KR101534141B1 - Rationale word extraction method and apparatus using genetic algorithm, and sentiment classification method and apparatus using said rationale word - Google Patents

Abstract

Disclosed is an extraction method of emotion rationale, which comprises the steps of: arbitrarily generating an initial chromosome group based on a word in a document (herein, attributes of each gene of a chromosome mean each word presented in a training data set of the document and an attribute value means whether or not the word is selected); evaluating each chromosome through an evaluation function (the evaluation function is for maximizing accuracy of classification of a document polarity) for the initial chromosome group; generating new chromosome by going through gene calculation based on an evaluation score of each chromosome; and generating an emotion rationale group by determining whether the generated chromosome satisfies termination conditions or not.

Description

TECHNICAL FIELD [0001] The present invention relates to a method and apparatus for extracting emotion based on genetic algorithms, and a method and apparatus for emotion classification using the emotion basis.

The present invention relates to a method and apparatus for extracting emotional basis, and more particularly, to a method and apparatus for performing emotional analysis by extracting evidence serving as a key among all sentences or words included in a document.

Conventional emotional analysis methods are classified into Lexicon-based Approach and Text Classification Approach. The emotion dictionary-based technique is a method of classifying whether a document is positive or negative according to the presence or absence of emotional vocabulary using a conventional emotional dictionary. Generally, the emotional dictionary considers the meaning of a commonly used vocabulary . Therefore, there is a problem that sensitivity can be lowered in sensitivity analysis for a specific domain, and performance results may be greatly changed depending on the domain. In addition, when constructing the emotionally false dictionary manually according to the domain, the time cost is very large and the result may be biased toward the domain. Next, the method using the machine learning expresses each word contained in the document as a feature, and performs emotional analysis using a machine learning algorithm. It is very important to set the criteria for attribute selection because the results can be very different depending on how the attribute selection is made. This requires prior knowledge of the domain experts. Therefore, both conventional methods require a lot of time and effort.
In particular, Korean Patent Laid-Open Publication No. 10-2013-0092342 discloses an emotional vocabulary dictionary creation system and a system for calculating the emotional intensity of a document using the emotional vocabulary dictionary system. Korean Patent Application Publication No. KR-10-2013-0075124 extracts emotional words from texts, Emotion analyzing apparatus for analyzing emotional state, which is also related to machine learning emotional analysis.

SUMMARY OF THE INVENTION An object of the present invention to solve the above problems is to provide a method and apparatus for extracting emotion evidence for performing emotion analysis of a document with high accuracy at low cost.

Another object of the present invention is to provide a method and apparatus for extracting emotional grounds that can minimize the cost of emotional analysis using a genetic algorithm, extract emotional bases that best represent the polarity of the document, And an apparatus.

According to an aspect of the present invention, there is provided a method of extracting emotion bases from at least one document, the method comprising: arbitrarily generating an initial chromosome set based on words in the document, Means an individual word appearing in a training data set, an attribute value indicates whether a word is selected, a function for maximizing the accuracy of classifying the polarity of the document, an evaluation function for the initial chromosome set, Evaluating each chromosome, generating a new chromosome by performing a gene operation based on the evaluation score of each chromosome, and generating an emotion basis set by determining whether the generated chromosome satisfies the termination condition .

Generating an emotion basis set by determining whether the generated chromosome satisfies the termination condition includes determining whether the generated chromosome satisfies the termination condition and, if the termination condition is satisfied, Repeating the step of evaluating the chromosome and the step of generating the new chromosome, and repeating the process of determining whether the end condition is satisfied when the end condition is not satisfied.

Each word appearing in the training data set is a word appearing both in the target document and the emotion dictionary data, and may have a positive or negative score according to the emotion dictionary data.

Each chromosome in the initial chromosome set can be expressed as a binary string indicating whether or not the candidate emotion based words are selected.

The step of generating a new chromosome based on the evaluation function of each of the chromosomes may include generating a new chromosome based on a plurality of chromosome binary strings in the chromosome set by using predetermined adjacent string pairs as a tournament selection, - selecting a candidate for the mating based on the consideration of the mismatch probability, performing a single-point mating by exchanging substrings at a predetermined mating point with the selected string pair, And mutating the individual attribute values.

