JP7396168B2 - Idea support devices, idea support systems and programs - Google Patents

Description

The present invention relates to an idea support device, an idea support system, and a program.

Learning is performed by processing large amounts of data using an electronic computer, and based on the results of this learning, it is now possible to extract complex rules and create patterns that were previously difficult to find by humans. Among the technologies related to text processing, there is a technology that expresses words as multidimensional vectors (Word2vec), quantitatively evaluates them, and processes similarities and regularities through calculations (for example, Patent Document 1).

In addition, key elements in a sentence can be converted as appropriate by acquiring synonyms and antonyms based on the distance between multiple terms indicated by the above multidimensional vector, or by acquiring synonyms and antonyms from dictionary information. There is a technology related to an ideation support device that generates a composite sentence in which different words are substituted (Patent Document 2).

JP 2019-20893 Publication JP2019-86815A

However, the numerical value indicated by the calculation result of the multidimensional vector does not necessarily indicate any term, and in some cases, an appropriate term may not be obtained. In addition, there is a problem in that simply converting to similar terms that have existed in the past may not expand the range of ideas very much.

An object of the present invention is to provide an idea support device, an idea support system, and a program that can support more diverse ideas.

In order to achieve the above object, the invention according to claim 1 of the present invention,
acquisition means for acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation means for performing a predetermined calculation on the plurality of acquired meaning vectors;
generation means for generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
This is an idea support device characterized by comprising:

Further, the invention according to claim 2 is the idea support device according to claim 1, which includes:
The reference object is characterized in that it is indicated by characters.

Further, the invention according to claim 3 is the idea support device according to claim 1 or 2,
The acquisition means is characterized in that each of the input objects is acquired from a plurality of input data.

Further, the invention according to claim 4 provides the idea support device according to any one of claims 1 to 3,
The input object is characterized in that it is an independent word in input data including a sentence.

Further, the invention according to claim 5 provides the idea support device according to any one of claims 1 to 4,
The output object is characterized in that it is image data or audio data.

Further, the invention according to claim 6 provides the idea support device according to any one of claims 1 to 5,
A plurality of peripheral objects each having a content indicated by a semantic vector whose distance from the output semantic vector satisfies a predetermined condition is identified, and the peripheral object is determined by the distance between the semantic vector of the peripheral object and the output semantic vector. It is characterized by comprising an output means that outputs the output object in association with the output object.

Further, the invention according to claim 7 is the idea support device according to claim 6,
The output object is image data,
The output means outputs a correlation diagram in which the peripheral objects are arranged around the generated image related to the output object, separated from the generated image by an interval corresponding to the distance.

Further, the invention according to claim 8 is the idea support device according to claim 6 or 7,
The output object is image data,
The output means outputs a correlation diagram in which the peripheral objects are arranged around the generated image related to the output object, with an angular difference depending on the distance of the meaning vector between the plurality of peripheral objects. shall be.

Further, the invention according to claim 9 is the idea support device according to claim 7 or 8,
The output means is characterized in that it outputs a peripheral image related to the peripheral object by displaying it on the correlation diagram.

Further, the invention according to claim 10 provides the idea support device according to any one of claims 7 to 9,
The output means is characterized in that it generates an intermediate object whose meaning vector is located between the peripheral object and the output object, and outputs an intermediate image related to the intermediate object displayed on the correlation diagram.

Furthermore, the invention according to claim 11 provides the idea support device according to any one of claims 7 to 10,
The output means selectively displays, on the correlation diagram, the peripheral objects whose size difference from the output meaning vector is within a reference range in some dimensions of a predetermined number or more out of the number of dimensions expressing the meaning vector. It is characterized by causing

Further, the invention according to claim 12 provides the idea support device according to any one of claims 7 to 11,
The output means is characterized in that, among the images included in the correlation diagram, images with corresponding text data set and those without corresponding text data can be identified in an additional display.

Further, the invention according to claim 13 is the idea support device according to claim 12,
The additional display includes a difference in frame line depending on whether or not the setting is applied to the image.

Further, the invention according to claim 14 is the idea support device according to claim 13,
The difference includes any one of line type, line thickness, line color, line shape, shadow, and three-dimensional display.

Further, the invention according to claim 15 provides the idea support device according to any one of claims 12 to 14,
The additional display includes a text display related to the set text data.

Further, the invention according to claim 16 is the idea support device according to claim 12,
The additional display includes the presence or absence or difference of an additional mark depending on the presence or absence of the text data for the image.

Moreover, the invention according to claim 17 is the idea support device according to claim 16,
The additional mark may include parentheses or a dot.

