JP2006190196A - Device and method for evaluating person - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform highly precise evaluation of users of an Internet community not only by unit of data but by unit of person. <P>SOLUTION: A person evaluation server 10 has a dictionary construction unit 10a which creates a dictionary data base 15 in which evaluation criteria are unified to be referred by every evaluation content and a person-based evaluation unit 10b. The person-based evaluation unit 10b refers to article data about an object for evaluation and the dictionary data base 15 and adopts an inner product value (for example, an affirmative inner product value) of an evaluation content having the maximum value among inner product values of dictionaries and articles of each evaluation content as a basic value. Then, the person-based evaluation unit 10b uses the basic value and registered user IDs to calculate evaluation values by unit of user. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a technique for evaluating a person using a communication log.

In recent years, with the spread of the Internet and the increase in the number of users, an Internet community for sharing various information on a network has been constructed (for example, see Patent Document 1). In the Internet community, an unspecified number of virtual people use services such as bulletin boards, chats, and web logs with posts to realize communication. For this reason, users of such services communicate with an unspecified number of people via the Internet. Therefore, in a community that does not have a mechanism for determining whether a communication partner can be trusted, there is a concern that the user cannot communicate with peace of mind.
JP 2002-163220 A

  In order to eliminate the above concerns, a method for evaluating a communication partner has been proposed. For example, for person evaluations based on text data such as text on bulletin boards and chats, postings from third parties, etc., the user inputs subjective evaluations for others or self, and each evaluation result is based on the accumulated evaluation results. Perform person evaluation. However, in such a person evaluation, since an objective scale is not defined, there is a problem that the accuracy and universality of the evaluation are lacking and the range of use as evaluation information is limited.

  In view of the above problems, it is effective to apply text mining that first unifies the evaluation scale and then extracts and evaluates the features that appear in the text. However, since the technology of text mining is intended to perform character search and trend extraction from text data, it is suitable for evaluation in units of data, but it is not suitable for evaluation in units of persons. is there.

  Therefore, an object of the present invention is to perform highly accurate evaluation on the Internet community users not only in data units but also in human units.

  The person evaluation device according to the present invention is a dictionary construction means for constructing a dictionary in which a feature word included in learning data and a score for the feature word are associated with each evaluation content according to a unified evaluation scale. And evaluation means for evaluating each user based on identification information and basic values of the plurality of users by referring to data (for example, articles) of the plurality of users constituting the learning data and the dictionary With.

  The person evaluation method according to the present invention relates to a feature word included in learning data and a score for the feature word, and builds a dictionary for each evaluation content according to a unified evaluation scale; A step of evaluating each user based on the identification information and the basic value of the plurality of users by referring to the data of the plurality of users constituting the learning data and the constructed dictionary.

  According to these inventions, feature words in the learning data and scores according to their appearance frequencies are registered for each evaluation content in the dictionary using a unified evaluation scale. The evaluation of the data constituting the learning data is performed based on the user identification information while referring to the dictionary. Therefore, by applying these inventions to the Internet community where the data input by the user is gathered, high-accuracy evaluation not only in data units but also in human units is realized for the user.

  It should be noted that the construction of the dictionary is a prior work for the user unit evaluation, and the dictionary construction and the person evaluation are performed independently. Therefore, the user who is the target of the person-unit evaluation is not limited to the user who is involved in the construction of the dictionary (for example, the user who wrote the learning data).

The dictionary construction means, as specific means for constructing a dictionary, decomposes the learning data into words, data classification means for classifying the learning data for each evaluation content, and the learning data classified by the data classification means. Based on data analysis means, score calculation means for assigning a score corresponding to the appearance frequency to a word obtained as a result of decomposition by the data analysis means, and a score of each word given by the score calculation means And feature word extracting means for extracting feature words from the learning data classified for each evaluation content.
According to the present invention, it is possible to construct a dictionary in which a plurality of feature words to be compared when data is evaluated are registered from learning data.

For example, the feature word extraction unit extracts, as a feature word, a word having an idf value (a value representing importance in all data) that is equal to or greater than a threshold among the words scored by the score calculation unit.
In addition, the feature word extraction unit may extract, as a feature word, a word whose total number of appearances in the learning data is greater than or equal to a threshold value among the words scored by the score calculation unit.
Furthermore, the feature word extraction means can exclude symbols (“”, (),...) Unnecessary for calculating the evaluation value from the feature words from the words extracted as feature words.