The termination condition can be determined by whether or not the variation amount of the evaluation score of the chromosome generated through the gene operation has a converging shape.

The evaluation score is a difference value between the positive emotion index - the sum of the positive scores of the training data set - and the negative emotion index - the sum of the negative scores of the training data set - to identify the polarity of the document . ≪ / RTI >

The evaluation score may be a ratio of the number of correctly classified documents to the total number of documents using the generated emotion basis set, a ratio of the number of correctly identified documents among the affirmative classified documents using the generated emotion basis set And a ratio of the number of correctly negated documents among the documents classified as negative using the ratio and the generated emotion basis set.

According to an aspect of the present invention, there is provided an apparatus for extracting an emotional basis of at least one document according to the present invention includes a chromosome set generator for randomly generating an initial chromosome set based on words in the document, Means an individual word appearing in a training data set in the document, and an attribute value indicates whether a word is selected. The evaluation function-evaluation function is a function that maximizes the accuracy of classifying the polarity of the document -, an evaluation unit for evaluating each chromosome through a chromosome, a gene operation unit for generating a new chromosome by performing a gene operation based on the evaluation score of each chromosome, and an emotion generation unit for judging whether the generated chromosome satisfies the termination condition And a basis set generation unit.

According to another aspect of the present invention, there is provided a method for extracting an emotional basis of at least one document according to the present invention, the method comprising: arbitrarily generating an initial chromosome set based on words in the document, Attribute indicates each word appearing in the training data set in the document, and the attribute value indicates whether the word is selected. The evaluation function-evaluation function for the initial chromosome set maximizes the accuracy of classifying the polarity of the document A step of evaluating each chromosome through a function of generating a chromosome, a step of generating a new chromosome by performing a gene operation based on the score of each chromosome, the step of generating an emotion basis set by judging whether the generated chromosome satisfies the termination condition And emotional classification of the test target document using the generated emotion basis set It can include.

The step of performing emotional classification may include calculating emotional analysis accuracy for evaluation using the generated emotion basis set through a ratio of the number of correctly classified documents to the number of documents.

The step of performing emotional classification includes classifying the emotion having a higher value into the emotion of the test object document by calculating the positive emotion index and the negative emotion index based on the emotion based set words included in the test object document .

According to another aspect of the present invention, there is provided an apparatus for extracting an emotional basis of at least one document according to the present invention, the apparatus comprising: an arbitrary generator for generating an initial chromosome set based on words in the document, Attribute indicates each word appearing in the training data set in the document, and the attribute value indicates whether the word is selected. The evaluation function-evaluation function is used to maximize the accuracy of classifying the polarity of the document And generating a new chromosome based on the evaluation score of each chromosome through a genetic operation and determining whether the generated chromosome satisfies the termination condition to generate an emotion basis set, And emotional classification of the test target document using the generated emotion basis set And an emotional classifier for performing the emotional classifier.

According to the emotion-based extracting method and apparatus of the present invention, it is possible to minimize the cost of emotional analysis using a genetic algorithm and to extract the emotional basis that best represents the polarity of the document.

In addition, since there is no need to manually build an emotional vocabulary dictionary without requiring expert knowledge of the domain, it is possible to perform low-cost and high-sensitivity emotional analysis.

1 is a block diagram showing a system to which an emotion-based reason extracting apparatus according to an embodiment of the present invention is applied;
FIG. 2 is a flowchart schematically illustrating an emotion-based extracting method according to an embodiment of the present invention;
FIG. 3 is a diagram for explaining a genetic algorithm of the emotion-based extracting method according to an embodiment of the present invention;
FIG. 4 is a diagram for explaining a process of generating an initial chromosome set in the emotion-based extracting method according to an embodiment of the present invention;
FIG. 5 is a diagram for explaining emotion evaluation and emotion classification of a specific document in the emotion-based extracting method according to an embodiment of the present invention;
FIG. 6 is a flow chart for explaining different flows of a training set and a test set of the emotion basis extraction method according to an embodiment of the present invention;
FIG. 7 is a block diagram schematically showing a configuration of an emotion-based reason extracting apparatus according to an embodiment of the present invention;
FIG. 8A is a diagram showing contents related to a data set for performance verification of the emotion-based extracting method according to an embodiment of the present invention; FIG.
FIG. 8B is a diagram showing contents related to parameters for performance verification of the emotion-based extracting method according to an embodiment of the present invention; FIG.
9 is a table for comparing accuracy differences between the three methods and the conventional method of the emotion-based extracting method according to an embodiment of the present invention;
10 is a graph for comparing accuracy differences between the three methods of the emotion-based extracting method and the conventional method according to an embodiment of the present invention.