Further, the invention according to claim 18 provides the idea support device according to any one of claims 1 to 16,
The predetermined operation includes at least one of four arithmetic operations, an inner product, and an outer product.

Moreover, the invention according to claim 19 is:
acquisition means for acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation means for performing a predetermined calculation on the plurality of acquired meaning vectors;
generation means for generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
This is an idea support system that is characterized by the following.

Furthermore, the invention according to claim 20 is:
to the computer,
an acquisition step of acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation step of performing a predetermined calculation on the plurality of acquired meaning vectors;
a generation step of generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
This is a program characterized by executing the following.

According to the present invention, there is an effect that more diverse ideas can be supported.

FIG. 1 is a block diagram showing the functional configuration of the ideation support system according to the present embodiment. FIG. 3 is a diagram illustrating an example of idea support. FIG. 3 is a diagram illustrating the correspondence between calculation results and outputs. FIG. 3 is a diagram illustrating the correspondence between calculation results and outputs. FIG. 3 is a diagram illustrating the correspondence between calculation results and outputs. FIG. 3 is a diagram illustrating an example of an image output pattern. It is a flowchart which shows the control procedure of idea support control processing of this embodiment. FIG. 7 is a diagram illustrating a modified example of an image output pattern.

Embodiments of the present invention will be described below based on the drawings.
FIG. 1 is a block diagram showing the functional configuration of the ideation support system 1 of this embodiment.
This idea support system 1 includes a server device 10 (idea support device), a database device 20, and a terminal device 30.

The server device 10 includes a control section 11 (acquisition means, calculation means, generation means, output means), a storage section 12, a communication section 13, and the like. The control unit 11 is a processor that includes a CPU (Central Processing Unit), a RAM (Random Access Memory), etc., and performs various arithmetic processes to centrally control the operation of the server device 10.

The storage unit 12 stores a program 121 executed by the control unit 11 and setting data. Furthermore, the storage unit 12 temporarily stores various data input from the terminal device 30 and its processed data. The storage unit 12 includes a nonvolatile memory such as an HDD (Hard Disk Drive) and/or a flash memory. Furthermore, the storage unit 12 may include a RAM or the like for temporarily storing large data and processing it at high speed. The program 121 includes programs related to text analysis, image recognition, voice recognition, and ideation support control processing to be described later. A trained model 1211 that is appropriately designed and trained can be used for each process related to image recognition, keyword extraction, and the like. The program 121 also includes an image generation processing unit 1212 that generates a new image from text data (here, it may be (a combination of) semantic vectors corresponding to the text data) based on learning data. For example, GAN (Generative Adversarial Nets) is known as the image generation processing unit 1212 that converts text (such as a sentence) into an image using a learning model. Here, one or more of various algorithms that have been developed in this regard may be adopted, and the correspondence may be such that image data can be directly generated from semantic vectors obtained by Word2vec etc. . Alternatively, a new image may be generated by combining a plurality of peripheral images with similar semantic vectors using another method.

The communication unit 13 controls transmission and reception of data with external devices according to a predetermined communication standard such as TCP/IP (Transmission Control Protocol/Internet Protocol). The communication unit 13 is connected to external devices via a network. The network includes the Internet, and is accessible from external devices such as the terminal device 30. The communication unit 13 may have a terminal that can directly communicate with an external device (such as a peripheral device) via a USB (Universal Serial Bus) or the like.

The database device 20 has a storage section 21 . The storage unit 21 stores and holds terminology data 211 in which a large number of objects expressed by text (characters), images, sounds, etc. are associated with their identification data (including values of meaning vectors described below). That is, the text objects (reference objects) of the terminology data 211 are used for learning and generating Word2vec, and are held in correspondence with the values of the meaning vectors obtained as a result of the learning. The terminology data 211 may be added according to usage, etc., and the value of the meaning vector may be updated according to the progress of learning. Furthermore, each reference object may be associated with and held with an image further indicating the content, if necessary. Further, the database device 20 may include a control unit that controls reading and writing of the storage unit 21 and a communication unit that controls communication with external devices. Note that the generation of the terminology data 211, the learning of Word2vec, GAN, etc. do not need to be performed by the server device 10, and those performed by another device may be copied and used.

The terminal device 30 receives input from the user and outputs the input to the user. The terminal device 30 includes a control section 31, a storage section 32, a display section 33, an operation reception section 34, a communication section 35, and the like. The control unit 31 is a processor that has a CPU, a RAM, etc., and controls the operation of the terminal device 30 in an integrated manner. The storage unit 32 includes a nonvolatile memory, an HDD, and the like, and stores various programs and data. The communication unit 35 controls data transmission and reception with external devices according to a predetermined communication standard. The predetermined communication standards include those that can communicate with the server device 10, such as TCP/IP.

