Artificial intelligence based fairy tale generation system and method

Abstract

The present invention relates to an artificial intelligence-based fairy tale generation system and method, and more specifically, to a system and method for removing bias and hate speech inherent in an artificial intelligence model that generates fairy tales so that children do not relearn prejudices through the fairy tales. An artificial intelligence-based fairy tale generation system according to one embodiment of the present invention may include a server that receives a suggestion word from a user through a user terminal, and a fairy tale generation model that generates candidate sentences based on the suggestion word and provides them to the user, and generates a fairy tale using a sentence selected by the user from among the candidate sentences. According to one embodiment of the present invention, an assimilation can be generated in which hate speech and negative expressions are removed.

Description

Technology Field

[0001] The present invention relates to an artificial intelligence-based fairy tale generation system and method, and more specifically, to a system and method for removing bias and hate speech inherent in an artificial intelligence model that generates fairy tales so that children do not relearn prejudices through the fairy tales. Background Technology

[0002] A fairy tale is a type of folktale that, as the word literally means, refers to a story for young children. In modern times, many stories are created to inspire children with dreams, hope, and potential for growth; however, before the modern era, many fairy tales containing somewhat cruel content were created to impart moral lessons to children.

[0003] Furthermore, conventional fairy tales contain social prejudices and stereotypes, so children who find it difficult to read books with a critical eye end up learning these prejudices and stereotypes as they are.

[0004] Therefore, there is a need for technology that can prevent children from learning social prejudices and stereotypes through fairy tales. Prior art literature

[0005] (Patent Document 0001) KR 10-2313203 B1 The problem to be solved

[0006] To solve the aforementioned problems, the present invention aims to build a refined dataset of expressions related to hate speech or negative emotions included in fairy tales and to provide a system and method capable of mitigating bias inherent in artificial intelligence. means of solving the problem

[0007] As an embodiment of the present invention, an artificial intelligence-based fairy tale generation system is provided.

[0008] An artificial intelligence-based fairy tale generation system according to one embodiment of the present invention may include a server that receives a suggestion word from a user through a user terminal, and a fairy tale generation model that generates candidate sentences based on the suggestion word and provides them to the user, and generates a fairy tale using a sentence selected by the user from among the candidate sentences.

[0009] An artificial intelligence-based fairy tale generation system according to one embodiment of the present invention may further include a database in which a plurality of fairy tale data, hate expression data, or negative emotion expression data are stored.

[0010] An artificial intelligence-based fairy tale generation system according to one embodiment of the present invention may further include a database processing model that filters hate expressions or negative emotional expressions included in a plurality of fairy tale data stored in a database.

[0011] In an artificial intelligence-based fairy tale generation system according to one embodiment of the present invention, the database processing model may further include a hate expression removal model that is trained using hate expressions stored in a database as training data, and is trained to remove hate expressions from fairy tale data stored in the database.

[0012] In an artificial intelligence-based fairy tale generation system according to one embodiment of the present invention, the database processing model may include a hate expression extraction unit that extracts hate expressions from fairy tale data stored in a database, a hate expression quantification unit that calculates and quantifies the amount of hate expressions included in the fairy tale data, and a hate expression filtering unit that filters fairy tales from the database in which the amount of hate expressions is greater than or equal to a preset value. A hate expression removal model may be further included.

[0013] In an artificial intelligence-based fairy tale generation system according to one embodiment of the present invention, the database processing model may include a negative emotion analysis unit that is trained using negative emotion expressions stored in a database as training data and is trained to extract negative emotions included in fairy tale data stored in the database, a negative emotion expression quantification unit that calculates and quantifies the amount of sentences related to negative emotion expressions included in fairy tales stored in the database, and a negative emotion expression filtering unit that removes the corresponding fairy tale from the database when the amount of sentences related to negative emotion expressions is greater than or equal to a preset value. A negative emotion expression removal model may be further included.

