Data visualization system, data visualization method, and data visualization program

Abstract

To provide a data visualization system that enables even a person without expert knowledge of an analysis target field to sufficiently understand the meaning of analysis results, and allows a person with expert knowledge to perform more advanced analysis and reduce analytical effort, by integrating and utilizing technologies of an integrated map and generative AI such as large language models.SOLUTION: A data visualization system 1 includes: an integrated map unit 2 configured to generate an integrated map by mapping, to a low dimension, models estimated based on a plurality of different multidimensional viewpoints of nonlinear data of the plurality of different viewpoints; a generative AI unit 3 configured to generate at least one of a summary sentence and an image picture on from the nonlinear data of the plurality of different viewpoints using generative AI such as a large language model; and a display unit 6 configured to, when a user designates an arbitrary location on the integrated map, display data of the integrated map corresponding to the designated location and at least one of the summary sentence and the image picture generated by the generative AI unit.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] This invention relates to a data visualization system that combines an integrated map with generative AI such as a large language model (LLM), and is applicable to knowledge discovery from big data. Specifically, it is suitable for systems that visualize and analyze data in marketing analysis, policy survey analysis, etc. [Background technology]

[0002] Surveys are often used in marketing analysis before a new product is launched. For example, prototypes are provided to contract monitors, who are asked to indicate which prototype they like best, the age group most likely to like it, and the region in which it is likely to sell. The survey results are then analyzed to make decisions such as which prototype should be commercialized and which target demographic to advertise to. However, analyzing thousands of pieces of data in marketing analysis can take an enormous amount of time and effort. Even when such work is outsourced to a specialized analysis company, the analysis often does not meet the user's expectations.

[0003] Patent Document 1, by the inventor of the present applicant, discloses an integrated map creation device that, for each piece of information viewed from a plurality of different viewpoints, estimates a model of the information as viewed from another viewpoint, and creates and displays the estimated model as an integrated map. This integrated map allows analysis while interactively manipulating a display that associates information from various viewpoints with information viewed from other viewpoints in accordance with the user's intentions, and can be used in the above-mentioned marketing analysis, policy questionnaire analysis, etc.

[0004] Furthermore, in recent years, semiconductor and artificial intelligence (AI) technologies have rapidly developed, and generative AI technologies such as large-scale language models have recently become a hot topic and are being put to practical use. For example, in chat systems that use large-scale language models, when a person asks a question, it is becoming possible to advance the conversation by quickly presenting relevant information and answers from huge amounts of big data. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Patent No. 5750804 Summary of the Invention [Problem to be solved by the invention]

[0006] The integrated map described in Patent Document 1, developed by the same inventor as the present applicant, allows users to visualize and analyze big data from multiple perspectives while viewing the screen, enabling analysis in a short time and in line with the user's intentions. However, even when using the integrated map, expert knowledge in the field of analysis and analytical effort are still required to fully grasp the meaning of individual numerical data and large amounts of text data. The present invention aims to solve this problem by combining and utilizing the above-mentioned integrated map with generative AI technologies such as large-scale language models, allowing even those without expert knowledge of the field of analysis to fully grasp the meaning of the analysis results, while enabling those with expert knowledge to perform more advanced analyses while reducing the analytical effort. [Means for solving the problem]

[0007] A data visualization system that combines an integrated map and generative AI such as a large-scale language model, comprising: an integrated map unit that generates an integrated map by mapping nonlinear data from a plurality of different perspectives onto a lower dimension using a model estimated based on a plurality of different multidimensional perspectives; a generative AI unit that generates at least one of summary text or image images from the nonlinear data from a plurality of different perspectives using generative AI such as a large-scale language model; a display unit that displays the integrated map and at least one of the summary text or image images; and an instruction unit that receives user instructions; when a user indicates any location on the integrated map using the instruction unit, the data on the integrated map corresponding to that location and at least one of the summary text or image images generated by the generative AI unit are displayed.