The operation reception unit 34 accepts an input operation from the outside (user, etc.) and outputs it to the control unit 31 as an input signal. The operation reception unit 34 includes, but is not particularly limited to, a keyboard, a mouse, a touch sensor provided over the display screen of the display unit 33, and the like, for example. The display section 33 has a display screen, and causes the display screen to display contents according to control commands from the control section 31. Although the display screen is not particularly limited, for example, it is a liquid crystal display screen (LCD). Further, the display section 33 may include an LED lamp (Light Emitting Diode) or the like that indicates a specific state.

Next, the operation contents of idea support in the idea support system 1 of this embodiment will be explained. The ideation support system 1 of this embodiment extracts objects from input data from the terminal device 30, and outputs different objects and different concepts based on the combination of the objects.

FIG. 2 is a diagram illustrating an example of idea support.
As shown in FIG. 2A, for example, when two sentences (a plurality of input data) are input from the terminal device 30 to the server device 10, the server device 10 extracts independent words (nouns, verbs, etc.) from these sentences. , adjectives, etc., and may include unique parts of speech depending on the language) are extracted and obtained as extracted objects. Note that words without specific content, such as "suru" and the auxiliary verb "araru", may be omitted as appropriate. Furthermore, it may be determined whether or not proper nouns such as place names, personal names, and names are used depending on whether the learning data (Word2vec and GAN) sufficiently include them.

Each extracted object is converted into a meaning vector that quantitatively indicates a direction according to the meaning of its content with reference to the term data 211. In the case of a simple compound word, the meaning vectors of the two original words may be added, or they may be divided into two words. A meaning vector corresponding to the extracted object is obtained by referring to the terminology data 211. Extracted objects for which no semantic vector has been defined may be excluded from processing.

Combinations are performed between the two sentence extraction objects as appropriate or by brute force. A predetermined arithmetic process that is predetermined or appropriately set is performed on each of the combined meaning vectors of the plurality of extracted objects (two), and an output meaning vector is calculated.

At this time, the output meaning vector is not necessarily the same as the meaning vector of other terms stored in the terminology data 211, and depending on the combination, the output meaning vector may be a term corresponding to the output meaning vector or a similar term (from the output meaning vector). Terms within a predetermined distance) are often not found in the terminology data 211 (not acquired). As shown in FIG. 2(b), by referring to the terminology data 211 and searching within a predetermined similarity range of the meaning vectors obtained by adding the meaning vectors representing "skydiving" and "sea", respectively. , where scuba diving is located. On the other hand, when "seawalk" and "B Island" are added to "skydiving", there are no corresponding terms within the predetermined similarity range.

3 to 5 are diagrams for explaining the correspondence between calculation results and outputs. Note that although a two-dimensional representation is used here for the sake of explanation, the actual meaning vector is multidimensional, and may have, for example, 100 to 1000 dimensions.
As shown in FIG. 3(a), if there is a term in the semantic vector R1 that is close to the output semantic vector M1, here there is a term whose distance (for example, Euclidean distance) from the output semantic vector M1 is within A1. In this case, the term may be displayed and output as is as text data, as shown in FIG. 3(b), regardless of other surrounding terms R2 to R5 (high diving, etc.).

On the other hand, as shown in FIGS. 4(a) and 4(b), the term ( If the reference object) does not exist (is not obtained from the terminology data 211), simply indicating the terms surrounding the output meaning vectors M21 and M22 does not adequately present the new concept. There is. The server device 10 generates and outputs images (non-text output objects) having the contents indicated by the output semantic vectors M21 and M22 using GAN or the like. Alternatively, as described above, the server device 10 uses the semantic vectors R21 to R23 and R25 to R27 (surrounding objects) indicating terms surrounding the output semantic vectors M21 and M22 to Based on this, images corresponding to the output semantic vectors M21 and M22 may be combined as appropriate. Note that in the case of an object that is expressed by sound rather than an image, audio data may be generated instead of image data.

