Content generation system, content generation method, and content generation program

Abstract

To provide a content generation system, a content generating method, and a content generation program for generating a variety of contents according to texts.SOLUTION: A supporting device 20 includes: a package storage unit 22 for storing a content package in which a generation variable for creating a content is set; an impression rating storage unit 23 for storing plural pieces of impression rating information when a content is generated by the content package; an impression information storage unit 25 for storing impression information of each content package calculated by analysing the impression rating information; and a control unit 21 for generating a content. The control unit 21 acquires a target text, calculates evaluation information, specifies a content package closer to the evaluation information of the impression information stored in the impression information storage unit 25, calls out the generation variable for the specified content package from the package storage unit 22, and generates a content by using the generation variable.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a content generation system, a content generation method, and a content generation program for generating a variety of content in accordance with text. [Background technology]

[0002] There are cases where content such as music and images are added to text. For this reason, technology has been considered for adding sound effects and music depending on the user's mood when making a call (for example, Patent Document 1). In the technology described in Patent Document 1, a voice call device analyzes the user's emotional state from sensor data or text data of an email and sets a mood mode and mood level. Then, sound effects or music are selected depending on the mood mode and mood level, and the selected sound effects or music is changed upon detecting an instruction from the user.

[0003] Furthermore, technology for selecting music that matches the user's tastes has also been considered (for example, Patent Document 2). In the technology described in Patent Document 2, the storage device of the server device holds ratings of music. Then, the content of the utterance acquired by the information processing terminal is acquired, and one or more music pieces related to the content of the utterance are selected. Furthermore, the rating of the selected music piece is updated based on the content of the user's response to the music piece played by the information processing terminal. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2012-4885 A [Patent Document 2] JP 2018-55440 A Summary of the Invention [Problem to be solved by the invention]

[0005] However, if pre-prepared content such as music is provided, the same content is provided depending on the situation, resulting in a problem of lack of variety. [Means for solving the problem]

[0006] A content generation system that solves the above problem includes a package storage unit that stores a content package in which generation variables for creating content are set, an impression rating storage unit that stores a plurality of pieces of impression rating information at the time of content generation using the content package, an impression information storage unit that stores impression information of each content package calculated by analyzing the impression rating information, and a control unit that generates content.The control unit acquires a target text, calculates rating information, identifies a content package that is close to the rating information from the impression information stored in the impression information storage unit, calls up generation variables of the identified content package from the package storage unit, and generates content using the generation variables. Effect of the Invention

[0007] According to the present invention, a variety of contents can be generated according to text. [Brief description of the drawings]

[0008] [Figure 1] FIG. 1 is a schematic diagram of a system according to an embodiment of the present invention. [Diagram 2] FIG. 2 is an explanatory diagram of a hardware configuration of the present embodiment. [Diagram 3] 4A is an explanatory diagram of an information storage unit of the present embodiment, in which FIG. 4A is an explanatory diagram of a package storage unit, FIG. 4B is an explanatory diagram of an impression evaluation storage unit, FIG. 4C is an explanatory diagram of an adjective dictionary storage unit, and FIG. 4D is an explanatory diagram of an impression information storage unit. [Figure 4] FIG. 4 is an explanatory diagram of a processing procedure for dictionary creation processing according to the embodiment. [Diagram 5] 5A is an explanatory diagram of an impression rating table, FIG. 5B is an explanatory diagram of impression rating information, and FIG. 5C is an explanatory diagram of the analysis results of principal component analysis, which are used in the present embodiment. [Figure 6]4A and 4B are explanatory diagrams of maps according to the present embodiment, in which FIG. 4A is a sound package map and FIG. [Figure 7] 5A and 5B are diagrams illustrating the processing procedure of the present embodiment, in which FIG. 5A illustrates the acquisition of synonyms of the first principal component, and FIG. 5B illustrates the acquisition of synonyms of the second principal component. [Figure 8] FIG. 4 is an explanatory diagram of a processing procedure of a sound reproduction process according to the embodiment. [Figure 9] FIG. 4 is an explanatory diagram of a text analysis process according to the present embodiment. [Figure 10] 6 is an explanatory diagram of a process for specifying a sound package according to a position on the axis of a principal component in the present embodiment. [Figure 11] 13A and 13B are explanatory diagrams of the processing procedures of classification processing and sound reproduction processing according to another embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] Hereinafter, an embodiment of a content generation system, a content generation method, and a content generation program will be described with reference to Figs. 1 to 10. In this embodiment, it is assumed that a service is provided for generating content according to the contents of a text created by a user. In this embodiment, automatic composition is performed to generate music as content. In this composition, music is generated using a sound package (content package) that combines multiple variable values ​​used in automatic composition, and is provided for the text. For this purpose, as shown in FIG. 1, a user terminal 10, a support device 20, and a natural language processing server 30 are used, which are connected via a network.