[0008] Furthermore, the data visualization system is a fusion of an integrated map and generative AI such as a large-scale language model, and includes an integrated map unit that generates an integrated map by mapping nonlinear data from a plurality of different perspectives onto a lower dimension using a model estimated based on a plurality of different multidimensional perspectives; a generative AI unit that generates at least one of summary text or images from the nonlinear data from a plurality of different perspectives using generative AI such as a large-scale language model; a display unit that displays the integrated map and at least one of the summary text or images; and an instruction unit into which instructions from a user are input, and when a user inputs words or text data from the instruction unit, the display unit displays at least one of the summary text or images generated by the generative AI unit, and the system also links the corresponding locations of the summary text or images displayed on the integrated map so that they are visually easy to understand. [Effects of the Invention]

[0009] According to the present invention, by combining and utilizing generative AI technologies such as integrated maps and large-scale language models, it is possible to achieve a synergistic effect between the functions of each. As a result, even those without specialized knowledge of the subject area of ​​analysis can fully understand the meaning of the analysis results, while those with specialized knowledge can perform more advanced analyses while reducing the effort required for analysis. [Brief explanation of the drawings]

[0010] [Figure 1] Configuration diagram of data visualization system 1 [Figure 2] A table of evaluation information stored in the aggregate data storage unit 4 [Figure 3] Figure showing an example of a display showing information from an integrated map [Figure 4] A diagram showing the changes in the maps of viewpoints 1 and 3 when a certain element of viewpoint 2 is identified in Fig. 3. [Figure 5] A diagram showing the changes in the maps of viewpoints 2 and 3 when a certain element of viewpoint 1 is identified in FIG. [Figure 6] A diagram for explaining the flow of operations during the learning phase [Figure 7] Flowchart explaining learning of numerical data (Fig. 7(a)) and learning of text data (Fig. 7(b)) [Figure 8] A diagram illustrating the operational flow of Use Case 1 [Figure 9] The first half of the figure explains Example 1 with a specific example. [Figure 10] The first half of the figure explains Example 1 with a specific example. [Figure 11] A diagram illustrating the operational flow of Use Case 2 [Figure 12] A diagram to explain use case 2 using a specific example DETAILED DESCRIPTION OF THE INVENTION

[0011] Hereinafter, embodiments of the present invention will be described, but the present invention is not limited to the following embodiments.

[0012] (Embodiment) FIG. 1 is a configuration diagram of a data visualization system 1 according to an embodiment of the present invention.

[0013] An overview of the data visualization system 1 will be described. The data visualization system 1 includes an integrated map unit 2 that displays an integrated map, a generation AI unit 3, and an instruction unit 5 that instructs the operation of the integrated map unit 2 and the generation AI unit 3 based on input information from a user. The data visualization system 1 also includes an aggregate data storage unit 4 that stores multidimensional array nonlinear data, and stores information such as questionnaires collected from contract monitors for use in analysis by the data visualization system 1. The integrated map unit 2 includes an integrated map generation unit 20 that generates an integrated map, and an integrated map control unit 25 that controls the integrated map generation unit 20 and displays the integrated map on the display unit 6. The generation AI unit 3 includes a generative AI model unit 30 that includes a large-scale language model 31 and an image generation model 32, and a generation AI control unit 33 that controls the operation of the generative AI model unit 30 based on input information from a user. For example, when a user specifies data for an arbitrary area via the instruction unit 5, the generation AI control unit 33 controls the generative AI model unit 30 to generate a summary sentence or an image for that area based on the data for the specified area. The generation AI model unit 30 is connected to the network information 7, and can access, for example, a large-scale database or various information to update the large-scale language model 31 and image generation model 32 with the latest information.

[0014] The data visualization system 1 can also be implemented by an electronic device such as a computer configured with a memory and a processor. The integrated map unit 2 and the generation AI unit 3 can also be implemented by software that operates in cooperation with a memory and a processor.

[0015] Next, the configuration and operation of the integrated map unit 2, which is composed of the integrated map generating unit 20 and the integrated map control unit 25, will be described.

[0016] The integrated map generator 20 not only creates a map by mapping multidimensional nonlinear data to a lower dimension, but also creates a single map by mapping a model estimated based on multiple different perspectives and taking into account the relationships between those perspectives into a lower dimension. That is, instead of creating a map based on different perspectives (e.g., customers, furniture ratings, and furniture impressions) using only information classified from those individual perspectives, the integrated map generator 20 creates a single map in which the information classified from each perspective is consistently and appropriately associated. Therefore, when viewing a single created map from a certain perspective, it is possible to refer to information that takes into account the relationships with other perspectives. Even with large-scale data in which multiple elements (e.g., customers "Person 1, Person 2,...,Person n") are intricately intertwined, it is possible to refer to information that takes into account the relationships between each element and the elements.