[0014] An artificial intelligence-based fairy tale generation system according to one embodiment of the present invention may further include an adjustment unit that adjusts the distance between specific words by adjusting the word embedding values ​​stored in the tokenizer of the fairy tale generation model.

[0015] As an embodiment of the present invention, an artificial intelligence-based fairy tale generation method is provided.

[0016] An artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may include an input step in which a server receives a suggested word from a user through a user terminal, a candidate sentence provision step in which candidate sentences are generated based on the suggested word and provided to the user using a fairy tale generation model, and a fairy tale generation step in which one of the candidate sentences is selected by a user through a user terminal and a fairy tale is generated using the selected sentence.

[0017] An artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may further include a database generation step of generating a database by storing a plurality of fairy tale data, hate expression data, or negative emotion expression data.

[0018] In an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention, the database generation step may further include a database processing step of filtering hate expressions or negative emotional expressions included in a plurality of fairy tale data stored in the database using a database processing model.

[0019] In an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention, the database processing step may involve removing hate expressions from fairy tale data stored in the database using a previously trained hate expression removal model that utilizes hate expressions stored in the database as training data.

[0020] In an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention, the database processing step may further include a hate expression extraction step in which a hate expression extraction unit extracts hate expressions from fairy tale data stored in a database, a hate expression quantification step in which a hate expression quantification unit calculates and quantifies the amount of hate expressions included in the fairy tale data, and a hate expression filtering step in which a hate expression filtering unit filters fairy tales in which the amount of hate expressions is greater than or equal to a preset value from the database.

[0021] In an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention, the database processing step may further include: a negative emotion analysis step in which negative emotions included in fairy tale data stored in the database are extracted using a negative emotion analysis unit that has been learned using negative emotion expressions stored in the database as training data; a negative emotion expression quantification step in which a negative emotion expression quantification unit calculates and quantifies the amount of sentences related to negative emotion expressions included in fairy tale data stored in the database; and a negative emotion expression filtering step in which a negative emotion expression filtering unit removes the corresponding fairy tale from the database when the amount of sentences related to negative emotion expressions is greater than or equal to a preset value.

[0022] An artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may further include a distance adjustment step of adjusting the distance between specific words by adjusting the word embedding value stored in the tokenizer of the fairy tale generation model by the adjustment unit.

[0023] As an embodiment of the present invention, a computer-readable recording medium is provided on which a program for implementing the above-described method is recorded. Effects of the invention

[0024] According to one embodiment of the present invention, an assimilation can be generated in which hate speech and negative expressions are removed.

[0025] By generating fairy tales using trained artificial intelligence models, there is an advantage in that fairy tales can be produced efficiently in terms of time and cost.

[0026] The effects obtainable from the present disclosure are not limited to those mentioned above, and other unmentioned effects will be clearly understood by those skilled in the art to which the present disclosure belongs from the description below. Brief explanation of the drawing

[0027] FIG. 1 is a configuration diagram of an artificial intelligence-based fairy tale generation system according to a first embodiment of the present invention. FIG. 2 is a configuration diagram of an artificial intelligence-based fairy tale generation system according to a second embodiment of the present invention. FIG. 3 is an example diagram of a database according to a second embodiment of the present invention. FIG. 4 is a block diagram of a hate speech removal model according to one embodiment of the present invention. FIG. 5 is a block diagram of a negative emotion expression removal model according to one embodiment of the present invention. FIG. 6 is a drawing for explaining an adjustment unit according to an embodiment of the present invention. FIG. 7 is a flowchart of an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention. Specific details for implementing the invention

[0028] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.

[0029] The terms used in this specification will be briefly explained, and the invention will be described in detail.

[0030] The terms used in this invention have been selected based on currently widely used general terms, taking into account their functions within the invention; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, the terms used in this invention should be defined not merely by their names, but based on their meanings and the overall content of the invention.

[0031] Throughout the specification, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "part" or "module" used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software. Additionally, throughout the specification, when a part is described as "connected" to another part, this includes not only cases where they are "directly connected" but also cases where they are connected "with other components in between."