[0010] (Hardware configuration description) 2, a hardware configuration of an information processing device H10 constituting the user terminal 10, the support device 20, and the natural language processing server 30 will be described. The information processing device H10 includes a communication device H11, an input device H12, a display device H13, a storage unit H14, and a processor H15. Note that this hardware configuration is an example, and it may be realized by other hardware.

[0011] The communication device H11 is an interface that establishes a communication path with another device and transmits and receives data, and is, for example, a network interface or a wireless interface.

[0012] The input device H12 is a device that accepts input from a user, etc., and is, for example, a mouse, a keyboard, etc. The display device H13 is, for example, a display that displays various information. The storage unit H14 is a storage device that stores data and various programs for executing various functions of the user terminal 10, the assistance device 20, and the natural language processing server 30. Examples of the storage unit H14 include a ROM, a RAM, and a hard disk.

[0013] The processor H15 uses the programs and data stored in the storage unit H14 to control each process in the user terminal 10, the support device 20, and the natural language processing server 30. Examples of the processor H15 include a CPU and an MPU. The processor H15 loads a program stored in a ROM or the like into a RAM and executes various processes for each service.

[0014] The processor H15 is not limited to a processor that performs software processing for all the processes it executes. For example, the processor H15 may be equipped with a dedicated hardware circuit (e.g., an application specific integrated circuit: ASIC) that performs hardware processing for at least a part of the processes it executes. That is, the processor H15 may be configured as a circuit including (1) one or more processors that operate according to a computer program (software), (2) one or more dedicated hardware circuits that execute at least a part of various processes, or (3) a combination thereof. The processor includes a CPU and memory such as RAM and ROM, and the memory stores program code or instructions configured to cause the CPU to execute processes. The memory, i.e., computer-readable medium, includes any available medium that can be accessed by a general-purpose or dedicated computer.

[0015] (System Configuration) Next, the system configuration of the user terminal 10, the assistance device 20, and the natural language processing server 30 will be described with reference to FIG.

[0016] The user terminal 10 is a computer terminal used by a user. In this embodiment, the users include subjects who fill out a questionnaire about their impressions of music automatically composed by the sound package, and people who wish to use the automatic composition service.

[0017] The support device 20 is a computer that automatically composes music based on a text and plays back the composed music. The support device 20 includes a control unit 21, a package storage unit 22, an impression evaluation storage unit 23, an adjective dictionary storage unit 24, and an impression information storage unit 25.

[0018] The control unit 21 includes control means (CPU, RAM, ROM, etc.) and performs processes (each process of a learning stage, a principal component analysis stage, a text analysis stage, a content generation stage, etc.) described below. By executing a content generation program for that purpose, the control unit 21 functions as a learning unit 211, a principal component analysis unit 212, a text analysis unit 213, and a content generation unit 214.

[0019] The learning unit 211 executes a process of learning impressions about music composed using each sound package. The principal component analysis unit 212 performs a process of analyzing the main impressions of each sound package. The text analysis unit 213 analyzes the text acquired from the user terminal 10 of the service user, and executes a process of identifying a sound package for composing music to be added to the text. The content generating unit 214 executes a process for automatic composition using a sound package corresponding to the text.

[0020] As shown in Fig. 3(a), a package management record 220 for a sound package that combines variable values ​​used for composing music is recorded in the package storage unit 22. This package management record 220 is recorded before the service is provided. In this package management record 220, data on a plurality of variables is recorded for a package ID.

[0021] The package ID data area records data related to an identifier for identifying each sound package. The variable data area stores data on the values ​​of the generation variables used when composing music with this sound package. The generation variables can be values ​​that determine the elements required for composing music (rhythm, time signature, scale, temperament, melody, etc.).

[0022] As shown in Fig. 3(b), the impression evaluation storage unit 23 records an impression evaluation management record 230 regarding each subject's impression of music composed by a sound package. In this embodiment, the impression of music is recorded when it is acquired by a questionnaire given to the subjects. The impression evaluation management record 230 records data regarding the subject ID, package ID, adjective, and evaluation value.

[0023] The subject ID data area records data regarding an identifier for identifying the subject who performed the impression evaluation. The package ID data area records data related to an identifier for identifying each sound package.