[0017] In this embodiment, the integrated map generating unit 20 creates an integrated map by repeating learning using the EM algorithm. Note that the EM algorithm is one method, and an integrated map may also be created by repeating learning using a gradient method with an objective function.

[0018] The integrated map generation unit 20 includes an initialization processing unit 21 that temporarily sets the placement position of elements of individual maps using the placement configuration of each element for each viewpoint obtained by learning the data in the aggregated data storage unit 4; a placement position calculation unit 23 that calculates the position of the individual map in the integrated map; an individual map information generation unit 22 that reads the placement position of the individual map and the data stored in the aggregated data storage unit 4 to generate an individual map and outputs the generated individual map to the placement position calculation unit 23; and an integrated map information generation unit 24 that reads the placement configuration for each viewpoint calculated by the placement position calculation unit 23 and the data stored in the aggregated data storage unit 4 to generate an integrated map, and an integrated map control unit 25 displays the integrated map on a display unit 6 in accordance with an instruction unit 5 to which instructions from a user are input.

[0019] The aggregated data storage unit 4 is a relational database, and stores, for example, information on the results of a furniture questionnaire given by customers. The aggregated data storage unit 4 stores evaluation information for each piece of furniture for each customer in a table format in association with the results. A specific example is shown in Figure 2.

[0020] As shown in Figure 2, the table contains customer information such as customer attributes (person, gender, age, and region). It also contains rating information such as furniture rating (1-5: rating "1" indicates a high rating, and the higher the number, the lower the rating) and information about the impression of the furniture (text information such as bright, cold, or solid). It also contains customer comments (free-form text information such as thoughts and daily thoughts).

[0021] For the sake of convenience, Figure 2 shows only evaluation data for five customers and three pieces of furniture, but the data volume is assumed to be large, including thousands of customers and hundreds of pieces of furniture.

[0022] The placement position calculation unit 23 calculates the position of the individual map in the integrated map. In the initial state, the placement position of the elements for each viewpoint is set by the initialization processing unit 21, so no processing is performed here. When the process moves from the initial state to the second or subsequent processing, the integrated map information generation unit 24 generates an integrated map, and the individual map information generation unit 22 generates an individual map, so the integrated map and the individual map are input, and the position of the individual map in the integrated map is calculated. At this time, for example, when calculating the position of an individual map of furniture created for each customer (for example, a map of furniture for person 1), the customer's position is calculated by finding the position where the individual maps of the furniture are similar.

[0023] The individual map information generating unit 22 generates an individual map by reading the position of the individual map calculated by the placement position calculating unit 23 (in the initial state, the placement position set by the initialization processing unit 21) and the data stored in the aggregated data storage unit 4. At this time, an individual map of customers for each piece of furniture (for example, a map of customers for furniture 1) is generated based on the customer position obtained by the calculation in the above example and the customer data for each piece of furniture stored in the aggregated data storage unit 4 (data in the column direction in FIG. 2).

[0024] The integrated map information generation unit 24 generates an integrated map by reading the positions of the individual maps for each viewpoint calculated by the placement position calculation unit 23 (in the initial state, the placement positions set by the initialization processing unit 21) and all data stored in the aggregate data storage unit 4. That is, for elements identified based on all placement positions for each element at each viewpoint calculated by the placement position calculation unit 23 (for example, in the above example, the placement position identified by the placement position of customer 1, the placement position of the evaluation of furniture 1, and the placement position of the impression of furniture 1 (placement position of customer 1 × placement position of the evaluation of furniture 1 × placement position of the impression of furniture 1)), actual data is input, and a single integrated map is generated that has learned all elements at all viewpoints.

[0025] The integrated map control unit 25 controls the display of the completed integrated map information on the display unit 6 in accordance with instructions from the instruction unit 5, which are instructions from the user. The display unit 6 is typically a display, but it can also output to a printer. The integrated map completed by the above processes has individual maps for each viewpoint, which are two-dimensional. Therefore, if there are two or more viewpoints, the integrated map becomes a multidimensional integrated map (four or more dimensions), which cannot be displayed as useful information. Therefore, it is necessary to represent the multidimensional integrated map in two dimensions. In addition, by fixing arbitrary parameters in accordance with instructions from the user, the necessary information can be output in an easy-to-read format. Furthermore, even if the number of viewpoints increases, the basic algorithm remains the same as that described above, and it is possible to create individual maps with more than two viewpoints by taking into account the placement positions of all other viewpoints when creating them.