  The above-mentioned evaluation means is a concrete means for evaluating each person from data, and calculates the inner product value of the feature vector of the feature word registered in the dictionary and the feature vector of the data to be evaluated for each evaluation content. And an evaluation value calculation means for calculating an evaluation value for each user using the identification information of the plurality of users and the calculated basic value.

  According to the present invention, the similarity between the feature word registered in the dictionary and the data to be evaluated can be estimated from the inner product value. Therefore, the person evaluation device registers feature words in the dictionary for each evaluation content, and classifies the evaluation target data for each user, thereby estimating the evaluation content of the data, that is, the user of the data based on a unified standard. Another evaluation is possible.

The inner product value difference calculating means calculates a difference between the inner product value of one evaluation content and the inner product value of another evaluation content as a basic value from the inner product value for each evaluation content calculated by the inner product value calculating means. It is good.
The inner product value of one evaluation content is, for example, the inner product value of the feature vector of the feature word registered in the “affirmation” dictionary and the feature vector of the evaluation target data, and the inner product value of the other evaluation content Is, for example, the inner product value of the feature vector of the feature word registered in the “No” dictionary and the feature vector of the evaluation target data.
Thus, by taking the inner product value of the specific evaluation content as a reference, the evaluation value calculated from the basic value is simplified, and an effect that the user can easily grasp the evaluation result can be expected.

The evaluation value calculation means calculates the evaluation value using, for example, the sum of the basic values corresponding to the user to be evaluated and the data.
More specifically, the evaluation value calculation means multiplies the sum of the basic values corresponding to the user and data to be evaluated by the number of data for each evaluation content corresponding to the user, and data corresponding to the user. The evaluation value can be calculated by dividing the total number. That is, for the same user, the sum of the basic values is multiplied with the ratio of the number of evaluation contents data used as the evaluation reference to the total number of data as a coefficient.

  According to the present invention, when calculating an evaluation value, there is a method of multiplying the sum of basic values by the ratio of the number of data obtained with a certain evaluation content (for example, affirmation, denial, or none) to the total number of data. Taken. Thereby, since all the posted data is reflected in the evaluation value for the user to be evaluated, the user of the Internet community can browse the evaluation result more efficiently and accurately.

  According to these inventions, since the person evaluation is performed in a unified standard and in units of persons, the evaluation accuracy and, consequently, the reliability with respect to the communication partner is improved. As a result, Internet community users can communicate with a large number of unspecified people by referring to the evaluation.

(First embodiment)
Hereinafter, a person evaluation system according to a first embodiment of the present invention will be described with reference to the accompanying drawings.
FIG. 1 is a diagram for explaining a functional configuration of a person evaluation system in the present embodiment. As shown in FIG. 1, the person evaluation system 1 includes at least a person evaluation server 10 (corresponding to a person evaluation apparatus), which will be described later, and portable terminals 20, 30, and 40 held by users of the Internet community. From the mobile terminals 20, 30, and 40, it is possible to browse articles posted on the community and write comments via the base station B and the network N.

  The person evaluation server 10 includes a dictionary construction unit 10a (corresponding to dictionary construction means) and a person unit evaluation unit 10b (corresponding to evaluation means). Further, the dictionary construction unit 10a includes a data classification unit 11 (corresponding to the data classification unit), a data analysis unit 12 (corresponding to the data analysis unit), a score calculation unit 13 (corresponding to the score calculation unit), and a feature word extraction. A unit 14 (corresponding to feature word extraction means) and a dictionary database 15 are provided. The human unit evaluation unit 10b includes an inner product value calculation unit 16 (corresponding to an inner product value calculation unit), an article classification unit 17, an article database 18, a user database 19, and an evaluation value calculation unit 110 (corresponding to an evaluation value calculation unit). ) And an evaluation value output unit 111. These components are connected to each other through a bus (not shown) so that signals can be input and output.

The data classification unit 11 classifies the learning data for each of the three types of evaluation contents: affirmation, denial, and none.
The data analysis unit 12 decomposes the classified learning data into words by morphological analysis.

The score calculation unit 13 assigns a score to the word obtained as a result of the decomposition so that the higher the appearance frequency, the higher the score.
The feature word extraction unit 14 extracts a characteristic word (feature word) for determining the evaluation content of the article from the learning data classified for each evaluation content based on the score of each given word. To do.

