KR101346115B1 - System and method for creating emotional word dictionary and computing emotional degrees of documents - Google Patents

Abstract

An emotional vocabulary dictionary generation and an emotional strength calculation system and method of a document using the same are provided. According to an embodiment of the present invention, an emotional vocabulary dictionary generation method and a method for calculating the emotional strength of a document using the same include: extracting a plurality of emotional words from a document set including one or more words and documents including an evaluation score in the emotional word extraction module; Extracting; Generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation scores; And in the emotion intensity calculation module, calculating an emotion intensity of a target document using the semantic classification matrix and the semantic weight matrix.

Description

SYSTEM AND METHOD FOR EMOTIONAL WORD DISTANCE AND COMPUTING EMOTIONAL DEGREES OF DOCUMENTS}

The present invention relates to a technique for generating an emotional vocabulary from a document and using it to calculate the emotional strength, ie, positive / negative strength, of the document.

In recent years, the rapid growth of the social web has exploded the frequency of expressing opinions on the Internet. It forms and reflects people's opinions. Therefore, more and more attempts have been made to analyze emotions in recent years. Now we must analyze the web to understand certain phenomena in society.

All the research so far has been focused on classification. For example, positive learning, subject classification, and methods using general statistics through machine learning. When reading some emotional texts and understanding human emotions, one of the most important functions is the average amount of a priori data shared within a language social group on key keywords. Therefore, for emotional analysis, it is important to extract important emotional words and how to analyze them.

Foreign researches on this include WordNet and SentiWordNet. But WordNet wasn't exactly the study for this, but SentiWordNet showed emotional strength, but in terms of methodology, it was too dependent on human intuition, and given only a value between 0.0 and 1.0, it was difficult to represent all a priori knowledge.

Today's corporate data analysts can't afford the amount of text data they need to analyze to understand customer responses. To do this, it is virtually impossible to analyze all textual data manually. This exposes the company to situations where it is not aware of current customer complaints and cannot respond appropriately. Of course, there are ratings in reviews, but there are a lot of problems to believe. Most review systems allow users to assign reviews and ratings at the same time, but this has a problem of discrepancy. Writing reviews and giving ratings is not a direct connection. In humans, this intermediate process is the "Human Decision Making Process". Therefore, if the machine can read and understand the text automatically like a human, it can have the advantage of fast processing of large review text comments.

It is an object of the present invention to provide a means for extracting an emotional word from a document such as a review, a review, and effectively determining the positive / negative emotional strength of the document.

According to an embodiment of the present invention, an emotional vocabulary dictionary generation method and a method for calculating the emotional strength of a document using the same include: extracting a plurality of emotional words from a document set including one or more words and documents including an evaluation score in the emotional word extraction module; Extracting; Generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation scores; And in the emotion intensity calculation module, calculating an emotion intensity of a target document using the semantic classification matrix and the semantic weight matrix.

In addition, the emotional vocabulary dictionary generation system according to an embodiment of the present invention and the emotional intensity calculation system of the document using the same, a plurality of emotional words from a document set consisting of one or more words and documents including an evaluation score for a predetermined object Emotion word extraction module for extracting; A matrix generation module for generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation scores; And an emotion intensity calculation module that calculates an emotion intensity of a target document using the semantic classification matrix and the semantic weight matrix.

According to the present invention, there is an advantage that an effective and accurate opinion analysis service can be provided by using a semantic network algorithm using a dictionary and an automatic extraction algorithm of a positive rate and negative strength. In other words, it is possible to contribute to the paradigm shift to build an effective customer response system by providing an analysis report by extracting important information from a large amount of review text data in a short time for the managers and data analysts of the company. There is.

1 is a block diagram illustrating an emotional vocabulary dictionary generation and emotional strength calculation system 100 of a document according to an embodiment of the present invention.
2 is a flowchart illustrating a method 200 for generating an emotional vocabulary dictionary and calculating an emotional strength of a document according to an embodiment of the present invention.
3 is a flowchart illustrating a detailed process of the emotion intensity calculation step 206 of the target document according to the present invention.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. However, this is merely an example and the present invention is not limited thereto.