[0024] The adjective data area stores data on adjectives used in impression evaluation. The rating data area stores data on the percentage of adjectives that apply to the impression of music composed using the sound package (impression rating value).

[0025] 3(c), the adjective dictionary storage unit 24 records adjective management information 240 about combinations of adjectives of principal components expressing a sound package. In this embodiment, since two principal components are used, a first adjective management record 241 and a second adjective management record 242 are recorded. This adjective management information 240 is recorded when a dictionary creation process is performed.

[0026] In the first adjective management record 241, data relating to adjectives, weighting values, synonyms, and correlation values ​​is recorded. In the adjective data area, data on adjectives (first adjectives) used as the first principal component identified by the principal component analysis is recorded. In this embodiment, the term "adjective" is used to mean a word that describes the impression of music, and for example, an adjective can be used.

[0027] In the weighting value data area, data on the weighting value in the first principal component for this first adjective is recorded. The synonym data area stores data on synonyms of the first adjective. The synonyms are identified by a so-called co-occurrence analysis.

[0028] In the correlation value data area, data on the degree of association of the synonym with the first adjective is recorded. In this embodiment, the degree of co-occurrence between the first adjective and the synonym in co-occurrence analysis can be used as the correlation value.

[0029] In the second adjective management record 242, data relating to adjectives, weighting values, synonyms, and correlation values ​​is recorded. In the adjective data area, data on adjectives (second adjectives) used as the second principal component identified by the principal component analysis is recorded. In this embodiment, adjectives are used as the second adjectives.

[0030] In the weighting value data area, data on the weighting value in the second principal component for this second adjective is recorded. The synonym data area stores data on synonyms of the second adjective. The synonyms are identified by a so-called co-occurrence analysis.

[0031] In the correlation value data area, data on the degree of association of the synonym with the second adjective is recorded. In this embodiment, the degree of co-occurrence between the second adjective and the synonym in co-occurrence analysis can be used as the correlation value.

[0032] As shown in Fig. 3(d), the impression information storage unit 25 records an impression management record 250 about the impression of music composed by a sound package. This impression management record 250 is recorded when a dictionary creation process is performed. The impression management record 250 records data about the package ID, the first principal component value, and the second principal component value.

[0033] The package ID data area records data related to an identifier for identifying each sound package. The data areas for the first principal component value and the second principal component value store data relating to values ​​representing each sound package using the first principal component and the second principal component, respectively.

[0034] The natural language processing server 30 is a computer system that provides information related to natural language. In this embodiment, the natural language processing server 30 identifies synonyms for a given word. Then, the natural language processing server 30 calculates an association value calculated by a co-occurrence analysis between the given word and the synonym. The natural language processing server 30 also calculates an association value calculated by a co-occurrence analysis between two words.

[0035] (Dictionary creation process) The dictionary creation process will be described with reference to Fig. 4. Here, each subject uses a user terminal 10 to access the support device 20. A number of subjects are asked to listen to music composed using a sound package and answer a questionnaire about their impressions.

[0036] For this reason, the following process is repeated for each subject and for each sound package. First, the control unit 21 of the support device 20 executes a sound generation process using a sound package (step S1-1). Specifically, the learning unit 211 of the control unit 21 provides the variable values ​​of the sound package to be processed to the content generation unit 214. In this case, the content generation unit 214 composes music using the variable values ​​of the sound package. Here, even when the variable values ​​of the same sound package are used, different music pieces are generated due to the arbitrariness of the variable values ​​of other elements. Then, the learning unit 211 plays the composed music on the user terminal 10.

[0037] Next, the control unit 21 of the support device 20 executes a process of acquiring impression evaluation information (step S1-2). Specifically, the learning unit 211 of the control unit 21 outputs a screen including an impression evaluation table to the user terminal 10 for a questionnaire for acquiring an impression of music.

[0038] As shown in FIG. 5(a), the impression rating table 500 has an input field for rating the impression of music in multiple stages for multiple pairs of opposing adjectives (adjectives). For example, between "dark" and "bright", a percentage (impression rating value) representing which impression of the music is closer to can be selected in the input field with 7 stages including the median ("0" in FIG. 5(a)). Note that the input is not limited to the median or 7 stages as long as it can identify the impression of the music. The test subjects who listen to the music input their impression rating value as their impression into the impression rating table 500 output to the user terminal 10. In this case, the learning unit 211 acquires the input impression rating value as a questionnaire response.