While the invention is susceptible to various modifications and alternative forms, specific embodiments thereof are shown by way of example in the drawings and will herein be described in detail.

It should be understood, however, that the invention is not intended to be limited to the particular embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another. For example, without departing from the scope of the present invention, the first component may be referred to as a second component, and similarly, the second component may also be referred to as a first component. And / or < / RTI > includes any combination of a plurality of related listed items or any of a plurality of related listed items.

It is to be understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, . On the other hand, when an element is referred to as being "directly connected" or "directly connected" to another element, it should be understood that there are no other elements in between.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the relevant art and are to be interpreted in an ideal or overly formal sense unless explicitly defined in the present application Do not.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In order to facilitate the understanding of the present invention, the same reference numerals are used for the same constituent elements in the drawings and redundant explanations for the same constituent elements are omitted.

Emotion basis extraction system

1 is a block diagram illustrating a system to which an emotion-based extracting apparatus according to an embodiment of the present invention is applied. 1, the emotion-based reason extracting system according to the embodiment of the present invention includes the emotion-based reason extracting apparatus 100 and the client devices 110-1, 110-2, ..., 110-N can do.

Referring to FIG. 1, the emotion-based extracting apparatus 100 receives an electronic document from at least one client device 110-1, 110-2, ..., 110-N using the Internet or a wired / wireless communication network . Here, the electronic document includes not only document files, but also various kinds of articles (for example, posts in a café or a blog, comments on a specific article, Internet articles, and the like) transmitted via the Internet, Mentions and the like.

The emotion-based reason extracting apparatus 100 receives a document from the client devices 110-1, 110-2, ..., 110-N and extracts the emotion basis of the document. At this time, a genetic algorithm is used to generate an emotion base set, and emotion words in the document are extracted through the above emotion base set, and its polarity is grasped. The emotion-based reason extracting apparatus 100 can separate the training data set and the test data set and use them for emotion-based extraction. The training data set may include a document required to generate an emotional basis set and polarity information of the document. On the other hand, the test dataset may refer to data for determining the emotional polarity of the document using the empirical basis set generated through the training data set. That is, the emotional polarity of the test dataset may be information that is ultimately desired to be acquired through the present invention.

The client devices 110-1, 110-2, ..., 110-N may be personal terminals that transmit documents to the emotion-based extracting apparatus 100. [ The personal terminal may include various terminals capable of transmitting and receiving document information to and from the emotion-based reason extracting apparatus 100 through wired / wireless communication such as a personal computer (PC), a tablet computer, a mobile terminal, and a PDA (Personal Digital Assistant).

Emotion basis extraction method

FIG. 2 is a flowchart for schematically explaining a method of extracting emotion evidence according to an embodiment of the present invention.

Referring to FIG. 2, an empirical basis extracting apparatus (not shown) compares the received document 202 with a previously stored emotional dictionary 204 to generate an initial chromosome set (S210). An initial chromosome set can be generated arbitrarily. Each gene on the chromosome represents a property, which can define one attribute for each word appearing in the training data set. That is, words of the data set are defined as chromosomes, and arbitrary words in the data set are first extracted to generate an initial chromosome set.

Then, when the construction of the initial chromosome set is completed, a combination of each chromosome, i.e., each word is evaluated through an evaluation function (S220). The purpose of the evaluation function is to maximize the accuracy of classifying the polarity of a document. To classify the polarity of a document, the positive and negative scores of the document are calculated using a set of attributes selected by each chromosome, that is, a set of words can do. Thus, it is possible to judge the polarity of the document having a larger score among the separately obtained positive and negative scores as the polarity of the document. The accuracy of the classification results can be used as an evaluation criterion.

Then, a new chromosome (generation) can be generated by performing a genetic algorithm based on the score of each chromosome after the evaluation function (S230). Here, the genetic algorithm includes a selection step S232, a crossing step S234, and a transition step S236. This will be described in more detail later.