Note that when elements of a large number of surrounding image data (peripheral images) are mixed, the resulting image may be difficult for the user (human being) to enlarge. On the other hand, for image data whose number is smaller than the number of dimensions, an accurate output semantic vector cannot be obtained by linearly combining a plurality of corresponding semantic vectors. Therefore, in this case, instead of a linear combination, a synthesis method is used in which the image to be mixed can be adjusted component by component, or the image does not need to exactly match the output semantic vector. For example, a predetermined number (sufficiently less than the number of dimensions, for example, several) of semantic vectors that are within a predetermined distance from the output semantic vector (the distance satisfies a predetermined condition) or are close to each other are identified and acquired in a limited manner. and may be added (synthesized) at a ratio depending only on each distance (absolute value) from the output semantic vector to the acquired semantic vector. In this case, the weight of each image can be determined so that the deviation of the meaning vector corresponding to the synthesized image from the original output meaning vector is not greater than the meaning vectors of surrounding terms. In addition, the peripheral objects may be a part of each component according to the number of dimensions of a multidimensional semantic vector, for example, a predetermined number of components in descending order of absolute value among the components of the output semantic vector, or a predetermined number of components whose absolute value is a predetermined value. Those with a difference from a larger component within a reference range may be selected in a limited manner. Alternatively, the object on which the calculation is performed may be used under predetermined conditions, such as when the semantic vectors are close.

Further, as shown in FIG. 4(a), there are cases where the distance between the terms that are acquired in a limited manner is relatively short. Here, for example, the meaning vector R21 of a suspension bridge and the meaning vector R22 of an aerial passage are located close to each other. In this way, if the selected terms are biased toward similar concepts, the generated image will not change significantly after all, so the meaning of generating the image will be diminished. Therefore, if the distance between the surrounding terms is close, only one representative term may be selected. As shown in FIG. 4(b), when the meaning vectors R25 to R27 of surrounding terms are scattered and the values of some components are scattered in positive and negative directions, an image that does not include both components may be created. or an image in which both components coexist.

For example, in the case of Fig. 4(a), image generation (acquisition) of scuba diving in seawater shown in Fig. 5(a) and a suspension bridge (aerial passage) with a good view shown in Fig. 5(b) are performed as peripheral images. ), and in addition to this, as shown in FIG. 5(c), as an image indicating the position of the meaning vector M21, an intermediate image such as a moving object or living thing may be used as an image showing the position of the meaning vector M21, with a sky blue or light blue space as the background. A typical image may be generated. The user may generate ideas by expanding the image as appropriate based on such images.

FIG. 6 is a diagram illustrating an example of an image output pattern.
The image to be output does not have to be only the image corresponding to the output semantic vector in FIG. 5(c). Each of the peripheral terms may be associated with the output meaning vector and output together.

As shown in FIG. 6(a), when there is no term indicating another meaning within the dashed line A3 indicating a predetermined distance from the coordinate M3 indicated by the output meaning vector, and there are six terms R31 to R36 in the vicinity, Image P30 is generated. In this case, as shown in FIG. 6(b), the generated image P30 is placed at the center, and the images P31, P33, P35, and P36 corresponding to the remaining four terms (indicating the contents) are placed at the above distance. A correlation diagram displayed at a position separated from the center by a corresponding distance (corresponding to the distance) may be output as an output image. Furthermore, the positions of the centers of images P31, P33, P35, and P36 with respect to image P30 may be arranged such that they are separated by an angular difference corresponding to the distance between the meaning vectors of terms R31, R33, R35, and R36. Here, the angular difference may be expressed as an angle according to cosine similarity, for example. Note that in a two-dimensional projection drawing, it is not possible to accurately show the relationship between all angular differences in a multidimensional space, so the angular intervals may be schematic. It may be possible to mainly reflect the distance difference in . Alternatively, a plurality of peripheral terms may be specified such that other peripheral terms that are commonly the shortest distance from a plurality of peripheral terms do not occur in the first place. Images P31, P33, P35, and P36 may be newly generated like image P30, or if they are stored in advance in association with terms, the stored images may be used as they are. .

In this way, by displaying the original images before compositing in an appropriate distance relationship, even in cases where it is difficult to grasp the concept from the generated image alone, the user can be guided more appropriately to the new image. We can guide you to the entrance.

FIG. 7 is a flowchart showing a control procedure by the control unit 11 of the idea support control process executed by the server device 10 of this embodiment. This idea support control process is started, for example, by specifying input data and acquiring an execution request from the terminal device 30.

When the ideation support control process is started, the control unit 11 (CPU) obtains input data (here, two pieces) (step S101). The control unit 11 performs text processing on each input data and extracts independent words. The control unit 11 refers to the terminology data 211 to remove unregistered terms from among the extracted terms, restores them to their basic form as necessary, or removes prefixes and suffixes to create objects. The word to be used is set (step S102).

