JP5875005B2 - Human emotion estimation device, human emotion estimation method, and human emotion estimation program - Google Patents

Description

  The present invention relates to a human emotion estimation device, a human emotion estimation method, and a human emotion estimation that estimate what kind of criteria a person evaluates others, and what impressions and emotions they have, and present the results. Regarding the program.

  With the development of image and voice recognition technology and the emergence of new physiological sensors, research to measure and estimate the inside information of people such as mood and emotion has become active. For example, a technique for detecting a facial expression of a person using an image recognition technique and estimating an emotion therefrom (for example, see Patent Document 1), a technique for extracting an emotion using information such as voice strength and shaking (for example, a patent) Document 2) has been proposed.

JP 2011-081445 A Japanese Patent No. 4704952

  However, the techniques for estimating emotions described in Patent Documents 1 and 2 are intended to acquire emotions that appear on the surface, such as human facial expressions and physiological responses, and interpretation of why these emotions occurred. There is a problem that can not be given.

  The present invention has been made in view of such circumstances, and can estimate evaluations and emotions that a user has with respect to a specified target, and can clarify the factors that cause them. It is an object to provide a human emotion estimation device, a human emotion estimation method, and a human emotion estimation program.

  According to the present invention, input means for inputting personal attribute information of a user, personal attribute information of an evaluation target, score information of psychological factors for the evaluation target, and score information of interpersonal emotion factors, and input by the input means Using the information storage means for storing information, the learning means for learning the relevance of the information stored in the information storage means, the learning model generated by the learning means, and the input An estimation means for estimating the expected value of the score information of the factor for which input from the means is missing, and an information presentation means for presenting information on the estimated result are provided.

  The present invention provides an account management means for managing the account information of the user, and inputs information fed back from the user with respect to the estimation result by the input means, and reconstructs the learning model for each user. And personality learning means for generating the learning model adapted to the user by associating with the account information.

  In the present invention, the information presentation means visualizes and presents the relationship between each factor of the personal attribute information, the psychological factor for the evaluation target, and the interpersonal emotion factor in the form of a directed or undirected graph, The information of the learning model is further presented.

  The present invention is characterized in that, based on the learning model constructed for each user, a criterion for evaluating the other by the user is evaluated.

  According to the present invention, input means for inputting personal attribute information of a user, personal attribute information of an evaluation target, score information of psychological factors for the evaluation target, and score information of interpersonal emotion factors, and input by the input means An interpersonal emotion estimation method performed by an interpersonal emotion estimation device including an information storage unit that stores information, the learning step generating a learning model that learns the relevance of information stored in the information storage unit, and the learning An estimation step for estimating the expected value of the score information of the factor for which the input from the input means is missing using the learning model generated by the step, and an information presentation step for presenting the information of the estimated result It is characterized by having.

  The present invention is a human emotion estimation program for causing a computer to function as the human emotion estimation device.

  According to the present invention, attention is paid to interpersonal emotions, and a learning model is constructed based on knowledge obtained from social psychology research to learn interpersonal emotions and the relationships between the factors. It is possible to estimate the evaluations and feelings that are held against and to clarify the factors that cause them.

It is a block diagram which shows the structure of the personal emotion estimation apparatus by one Embodiment of this invention. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of an antonym pair. It is explanatory drawing which shows an example of a personal attribute. It is explanatory drawing which shows an example of the psychological factor which comprises a personal feeling. It is explanatory drawing which shows the structural example of a graphical model. 4 is an explanatory diagram showing a subset S _i of information D. FIG. It is a block diagram which shows the structure of the modification of the personal emotion estimation apparatus shown in FIG. It is a flowchart which shows the processing operation before the operation | movement of the personal emotion estimation apparatus shown in FIG. It is a flowchart which shows the processing operation of the account management part 6 shown in FIG. It is a flowchart which shows the processing operation which the personal emotion estimation apparatus shown in FIG. 12 performs a personal emotion diagnosis.