When a new chromosome is generated, it is compared with the termination condition (S240). If the termination condition is not satisfied, the evaluation score is again calculated and the genetic algorithm is repeated. If the termination condition is satisfied, The chromosome set is extracted as an emotion based word (S250). Here, the termination condition may include a condition that the accuracy value set by the apparatus does not increase by more than the threshold value even if the repetition is performed, or whether the number of iterations of the gene operation has progressed to a preset number of times. Here, the definition of the accuracy will be described in detail with reference to FIG.

FIG. 3 is a diagram for explaining a genetic algorithm of the emotion-based extracting method according to an embodiment of the present invention.

Referring to FIG. 3, an emotion basis extracting device (not shown) is generated based on words included in the emotion dictionary 304 among words described in the training document 302. Here, each of the chromosomes 310 can represent a training data set and a word appearing respectively in the emotion dictionary. In addition, each word has a positive or negative score depending on the emotion dictionary. Each chromosome 310 in this set may be represented using a binary string character indicating whether to select the candidate emotion based words. That is, 1 means a selected word, and 0 means a word which is not selected. The selected words and their emotional scores may be used in the classifier learning stage (e.g., emotion-based extracting stage) and emotional classifying stage.

4 is a diagram for explaining an initial chromosome set generation process of the emotion-based extracting method according to an embodiment of the present invention.

Referring to FIG. 4, in the initial chromosome set generated through the emotion dictionary 404 and the training set 402, the candidate emotion bases are listed first. The candidate emotion bases may be words included in the emotion dictionary among the words included in the training set. That is, these words have positive or negative scores for each word, as shown in the table in the middle part of FIG. On the basis of this, each chromosome in the initial chromosome group can be generated in the form of a string by selecting whether or not the corresponding words are selected and expressing different choices of 1 or 0 bits.

Referring back to FIG. 3, the initial chromosome set can be composed of a predetermined number of chromosomes 310. For example, each chromosome 310 of n strings is randomly generated with a standard genetic algorithm. Then, the initial sensitivity classification is performed, and a conformity assessment is made through an appropriate evaluation function.

In a genetic algorithm (genetic algorithm), from the n chromosome strings in the set, certain adjacent chromosome pairs are selected so that they can be crossed based on the tournament selection. There are many ways to choose genetic algorithms such as loit wheel selection, fitness-proportional selection, etc. In the present invention, tournament selection can be used preferably. A tournament selection is a random selection of the determined number of entities in the set, selecting the most adaptable entity among them, and repeating this operation for the required number of times in the next generation.

Thereafter, with the selected chromosome, crossing is performed. Crossbreeding is mimicked in the form of a reproductive algorithm, and one-point crossing, simple crossing, multipoint crossing, multipoint crossing, uniform crossing, partial crossing, etc. can be performed. According to a preferred embodiment of the present invention, a chromosome set can be crossed through single point mating. That is, the device may perform mating by exchanging substrings at any mating point x with a selected pair of strings. Here, the mating point x can be determined at random.

Then, the mutation is performed, while the mutations continue to evolve as the mock evolution continues, and the reproduction and mating operators further evolve the cluster, which causes the chromosomes to resemble one another, which, if incorrect, results in a lack of gene diversity, Mechanism. That is, it is possible to prevent the specific bits of all the chromosomes from being fixed in the initial generation by changing the bits in the chromosomes based on the mutation probability. According to a preferred embodiment of the present invention, the mating operator randomly varies individual attribute characteristics in the chromosome string based on the mating probability constant P _m . Through this process, one cycle of gene operation is completed, and the chromosome is reconstructed.

FIG. 5 is a diagram for explaining emotional evaluation and emotion classification for a specific document in the emotion-based extracting method according to an embodiment of the present invention.

Referring to FIG. 5, the emotion basis extracting device (not shown) evaluates the emotion score of the document through the evaluation function, and thereby, the polarity can be grasped. As described above, the evaluation function is used to maximize the accuracy. In particular, in order to identify the polarity of the document, positive and negative feelings scores must be obtained, respectively.

As shown in FIG. 5, the positive scores and the negative scores are obtained as follows.