[0032] The present invention will be described in detail below with reference to the attached drawings.

[0033] First embodiment

[0034] FIG. 1 is a configuration diagram of an artificial intelligence-based fairy tale generation system according to a first embodiment of the present invention.

[0035] Referring to FIG. 1, an artificial intelligence-based fairy tale generation system according to a first embodiment of the present invention may include a user terminal (100), a server (200), and a fairy tale generation model (300).

[0036] The user terminal (100) is configured to receive a suggestion word from a user and transmit it to a server (200), and may further include an input unit for receiving a suggestion word from a user, a data receiving unit for receiving a generated story from the server (200), an output unit for providing the story received from the server (200) to the user, and a communication unit that connects the data receiving unit or the output unit to the server (200) via wired or wireless means.

[0037] According to an embodiment, the input unit may be implemented as a mouse, keyboard, or touchpad.

[0038] The server (200) can record words, sentences, or fairy tales selected or generated from the user terminal (100) and the fairy tale generation model (300), or transmit and receive them to the user terminal (100) or the fairy tale generation model (300).

[0039] For example, the server (200) can receive a suggested word input through the user terminal (100) and transmit it to the fairy tale generation model (300), and can receive a word, sentence, or fairy tale generated by the fairy tale generation model (300) and transmit it to the user terminal (100).

[0040] The fairy tale generation model (300) can generate candidate sentences based on a given word and generate a fairy tale using a sentence selected by the user among the candidate sentences. That is, the fairy tale generation model (300) can generate candidate sentences based on a given word transmitted from the server (200), transmit them to the user terminal (100) through the server (200), and generate a fairy tale by listing the sentences selected by the user among the candidate sentences through the user terminal (100).

[0041] According to an embodiment, the assimilation generation model (300) learns the correlation (i.e., distance) between words or sentences using words or sentences used in assimilation as training data, and can output a generated sentence (or paragraph) by listing words (or sentences) that are close in distance to the word (or sentence) input as a present word.

[0043] 2nd embodiment

[0044] FIG. 2 is a configuration diagram of an artificial intelligence-based fairy tale generation system according to a second embodiment of the present invention.

[0045] Referring to FIG. 2, an artificial intelligence-based fairy tale generation system according to a second embodiment of the present invention may include a user terminal (100), a server (200), a fairy tale generation model (300), a database (400), and a database processing model (500).

[0046] The user terminal (100), server (200), and fairy tale generation model (300) perform the same operation as the corresponding configuration described in the first embodiment.

[0047] Multiple fairy tales, hate expression data, or negative emotion expression data may be stored in the database (400).

[0048] According to an embodiment, the database (400) may store each word by classifying it by meaning or store each word by establishing interrelationships.

[0049] For example, each word can be classified based on which part of speech it belongs to among nouns, verbs, adjectives, adverbs, etc., or it can be classified to include information on whether each word can be used as a subject, object, predicate, etc. Additionally, the database (400) can store words by grouping them into word groups with similar meanings, or classify and store them as words indicating 'time', words indicating 'place', etc.

[0050] According to an embodiment, the database (400) may be implemented to classify and store each word based on ontology technology.

[0051] FIG. 3 is an exemplary diagram of a database (400) according to a second embodiment of the present invention.

[0052] For example, as illustrated in FIG. 3, the database (400) may store each word classified into fairy tales (410), hate speech (420), and negative emotional expressions (430). The 'fairy tales' classification (410) may include words such as Cinderella, Snow White, Little Mermaid, and The Little Match Girl. The 'hate speech' classification (420) may include words such as women / family, men, sexual minorities, race / nationality, age, region, religion, other hate, malicious comments / insults. The 'negative emotional expressions' classification (430) may include words such as depression, loneliness, sadness, anger, fear, withdrawal, lack, alienation.

[0053] The database processing model (500) can filter hateful expressions or negative emotional expressions included in multiple fairy tales stored in the database (400).