  Hereinafter, a human emotion estimation apparatus according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing the configuration of the embodiment. Reference numeral 1 is an input unit for a user to input information about the personal attributes of the user and the evaluation target, each psychological factor for the evaluation target, and the rating of the personal feeling. Reference numeral 2 denotes an information storage unit that stores information input by the input unit 2. Reference numeral 3 denotes a learning unit that constructs a learning model. Reference numeral 4 denotes an estimation unit that uses information stored in the information storage unit 2 and a model learned by the learning unit 3 to estimate an expected value of a factor whose input is missing. Reference numeral 5 denotes an information presentation unit that presents the learned graph structure and the expected value of the estimated missing factor on a display device or the like.

  The interpersonal emotion estimation device shown in FIG. 1 focuses on interpersonal emotions, constructs a mathematical model based on knowledge obtained from research in social psychology, and learns the interpersonal emotions and the relationships between the factors, thereby Estimates the evaluations and feelings held by a specified object, and clarifies what factors cause them.

  The concept of implicit personality, proposed by Bruner et al., Is widely supported as a framework for human evaluation of others. The tacit perception of personality is a belief system that links specific personal attributes and behavior with personality characteristics that individuals have cultivated through experience. For example, Japanese prefer to follow the right, type A is serious, etc.

  There is a great individual difference in the personality view of tacit jealousy, but it is reported that the elements that compose it are associated with the basic three dimensions of "personal friendliness", "social desirability" and "powerfulness" (Reference: Fumitoshi Hayashi, "A Study on Basic Dimensions of Interpersonal Cognitive Structure", Bulletin of Faculty of Education, Nagoya University, Department of Educational Psychology, vol. 25, pp. 233-247, Dec. 1978.). Personal friendliness falls into categories such as “kindness” and “warmth”, social desirability falls into “honesty”, “reason”, “intelligence”, and empowerment falls into categories such as “aggressiveness” and “motivation” .

  In addition, while tacit perceptions of personality vary widely among individuals, groups that share the same social background and living environment, such as the “baby boomer generation” and “busy madam”, have certain common values. For example, in marketing science, demographic attributes such as age, gender, nationality, and social categories such as occupation, educational background, income, etc. Called graphic attributes, target segmentation is performed to classify customers based on personal attributes such as demographic attributes and psychographic attributes. Similarly, personality can be grouped from individual attributes.

  The present invention constructs a mathematical model that incorporates these social psychological or marketing science findings. Specifically, each individual attribute and psychological factor is a random variable, and the relationship between them is expressed by a graph G (V, E). Here, V is a node set corresponding to personal attributes, psychological factors, and the like, and E is a set of edges representing the relationship between random variables.

  Psychological factors constituting the implicit personality view (hereinafter referred to as personality view factor) are expressed by an adjective pair of adjectives related to humanity based on the SD method (semantic-differential method). 2-7 is explanatory drawing which shows an example of an antonym pair. 2 to 7 show an example of an antonym pair for each of “appearance”, “strength”, “nature”, “extroversion”, “reason / emotion”, and “ability”.

  Personal attributes include elements that define social backgrounds and lifestyles such as age, gender and occupation, and elements related to living levels such as educational background and annual income. FIG. 8 is an explanatory diagram showing an example of personal attributes.

  In addition, psychological factors that constitute personal feelings (hereinafter referred to as personal feeling factors) are also referred to psychological knowledge, and several evaluation axes based on the SD method are set. FIG. 9 is an explanatory diagram illustrating an example of psychological factors that constitute interpersonal emotions.

The personality factor and the interpersonal emotion factor are expressed by discrete values (for example, five-level evaluation) or real values (for example, 0 to 1). At this time, the element X _i in the node set V = {X ₁ ,..., X _p } has an individual attribute such as “age” and “gender”, “speaks together—not talks”, “sympathetic” —not considerate It is a random variable corresponding to a personality factor such as “like” and an interpersonal emotion factor such as “like-dislike” and “respect-contempt”.

