1 to N. A weight representing which impression estimation model n is to be reflected as an estimation result on a priority basis is learned. At the time of estimation, an appropriate impression estimation unit 11-n is selected with the weight α according to the input evaluator tendency, and integrated posterior probabilities for input data can be obtained.

  In this embodiment, the weight estimation model update unit 131 and the impression estimation model update unit 132 are alternately updated, but when both are modeled by a neural network, the weight estimation model update unit 131 and the weight estimation model are used. Since the storage unit 16, the impression estimation model updating unit 132, and the impression estimation model storage unit 111-n can be expressed by one neural network, the values of the errors between y (j) and y '(j) can be simultaneously calculated. It becomes possible to update the weight estimation model and the impression estimation model.

<Impression estimation device>
As shown in FIG. 3, the impression estimation device 2 according to the embodiment includes N (22) impression estimation units 11-n (n = 1, 2,..., N), a weight estimation unit 12, and weight estimation. A model storage unit 16 and an estimated value integration unit 17 are provided. Each impression estimation unit 11-n includes an impression estimation model storage unit 111-n. The impression estimation method of the embodiment is realized by the impression estimation device 2 performing the processing of each step illustrated in FIG. 4.

  The impression estimation device 2 is configured by, for example, a special program read into a known or dedicated computer having a central processing unit (CPU: Central Processing Unit), a main storage device (RAM: Random Access Memory), etc. Device. The impression estimation device 2 executes each process, for example, under the control of the central processing unit. The data input to the impression estimation device 2 and the data obtained by each process are stored, for example, in the main storage device, and the data stored in the main storage device is read out to the central processing unit as necessary. It is used for other processing. At least a part of each processing unit of the impression estimation device 2 may be configured by hardware such as an integrated circuit. Each storage unit included in the impression estimation device 2 is, for example, a main storage device such as a random access memory (RAM), an auxiliary storage device configured by a semiconductor memory element such as a hard disk, an optical disk, or a flash memory It can be configured by middleware such as a relational database or key value store.

  The impression estimation model storage unit 111-n stores an impression estimation model n learned by the impression estimation model learning device 1.

  The weight estimation model storage unit 16 stores a weight estimation model learned by the impression estimation model learning device 1.

  Hereinafter, with reference to FIG. 4, the impression estimation method which the impression estimation apparatus 2 of embodiment performs is demonstrated.

  In step S11, the impression estimating unit 11-n calculates and outputs the posterior probability p (n) of the impression value for the input data x using the impression estimating model n stored in the impression estimating model storage unit 111-n. Do. The posterior probability p (n) of the impression value output from each impression estimation unit 11-n is input to the estimated value integration unit 17.

  In step S12, the weight estimation unit 12 obtains the weight α (n) from the input evaluator tendency q (k) using the weight estimation model stored in the weight estimation model storage unit 16. The weight α (n) output from the weight estimation unit 12 is input to the estimated value integration unit 17.

  The evaluator tendency q (k) input to the weight estimation unit 12 can give an arbitrary evaluator tendency. For example, if the evaluator tendency q (k) of a certain existing evaluator is given, it is possible to estimate the impression value of the evaluator having the same tendency as the evaluator. Also, if an evaluator tendency q (k) of a virtual evaluator set manually is given, it is possible to simulate an impression value by the virtual evaluator. Furthermore, it is also possible to obtain an average impression value by giving an average value of the evaluator tendency q (k) calculated from the entire evaluator.

  In step S17, the estimated value integration unit 17 calculates the following equation using the posterior probability p (n) received from the impression estimation unit 11-n and the weight α (n) received from the weight estimation unit 12, and integration is performed Calculate the a posteriori probability y.

  The estimated value integration unit 17 outputs the dimension of the highest posterior probability from the integrated posterior probability as an impression value c to be estimated. It is also possible to use the integrated posterior probability y itself as the impression value c.

  In the present embodiment, the evaluator tendency q (k) of the evaluator is given as auxiliary information for impression estimation, but other information can also be given to the weight estimation unit 12 as auxiliary information. For example, information on a predetermined environment may be combined with the evaluator tendency q (k) of the evaluator and given to the weight estimation unit 12 as auxiliary information. The information on the environment is information on the environment that may affect the evaluation of the impression, and, for example, in the case of automatically estimating the operator's impression (for example, preference, brightness, etc.) from the voice recording of the call center, the call center The evaluation of the impression may differ depending on the content of the reception (for example, failure reception, new reception, cancellation reception, customer consultation, etc.) or the skills of the person in charge of the call center (for example, primary receptionist, supervisor, etc.) A value representing the contents of acceptance of the content or the skill attribute is assumed. In the case of a scene where the impression of an image is to be evaluated, ambient brightness may affect the evaluation, so a value representing an attribute of an environment to be evaluated such as indoor or outdoor is assumed. That is, by providing information on the environment that can influence the evaluation of the impression as the auxiliary information, it becomes possible to perform the impression estimation reflecting the blurring of the evaluation due to the environment.

  As mentioned above, although embodiment of this invention was described, a specific structure is not restricted to these embodiment, Even if there is a change of design suitably etc. in the range which does not deviate from the meaning of this invention, Needless to say, it is included in the present invention. The various processes described in the embodiment are not only executed chronologically according to the order described, but may be executed in parallel or individually depending on the processing capability of the apparatus executing the process or the necessity.

[Program, recording medium]
When various processing functions in each device described in the above embodiments are implemented by a computer, the processing content of the function that each device should have is described by a program. By executing this program on a computer, various processing functions in each of the above-described devices are realized on the computer.

  The program describing the processing content can be recorded in a computer readable recording medium. As the computer readable recording medium, any medium such as a magnetic recording device, an optical disc, a magneto-optical recording medium, a semiconductor memory, etc. may be used.