[0054] That is, by using a database (400) in which fairy tales with hateful expressions and negative emotional expressions removed are stored through a database processing model (500), a fairy tale generation model (300) is trained, and by generating fairy tales using the trained fairy tale generation model (300), it is possible to generate fairy tales without social prejudice or stereotypes.

[0055] The database processing model (500) may further include a hate speech removal model (510) or a negative emotion expression removal model (520).

[0056] The hate speech removal model (510) is configured to filter hate speech from fairy tale data stored in the database (400), and can remove words corresponding to social prejudices or stereotypes from the database (400), or remove the relationship between specific words if the relationship between those words is due to social prejudices or stereotypes.

[0057] According to an embodiment, the hate speech removal model (510) is trained using hate speech stored in the database (400) as training data, and can be trained to remove hate speech from assimilation data stored in the database (400).

[0058] FIG. 4 is a block diagram of a hate speech removal model (510) according to one embodiment of the present invention.

[0059] Referring to FIG. 4, a hate speech removal model (510) according to one embodiment of the present invention may include a hate speech extraction unit (511) that extracts hate speech from an idiom stored in a database (400), a hate speech quantification unit (512) that calculates and quantifies the amount of hate speech included in the idiom, and a hate speech filtering unit (513) that filters idioms in the database (400) in which the amount of hate speech is greater than or equal to a preset value.

[0060] FIG. 5 is a block diagram of a negative emotion expression removal model (520) according to one embodiment of the present invention.

[0061] The negative emotional expression removal model (520) is configured to filter negative emotional expressions from fairy tale data stored in the database (400).

[0062] Referring to FIG. 5, a negative emotion expression removal model (520) according to one embodiment of the present invention may include a negative emotion analysis unit (521), a negative emotion expression quantification unit (522), and a negative emotion expression filtering unit (523).

[0063] The negative emotion analysis unit (521) is trained using negative emotion expressions stored in the database (400) as training data, and can be trained to extract negative emotions included in the fairy tale data stored in the database (400).

[0064] The negative emotion expression quantification unit (522) can calculate and quantify the amount of sentences related to negative emotion expressions included in the fairy tale stored in the database (400).

[0065] The negative emotion expression filtering unit (523) can remove the corresponding fairy tale from the database (400) if the amount of sentences related to negative emotion expression is greater than or equal to a preset value.

[0067] Third embodiment

[0068] The artificial intelligence-based fairy tale generation system according to the third embodiment of the present invention may further include an adjustment unit (600) in the first or second embodiment.

[0069] The user terminal (100), server (200), animation generation model (300), database (400), and database processing model (500) perform the same operation as the corresponding configuration described in the first or second embodiment.

[0070] The adjustment unit (600) can adjust the distance between specific words by adjusting the word embedding values ​​stored in the tokenizer of the assimilation generation model (300). Since the tokenizer of the language model stores word embedding values ​​that quantify the spatial positions of words (i.e., tokens), the adjustment unit (600) can mitigate the bias formed in the assimilation generation model (300) by adjusting the word embedding values ​​stored in the tokenizer.

[0071] For example, in the assimilation generation model (300) learned using a database, the words 'kitchen', 'shoes', 'cooking', 'cosmetics', and the word 'woman' are formed such that the distance between the words is closer than the distance between the words 'man' and the word 'woman', so the adjustment unit (600) can adjust the words 'kitchen', 'shoes', 'cooking', and 'cosmetics' to be at the same distance from the word 'woman' and the word 'man', respectively.

[0072] FIG. 6 is a drawing for explaining an adjustment unit (600) according to one embodiment of the present invention.

[0073] Referring to FIG. 6, according to an embodiment, the adjustment unit (600) may adjust the word embedding value using the formulas of FIG. 6. In the formulas of FIG. 6, e woman and e man Each can form a word set consisting of female-oriented words and male-oriented words, and g represents the gender axis of the word set. That is, the adjustment unit (600) can mitigate the bias formed in the assimilation generation model (300) by adjusting the gender axis of a specific word set.