In the following description, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. The following terms are defined in consideration of the functions of the present invention, and may be changed according to the intention or custom of the user, the operator, and the like. Therefore, the definition should be based on the contents throughout this specification.

The technical idea of the present invention is determined by the claims, and the following embodiments are merely a means for effectively explaining the technical idea of the present invention to a person having ordinary skill in the art to which the present invention belongs.

1 is a block diagram illustrating an emotional vocabulary dictionary generation and emotional strength calculation system 100 of a document according to an embodiment of the present invention. As shown, the emotional vocabulary dictionary generation and emotional intensity calculation system 100 of the document according to an embodiment of the present invention is the emotional word extraction module 102, the matrix generation module 104 and the emotional intensity calculation module 106 It includes.

The emotion word extraction module 102 extracts a plurality of emotion words from a document set including one or more words and documents including an evaluation score for a predetermined object.

The document set is a set of a plurality of documents, each document including one or more sentences for expressing the opinions or feelings of the author for a specific object and evaluation score information for the specific object. For example, the documents may be reviews on a movie or music and a rating for the movie. For example, on a movie-related website, each user may assign a score for the movie to a value between 1 and 10, along with a review that records the feelings of the movie they watched, and the like. The above rating may be. Of course, depending on the website, the score may be classified into five sections instead of ten sections, or even ten sections may have various rating methods such as -5 to 5 instead of 1 to 10. That is, the present invention is not limited to a specific scoring method, and it is noted that any rating (or scoring) system may be used in the present invention through an appropriate scaling process.

The emotion word is a word extracted from the documents belonging to the document set as described above, and means a word for representing the emotion of the author. For example, words such as "fun", "fun", "fear", "sad" and the like can be the emotional words in the present invention. The emotion word extraction module 102 may include a candidate list of emotion words therein, and search for documents included in the document set and extract the words included in the list as an emotion word. Algorithms for extracting a specific word from a document are well known in the art, so a detailed description thereof will be omitted herein.

Next, the matrix generation module 104 generates a Semantic Classification Matrix (SCM) and a Semantic Weight Matrix (SWM) from the extracted emotional words and the evaluation scores. The SCM and SWM will be described in detail below.

Next, the emotion intensity calculation module 106 calculates the emotion intensity of the target document using the semantic classification matrix and the semantic weight matrix.

2 is a flowchart illustrating a method 200 for generating an emotional vocabulary dictionary and calculating an emotional strength of a document according to an embodiment of the present invention.

As illustrated, the method 200 for generating an emotional vocabulary dictionary and calculating the emotional strength of a document according to an embodiment of the present invention includes a plurality of emotional words from a document set consisting of documents including one or more words and an evaluation score. Extracting 202, generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation score 204 and the semantic classification matrix and the semantic Computing the emotional intensity of the target document using the weight matrix (206).

Hereinafter, the detailed configuration of each step will be described.

Extract Emotion Words from Document Set (202)

This step is to retrieve documents included in the document by searching for documents belonging to the document set. As described above, the step may be configured to search for the documents included in the document set and extract the words included in the preset emotion word candidate list as the emotion words.

Semantic  Classification matrix ( SCM ) And Semantic  Weighting matrix ( SWM ) Generation (204)

This step calculates a Semantic Classification Matrix (SCM) and a Semantic Weight Matrix (SWM) of emotional words extracted from a document set.

SCM is an n-row 4-column (nx 4) matrix, with each row corresponding to each sentiment word extracted from the document set. That is, the n value is equal to the number of words extracted from the document set. Each column of the SCM is called a Semantic Classification Feature (SCF), and the SCF includes the following values.

SCF = {AT, PR, ONT, P}

AT (Attitude Type) is an emotion classification type of the corresponding emotion word. For example, White's Classification may be used. The white classification is classified into 24 kinds of human emotions according to their kinds.