  Further, the distribution of this program is carried out, for example, by selling, transferring, lending, etc. a portable recording medium such as a DVD, a CD-ROM, etc. in which the program is recorded. Furthermore, this program may be stored in a storage device of a server computer, and the program may be distributed by transferring the program from the server computer to another computer via a network.

  For example, a computer that executes such a program first temporarily stores a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. Then, at the time of execution of the process, this computer reads the program stored in its own storage device and executes the process according to the read program. Further, as another execution form of this program, the computer may read the program directly from the portable recording medium and execute processing according to the program, and further, the program is transferred from the server computer to this computer Each time, processing according to the received program may be executed sequentially. In addition, a configuration in which the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes processing functions only by executing instructions and acquiring results from the server computer without transferring the program to the computer It may be Note that the program in the present embodiment includes information provided for processing by a computer that conforms to the program (such as data that is not a direct command to the computer but has a property that defines the processing of the computer).

  Further, in this embodiment, although the present apparatus is configured by executing a predetermined program on a computer, at least a part of the processing contents may be realized as hardware.

1 Impression estimation model learning device 2 Impression estimation device 11 Impression estimation unit 111 Impression estimation model storage unit 12 Weight estimation unit 13 Model learning unit 14 Learning data storage unit 15 Evaluator tendency data storage unit 16 Weight estimation model storage unit 17 Integration of estimated values Department

Claims (8)

A learning data storage unit that stores a learning database in which learning data, a correct value of an impression value given to the learning data, and an evaluator who gave a correct value of the impression value are associated;
An evaluator tendency data storage unit storing an evaluator tendency database in which an evaluator and an evaluator tendency representing an appearance probability for each impression value in the evaluator are associated;
An impression estimation model storage unit for storing an impression estimation model for estimating a posterior probability of an impression value from the learning data;
A weight estimation model storage unit that stores a weight estimation model that estimates a weight from the evaluator tendency;
An impression estimating unit that estimates the posterior probability of the impression value from the learning data using the impression estimation model for each group into which the evaluator tendency is classified;
A weight estimation unit that estimates the weights from the evaluator tendency associated with the learning data using the weight estimation model;
A weight estimation model updating unit that updates the weight estimation model from the difference between the posterior probability of the impression value weighted by the weight and the correct value of the impression value given to the learning data;
An impression estimation model updating unit that updates the impression estimation model using the weight as a learning weight from the posterior probability of the impression value;
Impression estimation model learning device including. The impression estimation model learning device according to claim 1, wherein
The impression estimating unit determines the number of clusters based on the evaluator tendency database, and the impression estimation model is used for each group in which the evaluator tendency is classified into the number of clusters. To estimate the probability,
Impression estimation model learning device. The impression estimation model learning device according to claim 1 or 2, wherein
The evaluator tendency database associates the evaluator and the evaluator tendency with information about the environment in which the evaluator was evaluated,
The weight estimation model estimates the weight from at least information on the environment,
The weight estimation unit estimates the weight from at least information on the environment associated with the learning data using the weight estimation model.
Impression estimation model learning device. An impression estimation model storage unit that stores an impression estimation model learned by the impression estimation model learning device according to any one of claims 1 to 3;
A weight estimation model storage unit storing a weight estimation model learned by the impression estimation model learning device according to any one of claims 1 to 3.
An impression estimation unit for estimating a posterior probability of an impression value from input data using the impression estimation model for each group into which an evaluator tendency representing the appearance probability for each impression value in the evaluator is classified;
A weight estimation unit that estimates a weight from the evaluator tendency input using the weight estimation model;
An integrated value estimating unit for obtaining an impression value of the input data based on an integrated posterior probability which is a posterior probability of the impression value weighted by the weight;
Impression estimation device including. The impression estimation apparatus according to claim 4, wherein
The integrated value estimation unit obtains a dimension having the highest value among the integrated posterior probabilities as an impression value of the input data.
Impression estimation device. The learning data storage unit stores a learning database in which the learning data, the correct value of the impression value given to the learning data, and the evaluator who gave the correct value of the impression value are associated with each other,
The evaluator tendency data storage unit stores an evaluator tendency database in which an evaluator and an evaluator tendency representing an appearance probability for each impression value in the evaluator are associated with each other,
The impression estimation model storage unit stores an impression estimation model that estimates the posterior probability of the impression value from the learning data,
The weight estimation model storage unit stores a weight estimation model for estimating the weight from the evaluator tendency,
The impression estimation unit estimates the posterior probability of the impression value from the learning data using the impression estimation model for each group into which the evaluator tendency is classified.
A weight estimation unit estimates the weight from the evaluator tendency associated with the learning data using the weight estimation model;
The weight estimation model updating unit updates the weight estimation model from the difference between the posterior probability of the impression value weighted by the weight and the correct value of the impression value given to the learning data,
The impression estimation model updating unit updates the impression estimation model using the weight as a learning weight from the posterior probability of the impression value.
Impression estimation model learning method. An impression estimation model learned by the impression estimation model learning method according to claim 6 is stored in the impression estimation model storage unit,
A weight estimation model learned by the impression estimation model learning method according to claim 6 is stored in the weight estimation model storage unit,
The impression estimating unit estimates the posterior probability of the impression value from the input data using the impression estimation model for each group into which the evaluator tendency representing the appearance probability for each impression value in the evaluator is classified.
A weight estimation unit estimates a weight from the evaluator tendency input using the weight estimation model;
The integrated value estimation unit obtains an impression value of the input data based on an integrated posterior probability which is a posterior probability of the impression value weighted by the weight.
Impression estimation method.   A program for causing a computer to function as the impression estimation model learning device according to any one of claims 1 to 3 or the impression estimation device according to claim 4 or 5.

Images