The element e (X _i , X _j ) in the edge set E = {e (X _i , X _j )} indicates that X _i has some influence on X _j . For example, it corresponds to the phenomenon that the personality factor X _i “speaks fits-not talks” affects the interpersonal emotion factor X _j “likes-dislikes”. e (X _i , X _j ) may be undirected or directed, and represents a correlation if it is an undirected edge and a causal relationship if it is a directed edge. When e (X _i , X _j ) is directed, X _i is called the parent of X _j .

If a directed acyclic graph is assumed as the structure of the graph G (V, E), a Bayesian network model is assumed, and if an undirected graph is assumed, a linear Markov model or a factor model is obtained. Hereinafter, a Bayesian network model B (G (V, E), θ) is assumed, and a parent set of the node X _i is Pa _i . That is, E = {e (Pa _ij , x _i ) | Pa _ij ∈ Pa _i , i = 1,..., P). FIG. 10 is an explanatory diagram showing a structural example of the graphical model, that is, a configuration example of the parent set Pa _i for each node X _i . The node is composed of four types: a user personal attribute, an evaluation target personal attribute, a user personality factor, and an interpersonal emotion factor.

  The user's personality factor is influenced by the user's personal attributes and the personal attributes of the evaluation target. There is also a causal or co-occurrence relationship between the user's personality factors. Those psychological factors are finally summarized in the three dimensions of “personal friendliness”, “social desirability”, and “strength”. Interpersonal feelings are formed based on this personality model.

  As described above, the personal feeling has a layered structure in which the personal attribute of the user and the personal attribute to be evaluated are the top layer, the personality factor of the user is the middle layer, and the personal emotion factor is the bottom layer.

ここで、θ＝（θ_１，…，θ_ｐ）はモデル中のパラメータ、π（θ_ｉ）はその事前確率である。 Given n observation data D = {x ₁ ,..., X _n } for X ₁ ,..., X _p , the likelihood L (B of the Bayesian network model B (G (V, E), θ) | D) is as follows.

Here, θ = (θ ₁ ,..., Θ _p ) is a parameter in the model, and π (θ _i ) is its prior probability.

と表すことができる。ここでＮ_ｉｊｋはＤ中の、Ｘｉ＝ｋかつＰａ_ｉがｊ番目のパターンをとった観測データ数である。 Specific examples of theta, for example, a discrete value of the range of all the random variable _{X i} is 1 ≦ _{X i} ≦ _{r i, X i} conditional probability _p of | a _(X i Pa _i) follows a multinomial distribution . Further, if the number of patterns that the parent observation vector Pa _i can take is q _i and the probability that X _i = k when Pa _i takes the jth pattern is θ _ijk , θ = {θ _ijk }. . At this time, the likelihood l (B | D) of the model is

It can be expressed as. Here, N _ijk is the number of observation data in _X where Xi = k and Pa _i takes the jth pattern.

を考える。つまり、Ｓ_ｉは、Ｘ_ｉとその親集合Ｐａ_ｉの欠損パターン毎にＤを分割したものである（図１１参照）。 The learning unit 3 learns parameters (Pa _i , θ) from the information D stored in the information storage unit 2. Here, since the user responds to any item among personal attributes, personality factors, and interpersonal emotion factors, the accumulated data includes corresponding deficiencies. The tendency of data loss depends on what viewpoint the user normally judges others, and is not related to the magnitude of the value. Therefore, the data loss tendency has a property of MAR (Missing at Random). . A subset of information D

think of. That is, S _i is obtained by dividing D for each missing pattern of X _i and its parent set Pa _i (see FIG. 11).