[0074] In the formula of Fig. 6, e is the vector of words with bias, is the vector calculated for bias removal, is the vector of words with bias removed, means a variable.

[0076] In relation to the method according to one embodiment of the present invention, the details of the aforementioned system may be applied. Hereinafter, details identical to the details of the aforementioned system in relation to the method are omitted.

[0077] FIG. 7 is a flowchart of an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention.

[0078] Referring to FIG. 7, an artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may include an input step (S100), a candidate sentence provision step (S200), and a fairy tale generation step (S300).

[0079] In the input step (S100), the server (200) can receive a suggested word from the user through the user terminal (100).

[0080] In the candidate sentence stage (S200), a candidate sentence can be generated based on the suggested word using an assimilation generation model (300) and provided to the user.

[0081] In the fairy tale generation step (S300), one of the candidate sentences is selected from the user through the user terminal (100), and a fairy tale can be generated using the selected sentence.

[0082] An artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may further include a database generation step of generating a database (400) by storing a plurality of fairy tale data, hate expression data, or negative emotion expression data.

[0083] The database creation step may further include a database processing step that filters hateful expressions or negative emotional expressions included in multiple fairy tale data stored in the database (400) using a database processing model (500).

[0084] The database processing step may involve removing hate expressions from the assimilation data stored in the database (400) using a previously trained hate expression removal model (510) that uses hate expressions stored in the database (400) as training data.

[0085] According to an embodiment, the database processing step may further include a hate expression extraction step in which a hate expression extraction unit (511) extracts hate expressions from fairy tale data stored in the database (400), a hate expression quantification step in which a hate expression quantification unit (512) calculates and quantifies the amount of hate expressions included in the fairy tale data, and a hate expression filtering step in which a hate expression filtering unit (513) filters fairy tales in the database (400) in which the amount of hate expressions is greater than or equal to a preset value.

[0086] According to an embodiment, the database processing step may further include a negative emotion analysis step in which negative emotions included in fairy tale data stored in the database (400) are extracted using a negative emotion analysis unit (521) that has been learned using negative emotion expressions stored in the database (400) as training data; a negative emotion expression quantification step in which a negative emotion expression quantification unit (522) calculates and quantifies the amount of sentences related to negative emotion expressions included in the fairy tale data stored in the database (400); and a negative emotion expression filtering step in which a negative emotion expression filtering unit (523) removes the fairy tale from the database (400) when the amount of sentences related to negative emotion expressions is greater than or equal to a preset value.

[0087] An artificial intelligence-based fairy tale generation method according to one embodiment of the present invention may further include a distance adjustment step in which the distance between specific words is adjusted by adjusting the word embedding value stored in the tokenizer of the fairy tale generation model by the adjustment unit (600).

[0088] Meanwhile, the above-described method can be written as a program executable on a computer and can be implemented on a general-purpose digital computer that operates the program using a computer-readable recording medium. In addition, the structure of the data used in the above-described method can be recorded on a computer-readable recording medium through various means. The computer-readable recording medium includes storage media such as magnetic storage media (e.g., ROM, RAM, USB, floppy disk, hard disk, etc.) and optical reading media (e.g., CD-ROM, DVD, etc.).

[0089] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.

[0090] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention. Explanation of the symbols

[0091] 100: User terminal 200: Server 300: Assimilation generation model 400: Database 410: Fairy Tale Classification 420: Classification of Hate Speech 430: Classification of Negative Emotion Expressions 500: Database processing model 510: Hate Speech Removal Model 511: Hate Speech Extraction Unit 512: Hate Speech Numerical Section 513: Hate Speech Filtering Section 520: Negative Emotion Removal Model 521: Negative Emotion Analysis Department 522: Numerical part of negative emotion expression 523: Negative Emotion Expression Filtering Section