The control unit 11 refers to the terminology data 211 and acquires the meaning vector of each object (step S103; operation as an acquisition means, acquisition step). The control unit 11 sets a combination of objects for each input data (step S104). The combination may be a round robin. The control unit 11 calculates an output semantic vector by performing a predetermined calculation using the semantic vector for each set combination (step S105; operation as a calculation means, calculation step). The operation does not have to be simple addition, and may be various arithmetic operations or vector operations (eg, cross product). In addition, inner products, trigonometric functions, and the like may be used as coefficients of semantic vectors to be calculated. Further, the control unit 11 may obtain a plurality of types of calculation results for one combination.

The control unit 11 determines whether there is a term having a meaning vector within a predetermined distance from each of the obtained output meaning vectors (step S106). If it is determined that there is one (“YES” in step S106), the closest term is determined as the corresponding term and output as is (step S121). As described above, the control unit 11 may output simply defined corresponding terms in text. Then, the control unit 11 ends the ideation support control process.

If it is determined that there is no term having a meaning vector within the predetermined distance (“NO” in step S106), the control unit 11 extracts a predetermined number of surrounding terms (step S107). As mentioned above, extraction may be performed in order of decreasing distance, but if there are terms whose distance is closer than the standard between neighboring terms, it is possible to omit some or all of the terms that appear after the second term. good. Further, if there is no predetermined number of terms within a predetermined distance range, the number of terms less than the predetermined number may be extracted.

The control unit 11 generates or acquires an image related to each term (step S108). The control unit 11 generates an image indicating the content of the output semantic vector (step S109; operation as a generating means including step S106, generation step). As described above, GAN or the like may be used to generate the image.

The control unit 11 determines the positional relationship on the display based on the difference in distance and direction between the output meaning vector and each meaning vector of the peripheral term (step S110). The control unit 11 generates an output image in which the generated image and the image corresponding to the peripheral term are displayed in the above-mentioned positional relationship, and outputs it to the terminal device 30 (step S111; operation as an output means). Then, the control unit 11 ends the ideation support control process.

[Modified example]
FIG. 8 is a diagram illustrating a modified example of the image output pattern.
In the above, the generated image (output object) corresponding to the output semantic vector and the image (surrounding object) corresponding to the peripheral term are displayed in the output image, but the generated image and images related to other peripheral terms are displayed. There are often large gaps between the two. Therefore, by changing the weighting described above, data of the intermediate image P45 (intermediate object) located between the output meaning vector (output object) and the meaning vector of the peripheral term (surrounding object) is generated, and the data is also displayed and output. It's okay. In this case, the intermediate image P45 becomes closer to a specific image that is closer to the content of the peripheral term, making it easier for the user to judge the degree of change in the image.

Furthermore, in the output images of this modified example, an additional display is made so that it can be identified whether or not text data corresponding to each image is set. For example, in the images P41 to P44 corresponding to peripheral terms, the peripheral terms are written in parentheses (text display), indicating that the images directly correspond to the terms. On the other hand, for the generated image P40 and the intermediate image P45, a specific display is performed that makes it possible to identify that there is no directly corresponding term. Here, the frames of these images are displayed three-dimensionally with shadows to indicate that they are generated images. In addition, when changing the frame, change the frame's line type (solid line, broken line, dotted line, etc.), line thickness, line color, line shape (straight line, wavy line, broken line, various decorative lines, etc.), etc. It may be different from the text data set. Alternatively, instead of a frame line, additional indicators such as dots or underlines may be added to the side, top, or bottom of the generated image or intermediate image, and instead of parentheses including the above text display, brackets may be used. may be shown, or it may be shown in parentheses that there is no text data. In addition, various characteristic marks may be attached.

As described above, the server device 10, which is the idea support device of this embodiment, includes the control unit 11. The control unit 11, as an acquisition means, acquires, for each of a plurality of input objects, a semantic vector indicating a direction according to the meaning of the object, and as a calculation means, performs a predetermined operation between the acquired plurality of meaning vectors, and generates a generation means. If a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained, a non-text output object is generated according to the output semantic vector.
In this way, if the output meaning vector obtained by the calculation does not almost exactly indicate a certain term, the server device 10 outputs a non-text output object that corresponds to the content indicated by the output meaning vector or content sufficiently close to this. generate. In this way, by not being tied to linguistic expressions, it is possible to further expand the range of expressions, present outputs that better fit the calculation results to the user, and encourage image arousal. Therefore, the server device 10 can support more diverse ideas for those who obtain the output results, such as users.

Further, the reference object is indicated by characters. By easily associating and retaining terms and meaning vectors, the server device 10 can easily and appropriately refer to the terms stored in the terminology data 211.