ここで、ｑ_ｉ ^ｏｂｓはＤの部分集合Ｓ_ｘにおいて、欠損している親を無視し、値が観測されていた親集合のみで構成されるパターン数を示す。またＮ_ｉｊｋ ^ｏｂｓはＳ_ｘ内でそのようなパターンが観測されたデータの個数を示す。 At this time, the likelihood l (D | B) can be rewritten as follows under the assumption of MAR.

Here, q _i ^obs represents the number of patterns composed of only the parent set in which the missing parent is ignored in the subset S _x of D, ignoring the missing parent. N _ijk ^obs represents the number of data in which such a pattern is observed in S _x .

ここでａ_ｉｊｋはＮ_ｉｊｋ ^ｏｂｓに対応するハイパーパラメータであり、Γ（）はガンマ関数である。 For the prior probability π (θ), a Dirichlet distribution which is a conjugate distribution of a multinomial distribution is assumed.

Here, a _ijk is a hyper parameter corresponding to N _ijk ^obs , and Γ () is a gamma function.

この対数をとり、両辺を偏微分することにより、事後分布を最大化するθ_ｉｊｋの推定値＾θ_ｉｊｋ（＾はθ_ｉｊｋの上に付く）は

で求めることができる。 The posterior distribution is given by the product of the likelihood and the prior probability. That is,

Taking this logarithm and partial differentiation of both sides, the estimated value of θ _ijk that maximizes the posterior distribution ^ θ _ijk (^ is attached to θ _ijk ) is

Can be obtained.

Dirichlet distribution is a beta distribution for multivariate, the prior probability of θ _ijk can be interpreted as follows the beta distribution, which is defined by hyper-parameters a _ijk each, "Which personal attributes and psychological factors is likely to be both answer to a _ijk Can put in prior knowledge. For example, a matrix is generated in which 1 is set when there is a response to each item and 0 is set when there is no response, and the component score of each item is calculated using a method such as quantification type III. For the same component, it is considered that there is a high correlation between items showing a high component score, and a high a _ijk is given.

G (V, E) were formulated the model as is known so far, but actually E, that is, the parent set Pa _i for each X _i is generally unknown. As the learning algorithm, the following algorithms (1) and (2) can be used.
(1) Greedy algorithm for bottom-up searching for a graph structure that minimizes an information criterion, for example, BIC = −2 logπ (D, θ | B) − | θ | log (n), considering a statistical model selection problem (2) As a graph-theoretic approach, IC (Inductive Causation) algorithm (reference: Manabu Kuroki, “Statistical causal reasoning-model, reasoning, guess”, Kyoritsu Publishing Co., Ltd., March 1, 2009), GS (Grow -Shrink) algorithm (literature: D. Margaritis and S. Thrun. “Bayesian network induction via local neighborhoods.” In SA Solla, TK Leen, and K.-R. Muller, editors, Proceedings of Conference on Neural Information Processing Systems ( NIPS-12). MIT Press, 1999.)

つまりＢＩＣ_ｉは、Ｘ_ｉとその親集合Ｐａ_ｉに対するモデルのあてはまりの良さを示す。これをスコア関数とする貪欲アルゴリズムを以下に示す。 Hereinafter, (1) the greedy algorithm will be described. First, the BIC is modified as follows to obtain BIC _i .

That is, BIC _i indicates the goodness of fit of the model for X _i and its parent set Pa _i . The greedy algorithm using this as a score function is shown below.

Step 1) For any two node pairs (X _i , X _j ), BIC _i is calculated as Pa _i = X _i . The upper M pieces are assumed to be the temporary parent set CP _i of X _i .
Step 2) Set Pa _i = φ (empty set).
Step 3) For all nodes Xi (1 ≦ i ≦ p)
Step 3-1: For all nodes CP _ij εCP _i (1 ≦ j ≦ M)
If (i) ¬ (CP _ij ∈ Pa _i ), then CP _ij ∈ Pa _i is set (append).
(Ii) If CP _ij ∈ Pa _i , set ¬ (CP _ij ∈ Pa _i ) (delete).
(Iii) If ¬ (CP _ij ∈ Pa _i ) and X _i ∈ Pa _i , then CP _ij ∈ Pa _i , ¬ (X _i ∈ Pa _i ) (reverse).
Step 3-2: For the new parent set Pa _i , ^ θ _ijk (^ is attached to θ _ijk ) is obtained, and BIC _i is calculated.
Step 3-3: Adopt the change that gave the best score (minimum BIC _i ) and update Pa _j .