[0026] Below is an example of the integrated map information displayed.

[0027] Figure 3 simultaneously displays the integrated map as viewed from viewpoint 1 (customer map), as viewed from viewpoint 2 (furniture evaluation map), and as viewed from viewpoint 3 (furniture impression map). Each map (each map viewed from a different viewpoint is considered a single-viewpoint map) is information from the integrated map modeled with each viewpoint in mind, so a single-viewpoint map from one viewpoint is a map that takes into account the other viewpoints. Furthermore, in each single-viewpoint map, elements with high similarity are placed close to each other. For example, in the customer map, a "man in his 60s" is more similar to a "woman in his 50s" than to a "man in his 40s," and in the furniture evaluation map, "furniture 1" is more similar to "furniture 2" than to "furniture 3."

[0028] Figure 4 shows how the maps for viewpoint 1 and viewpoint 3 change when a certain element in viewpoint 2 is identified in Figure 3. In this example, when the unit for "furniture 5" is clicked and identified in viewpoint 2, the maps for viewpoint 1 and viewpoint 3 corresponding to "furniture 5" also change.

[0029] The maps for viewpoint 1 and viewpoint 3 show correlations with "Furniture 5" using a shading scheme, with darker shades indicating stronger correlations. In this example, looking at the map for viewpoint 1 (customers), we can see that the correlation with "women in their 20s" is highest, followed by "men in their 20s." Looking at the map for viewpoint 3 (impressions of furniture), we can see that the correlations with "cold" and "bright" are highest, in that order.

[0030] Next, Figure 5 shows how the maps for Viewpoint 2 and Viewpoint 3 change when a certain element in Viewpoint 1 is identified in Figure 3. In this example, when "Man in his 40s" is clicked and identified in Viewpoint 1, the maps for Viewpoint 2 and Viewpoint 3 corresponding to "Man in his 40s" also change. In Viewpoint 2 and Viewpoint 3, the correlation with "Man in his 40s" is shown by shade, with a darker display indicating a stronger correlation. In this example, looking at the map for Viewpoint 2 (furniture evaluation), it can be seen that the correlation with "Furniture 2" and "Furniture 6" is highest, while looking at the map for Viewpoint 3 (furniture impression) it can be seen that there is a high correlation with "solid," "classic," and partially with "warm."

[0031] A unit corresponds to a grid, which is the smallest unit of information, but an area spanning multiple grids may also be a unit.

[0032] Although the integrated map in this embodiment is realized using the EM algorithm, other techniques such as SOM (Self-Organization Maps), GTM (Generative Topographic Map), topology-preserving mapping, manifold learning, or nonlinear subspace methods can also be used.

[0033] For example, when a user scans a single-viewpoint map using a mouse or touch panel, the distribution display of other maps may change in an animated manner. Also, even if the user does not scan the single-viewpoint map, adjacent elements may be selected at predetermined time intervals (information on adjacent elements may be read sequentially), and in this case, maps from other viewpoints may be animated and displayed at predetermined time intervals in accordance with the selected element.

[0034] Next, we will explain the configuration and operation of the data visualization system 1 when it is used in combination with an integrated map and generative AI such as a large-scale language model.

[0035] In Figure 1, the generative AI model unit 30 includes a large-scale language model 31 and an image generation model 32. The large-scale language model 31 is trained on related sentences and can summarize specified sentences using the trained model. The image generation model 32 is trained on related sentences and images and can generate images from specified sentences using the trained model.

[0036] The generation AI control unit 33 instructs the generation AI model unit 30 to generate summary text and images, and displays the generated summary text and images on the display unit 6. Furthermore, the generation AI control unit 33 cooperates with the integrated map control unit 25 to receive information on the integrated map generated by the integrated map generation unit 20 and cause the generation AI model unit 30 to learn the information. The generation AI control unit 33 also converts the summarized text generated by the generation AI model unit 30 into a form that can be understood by the integrated map generation unit 20 and passes it to the integrated map control unit 25.

[0037] Below, we will explain the operation of this data visualization system 1 when it is used to integrate a generative AI such as a large-scale language model into an integrated map, dividing it into (1) the learning phase and (2) the usage phase.

[0038] (1) Learning Phase FIG. 6 is a diagram for schematically explaining the flow of operations in the learning phase.

[0039] FIG. 6 shows the flow of learning numerical data and text data by the integrated map unit 2 and the generation AI unit 3 working together. <Learning Numerical Data> Numerical data learning will be described with reference to FIG. 6 and FIG. 7(a).