[0039] 5(b), an impression rating value for each package ID (Pack) is obtained from each subject's questionnaire response 510. Then, an impression rating management record 230 recording the subject ID is generated and recorded in the impression rating storage unit 23. Then, for each subject, the above process is repeated for each sound package.

[0040] Next, the control unit 21 of the support device 20 executes a principal component analysis process (step S1-3). Specifically, the learning unit 211 of the control unit 21 performs a principal component analysis of the evaluation values ​​of the impression evaluation management record 230 to reconstruct a principal component axis that plots components that represent the variance of the impression of music composed by each sound package. In this case, the principal component analysis calculates a plurality of adjectives (variables) that express each sound package and a weighting value for each adjective.

[0041] Specifically, the principal component analysis process will be described with reference to FIG. As shown in Fig. 5(c), an analysis result 520 is obtained by principal component analysis. In this analysis result 520, weighting values ​​(weight 11 to weight mn) of each adjective (1 to n) are calculated for new axes (1 to m) effective for distinguishing each sound package. Furthermore, the contribution rate (c1 to cm) of each new axis to the distinction of each sound package is calculated.

[0042] Next, the learning unit 211 determines the number of axes based on the cumulative contribution rate of the principal component axes. Here, the contribution rates of each principal component axis calculated by the principal component analysis are added in descending order to calculate the cumulative contribution rate. Then, the principal component axis whose cumulative contribution rate exceeds a reference value (e.g., 0.8) is specified as a new axis to be used for evaluation. In this embodiment, it is assumed that two axes (new axis 1 and new axis 2) of the first and second principal components are specified as the new axes.

[0043] Next, the control unit 21 of the support device 20 executes a mapping process (step S1-4). Specifically, the learning unit 211 of the control unit 21 calculates the first principal component value and the second principal component value of each sound package using the new axis 1 (first principal component) and the new axis 2 (second principal component). Then, the learning unit 211 generates an impression management record 250 in which the package ID, the first principal component value, and the second principal component value are recorded, and records the impression management record 250 in the impression information storage unit 25.

[0044] Here, using Fig. 6(a), mapping using the principal component values ​​of each sound package on new axis 1 (first principal component) and new axis 2 (second principal component) will be explained. In sound package map 530 of new axis 1 and new axis 2, each sound package is placed at coordinates (first principal component value, second principal component value). In sound package map 530, the package ID of each sound package is indicated by a circled number.

[0045] Next, the control unit 21 of the support device 20 executes an extraction process of an adjective group of the first and second principal components (step S1-5). Specifically, the learning unit 211 of the control unit 21 specifies weighting values ​​in the first and second principal components of each adjective in the analysis result 520 of the principal component analysis. Then, the learning unit 211 specifies adjectives whose absolute values ​​of the weighting values ​​are equal to or greater than a predetermined threshold value, and generates an adjective group.

[0046] In this case, as shown in Fig. 6(b), in adjective map 540, which has two axes representing the weighting values ​​of new axis 1 and new axis 2, each adjective is placed at the coordinates (weighting value of new axis 1, weighting value of new axis 2). For example, referring to Fig. 5(c), the weighting value of "adjective 1" on new axis 1 is "weight 11," and the weighting value of new axis 2 is "weight 21," so in adjective map 540, "adjective 1" is placed at the coordinates (weight 11, weight 21).

[0047] Next, the control unit 21 of the support device 20 executes an extraction process of a thesaurus of the adjective group (step S1-6). Specifically, the learning unit 211 of the control unit 21 acquires synonyms of the adjectives in each principal component from the natural language processing server 30.

[0048] For example, as shown in FIG. 7(a), synonyms (1A to 1C), synonyms (2A, 2B), and synonyms (7A to 7D) are obtained for the adjective group (1, 2, 7) of the first principal component.

[0049] Also, as shown in FIG. 7(b), synonyms (3A to 3C), synonyms (5A, 5B), and synonyms (10A, 10B) are obtained for the group of adjectives (3, 5, 10) in the second principal component.

[0050] Furthermore, the learning unit 211 acquires from the natural language processing server 30 the association values ​​of each synonym for each adjective. Then, the learning unit 211 generates a first adjective management record 241 and a second adjective management record 242 in which the weighting values, synonyms, and related values ​​are recorded for each adjective in the first principal component and the second principal component, respectively, and records them in the adjective dictionary storage unit 24.

[0051] (Sound playback processing) Next, the sound reproduction process will be described with reference to Fig. 8. A service user who wishes to add music to text accesses the assistance device 20 using the user terminal 10.