  The dictionary database 15 is constructed for each evaluation scale as shown in FIG. Further, the feature word and the evaluation value are stored in the dictionary of each evaluation scale in association with each of the three types of evaluation contents. For example, in the dictionary 151 in which “common factor” is set as the evaluation scale, the score “5.28” is assigned to the feature word “a” whose evaluation content is “Yes”, and the score is also assigned to the feature word “b”. “2.24” is assigned. In addition, in the dictionary 152 in which “self-recognition factor” is set as the evaluation scale, the score “2.30” is assigned to the feature word “i” whose evaluation content is “No”, and the feature word “j” is also assigned to the feature word “j”. Score “3.71” is given.

The inner product value calculation unit 16 calculates the inner product value of the feature vector of the feature word registered in the dictionary database 15 and the feature vector of the data to be evaluated as a basic value for each evaluation content.
The article classification unit 17 classifies all articles to be evaluated for each user ID based on the user ID attached to the article.

  In the article database 18, as shown in FIG. 3, the inner product value in units of articles is stored for each evaluation content. For example, for an article having the article ID “a-29”, “ogata” is stored as the ID of the user who posted the article, and the content of the article “Today went to ○○…” Is stored. Further, three types of inner product values calculated by the inner product value calculation unit 16 in association with these pieces of information, that is, “5.28” which is a positive inner product value and “3.26” which is a negative inner product value. , “0.89”, which is the inner product value without, is stored. The article database 18 also includes a difference between an inner product value of a certain evaluation content (eg, positive inner product value “5.28”) and an inner product value of another evaluation content (eg, negative “3.26”). A value (for example, 2.02) is stored.

  An example of data storage in the user database 19 is shown in FIG. The user database 19 stores evaluation values calculated for each evaluation scale in units of user IDs. In the present embodiment, as described above, “sympathy factor” and “self-recognition factor” are assumed as evaluation scales, and “affirmation”, “denial”, and “none” are assumed as evaluation contents. For example, as an evaluation value for a person whose user ID is “ogata”, “3.29” is used when the “sympathy factor” is used as the evaluation scale, and “0.98” is used when the “self-recognition factor” is used. "" Is stored.

The evaluation value calculation unit 110 calculates an evaluation value for each user using a plurality of user IDs and the calculated basic value. The details of the evaluation value calculation processing will be described later in the explanation of the operation, but it is effective to unify the criteria by a unified evaluation scale in order to perform the evaluation for each person from the article content. An EQ (Emotional Intelligence Quotient) model can be adopted to unify the evaluation scale. EQ is based on 8 factors including intra-intelligence (self-recognition, stress coexistence, energy creation, etc.), interpersonal intelligence (self-expression, assertion, interpersonal relationship, etc.), and situation judgment intelligence (interpersonal acceptance, empathy, etc.). Constructed and an indicator of the individual's overall ability. By selecting, for example, a sympathetic factor, a self-recognition factor, or the like as an evaluation scale from factors constituting the EQ, a human evaluation with a standardized standard is realized.
As an evaluation model, besides the EQ, a “five factor model (reference: personality psychology—personality from the perspective of Big Five and clinical practice—Yoshihiko Tanno, first published in April 2003, Science Inc.)” can be adopted.

  The evaluation value output unit 111 transmits the evaluation value calculated by the evaluation value calculation unit 110 together with the user ID to the mobile terminal that has made the presentation request via the network N and the base station B. The user of the portable terminal that has received the evaluation value refers to the presented evaluation value together with the user ID to know the person evaluation for the user.

Subsequently, the operation of the person evaluation system 1 according to the present embodiment will be described with reference to a flowchart, and each step constituting the person evaluation method according to the present invention will be described.
First, the process of constructing a dictionary will be described with reference to the flowchart of FIG. This dictionary is referred to in order to quantify the characteristic degree in the evaluation for each person.
In S1, as pre-preparation, learning data composed of a plurality of articles is classified into three data for each evaluation content. This classification is performed visually by multiple persons. For example, regarding the content of the text of each article constituting the learning data, if more than two-thirds of a plurality of persons determine that the evaluation content is “present”, the article is classified as “affirmed”. On the other hand, if more than two-thirds of the plurality of people judge that the evaluation content is “none”, the article is classified as “negative”. Articles for which other judgments have been made are classified as “none”.