Further, the control unit 11 functions as an acquisition means to acquire input objects from a plurality of input data. By performing arithmetic processing by combining the contents of a plurality of independent or defined input data according to concepts, the server device 10 can more flexibly expand the range of output objects and support a wide range of ideas from users. I can do it.

Further, the input object is an independent word within the input data including a sentence. That is, the server device 10 performs natural language processing that extracts independent words from the input text and determines the input object. In other words, for input data that can be entered with little effort on the part of the user, an appropriate input object can be easily determined through low-load processing, so processing can be easily performed even on business-use PCs that are not particularly capable. It is.

Further, the output object is image data or audio data. In contrast to input, output is expressed in images and sounds that are rich in information, regardless of language, allowing the user who obtains the output results to use their imagination widely and stimulate flexible ideas.

Further, the control unit 11, as an output means, identifies a plurality of peripheral objects whose contents are each indicated by a semantic vector whose distance from the output semantic vector satisfies a predetermined condition, and identifies the peripheral objects as the semantic vector of the peripheral object. It is output together with the output object in correspondence with the distance from the output semantic vector.
The non-text data generated according to the output semantic vector tends to be abstract images, and the content is often too complicated for the user to understand on its own, so a display that allows the relationship with surrounding objects to be visually recognized is necessary. By outputting , it is possible to appropriately support the user's idea recall.

Further, the output object is image data, and the control unit 11 generates, as an output means, a correlation diagram in which peripheral objects are arranged around the generated image related to the output object, separated from the generated image by distances corresponding to the respective distances. Output.
When the output object is image data, the server device 10 can make it easier for the user to visualize various ideas by visualizing the relationship with the above-mentioned peripheral objects as a correlation diagram.

Further, the control unit 11 outputs, as an output means, a correlation diagram in which peripheral objects are arranged around the generated image related to the output object with an angular difference according to the distance of the semantic vector between the plurality of peripheral objects. In this way, by visualizing the difference in direction of surrounding objects with respect to the generated image, it is possible for the user to easily enlarge the image while considering the cut and direction.

Further, the control unit 11, as an output means, displays peripheral images related to peripheral objects in a correlation diagram and outputs the resultant image. As mentioned above, when the generated image is complex and difficult to understand, by displaying images of surrounding objects that are assumed to be somewhat close to the generated image, it is easier for the user to understand the details of the generated image. This can make it easier to find clues to expand your ideas.

Further, the control unit 11, as an output means, generates an intermediate object whose meaning vector is located between the peripheral object and the output object, displays an intermediate image related to the intermediate object in a correlation diagram, and outputs the intermediate image. It may be difficult to grasp the image from just the surrounding image as shown above, so by displaying an intermediate image between the surrounding image and the generated image, it is helpful for users who are not good at freely expanding their ideas from images. It is also possible to support ideas that are easier to understand.

Further, the control unit 11, as an output means, selectively draws a correlation diagram of peripheral objects whose size difference from the output semantic vector is within a reference range in some dimensions that are greater than or equal to a predetermined number of dimensions expressing the semantic vector. to be displayed. Even if there are a large number of similar surrounding objects, the display may become complicated, so in particular, a correlation diagram should be used for surrounding objects whose characteristic parts, that is, components with large absolute values, are somewhat similar to the output semantic vector. By displaying the image in a similar manner, it is possible to easily grasp the commonality of the characteristics of the generated image and the surrounding images, and accordingly, it is possible to more appropriately provide the user with the image attachment.

Further, as an output means, the control unit 11 performs an additional display that allows identification of images included in the correlation diagram with corresponding text data set and those not set. In other words, for images generated according to meaning vectors other than terms for which meaning vectors have been defined using Word2vec, a correlation diagram can be created to show that, and new concepts without linguistic expression can be created. It can be emphasized that this is a diagram showing . In addition, it is possible to easily identify the part that expands ideas and the part that is the basis thereof, without being bound by language, and this can help the user determine the direction in which to expand ideas.

Further, the additional display includes differences in frame lines depending on whether or not settings are made for the image. Therefore, it is easy for the user to easily identify the difference when viewing the images.

Further, the above-mentioned differences include differences in any one of line type, line thickness, line color, line shape, shadow, and three-dimensional display. Since discrimination can be easily performed based on such a simple difference in frame lines, useful information can be efficiently provided to the user without the need for generating a correlation diagram.

Further, the additional display includes a text display related to the set text data. That is, if a term corresponding to a peripheral image is set, a text display of the term can be placed alongside the peripheral image. This allows the user to immediately grasp the characteristics shown by the surrounding images. In addition, users can easily expand their ideas by using text and images together.