  Based on the learning algorithm described above, the learning unit 3 uses the information stored in the information storage unit 2 to learn the relationship between individual attributes and psychological factors, and stores the parameters Pa and θ.

  The information accumulating unit 2 stores information acquired in advance through questionnaire surveys or the like, and sequentially accumulates information that is input every time the user uses the personal emotion estimation apparatus shown in FIG. The learning unit 3 re-learns the model using information accumulated so far at a timing designated by the administrator or by feedback from the user.

  Next, the operation of the human emotion estimation apparatus shown in FIG. 1 will be described. Through the input unit 1, the user inputs his / her personal attributes to be evaluated, each psychological factor for the evaluation target, and the score of the personal feeling. For personal attributes, information on gender, age, occupation, other hobbies and lifestyles is selected from the presented options. As described above, personality factors and interpersonal factors are answered using a five-level evaluation or a continuous value of 0 to 1. For example, “speak to the subject does not fit” for the subject is 2, “like as a lover to hate” is 5, and so on. At this time, it is not necessary to input all the elements, and the input factor is interpreted as an explanatory variable, and the factor that is not input is interpreted as an objective variable (estimation target).

  Next, the estimation unit 4 uses the information stored in the information storage unit 2 and the model learned by the learning unit 3 to estimate the expected value of the factor whose input is missing. When there is no cycle in the graph structure, the probability distribution of the missing factor is sum-product method (reference: FR Kschischang, BJ Frey, and HA Loeliger, "Factor Graphs and the Sum-Product Algorithm," IEEE Trans. On Information Theory, vol. 47, no. 2, pp.498-519, Feb. 2001.). Loop belief propagation (reference: KP Murphy, Y. Weiss, MI Jordan, “Loopy Belief Propagation for Approximate Inference: An Empirical Study,” Proceedings of Uncertainty in AI, pp. 467-475, 1999 ) Etc., in many cases, a solution can be obtained approximately.

として、Ｘ_ｉの周辺ノードに対して伝播するメッセージを求めることができる。 Hereinafter, calculation of an expected value by Loopy belief propagation will be described. A parent set of a certain node X _i is Pa _i , and a child set is Ch _i . From the definition of the Bayesian network, if the probability that Pa _i can take is known, the expected value ^ pa _ij (^ is attached to pa _ij ) can be obtained, and the conditional probability p (x _ij | pa _ij ) of X _i Can be requested. Here, the probability that the parent Pa _ij is a is defined as a message from the parent Pa _{ij to the} child X _i and is represented by π _j (a). Conversely, a message sent to the parent Pa _ij when the child X _i is a is λ _j (a). At this time,

As a result, a message to be propagated to the peripheral nodes of X _i can be obtained.

Loopy belief propagation is successively adapt it to the peripheral for all nodes _{X i,} and updates the probability value to converge, the conditional probability _p for all the nodes _{X i} | obtain (x _{ij pa} ij).

  Next, the information presentation unit 5 presents the learned graph structure and the estimated expected value of the missing factor on a display device or the like. Thereby, the user can know what feelings are held for the evaluation object and what evaluation factors are based on the feelings. It is not necessary to show the results for all the factors in the information presenting unit 5, but only factors with expected values that are equal to or higher than a predetermined threshold are extracted, and only the edges between them are visualized, and factors that are estimated to have a strong influence Only sets and relationships between them can be presented. The user can also adjust this threshold and view the results at various levels.