[0052] First, the control unit 21 of the assistance device 20 executes a text acquisition process (step S2-1). Specifically, the text analysis unit 213 of the control unit 21 outputs a text input screen to the user terminal 10. Then, the service user inputs desired text into the text input screen. Next, the text analysis unit 213 acquires the target text to which music is to be added from the user terminal 10. Then, the text analysis unit 213 extracts words (expression elements) included in the acquired text.

[0053] Next, the control unit 21 of the support device 20 executes a text analysis process (step S2-2). Specifically, the text analysis unit 213 of the control unit 21 calculates an evaluation value of the relevance between the extracted word and the adjectives and synonyms recorded in the adjective dictionary storage unit 24.

[0054] The processing for the first principal component will be described with reference to FIG. 9. Here, the text analysis unit 213 acquires weighting values ​​(weight 11, weight 31, weight 71) for the adjectives (1, 3, 7) from the adjective dictionary storage unit 24. Furthermore, the text analysis unit 213 acquires synonyms and correlation values ​​(first correlation values) for the adjectives from the adjective dictionary storage unit 24. In this case, the synonyms of each adjective use the same weighting value as the adjective. Next, the text analysis unit 213 acquires correlation values ​​(second correlation values ​​as evaluation information) between each extracted word (extracted word) and the adjectives and synonyms recorded in the adjective dictionary storage unit 24 from the natural language processing server 30.

[0055] Then, the text analysis unit 213 substitutes the weighting value, the first correlation value, and the second correlation value into a predetermined evaluation function for each adjective and synonym to calculate an evaluation value. For example, the evaluation value may be a product of the weighting value, the first correlation value, and the weighting value.

[0056] Then, the text analysis unit 213 calculates a first evaluation value for the first principal component using the evaluation values ​​of each of the adjectives and synonyms. For example, the highest value among the evaluation values ​​of each of the adjectives and synonyms is calculated as the first evaluation value. Similarly, a second evaluation value is calculated for the second principal component.

[0057] Next, the control unit 21 of the support device 20 executes a sound package specification process (step S2-3). Specifically, the text analysis unit 213 of the control unit 21 uses the first evaluation value and the second evaluation value to specify a close sound package according to the position on the axis of the principal component.

[0058] 10, a position 531 is identified by coordinates (first evaluation value, second evaluation value) in the sound package map 530. Then, the text analysis unit 213 identifies the package ID of the coordinates closest to the position 531.

[0059] Next, the control unit 21 of the support device 20 executes a sound generation process using the sound package (step S2-4). Specifically, the text analysis unit 213 of the control unit 21 acquires the variable values ​​of the identified package ID from the package storage unit 22. Next, the content generation unit 214 composes music using the acquired variable values. Then, the content generation unit 214 plays the composed music on the user terminal 10.

[0060] As described above, according to this embodiment, the following effects can be obtained. (1) According to this embodiment, the sound package executes a sound generation process (step S1-1) and an impression evaluation information acquisition process (step S1-2). This makes it possible to acquire each subject's impression of the music composed by the sound package.

[0061] (2) According to this embodiment, the control unit 21 of the support device 20 executes a principal component analysis process (step S1-3), which makes it possible to obtain information about the main impressions that express the music composed by the sound package.

[0062] (3) According to this embodiment, the control unit 21 of the support device 20 executes an extraction process of adjective groups of the first and second principal components (step S1-5) and an extraction process of a thesaurus of the adjective groups (step S1-6). This makes it possible to identify expression elements that represent the subject's impression by adjectives and synonyms.

[0063] (4) According to this embodiment, the control unit 21 of the support device 20 executes a text acquisition process (step S2-1) and a text analysis process (step S2-2). This makes it possible to identify expression elements for creating content to be added to the text.

[0064] (5) According to this embodiment, the control unit 21 of the support device 20 executes a process of identifying a sound package (step S2-3) and a process of generating sound by the sound package (step S2-4). This allows for a composition that is related to the expression elements and impressions contained in the text. In this case, the composition is performed for the text using the variable values ​​of predetermined elements (generated variables). Therefore, unlike the case where music is prepared in advance, a variety of content can be provided due to the arbitrariness of the variable values ​​of other elements.

[0065] This embodiment can be modified as follows: This embodiment and the following modifications can be combined with each other to the extent that there is no technical contradiction. In the above embodiment, automatic composition is performed to generate music as content. The content is not limited to music that appeals to the sense of hearing, but may be any content that appeals to the five human senses. For example, images (sight), scents (olfaction), textures (touch), tastes (gustation), etc. to be added to text may be generated.