(1) basic human emotion classification

"affect_DIS / INCLINATION_fear"

"affect_DIS / INCLINATION_desire"

"affect_UN / HAPPINESS_misery"

"affect_UN / HAPPINESS_antipathy"

"affect_UN / HAPPINESS_cheer"

"affect_UN / HAPPINESS_affection"

"affect_IN / SECURITY_disquiet"

"affect_IN / SECURITY_surprise"

"affect_IN / SECURITY_confidence"

"affect_IN / SECURITY_trust"

"affect_DIS / SATISFACTION_ennui"

"affect_DIS / SATISFACTION_displeasure"

"affect_DIS / SATISFACTION_interest"

"affect_DIS / SATISFACTION_pleasure"

(2) Classification of emotions according to social standards

"judgement_NORMALITY"

"judgement_CAPACITY"

"judgement_TENACITY"

"judgement_VERACITY"

"judgement_PROPRIETY"

(3) classification of emotions according to aesthetic criteria;

"appreciation_REACTION_impact"

"appreciation_REACTION_quality"

"appreciation_COMPOSITION_balence"

"appreciation_COMPOSITION_complexity"

"appreciation_VALUATION"

In the above example, the first category (e.g. affect, judgement, etc.) of each emotion classification is a major category, the second category (NORMALITY, CAPACITY, etc.) is a middle category, and the third category (ennui, displeasure, interest, etc.) is a small category.

In the present invention, the emotional words are grouped together with words having the same or similar meanings to form a plurality of synonyms, and each set of synonyms corresponds to one of the white classifications. Then, when emotion words are extracted from the document set, the AT value is calculated by finding out the synonym set to which the extracted emotion words belong and the white classification associated with (corresponding) the synonym set. At this time, the AT value may be a number assigned to the corresponding white classification. For example, different numbers may be assigned to each of the white classifications, and the assigned numbers may be used as AT values of respective emotion words.

Meanwhile, in the present invention, the white classification is used to calculate the AT value. However, this is only an example, and any classification related to the emotion classification may be used.

Next, PR (PRototypicality) is a circularity of the emotion word, it can be expressed as an integer between 0 and k-1. K is the number of words in the synonym set to which the corresponding emotional word belongs. The PR value is a value indicating how close the emotion word is to the prototypical meaning of the emotion to which the synonym set belongs, and is determined by an SV value to be described later. That is, the PR value of the word is determined by sequentially assigning integers from 0 to k-1 in order of decreasing SV value in the synonym set.

ONTology (ONTology) is an ontology value of the emotion word, and means an ontology node value assigned to the word in the ontology tree including the emotion word. One ONT value is assigned to each synonym set. That is, to obtain the ONT value of a specific emotion word, a synonym set to which the corresponding emotion word belongs is obtained, and a node value of the word is obtained by searching the ontology tree for a word having a PR value of 0 among words in the synonym set.

The ontology tree may vary depending on the document set to be processed and the type of emotional word. For example, if the document set is related to a movie, the ontology tree associated with the movie may be used. If the document set is an automobile test vehicle, the ontology tree associated with the vehicle may be used.

P (Polarity) is a value indicating the polarity of the emotion word, and has a value of +1 when the emotion word is used in a positive sense and -1 when used in a negative sense. The P value can be obtained as follows.

First, the rating given to each document is divided into a positive rating and a negative rating. For example, if the rating given to each document is between 1 and 10, it can be classified into negative ratings from 1 to 5 and positive ratings from 6 to 10. The classification may be performed, for example, by obtaining a median (or average) value of each of the ratings, dividing a score lower than the calculated median (or average) into a negative value and a high score by a positive value.