In order to reduce each person's judgment error as much as possible, it is desirable to make the above judgment from the following viewpoints.
・ It can be read from the main text that the intention and the situation of the other party are accurately picked up ・ The person who is speaking in the position of the other party based on the above

In S2, morphological analysis is performed on the text data corresponding to the body text in the article classified as “Yes” in S1. Since morphological analysis is one of well-known and commonly used natural language processing techniques, detailed description and illustration thereof will be omitted, but in S2, sentences written in natural language are converted into morphemes (in this embodiment, words). The work is done to divide the line into parts and identify the part of speech. As a result of the morphological analysis, a part-of-speech decomposed word is extracted from the text.
The morphological analysis is similarly executed for articles classified as “Negative” and “None”, and as a result, words are extracted for all articles classified for each evaluation content.

  In S3, the ease of appearance is quantified (score calculation) based on the appearance frequency for each word extracted in S2. This process is similarly performed for all articles classified by evaluation contents, that is, data of “affirmation”, “denial”, and “none”. The difference in the scores given to the words serves as a reference for determining the evaluation contents of the data including the feature words. Therefore, in order to facilitate the extraction of feature words, the score of the word is likely to be different for each classification, and a high score is assigned to a characteristic word, and a low score is given to a non-character word. Is desirable.

また、学習用データの記事総数をＭ、単語ｉを含む記事の数をＶｉとすると、全記事からみた単語ｉの重要度は、下記数式（２）のように表すことができる。

その結果、下記数式（３）を得る。

From the above viewpoint, it is preferable to use the tfidf method as a scoring method. Assuming that the total number of occurrences of the word i in the category k (affirmation, negation, none) is wik, the appearance frequency tfik of the word i in the category k is expressed by the following equation (1) using the total number of feature words in the category k. Can be represented.

Further, when the total number of articles in the learning data is M and the number of articles including the word i is Vi, the importance of the word i viewed from all the articles can be expressed as the following formula (2).

As a result, the following mathematical formula (3) is obtained.

In S4, based on the score of each word calculated in S3, feature words are extracted from the three data classified for each evaluation content. The feature word can be extracted from the following viewpoints, for example.
1. A word having an idfi value (corresponding to the idf value) equal to or greater than a predetermined threshold is extracted as a feature word.
2. Words whose total number of appearances in all data is a predetermined number or more are extracted as feature words.
3. Symbols that have no meaning as an evaluation scale (“”, (),..., Etc.) are excluded from feature words.

In conventional text mining, only parts of speech (nouns and verbs) that are highly relevant to the subject of a sentence are treated as feature words to be extracted, so other parts of speech are not necessarily recorded in the constructed dictionary. However, since this embodiment is intended to determine the internal characteristics of users in the Internet community, parts of speech other than nouns and verbs are also extracted.
The words extracted in S2 include words that are not necessary for determining the evaluation contents or that have an adverse effect on the determination. However, the above processing is performed to screen the words. . Then, the feature word extracted in S4 and its score are stored in the database, thereby constructing a dictionary (S5).

Next, an evaluation process for each person using the constructed dictionary will be described with reference to the flowchart of FIG.
As a premise of the evaluation process, first, for the data to be evaluated, it is determined whether the evaluation content of the data is “Yes”, “No”, or “None”. Next, the characteristic words included in the evaluation target data are digitized (scored) with reference to the dictionary of the determined evaluation contents. The digitization of feature words is performed for each article using the inner product value. That is, by calculating the inner product value of the feature vector of the feature word (above tfidfik) registered in the dictionary and the feature vector of the article to be evaluated (above tfik), the similarity between the dictionary and the article can be calculated. Estimate and digitize. Digitization is performed using three types of dictionaries with different evaluation contents for each article.

  In S11, an inner product value between the “affirmation” dictionary and the article, an inner product value between the “no” dictionary and the article, and an inner product value between the “none” dictionary and the article are calculated. These inner product values are used for calculation of evaluation values as basic values of the corresponding evaluation contents. Referring to FIG. 3 again, in the present embodiment, for the article having ID “a-29”, the inner product values for each evaluation content are “5.28”, “3.26”, and “0.89”, respectively. Therefore, the basic values (yik described later) are these values.

  However, among the three types of inner product values, only one type of inner product value can be representatively used as a basic value. That is, the inner product value of the evaluation content having the maximum value among the inner product values of the dictionary of each evaluation content and the article can be adopted as the basic value. For example, for an article having ID “a-29”, the inner product values for each evaluation content are “5.28”, “3.26”, and “0.89”, respectively, as described above. Therefore, the inner product value “5.28” of “affirmation” that is the maximum value is used as one of the basic values of the user “ogata”. In this case, an evaluation value (Hk to be described later) presented to the user is also one type regarding affirmation.