[0040] (Operation of S701 and S702) As shown in Figure 6, the integrated map unit 2 and the generation AI unit 3 need to learn the same data. Specifically, they each learn the numerical data of the table shown in Figure 2 stored in the aggregate data storage unit 4 (S601). At this time, the generation AI unit 3 reads the numerical data in learning and expresses it in text format as pairs of items and numerical values ​​(S602). For example, (furniture ratings) are converted and expressed in text format such as "furniture 1·1" and "furniture 2·5". The large-scale language model 31 learns the data converted into text format.

[0041] (S703 operation) The integrated map generation unit 20 reads the numerical values ​​corresponding to each square of the integrated map and converts them into text format (S603). The method of converting numerical data into text is the same as that in S702; for example, (furniture ratings) is converted into text format such as "furniture 1·1" and "furniture 2·5." If the furniture rating map is a map with, for example, 50 columns and 20 rows, each square corresponds to a furniture rating value, and 50 x 20 pieces of numerical data are converted into text format. The converted text data is passed from the integrated map control unit 25 to the generation AI control unit 33, and the generation AI control unit 33 instructs the generation AI model unit 30 to learn the text data using the large-scale language model 31.

[0042] Through the above learning operation, the large-scale language model 31 becomes able to verbalize the visual meaning of each unit in the integrated map. <Text data learning> The learning of text data will be explained with reference to FIG. 6 and FIG. 7(b).

[0043] (S711 operation) The large-scale language model 31 of the generation AI model unit 30 learns the text data such as customer comments (S604). The generation AI control unit 33 instructs the generation AI model unit 30 to learn and summarize the text data stored in the aggregate data storage unit 4 (data written in the comment field in Figure 2).

[0044] (S712 operation) The generation AI control unit 33 vectorizes the summarized sentences (text data) using doc2vec. By vectorizing using doc2vec, correlations such as similarity or distance in meaning between multiple sentences can be quantified.

[0045] (S713 operation) The vectorized data is passed from the generation AI control unit 33 to the integrated map control unit 25, and the integrated map control unit 25 instructs the integrated map generation unit 20 to learn the vectorized data (S605).

[0046] (S714 operation) The integrated map generation unit 20 passes information about which unit contains the text corresponding to each customer comment ("Comment (what they think, what they keep in mind on a daily basis, etc.)" in Figure 2) to the generation AI control unit 33 via the integrated map control unit 25, and the generation AI control unit 33 conveys this information to the generation AI model unit 30 for learning.

[0047] In this embodiment, text data is vectorized using doc2vec, but other vectorization methods may also be used for vectorization.

[0048] (2) Use Phase Usage example 1: <Displaying customer comments, their summary text, and images for a specified location (unit, etc.) on the integrated map> 8 to 10, an example will be described in which the location (unit, etc.) clicked by the user on the integrated map is displayed on the screen as generated by the integrated map unit 2 and the generation AI unit 3.

[0049] Figure 8 is a diagram that illustrates the operational flow of Usage Example 1. When a user specifies the location of the content they want to know on the integrated map using a mouse or the like (S801), the information generated by the integrated map unit 2 is displayed (S802). Based on the information of the specified location on the integrated map sent from the integrated map unit 2 (S803), the generation AI unit 3 displays the generated information (S804). This will be explained using the specific example in Figure 9. Figure 9 shows the state in which the user has designated "men in their 40s" and the correlation for "men in their 40s" is displayed on the integrated map using the shading of viewpoints 2 and 3 for "men in their 40s." The state is then shown in which the user has further designated and clicked on a specified area or unit (square) on the (furniture impression) map on the integrated map using the instruction unit 5. In Figure 9, areas with a "bright" or "warm" impression are designated. Then, customer comments, summary text, and images are displayed as shown in Figure 10.

[0050] The corresponding customer comment, "They value old traditions. They see the products in the store before they buy...," is displayed on the integrated map screen (S1001), and the generation AI screen displays the summary sentence, "Conservative and solid," along with an image of furniture that evokes the image of that sentence (S1002).

[0051] In Figure 8, the customer comments from the integrated map unit 2, the summarized text from the generation AI unit 3, and the image images are displayed on separate displays, but they may be displayed together on a single display. Also, the user may be able to select whether to display the summarized text or the image images using the instruction unit 5. Furthermore, the text displayed on the display may be explained by voice.