  Next, with reference to FIG. 12, the modification of the interpersonal emotion estimation apparatus shown in FIG. 1 is demonstrated. 12 is a block diagram showing a configuration of a modification of the interpersonal emotion estimation apparatus shown in FIG. In this figure, the same parts as those in the apparatus shown in FIG. The apparatus shown in this figure is different from the apparatus shown in FIG. 1 in that an account management unit 6 and a personal learning unit 7 are newly provided. In the interpersonal emotion estimation apparatus shown in FIG. 12, when the result presented to the information presentation unit 5 does not match the user's own judgment, the user inputs information via the input unit 1 and performs feedback based on this input information. Is.

  Next, with reference to FIG. 13, the processing operation before operation of the human emotion estimation apparatus shown in FIG. 12 will be described. FIG. 13 is a flowchart showing a processing operation before operation of the human emotion estimation apparatus shown in FIG. First, the user accumulates data collected in advance in the information accumulation unit 2 via the input unit 1 (step S1). In response, the learning unit 3 learns parameters from the data stored in the information storage unit 2 to obtain a general model (step S2).

  Next, the processing operation of the account management unit 6 shown in FIG. 12 will be described with reference to FIG. FIG. 14 is a flowchart showing the processing operation of the account management unit 6 shown in FIG. First, the user creates a new account via the input unit and registers personal information. Then, the account management unit 6 assigns general parameters generated by the learning unit 3 to the registered account (step S12).

  Next, with reference to FIG. 15, the processing operation in which the human emotion estimation apparatus shown in FIG. 12 performs human emotion diagnosis will be described. FIG. 15 is a flowchart showing a processing operation in which the human emotion estimation apparatus shown in FIG. First, the user inputs arbitrary items among personal information to be diagnosed, evaluation of characteristic words, and emotions he / she currently has via the input unit 1 (step S21). Subsequently, the estimating unit 4 calculates an estimated value for the attribute that has not been input, and notifies the user of the expected value of each attribute through the information presenting unit 5 (step S22). The user determines whether the diagnosis result is satisfactory or feels uncomfortable (step S23). As a result of this determination, if the user feels uncomfortable with the diagnosis result, the user feeds back a correct value for the wrong diagnosis result via the input unit 1 (step S24). In response to this, the personality learning unit 7 receives the feedback from the user, locally learns and stores the new model structure and parameters (step S25).

  Thus, feedback is given as a score (numerical value) for a factor. For example, when a certain input is received, the score for a factor of “sincere-unfaithful” from the user to the evaluation target is estimated as 1. If the user sees this result and feels that this value is inaccurate, the user inputs a correct score (for example, 3). Based on the input information, the personality learning unit 7 re-learns the weights of the edges around the factor, and holds them associated with the user account. Thereby, an edge weight set adapted for each user is learned, and estimation reflecting the individuality of the user can be performed. Also, by visualizing the learned edge weights and presenting them to the user, the user can know what axis he / she evaluates on a daily basis.

  As described above, each of the personal attribute, the interpersonal evaluation factor, and the interpersonal emotion factor is a random variable, the random variable is represented by a node, and the relationship between the random variables is represented by an edge (FIG. 10). It is generated by learning using the user and the personal attributes of the evaluation target (FIG. 8), and the psychological factors (FIGS. 2 to 7) and the personal feelings (FIG. 9) for the evaluation target by the user. Then, for the graph obtained by learning, the input personal attributes and the scores of psychological factors and interpersonal feelings are used as explanatory variables, and the personal attributes that were not input and the scores of psychological factors and interpersonal emotions are estimated. As an objective variable, an expected value of the objective variable can be calculated and presented. This makes it possible to visualize the psychological factors that make up the implied personality view, which is a measure by which a person evaluates others, and to provide what emotions a person has and why Become.