Further, the additional display includes the presence or absence of an additional mark or a difference depending on the presence or absence of text data for the image. Instead of or in addition to the above-mentioned display of the image frame, it may be possible to determine whether or not text data is present by attaching an additional mark to the image. Since the additional display is placed further away from the image than in the frame adjacent to the image, it is easier for the user to see it separately from the image itself, so it is less likely to cause bias when expanding ideas. Although the occurrence of bias is not necessarily a bad thing when it comes to expanding and expanding ideas, generated images can also be presented appropriately to users who want to separate them.

Further, the additional mark includes parentheses or a dot. By attaching such a simple additional mark, it is possible to indicate that the image (including intermediate images) has been generated clearly with low load and without taking up much space.

Further, the predetermined operation includes at least one of four arithmetic operations, an inner product, and an outer product. That is, it is sufficient to perform appropriate vector calculations, and there is no particular limitation in order to expand the idea.

Furthermore, an image corresponding to the computed meaning vector may be generated by an ideation support system having the above-mentioned configurations in a distributed manner. Semantic vector computation and image generation do not necessarily require large computer resources, but may be split into separate computers for each process. Alternatively, when the server device acquires a large number of instructions from a plurality of terminal devices 30, the processing may be shared among a plurality of processing devices to reduce and equalize the load.

In addition, by installing and executing the program 121 on a computer to execute each of the above processes through software processing, the user's ideas can be easily supported using general-purpose processing on a normal computer without the need for special hardware. be able to.

Note that the present invention is not limited to the embodiments described above, and various changes are possible.
For example, in the embodiment described above, the server device 10 independently performs the above processes as an idea support device, but a plurality of electronic devices may perform distributed processing to perform the above processes as an idea support system.

Further, although the arithmetic processing is normally performed between the semantic vectors of two objects, the arithmetic processing may be performed using the semantic vectors of three or more objects.

Further, the reference range, distance, etc. related to the selection of a term corresponding to the output meaning vector and the extraction of peripheral terms do not need to be fixed conditions. It may be adjusted as appropriate depending on the excess or deficiency of the relevant peripheral terms. Furthermore, different criteria may be set in advance for each server device 10 (idea support system 1) depending on the number of dimensions of the meaning vector.

In addition, in the above embodiment, it is assumed that specific arithmetic processing is predetermined, but the predetermined arithmetic processing is not just one, and multiple outputs can be obtained by multiple types of arithmetic processing. It's okay to be hit. Alternatively, the type of arithmetic processing may be selected randomly or according to predetermined conditions.

Furthermore, in the above embodiment, some of the peripheral terms that are close to each other may be omitted, but the peripheral terms may be biased as a result, such as simply in order of proximity. .

Furthermore, in the above embodiment, the database device 20, which is an external device, is accessed to obtain the semantic vector corresponding to the term (object) and the term (object) corresponding to the semantic vector. It may be stored in the storage unit 12 of the server device 10.

Furthermore, it may be arbitrarily determined which position between the output semantic vector and the semantic vectors related to peripheral objects is indicated by the semantic vector corresponding to the intermediate image. For example, the user looks at the output result and outputs a command to regenerate the correlation diagram by moving the image closer to either side, and in accordance with the command, the server device 10 regenerates and outputs the correlation diagram with the intermediate image changed. It's okay.

Furthermore, in the above embodiment, the input data is described as being a sentence, but it may be something like a memo or a heading that does not completely form a sentence, or on the contrary, it may be a sentence that contains multiple sentences. There may be. Alternatively, non-text data may be input and converted into text using image or voice recognition technology. Alternatively, only a plurality of independent word keywords may be input from the beginning and used as they are.

Further, in the above embodiment, the semantic vector is calculated by combining input objects included in a plurality of input data, but the present invention is not limited to this. Multiple input objects may be obtained from a single input data. In this case, if there is no need to stick to a specific concept or if you want to remove the output from the purpose of the input data, you can use an operation such as cross product to obtain an output semantic vector in a completely different direction from the semantic vector of the input object. One may be selectively set.

In addition, in the above embodiment, the case where images are output has been mainly explained, but even when outputting audio, a link for outputting audio is placed at each position in the above correlation diagram, and by selecting the link, Image data that can output audio may also be output.

Further, in the above embodiment, peripheral images are displayed in an appropriate positional relationship in the correlation diagram, but the positional relationship does not need to involve a difference in distance. Alternatively, the difference in direction may not be shown, and a list of surrounding images may be displayed in an array together with numerical values indicating the distance or without indicating the distance. Alternatively, the name of the peripheral object may only be displayed in text without showing the peripheral image.