D stands for a document, which consists of a set of words. Where P and N may mean a set of positive and negative emotion words. For example, w? P means that w is a positive word. Further, w? D may mean that w is a word belonging to a specific document. Thus, two scores, positive scores of the whole document, Score _p (D), negative score, and Score _n (D) can be expressed respectively. This can be calculated as the summation of each emotion score for a particular document, and a label indicating the higher score can be assigned to the label of that document. In the training process, the optimization problem for this evaluation function can be defined as follows. That is, the first may be the maximization of the accuracy, and the second may be the maximum of the gap between the positive and negative feelings of the document. This gap can be expressed as G (D).

By maximizing the difference between the positive and negative emotion scores of the documents, the most influential emotive evidence word can be found. In addition, the accuracy or accuracy of the positive and negative training sets can also be considered as an evaluation value, to prevent emotion bases from being extracted without excluding certain emotion base words.

The evaluation value can be calculated by not only the accuracy and G (D) but also other values through the two values. For example, G (D) is classified as an affirmative document based on emotion based on the total number of positive and actual fraudulent documents calculated through the number of affirmative documents accurately classified based on emotion based on the total number of actual affirmative documents There may be a first accuracy that is expressed as a product of three values of negative precision, which is calculated through the number of errors. Also, the second accuracy, expressed as the product of G (D) and the general accuracy value (the general accuracy value may mean the number of documents accurately classified by emotion basis versus the total number of documents), and only the general accuracy value A third accuracy may be present. It is possible to find a condition for maximizing the first to third accuracy values and to extract the emotion basis.

Further, in the emotion classification, the best chromosome is extracted through a plurality of gene arithmetic operations through the evaluation value, and the emotion based on the extracted chromosome is used for emotion classification (hereinafter referred to as " Classification can be performed. At this time, the general classification accuracy for evaluation using the best chromosome can be calculated as follows.

Classification Accuracy = number of correctly classified documents / total number of documents

Additionally, when performing a sensitive classification of only a portion of the sentences in the document, the document may include a mix of positive and negative sentiments. Thus, if we identify the best key sentences in the document, we can improve the performance of emotion classification. Also, in order to identify the best main sentence, it is possible to calculate the emotion score for each emotion word of the sentence, and to have the highest emotion score as the best sentence. Thus, only the most important sentences (the number of key sentences can be a preset value) can only be used for emotional classification. As a result, the result can maintain the same performance level of emotional classification, while keeping only certain sentences in the document.

6 is a flowchart for explaining different flows of a training set and a test set of the emotion basis extraction method according to an embodiment of the present invention.

Referring to FIG. 6, the training set 602 removes unnecessary words such as data cleaning, that is, idle words, through the data preprocessing (S610), and separates the words in each document by segmentation.

Then, the emotion dictionary 606 is used to generate an initial data set through data representation, that is, a binary string.

After the initial data set generation, scores are given to each object of the initial data set based on the evaluation function through a training process (S630) including a genetic algorithm and an emotion basis extraction step, and a better object is generated through the evolution process , This process is repeated to generate a final emotion bases set. At this time, when only the selected emotion basis is used, when the repetition process is performed so that the best result (e.g., accuracy (including the first to third accuracy)) can be obtained, The set can be confirmed.

The test set 604 is subjected to data preprocessing (S610) and data representation (S620) in the same manner. In the test process (S640), the classifier and emotion classification are performed, Can be used to perform emotional classification.

Then, in the evaluation step S650, cross validation can be performed. For example, if there are 100 data, if 5-fold cross validation, 80 test data are divided into training data and 20 test data are divided into 5 test data. The performance of the sensitivity classification method according to the present invention can be verified.

Emotion basis extracting device

FIG. 7 is a block diagram schematically showing a configuration of an emotion-based reason extracting apparatus according to an embodiment of the present invention. As shown in FIG. 7, the emotion-based reason extracting apparatus 700 according to an embodiment of the present invention may include an emotion-based set generating unit 710 and an emotion classification performing unit 720.