[0066] In the above embodiment, a user terminal 10, a support device 20, and a natural language processing server 30 are used, which are connected via a network. The hardware configuration is not limited to this. For example, a thesaurus may be provided in the support device 20.

[0067] In the above embodiment, the control unit 21 of the support device 20 executes a principal component analysis process (step S1-3). As long as the impression of each package can be identified, the analysis method is not limited to the principal component analysis. For example, factor analysis may be used. Also, a model that predicts the likelihood of each package using adjectives and synonyms as input may be created by machine learning.

[0068] In the above embodiment, co-occurrence analysis is used to identify synonyms and calculate correlation values, but the analysis method is not limited to this. In the above embodiment, the control unit 21 of the support device 20 executes a mapping process (step S1-4). Here, the first principal component and the second principal component are used, but the number of axes of the principal components is not limited to "2".

[0069] In the above embodiment, the control unit 21 of the support device 20 executes a text analysis process (step S2-2). Here, the evaluation value is calculated by multiplying the weighting value, the first correlation value, and the weighting value. The method of calculating the evaluation value is not limited to multiplication, and any method can be used as long as it can calculate the relevance between the expression element representing each sound package and the extracted word.

[0070] Furthermore, the text analysis unit 213 calculates a first evaluation value for the first principal component using the evaluation values ​​of each adjective and synonym. In this case, in the above embodiment, the highest value among the evaluation values ​​of each adjective and synonym is calculated as the first evaluation value. Similarly, a second evaluation value is calculated for the second principal component. Here, the first evaluation value and the second evaluation value may be determined by any method that can obtain values ​​corresponding to various principal components for expressions included in the text, such as by using a predetermined function.

[0071] In the above embodiment, the control unit 21 of the support device 20 executes a process of identifying a sound package (step S2-3) and a process of generating a sound using the sound package (step S2-4). Here, the composition may be performed using variable values ​​according to the situation when the service is used. In this case, in the dictionary creation process, a principal component analysis is performed including situation information (e.g., time period, season, etc.) when the subject answered the questionnaire.

[0072] In the above embodiment, the control unit 21 of the support device 20 executes a text acquisition process (step S2-1). Here, the control unit 21 of the support device 20 acquires the target text from the user terminal 10. The method of acquiring the text is not limited to this. For example, the control unit 21 of the support device 20 or the user terminal 10 may collect the voice of the service user and generate the target text by voice recognition.

[0073] In the above embodiment, music is generated according to the text contents based on the text created by the user. Alternatively, an adjective may be added to the newly acquired music. In this case, the music to which the adjective is added (newly acquired music) is input to the support device 20. In this case, the control unit 21 identifies the musical elements (rhythm, time signature, scale, temperament, melody, etc.) of the newly acquired music. Then, the control unit 21 compares the musical elements of the newly acquired music with the musical elements of the sound package to identify a sound package whose statistical values ​​of each musical element are close to the newly acquired music. Next, the control unit 21 identifies an adjective in the adjective map 540 using the first principal component value and the second principal component value of the sound package whose musical elements are close. This allows the adjective to be added to the newly acquired music.

[0074] In the above embodiment, the control unit 21 of the support device 20 executes a text acquisition process (step S2-1), a text analysis process (step S2-2), and a sound package specification process (step S2-3). Here, the text analysis unit 213 of the control unit 21 uses the coordinates (first evaluation value, second evaluation value) calculated by the text analysis process to specify a sound package located close to the new axis coordinate space. Here, instead of specifying one sound package, multiple sound packages may be specified so that the user can select from among them. In this case, the control unit 21 of the support device 20 clusters the sound packages in advance in the new axis coordinate space. Furthermore, a representative point (for example, a center of gravity) of the cluster is specified for each cluster. Then, the control unit 21 specifies a cluster corresponding to the coordinates (first evaluation value, second evaluation value) in the sound package specification process (step S2-3). Here, the distance between the coordinates (first evaluation value, second evaluation value) and the representative point of each cluster is calculated, the representative point having the closest distance to the evaluation value is specified, and the cluster of this representative point is specified.

[0075] Next, the control unit 21 outputs the sound packages included in the identified cluster as selection candidates to the user terminal 10. Then, when a sound package candidate is selected in the user terminal 10, the control unit 21 of the support device 20 executes a sound generation process using the sound package, similar to step S1-1.