Further, in the above embodiment, the generated image may be made distinguishable by different frame lines or additional marks, but other methods may be used, such as markings, diagonal lines, hatches, etc., within the image. Alternatively, the shape of the image itself may be made different. Furthermore, the size of the generated image may be different from the size of other images.

Furthermore, in the above explanation, the storage unit 12 made of a non-volatile memory including an HDD was used as an example of a computer-readable medium for the program 121 related to the processing operation of the control unit 11 according to the present invention. Not limited. As other computer-readable media, it is possible to apply portable recording media such as CD-ROMs and DVD discs. Further, a carrier wave is also applied to the present invention as a medium for providing data of the program according to the present invention via a communication line.
In addition, the specific configuration, contents and procedures of processing operations, etc. shown in the above embodiments can be changed as appropriate without departing from the spirit of the present invention. The scope of the present invention includes the scope of the invention described in the claims and its equivalents.

1 Ideation support system 10 Server device 11 Control unit 12 Storage unit 121 Program 1211 Learned model 1212 Image generation processing unit 13 Communication unit 20 Database device 21 Storage unit 211 Terminology data 30 Terminal device 31 Control unit 32 Storage unit 33 Display unit 34 Operation Reception Department 35 Communication Department

Claims (20)

acquisition means for acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation means for performing a predetermined calculation on the plurality of acquired meaning vectors;
generation means for generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
An idea support device characterized by comprising: 2. The ideation support device according to claim 1, wherein the reference object is indicated by a character. 3. The ideation support device according to claim 1, wherein the acquisition means acquires each of the input objects from a plurality of input data. 4. The ideation support device according to claim 1, wherein the input object is an independent word in input data including a sentence. 5. The ideation support device according to claim 1, wherein the output object is image data or audio data. A plurality of peripheral objects each having a content indicated by a semantic vector whose distance from the output semantic vector satisfies a predetermined condition is identified, and the peripheral object is determined by the distance between the semantic vector of the peripheral object and the output semantic vector. The ideation support device according to any one of claims 1 to 5, further comprising an output means that outputs the output object in association with the output object. The output object is image data,
7. The output means outputs a correlation diagram in which each of the peripheral objects is arranged around the generated image related to the output object, separated from the generated image by an interval corresponding to the distance. idea support device. The output object is image data,
The output means outputs a correlation diagram in which the peripheral objects are arranged around the generated image related to the output object, with an angular difference depending on the distance of the meaning vector between the plurality of peripheral objects. The idea support device according to claim 6 or 7. 9. The ideation support device according to claim 7, wherein the output means outputs a peripheral image related to the peripheral object by displaying it on the correlation diagram. The output means generates an intermediate object whose meaning vector is located between the peripheral object and the output object, and outputs an intermediate image related to the intermediate object displayed on the correlation diagram. The idea support device according to any one of items 7 to 9. The output means selectively displays, on the correlation diagram, the peripheral objects whose size difference from the output meaning vector is within a reference range in some dimensions of a predetermined number or more out of the number of dimensions expressing the meaning vector. The ideation support device according to any one of claims 7 to 10, characterized in that: 12. The output device according to claim 7, wherein the output means performs an additional display that allows identification of images included in the correlation diagram with corresponding text data set and those without the corresponding text data set. The idea support device according to any one of the items. 13. The idea support device according to claim 12, wherein the additional display includes a difference in frame line depending on whether or not the setting is applied to the image. 14. The ideation support device according to claim 13, wherein the difference includes a difference in any one of line type, line thickness, line color, line shape, shadow, and three-dimensional display. 15. The ideation support device according to claim 12, wherein the additional display includes a text display related to the set text data. 13. The idea support device according to claim 12, wherein the additional display includes the presence or absence or difference of an additional mark depending on the presence or absence of the text data for the image. 17. The ideation support device according to claim 16, wherein the additional mark includes a parenthesis or a dot. 17. The ideation support device according to claim 1, wherein the predetermined operation includes at least one of four arithmetic operations, an inner product, and an outer product. acquisition means for acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation means for performing a predetermined calculation on the plurality of acquired meaning vectors;
generation means for generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
An idea support system characterized by: to the computer,
an acquisition step of acquiring a meaning vector indicating a direction according to the meaning of the object for each of the plurality of input objects;
a calculation step of performing a predetermined calculation on the plurality of acquired meaning vectors;
a generation step of generating a non-text output object according to the output semantic vector when a reference object represented by a semantic vector within a range that satisfies a predetermined distance condition from the output semantic vector obtained by the calculation is not obtained; and,
A program characterized by executing.

Images