Subsequently, a process for classifying all articles to be evaluated for each user ID is performed in order to perform a tabulation based on a user ID from a numerical result in a data unit (S12). .
In S13, the person evaluation server 10 uses the evaluation value calculation unit 110 to apply the basic value of articles having the same user ID to a predetermined evaluation formula, and calculates an evaluation value for each person. In calculating the evaluation value, the ratio of the number of articles for each evaluation content in the total number of posted articles may be evaluated as a coefficient, and the value may be incorporated into the evaluation formula.

ここで、ｙｉｋは、人物ｋの記事ｉにおける基礎値（本第１の実施形態では、内積値そのもの）である。また、Ｎｋは、人物ｋの発信した、ある評価内容の記事数（例えば、“肯定”の評価内容を得た記事数）であり、Ａｋは人物ｋの発信した記事総数である。 In this case, the evaluation value Hk is calculated from the following evaluation formula (4) using yik, Nk, and Ak.

Here, yik is a basic value in the article i of the person k (inner product value itself in the first embodiment). Further, Nk is the number of articles with a certain evaluation content (for example, the number of articles having an evaluation content of “Yes”) transmitted by the person k, and Ak is the total number of articles transmitted by the person k.

  The evaluation value Hk obtained in S14 is displayed on the terminal of each user who uses the Internet community together with the corresponding user ID in response to the request (S15). The terminals on which the evaluation values are displayed are, for example, portable terminals 20, 30, and 40.

  As described above, the calculation of the evaluation value with reference to the dictionary constructed using the unified evaluation scale in the person evaluation system 1 of the first embodiment contributes to a person evaluation that is fair and highly reliable. . Moreover, the calculation of the evaluation value for each user ID using the inner product value corresponding to the evaluation content contributes to the evaluation not only for the article but also for each person. Therefore, a user of the Internet community to which the person evaluation system 1 is applied can communicate with the unspecified and many people with confidence by referring to the evaluation value presented on the mobile terminal. .

(Second Embodiment)
Next, a second embodiment of the present invention will be described with reference to FIG.
In the first embodiment, the inner product value itself is used as a basic value when executing the person-unit evaluation process. However, in the present embodiment, the difference between the inner product values of articles having different evaluation contents is calculated. For this reason, the person evaluation server in the present embodiment further includes an inner product value difference calculation unit 112 as shown in FIG. In FIG. 7, the configuration of the person evaluation server in the second embodiment is similar to the configuration of the person evaluation server 1 described in detail in the first embodiment. The description thereof is omitted, and differences from the first embodiment are described in detail.

  Although the detailed function will be described later in the explanation of the operation, the inner product value difference calculating unit 112 (corresponding to the inner product value difference calculating unit) which is a specific component is an inner product for each evaluation content calculated by the inner product value calculating unit 16. Among the values, a difference between an inner product value with an evaluation content of “affirmation” and an inner product value with an evaluation content of “no” is calculated as a basic value.

  Next, a person unit evaluation process in the second embodiment will be described with reference to FIG. In addition, each step constituting the person evaluation method according to the present invention will be described. The personal unit evaluation process is basically the same as the personal unit evaluation process (see FIG. 6) described in detail in the first embodiment. Specifically, S21 and S23 to S25 in FIG. 8 correspond to S11 and S12 to S14 shown in FIG.

Hereinafter, with reference to the flowchart of FIG. 8, S22 which is the difference of each person evaluation process is demonstrated. That is, in S22, the inner product value difference calculation unit 112 registers the inner product value of the feature vector of the feature word registered in the “affirmation” dictionary and the feature vector of the article to be evaluated, and the “no” dictionary. The difference between the inner product value of the feature vector of the feature word and the feature vector of the article is calculated as a basic value. Instead of the “negative” dictionary, an inner product value between the “none” dictionary and the article may be used.
Thereafter, the evaluation value calculation unit 110 substitutes the basic value calculated in S22 into yik in the evaluation formula (4), and thereafter calculates the evaluation value by the same method as in the first embodiment.

  As described above, according to the person evaluation system 1 in the second embodiment, the inner product value is used to estimate the similarity between the positive dictionary and the article to be evaluated, and the difference between them is reflected. The evaluation value can be presented to the user of the mobile terminal 20, 30, 40. This evaluation value is an evaluation value based on a positive evaluation content, and the evaluation value to be presented is this one type. Therefore, the user can easily evaluate the person to be evaluated with reference to this evaluation value. It is possible to grasp quickly.