Claims

Claim 1 A server that receives a keyword input from a user through a user terminal; an animation generation model that learns the correlation between words or sentences used in an animation, generates candidate sentences based on the keyword and provides them to the user, and generates an animation by listing words or sentences that are close to the word or sentence input as the keyword using a sentence selected by the user among the candidate sentences; a database storing multiple animation data, hate speech data, or negative sentiment expression data; a database processing model that includes a hate speech removal model and a negative sentiment expression removal model, and filters hate speech or negative sentiment expressions included in the multiple animation data stored in the database; and an adjustment unit that adjusts the distance between specific words by adjusting the word embedding values ​​stored in the tokenizer of the animation generation model, wherein the hate speech removal model includes a hate speech extraction unit that extracts hate speech from the animation stored in the database; and a hate speech quantification unit that calculates and quantifies the amount of hate speech included in the animation. An AI-based fairy tale generation system comprising a hate speech filtering unit that filters out fairy tales in a database in which the amount of hate speech is greater than or equal to a preset value, and an adjustment unit that mitigates bias formed in the fairy tale generation model by adjusting the axis of a specific word set. Claim 2 delete Claim 3 delete Claim 4 In claim 1, the hate speech removal model is an artificial intelligence-based fairy tale generation system that is trained using hate speech stored in the database as training data, and is trained to remove hate speech from fairy tale data stored in the database. Claim 5 delete Claim 6 An artificial intelligence-based fairy tale generation system according to claim 1, wherein the database processing model further comprises: a negative emotion analysis unit that is trained using negative emotion expressions stored in the database as training data and is trained to extract negative emotions included in fairy tale data stored in the database; a negative emotion expression quantification unit that calculates and quantifies the amount of sentences related to negative emotion expressions included in fairy tales stored in the database; and a negative emotion expression filtering unit that removes the corresponding fairy tale from the database when the amount of sentences related to negative emotion expressions is greater than or equal to a preset value. Claim 7 delete Claim 8 The method comprises: an input step in which a server receives a suggested word from a user through a user terminal; a candidate sentence provision step in which candidate sentences are generated based on the suggested word and provided to the user using an assimilation generation model; an assimilation generation step in which one of the candidate sentences is selected by the user through a user terminal and an assimilation is generated using the selected sentence; a database creation step in which a database is created by storing multiple assimilation data, hate expression data, or negative emotion expression data; a database processing step in which hate expressions or negative emotion expressions included in the multiple assimilation data stored in the database are filtered using a database processing model; and a distance adjustment step in which the distance between specific words is adjusted by adjusting the word embedding values ​​stored in the tokenizer of the assimilation generation model by an adjustment unit, wherein the database processing step includes: a hate expression extraction step in which a hate expression extraction unit extracts hate expressions from the assimilation data stored in the database; and a hate expression quantification step in which a hate expression quantification unit calculates and quantifies the amount of hate expressions included in the assimilation data. An AI-based fairy tale generation method, wherein a hate speech filtering unit includes a hate speech filtering step in which the amount of hate speech is greater than or equal to a preset value in a database, and the distance adjustment step is a step in which the adjustment unit mitigates bias formed in the fairy tale generation model by adjusting the axis of a specific word set. Claim 9 delete Claim 10 delete Claim 11 In claim 8, the database processing step is to remove hate expressions from the assimilation data stored in the database using a previously trained hate expression removal model that uses hate expressions stored in the database as training data, an artificial intelligence-based assimilation generation method. Claim 12 delete Claim 13 In claim 8, the database processing step further comprises: a negative emotion analysis step in which negative emotions included in fairy tale data stored in the database are extracted using a previously learned negative emotion analysis unit utilizing negative emotion expressions stored in the database as training data; a negative emotion expression quantification step in which a negative emotion expression quantification unit calculates and quantifies the amount of sentences related to negative emotion expressions included in the fairy tale data stored in the database; and a negative emotion expression filtering step in which a negative emotion expression filtering unit removes the corresponding fairy tale from the database when the amount of sentences related to negative emotion expressions is greater than or equal to a preset value. Claim 14 delete Claim 15 A computer-readable recording medium having a program recorded thereon for implementing the method of claim 8, 11, or 13.

Images