[0076] Also, if the distance between the coordinates (first evaluation value, second evaluation value) calculated by the text analysis process and the representative point of each cluster is equal to or greater than a threshold, it may be determined that the sound package is outside the cluster. Also, if the distances from the representative points of multiple clusters are equal within a predetermined range, it may be determined that the sound package is located approximately on the boundary between the clusters. In such a case, a sound generation package corresponding to a point group that is close to the evaluation value within a predetermined threshold may be identified as a selection candidate for the user. In addition to a clustering method in which all sound generation packages belong to one of the clusters, there is also a clustering method that allows sound generation packages that do not belong to any cluster. In the latter clustering method, if the coordinates (first evaluation value, second evaluation value) are closer to a sound generation package that does not belong to any cluster than the representative point of each cluster, this sound generation package may be recommended.

[0077] In the above embodiment, the control unit 21 of the support device 20 executes a text acquisition process (step S2-1), a text analysis process (step S2-2), and a sound package specification process (step S2-3). Here, the text type may be classified according to the contents of the text, and the content may be selected according to this classification to execute the sound playback process. As the text type, for example, classifications such as news, novels, folk tales, and online articles may be used. In addition, as the text type, entities and attributes that the text deals with, or topics that are evoked from a group of multiple nouns included in the text may be used. Furthermore, at least one of the entities, attributes, and topics that each text deals with may be determined by machine learning.

[0078] The classification process and sound reproduction process will be described with reference to FIG. 11. Here, the control unit 21 is provided with a learning unit that creates a sentence type prediction model that predicts the sentence type from the text by machine learning. Furthermore, a package type table is prepared for identifying a sound package according to the sentence type. In this package type table, applicable sentence types may be recorded in association with each package ID of the sound package, or in association with a cluster of the sound package. First, in the classification process, the control unit 21 of the support device 20 executes a process of acquiring teacher information (step S3-1). Specifically, the learning unit of the control unit 21 acquires teacher information that associates a text with a sentence type. This sentence type may be set manually or may be set using a resource (source) from which the text was acquired.

[0079] Then, the control unit 21 of the assistance device 20 executes a process of generating a sentence type prediction model (step S3-2). Specifically, the learning unit of the control unit 21 generates a sentence type prediction model (sentence type prediction information) that predicts a sentence type from a text by machine learning. Next, in the sound reproduction process, the control unit 21 of the support device 20 executes a text acquisition process (step S4-1) in the same manner as in step S2-1. Next, the control unit 21 of the assistance device 20 executes a sentence type prediction process (step S4-2). Specifically, the text analysis unit 213 of the control unit 21 inputs the acquired text to a sentence type prediction model to predict the sentence type. Next, the control unit 21 of the assistance device 20 executes the text analysis process (step S4-3), similarly to step S2-2. Next, the control unit 21 of the support device 20 executes a process of identifying available sound package candidates (step S4-4). Specifically, the text analysis unit 213 of the control unit 21 uses a package type table to identify sound package candidates corresponding to the predicted text type. Next, the control unit 21 of the support device 20 executes a sound package specification process similar to step S2-3 (step S4-5). Here, the text analysis unit 213 of the control unit 21 specifies a sound package that is close to the sound package candidate specified using the package type table, according to the position on the axis of the principal component, using the first evaluation value and the second evaluation value.

[0080] Then, the control unit 21 of the support device 20 executes a sound generation process using the sound package in the same manner as in step S2-4 (step S4-6). Even when the same adjective is used in different texts, the adjective may be used differently depending on the type of text, but it is possible to specify an appropriate sound package depending on the type of text.

[0081] Also, a package type table for identifying a sound package according to a text type may be created by machine learning. In this case, teacher information in which a text type is associated with a sound package is prepared. Then, the control unit 21 of the support device 20 generates a sound package classification model for predicting an applicable text type from the variable value of the sound package in the teacher information by machine learning. Then, the control unit 21 inputs the variable value of the sound package that has not yet been classified into the sound package classification model, predicts the applicable text type, and creates a package type table. The control unit 21 of the support device 20 may also divide the input text into individual sentences and estimate a sound generation package for each sentence according to the object, attribute, subject, etc. of the sentence. In this case, a sound generation package or cluster according to the object, attribute, subject, etc. of each sentence may be identified and a sound generation package or cluster recommended for the entire text may be output together.