The emotion basis set generation unit 710 may include a preprocessing unit 715, an evaluation index calculation unit 716, a gene operation unit 717, and an emotion basis extraction unit 718. [

The preprocessing unit 715 may include an input unit 711, a word extraction unit 712, and an aggregation generation unit 713. An input unit 711 inputs a set of target documents including a training set and a test set. The inputted document is subdivided into words through the word extracting unit 712, where only the pure words are left through processes such as elimination of abolition. The set generation unit 713 generates an initial chromosome set based on the words included in the emotion dictionary among the words described in the training document. For test documentation, you can also create a chromosome set. Each chromosome can represent a set of training data and words that appear in the emotion dictionary, respectively. In addition, each word has positive or negative scores according to the emotion dictionary, and each chromosome in the set can be expressed using a binary string character indicating whether to select the candidate emotion based words.

The evaluation index calculating unit 716 calculates a fitness evaluation value for each chromosome through an evaluation function based on the emotion dictionary.

The gene operation unit 717 selects, from the generated chromosome strings, predetermined adjacent chromosome pairs so that they can be mated based on the tournament selection. A tournament selection is a random selection of the determined number of entities in the set, selecting the most adaptable entity among them, and repeating this operation for the required number of times in the next generation. Then, the gene operation unit 717 carries out mating with the selected chromosome. According to a preferred embodiment of the present invention, a chromosome set can be crossed through single point mating. The gene operation unit 717 can perform mating by exchanging substrings at an arbitrary mating point x with a selected pair of strings. The mating point x can be determined at random. The gene operation unit 717 further evolves the set of reproduction and mating operators while the mock evolution continues, thereby causing the chromosomes to resemble each other. If wrong, the gene lacks diversity. Therefore, a mutation which is a mechanism for escaping unwanted harm Can be performed. The gene operation unit 717 can change the bits in the chromosomes based on the mutation probability and prevent specific bits of all the chromosomes from being fixed in the initial generation. According to a preferred embodiment of the present invention, the mating operator of the gene operation unit 717 randomly changes the individual attribute characteristics in the chromosome string based on the mating probability constant P _m . Through this process, one cycle of gene operation is completed, and the chromosome is reconstructed.

The emotion basis extracting unit 718 determines whether or not the newly generated chromosome set satisfies the end condition through the gene calculating unit 717. If the chromosome set is not satisfied, the emotion basis extracting unit 718 sends the evaluation result to the evaluation index calculating unit 716 again. It can be determined as a set of evidence.

The emotion classification performing unit 720 performs emotion classification on the basis of the emotion basis extracted by the emotion basis extracting unit 718 for the words in the test document generated as a chromosome set through the set generating unit 715. [

Simulation result

FIG. 8A is a diagram illustrating contents related to a data set for performance verification of the emotion-based reasoning extracting method according to an embodiment of the present invention, FIG. 8B is a flowchart illustrating a performance verification of the emotion-based reasoning extracting method according to an embodiment of the present invention And Fig.

8A to 8B, simulation was performed based on multi-domain review data, and after filtering, 6000 review data selected randomly were used as a parameter. As a parameter, the chromosome set size was 500, 1.0, the variation probability was 0.015, and the number of repetitions of generation was set to 1,000.

9 is a table for comparing accuracy differences between the three methods of the emotion-based extracting method and the conventional method according to an embodiment of the present invention.

Referring to FIG. 9, in the first method, which is based on the difference between the negative score and the positive score and the positive and negative accuracy, which is described by the accuracy G (D), the accuracy is calculated based on the difference value and the general accuracy As a result of averaging the affirmative documents, the irregular documents and the average through the third method calculated using only the calculated second method and the general accuracy, the conventional method baseline (sensitivity classification as a general emotional dictionary) The accuracy of the first through third schemes is 82.9%, 81.4%, and 85.1%, respectively, which is significantly better than the conventional method.

10 is a graph for comparing accuracy differences between the three methods of the emotion-based extracting method and the conventional method according to an embodiment of the present invention.

Referring to FIG. 10, the contents of the table of FIG. 9 are shown in a graph. It can be seen that the accuracy results of the first to third methods according to the present invention are much better than those of the conventional baseline method.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the inventions as defined by the following claims It will be understood that various modifications and changes may be made thereto without departing from the spirit and scope of the invention.