Next, in order to give a polarity of a specific emotional word, all the documents in which the emotional word appears are compared, and the ratings in the extracted document are compared. If the scores of the documents including the emotional word are more negative than the positive case, the P value of the emotional word may be set to -1 and vice versa to +1. Alternatively, the average of the ratings of the documents in which the corresponding emotional word appears may be calculated, and the P value may be determined according to whether the average value belongs to a negative rating or a positive rating. That is, the average score (evaluation score) of the documents including the emotional word may be set to +1 if the average is higher than the evaluation score of the entire document, and to -1 if the average is low.

Next, SWM is an n-row 3-column (nx 3) matrix, with each row corresponding to each sentiment word extracted from the document set. That is, the n value is equal to the number of words extracted from the document set. Each column of the SCM is called a SWF (Semantic Weight Feature), and the SWF includes the following values.

SWF = {TFIDF, ZT, SV}

TFIDF (Term Frequency-Inverse Document Frequency) is a statistical measure of how important a word is in a particular document when there is a group of documents. The term frequency (TF) is a value that indicates how often a particular word appears in a document. The higher this value, the more important the document may be considered. However, if the word itself is often used in a family of documents, this means that the word is common. This is called DF (document frequency), and the inverse of this value is called inverse document frequency (IDF). TFIDF is the product of TF and IDF. That is, in the present invention, the TFIDF of each emotion word is a value obtained by multiplying the frequency of the emotion word by the inverse of the frequency of the document in which the emotion word appears.

ZT is a value obtained by normalizing the TFIDF value of the corresponding emotional word in the range of [0.0, 1.0] according to the Z-transform.

The SV (Semantic Value) is a meaning value of an emotional word and is obtained through the following process.

First, the words included in the synonym set to which the emotional words belong are sorted according to the TFIDF value. At this time, the TFIDF value of the word having the highest TFIDF value becomes a representative value of the set of synonyms. When the words are sorted, the frequency of the rating of the document in which each word appears is calculated, and the highest frequency is set as the initial value of the word. The initial value is scaled in the range of positive [0.5, 10.5], negative [-10.5, -0.5] depending on whether the rating is positive or negative. In this case, whether the rating is positive or negative may be determined depending on whether the corresponding rating is higher or lower than the average of the ratings of the entire document. Alternatively, a predetermined reference value (for example, 5 points) may be set, and a positive rating may be set to a higher rating than the reference value and a negative rating to a lower rating.

Next, add the Z-normalized value of the TFIDF value of the word to the scaled initial value. Through this process, each word in the set of synonyms has a single continuous distribution, that is, between [-11.0, +11.0], because the Z-normalized value is based on the grade point .5. This is because it has a range of [0.0, 1.0]. Thereafter, the calculated value is scaled again to a range of [−5.5, 5.5] to generate an SV value of each word in the synonym set.

Calculate Emotional Strength of Target Document (206)

After generating the SCM and SWM through the above-described steps, the emotional strength of the target document is calculated using the SCM and SWM.

3 is a flowchart illustrating a detailed process of the emotion intensity calculation step 206 of the target document according to the present invention.

First, an emotional word is extracted from the target document (302). Regarding a method of extracting a word from a specific document, various methods are well known in the art, and thus detailed description thereof will be omitted.

Next, a classification and weight matrix of the emotion words extracted using the pre-generated SCM and SWM is generated (304). Next, a CWM (Classification and Weight Matrix) of the document is generated using the extracted emotion word. The CWM is a matrix generated by using the above-described values of SCM and SWM, and is an n-row 4-column matrix composed of the following values (where n is the number of extracted emotional words).

{AT, PR, P ', SV'}

Among these, AT and PR values use AT and PR values in SCM as they are, and P 'and SV' are values obtained by scaling P and SV values by the following equation.

[Equation 1]

&Quot; (2) "

The generated CWM is then separated into a positive matrix and a negative matrix (306). In this case, the positive matrix is a matrix including only words having a P value of +1 among the emotional words included in the CWM, and the negative matrix is a matrix including only words having a P value of −1 among the emotional words included in the CWM.