[0082] In the above embodiment, the control unit 21 of the support device 20 executes a sound generation process using a sound package (step S2-4). Here, feedback on the sound package identified in step S2-3 may be obtained from the user. In this case, the user's reaction to the sound package recommended based on the text (e.g., like, dislike, no reaction, etc.) is obtained and stored as teacher data in association with each evaluation value and package ID. Then, the control unit 21 of the support device 20 generates a reaction prediction model that predicts the reaction from the user by machine learning. When new text is obtained, the control unit 21 of the support device 20 uses the reaction prediction model to identify and recommend a sound package that predicts a favorable reaction. Furthermore, in this case, the control unit 21 of the support device 20 may calculate the text type, object, attribute, and subject as feature amounts from the text, and perform machine learning by combining the feature amounts with the user's reaction to each sound package.The control unit 21 of the support device 20 then uses a reaction prediction model to identify and recommend a sound package that predicts a favorable reaction according to the text type, object, attribute, and subject. [Explanation of symbols]

[0083] 10...user terminal, 20...assistance device, 21...control unit, 211...learning unit, 212...principal component analysis unit, 213...text analysis unit, 214...content generation unit, 22...package storage unit, 23...impression rating storage unit, 24...adjective dictionary storage unit, 25...impression information storage unit, 30...natural language processing server

Claims

1. a package storage unit that stores a content package in which a generation variable for creating the content is set; an impression evaluation storage unit that stores impression evaluation information consisting of a plurality of words that evaluate impressions of content generated using the content package; an impression information storage unit that stores impression information consisting of component values ​​of each content package calculated by analyzing the impression rating information; A content generation system including a control unit connected to a user terminal and generating content, The control unit: acquiring a target text from the user terminal, extracting words contained in the target text, and calculating an evaluation value which is evaluation information regarding a correlation with the words contained in the impression evaluation information; Identifying a content package that is close to the evaluation value among the component values ​​of the impression information stored in the impression information storage unit; A content generation system characterized by retrieving generation variables of the specified content package from the package storage unit, generating content using the generation variables, and outputting the content to the user terminal.

2. 2. The content generating system according to claim 1, wherein the control unit calculates a principal component value of each content package by performing a principal component analysis using the impression rating information.

3. The impression evaluation information is an adjective expressing an impression of the content, The control unit: weighting the adjectives according to impression information of each content package; 3. The content generation system according to claim 1, further comprising: a processor for calculating evaluation information of the target text by comparing the adjective with the target text and using the weighting.

4. The content generation system according to claim 3 , wherein the control unit calculates an evaluation value of the target text based on synonyms of the adjective and the target text.

5. The content generation system according to claim 4 , wherein the control unit calculates an evaluation value of the target text using a degree of co-occurrence calculated by a co-occurrence analysis between the adjective and a synonym.

6. The control unit: Identifying a sentence type for the target text using sentence type prediction information; Identifying available candidate content packages for said document type; 6. The content generation system according to claim 1, further comprising: identifying a content package that is close to the evaluation information from among the content package candidates.

7. a package storage unit that stores a content package in which a generation variable for creating content is set; an impression evaluation storage unit that stores impression evaluation information consisting of a plurality of words that evaluate impressions of content generated using the content package; an impression information storage unit that stores impression information consisting of component values ​​of each content package calculated by analyzing the impression rating information; A method for generating content using a content generation system including a control unit connected to a user terminal and generating content, the method comprising: The control unit: acquiring a target text from the user terminal, extracting words contained in the target text, and calculating an evaluation value which is evaluation information regarding a correlation with the words contained in the impression evaluation information; Identifying a content package that is close to the evaluation value among the component values ​​of the impression information stored in the impression information storage unit; A content generating method comprising the steps of: retrieving a generation variable of the specified content package from the package storage unit; generating content using the generation variable; and outputting the content to the user terminal.

8. a package storage unit that stores a content package in which a generation variable for creating the content is set; an impression evaluation storage unit that stores impression evaluation information consisting of a plurality of words that evaluate impressions of content generated using the content package; an impression information storage unit that stores impression information consisting of component values ​​of each content package calculated by analyzing the impression rating information; A program for generating content using a content generation system including a control unit connected to a user terminal and generating content, The control unit, acquiring a target text from the user terminal, extracting words contained in the target text, and calculating an evaluation value which is evaluation information regarding a correlation with the words contained in the impression evaluation information; Identifying a content package that is close to the evaluation value among the component values ​​of the impression information stored in the impression information storage unit; A content generation program characterized by calling up generation variables of the specified content package from the package storage unit, generating content using the generation variables, and functioning as a means for outputting the content to the user terminal.

Images