In addition, the aspect described in each said embodiment is a suitable example of the person evaluation apparatus which concerns on this invention, and this invention is not limited to this aspect.
For example, in the above embodiment, the server device having the dictionary construction function and the server device having the human unit evaluation function are described and illustrated as an integrated device, but these functions are distributed to two or more server devices. The physical configuration and operation mode of the devices constituting the system may be arbitrary.
The mobile terminal is, for example, a mobile phone, but is not limited thereto, and may be a PDA (Personal Digital Assistants) or a PC (Personal Computer).

It is a figure which shows the functional structure of the person evaluation system in 1st Embodiment. It is a figure which shows the example of data storage in a dictionary database. It is a figure which shows the example of data storage in an article database. It is a figure which shows the example of data storage in a user database. It is a flowchart for demonstrating the dictionary construction process in 1st Embodiment. It is a flowchart for demonstrating the person unit evaluation process in 1st Embodiment. It is a figure which shows the functional structure of the person evaluation system in 2nd Embodiment. It is a flowchart for demonstrating the person unit evaluation process in 2nd Embodiment.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1, 2 ... Person evaluation system, 10 ... Person evaluation server, 10a ... Dictionary construction unit, 10b ... Person unit evaluation unit, 11 ... Data classification part, 12 ... Data analysis part, 13 ... Score calculation part, 14 ... Feature word extraction , 15 ... Dictionary database, 16 ... Inner product value calculation unit, 17 ... Article classification unit, 18 ... Article database, 19 ... User database, 110 ... Evaluation value calculation unit, 111 ... Evaluation value output unit, 112 ... Inner product value difference calculation Part, B ... base station, N ... network, 20, 30, 40 ... mobile terminal

Claims (10)

A dictionary construction means for constructing a dictionary in which a feature word included in the learning data and a score for the feature word are associated for each evaluation content according to a unified evaluation scale;
An evaluation means for evaluating each user based on identification information and basic values of the plurality of users by referring to the data of the plurality of users constituting the learning data and the dictionary; Person evaluation device. The dictionary construction means includes
Data classification means for classifying the learning data for each evaluation content;
Data analysis means for decomposing the learning data classified by the data classification means into words;
Score calculation means for assigning a score corresponding to the appearance frequency to a word obtained as a result of decomposition by the data analysis means;
2. A feature word extraction unit that extracts a feature word from the learning data classified for each evaluation content based on the score of each word given by the score calculation unit. The person evaluation device described in 1.   3. The person evaluation device according to claim 2, wherein the feature word extraction unit extracts, as a feature word, a word having an idf value equal to or greater than a threshold among the words given a score by the score calculation unit.   3. The feature word extraction unit extracts, as a feature word, a word whose total number of appearances in the learning data is greater than or equal to a threshold among the words scored by the score calculation unit. Person evaluation device.   3. The person evaluation device according to claim 2, wherein the feature word extraction unit excludes, from the feature words, unnecessary symbols for calculating an evaluation value from words extracted as feature words. The evaluation means includes
Inner product value calculating means for calculating the inner product value of the feature vector of the feature word registered in the dictionary and the feature vector of the data to be evaluated as a basic value for each evaluation content;
The person evaluation apparatus according to claim 1, further comprising: an evaluation value calculation unit that calculates an evaluation value for each user using the identification information of the plurality of users and the calculated basic value.   It further comprises an inner product value difference calculating means for calculating a difference between an inner product value of one evaluation content and an inner product value of another evaluation content as a basic value from the inner product value for each evaluation content calculated by the inner product value calculating means. The person evaluation device according to claim 6.   The person evaluation apparatus according to claim 6, wherein the evaluation value calculation unit calculates the evaluation value by using a sum of the basic values corresponding to a user to be evaluated and data.   The evaluation value calculation means multiplies the sum of the basic values corresponding to the user and data to be evaluated by the number of data for each evaluation content corresponding to the user, and divides the total number of data corresponding to the user The person evaluation device according to claim 6, wherein the evaluation value is calculated. Associating a feature word included in the learning data with a score for the feature word and constructing a dictionary for each evaluation content according to a unified evaluation scale;
Performing evaluation in units of users based on the identification information and basic values of the plurality of users by referring to the data of the plurality of users constituting the learning data and the constructed dictionary. Character evaluation method.

Images