Claims (13)

A method for extracting an emotional basis of at least one document,
Arbitrarily generating an initial chromosome set based on words in the document, wherein each attribute of each gene of the chromosome refers to each word appearing in the training data set in the document, and wherein the property value indicates whether the word is selected;
Evaluating each chromosome through the function of maximizing the accuracy of classifying the polarity of the document;
Generating new chromosomes based on the evaluation score of each chromosome through gene calculation;
And generating an emotion basis set by determining whether the generated chromosome satisfies the termination condition. The method according to claim 1, wherein the step of generating an emotion basis set by determining whether the generated chromosome satisfies an end condition
Determining whether the generated chromosome satisfies the termination condition; And
And if the end condition is satisfied, the step of adding to the emotion basis set and repeating the chromosome evaluation step and the new chromosome generating step when the end condition is not satisfied, And repeating the step of extracting the emotional basis of the document. The method according to claim 1,
Wherein each word appearing in the training data set is a word that appears in both the target document and the emotion dictionary data and has a positive or negative score according to the emotion dictionary data. The method according to claim 1,
Wherein each chromosome in the initial chromosome set is expressed as a binary string indicating whether or not candidate emotion based words are selected. 5. The method according to claim 4, wherein the step of generating a new chromosome by performing a gene operation based on an evaluation function of each chromosome comprises:
Selecting, for a plurality of chromosomal binary strings in the chromosome set, for predetermined mating based on predetermined adjacent string pairs based on tournament selection - taking score points into consideration when selecting a tournament;
Performing a single-point mating with the selected string pair by exchanging substrings at a predetermined mating point; And
And varying individual attribute values in a string in which mating is performed based on a variation probability constant. The method according to claim 1,
Wherein the termination condition is determined by whether a change amount of an evaluation score of a chromosome generated through a gene operation has a convergence form. The method according to claim 1,
The evaluation score is a difference value between the positive emotion index - the sum of the positive scores of the training data set - and the negative emotion index - the sum of the negative scores of the training data set - to identify the polarity of the document And extracting the emotional basis of the document. 2. The method according to claim 1,
A ratio of the number of correctly classified documents to the total number of documents using the generated emotion basis set;
A ratio of the number of correctly positive documents among the documents classified as affirmative using the generated emotion basis set; And
And a ratio of the number of correctly denied documents among the documents classified as negative by using the generated emotion basis set. An apparatus for extracting an emotional basis of at least one document,
A chromosome set generator for arbitrarily generating an initial chromosome set based on words in the document, wherein attributes of respective genes of the chromosome means respective words appearing in a training data set in the document, Indicates -;
An evaluation unit for evaluating each chromosome through the initial chromosome set, an evaluation function-evaluation function for evaluating each chromosome through a function for maximizing the accuracy of classifying the polarity of the document;
A gene operation unit for generating a new chromosome through gene operation based on the score of each chromosome;
And an emotion basis set generation unit for generating an emotion basis set by determining whether the generated chromosome satisfies a termination condition. A method for performing emotional analysis by extracting an emotional basis of at least one document,
Arbitrarily generating an initial chromosome set based on words in the document, wherein each attribute of each gene of the chromosome refers to each word appearing in the training data set in the document, and wherein the property value indicates whether the word is selected;
Evaluating each chromosome through the function of maximizing the accuracy of classifying the polarity of the document;
Generating new chromosomes based on the evaluation score of each chromosome through gene calculation;
Generating an emotion basis set by determining whether the generated chromosome satisfies an end condition; And
And performing emotional classification of a document to be tested using the generated emotion basis set. [11] The method of claim 10,
And calculating the sensitivity analysis accuracy for evaluation using the generated emotion basis set through a ratio of the number of documents classified correctly to the number of documents. [11] The method of claim 10,
Calculating a positive emotion index and a negative emotion index based on the emotion based set words included in the test object document and classifying the emotion having a higher value as emotion of the test target document . An apparatus for performing emotional analysis by extracting an emotional basis of at least one document,
Wherein an attribute of each gene in the chromosome refers to each word appearing in a training data set in the document and an attribute value indicates whether a word is selected; In the initial chromosome set, the evaluation function - evaluation function evaluates each chromosome through a function that maximizes the accuracy of classifying the polarity of the document - and generates a new chromosome based on the evaluation score of each chromosome An emotion based set generation unit for generating an emotion based set by determining whether the generated chromosome satisfies an end condition; And
And an emotional classifier for performing emotional classification of a document to be tested using the generated emotion bases.

Images