Next, singular value decomposition is performed on the generated positive matrix and the negative matrix (308). The sigma matrix can then be obtained from the respective matrices (positive matrix and negative matrix). In this case, let the values of the diagonal components ((1, 1), (2, 2), (3, 3), ...) of each sigma matrix be Σ _i (i is the number of rows or columns of the sigma matrix).

Next, the emotion intensity of the target document is calculated using the sigma matrix of the positive matrix and the negative matrix generated by the singular value decomposition and the length of the target document (310). If the maximum value of the Σ _i values is max (Σ _i ), the emotional strength of the target document is determined as in Equation 3 below.

&Quot; (3) "

Where length _doc is the length of the target document and θ is the emotional strength of the document. This is generally called a normalization factor and is defined as follows in the present invention. As shown in the above equation, by dividing max (Σ _i ) by the log value of the length of the target document, it is possible to prevent the emotional intensity of the document according to the document length. If max (Σ _i ) is not divided by the logarithm of the length of the target document as described above, there is a concern that the longer the document length, the stronger the emotional document, and the shorter the document, the wrong document is expressed. Because.

In the above equation, when Σ _i value is derived from the positive matrix, the θ value is negative emotion intensity of the target document, and is derived from the negative matrix.

On the other hand, the value of θ may be normalized to a range between [0, 1] by using a classification function as in the following equation.

&Quot; (4) "

In the above equation, P is a normalized emotion intensity, α and β are adjustment coefficients, which are values that can be appropriately set in consideration of characteristics of the target document and the like.

On the other hand, an embodiment of the present invention may include a computer-readable recording medium including a program for performing the methods described herein on a computer. The computer-readable recording medium may include a program command, a local data file, a local data structure, or the like, alone or in combination. The media may be those specially designed and constructed for the present invention or may be known and available to those of ordinary skill in the computer software arts. Examples of computer readable media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs and DVDs, magneto-optical media such as floppy disks, and magnetic media such as ROMs, And hardware devices specifically configured to store and execute program instructions. Examples of program instructions may include machine language code such as those generated by a compiler, as well as high-level language code that may be executed by a computer using an interpreter or the like.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is clearly understood that the same is by way of illustration and example only and is not to be construed as limiting the scope of the present invention. I will understand.

Therefore, the scope of the present invention should not be limited to the above-described embodiments, but should be determined by equivalents to the appended claims, as well as the appended claims.

100: emotional vocabulary dictionary generation and emotional strength calculation system of the document
102: emotion word extraction module
104: matrix generation module
106: Emotional Strength Calculation Module

Claims (15)