  For new users, the general model learned by the learning unit 3 using the input information of all other users accumulated in the information accumulating unit 2 can be used. It is possible to generate a model reflecting the user's individuality by performing adaptive learning according to the situation. In addition, the user does not need to input all information when inputting his / her own personal information, partner information, personal evaluation and personal feeling for the partner. The interpersonal emotion estimation apparatus according to the present embodiment can estimate information that the user did not input using the estimation unit 4 and present it to the user using the information presentation unit 5. The user can check the presented information and, if necessary, make corrections via the input unit 1 to perform relearning using the feedback from the user and update the model.

  The program for realizing the functions of the processing units in FIGS. 1 and 12 is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read into a computer system and executed. Emotion estimation processing may be performed. Here, the “computer system” includes an OS and hardware such as peripheral devices. The “computer system” includes a WWW system having a homepage providing environment (or display environment). The “computer-readable recording medium” refers to a storage device such as a flexible medium, a magneto-optical disk, a portable medium such as a ROM and a CD-ROM, and a hard disk incorporated in a computer system. Further, the “computer-readable recording medium” refers to a volatile memory (RAM) in a computer system that becomes a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. In addition, those holding programs for a certain period of time are also included.

  The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, what is called a difference file (difference program) may be sufficient.

  As mentioned above, although embodiment of this invention has been described with reference to drawings, the said embodiment is only the illustration of this invention, and it is clear that this invention is not limited to the said embodiment. is there. Accordingly, additions, omissions, substitutions, and other changes of the components may be made without departing from the technical idea and scope of the present invention.

  It is also applicable to uses where it is indispensable to estimate what kind of criteria a person evaluates others, what kind of impression or emotion they have, and present the results.

  DESCRIPTION OF SYMBOLS 1 ... Input part, 2 ... Information storage part, 3 ... Learning part, 4 ... Estimation part, 5 ... Information presentation part, 6 ... Account management part, 7 ... Individual Sex learning department

Claims (5)

Input means for inputting personal attribute information of the user, personal attribute information of the evaluation target, score information of the psychological factor for the evaluation target, and score information of the interpersonal emotion factor,
Information storage means for storing information input by the input means;
Account management means for managing the account information of the user;
Learning means for generating a learning model learning the relevance of the information stored in the information storage means;
Using the learning model generated by the learning means, an estimation means for estimating an expected value of the score information of a factor for which an input from the input means is missing;
Information presenting means for presenting information of the estimated result ;
The information fed back from the user with respect to the estimated result is input by the input means, the learning model for each user is reconstructed, and related to the account information, thereby being adapted to the user. An interpersonal emotion estimation device comprising personality learning means for generating a learning model . The information presenting means includes
Further presenting the information of the learning model by visualizing and presenting the relationship between each factor of the personal attribute information, the psychological factor for the evaluation object and the interpersonal emotion factor in the form of a directed or undirected graph The interpersonal emotion estimation apparatus according to claim 1 . Based on the learning model for each of the users, interpersonal emotion estimation apparatus according to claim 1, wherein the diagnosing whether the evaluated others in any reference the user. Input means for inputting personal attribute information of a user, personal attribute information to be evaluated, psychological factor score information for the evaluation target, and personal emotion factor score information, and storing information input by the input means An interpersonal emotion estimation method performed by an interpersonal emotion estimation device comprising information storage means and account management means for managing account information of the user ,
A learning step for generating a learning model that learns the relevance of information stored in the information storage means;
Using the learning model generated by the learning step, an estimation step for estimating an expected value of the score information of the factor for which the input from the input means is missing,
An information presentation step for presenting information of the estimated result ;
The information fed back from the user with respect to the estimated result is input by the input means, the learning model for each user is reconstructed, and related to the account information, thereby being adapted to the user. An interpersonal emotion estimation method comprising: a personality learning step for generating a learning model . A human emotion estimation program for causing a computer to function as the human emotion estimation device according to any one of claims 1 to 3 .

Images