In the emotion word extraction module, extracting a plurality of emotion words from a document set consisting of documents including one or more words and an evaluation score;
Generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation scores; And
In the emotion intensity calculation module, calculating the emotional vocabulary of the target document from the component value corresponding to the emotion word included in the target document of the component values of each of the semantic classification matrix and the semantic weight matrix And method for calculating emotional strength of documents.
The method according to claim 1,
The semantic classification matrix is an N-row 4-column matrix including an emotion classification type (AT), a circularity (PR), an ontology value (ONT), and a polarity (P) of each of the extracted emotion words, where N is an extracted emotion. Number of words), and the semantic weight matrix is N rows including inverse document frequency (TFIDF), Z-normalized value of the inverse document frequency, and semantic value (SV) in the document of each of the extracted emotional words. A method of generating an emotional vocabulary dictionary, which is a three-column matrix, and a calculation of the emotional strength of a document.
The method according to claim 2,
The emotion classification type is a classification value obtained by classifying the extracted emotion words according to White's Classification.
The method according to claim 2,
And the ontology value is an ontology node value assigned to the emotion word in the ontology tree including the extracted emotion word.
The method according to claim 2,
The polarity value is compared with the average of the evaluation scores of the documents containing the extracted emotional word and the total evaluation score of the entire document included in the document set, the average of the evaluation score of the documents containing the emotional word is the evaluation of the entire document A method of generating an emotional vocabulary dictionary and calculating an emotional strength of a document, which is set to +1 if the score is higher than -1 and to -1 if the score is lower than the score.
The method according to claim 2,
The meaning value is a document including each of the emotional words aligned by sorting the emotional words included in the synonym set to which the emotional words extracted belong according to the size of the inverse bibliographic frequency value in the document set of the respective emotional words. Sets an initial semantic value according to the evaluation scores of the subjects, and generates an emotional vocabulary dictionary, which is determined by scaling the initial semantic value to a predetermined range by adding a Z-normalized value of the inverse bibliographic frequency of each emotional word. How to calculate the emotional strength of documents.
The method of claim 6,
An initial meaning value of each emotion word is determined by the most frequent evaluation score among evaluation scores of the documents in the document set including each emotion word.
The method of claim 6,
The circularity values are arranged in the order that the emotional words included in the extracted emotional word belong to the synonym set belonging to each other in the order of decreasing meaning value SV, and the integers from 0 to k-1 in the sorted emotional words. Wherein k is an integer value assigned to the extracted emotional word when the emotional word is included in the synonym set.
The method according to claim 1,
Calculating the emotion intensity of the target document,
In the emotion intensity calculation module, extracting an emotion word from the target document;
Generating, by the emotion intensity calculation module, a classification and weight matrix from component values corresponding to emotion words extracted from the target document among component values of each of the semantic classification matrix and the semantic weight matrix;
In the emotion intensity calculation module, separating the classification and weight matrix into a positive matrix and a negative matrix;
In the emotion intensity calculation module, performing singular value decomposition for each of the positive matrix and the negative matrix; And
Calculating an emotional vocabulary of the target document by normalizing a value derived from the diagonal component values of the sigma matrix of each of the positive matrix and the negative matrix generated by the singular value decomposition to the length of the target document. How to create a dictionary and calculate the emotional strength of a document.
The method of claim 9,
The classification and weighting matrix includes N rows including an emotion classification type (AT), a circularity (PR), a scaled polarity (P ′), and a scaled semantic value (SV ′) of each emotion word extracted from the target document. A method of generating emotional vocabulary dictionaries, wherein N is the number of emotional words extracted, and a four-column matrix.
The method of claim 10,
The scaled polarity P 'is represented by the following equation

Where P is the polarity stored in the semantic classification matrix and PR is the circularity value stored in the semantic classification matrix.
A method for generating an emotional vocabulary dictionary and calculating the emotional strength of a document.
The method of claim 10,
The scaled meaning value SV ′ is represented by the following equation

Where SV is a semantic value stored in the semantic weight matrix and PR is a circularity value stored in the semantic classification matrix.
A method for generating an emotional vocabulary dictionary and calculating the emotional strength of a document.
The method of claim 9,
The positive matrix is a matrix composed of words having a polarity (P) value of +1 in the semantic classification matrix among the emotional words included in the classification and weight matrix, and the negative matrix is an emotion included in the classification and weight matrix. A method of generating an emotional lexicon and calculating emotional intensity of a document, wherein the word is a matrix composed of words having a polarity (P) value of -1 in the semantic classification matrix.
The method of claim 9,
The emotional strength of the target document is represented by the following equation.

Where i is the number of rows of the sigma matrix calculated from the positive matrix or the negative matrix, Σ _i is the diagonal component of the sigma matrix, max (Σ _i ) is the maximum value of the diagonal component of the sigma matrix, and length _doc is the object Document length)
A method of calculating the emotional vocabulary and calculating the emotional strength of the document.
An emotion word extraction module for extracting a plurality of emotion words from a document set including one or more words and documents including an evaluation score for a predetermined object;
A matrix generation module for generating a semantic classification matrix and a semantic weight matrix from the extracted emotional words and the evaluation scores; And
An emotional vocabulary dictionary generation module comprising an emotion intensity calculation module for calculating an emotion intensity of the target document from component values corresponding to emotion words included in the target document among the component values of each of the semantic classification matrix and the semantic weight matrix; Emotional strength calculation system.

Images