[0052] In the above-described use case 1, users can see the raw customer comments as they are, as well as the summary text and images generated by the generation AI. Therefore, even those without specialist knowledge can easily understand the meaning of the analysis results from the summary text and images. Furthermore, those with specialist knowledge can perform advanced analysis, such as analyzing raw customer comments, while referring to the summary text and images.

[0053] Use case 2: When a question is asked to the data visualization system, the generative AI displays the answer and also visually displays the area or unit corresponding to the answer on the integrated map. The operation of Usage Example 2 will be explained using Figures 11 and 12. Figure 11 is a diagram that schematically explains the flow of operation. When a user asks this data visualization system a question (S1101), the summary text and image generated by the generation AI unit 3 are displayed (S1102). The text summarized by the generation AI unit 3 is vectorized using doc2vec and sent to the integrated map unit 2 (S1103), and the integrated map unit 2 visually displays the area or unit (grid) on the integrated map that corresponds to the answer to the question (S1104).

[0054] A specific example will be explained using Figure 12. The user issues a question via the instruction unit 5, such as, "I've renovated my apartment, so I want to buy a bed. I want a Scandinavian-style design. I also want to incorporate the latest trends" (S1201). The generation AI unit 3 generates a summary and an image of the bed. The generated summary and image are displayed on the screen (S1202). The summary, in this example, "A light wood-grain bed," is then vectorized using doc2vec. The vectorized data is sent to the integrated map unit 2 (S1203). The integrated map unit 2 compares the sent vectorized data with regions or units of data that have a certain level of correlation, and highlights the extracted regions or units to provide a visual understanding to the user (S1204). The user can click on a region or unit of interest to learn more detailed information about the specific furniture item (in this example, a bed).

[0055] In the above-described use case 2, the user not only obtains the generative AI's answer based on the question, but also visually identifies areas and units on the integrated map that are highly correlated with the question content, enabling analysis by combining the information from the generative AI with the information from the integrated map.

[0056] In this example, based on the user's question, the answer from the generation AI is displayed and the area or unit on the integrated map that corresponds to the answer is visually indicated. However, it is also possible to automatically input the comment data of the selected customer from the aggregate data storage unit 4, so that the answer from the generation AI is displayed and the area or unit on the integrated map that corresponds to the answer is visually indicated.

[0057] In Fig. 11, the display from the integrated map unit 2 and the summarized text and image from the generation AI 3 are displayed on separate displays, but they may be displayed together on a single display. The user may be able to select whether to display the summarized text or the image from the instruction unit 5. The text displayed on the display may also be explained by voice.

[0058] Use case 3: <Summarizing collected comments from surveys, etc. as training data for the integrated map> The sentences such as questionnaires collected from contract monitors and stored in the aggregate data storage unit 4, for example customer comments, are summarized using a large-scale language model.

[0059] Specifically, the functions of the large-scale language model 31 are used to summarize the sentences in the "Comments" column of the table shown in Figure 2 before learning in the integrated map unit 2. In response to an instruction from the user's instruction unit 5, the generation AI control unit 33 instructs the large-scale language model 31 to read the comments stored in the aggregated data storage unit 4 and create a summarized sentence. The summarized sentence is sent to the aggregated data storage unit 4.

[0060] The summarized text may be overwritten in the original "Comments" field, but it is preferable to create a separate "Summary" field and write the summary in that area.

[0061] Furthermore, when learning in the integrated map section 2, the user may be able to select, by instruction from the instruction section 5, whether to use the text in the original "comment" column or the text in the "summary" column.

[0062] In the above-described use example 3, by using the "summary" generated by the large-scale language model 31 as learning data in the integrated map unit 2, it is expected that unclear text data will be reduced and learning accuracy will be improved. This also has the effect of making the text display on the integrated map easier to understand.

[0063] As explained in detail above, the present invention combines generative AI technologies such as integrated maps and large-scale language models, enabling the synergistic effects of each function to be achieved. As a result, even those without specialized knowledge of the field being analyzed can fully grasp the meaning of the analysis results, while those with specialized knowledge can perform more advanced analyses while reducing the effort required for analysis. The effects of combining the integrated map (multi-perspective model) of this invention with generative AI technology such as large-scale language models, and the key points of how to achieve this, will be explained in more detail in (1) to (4) below. (1) It is difficult to verbalize a multidimensional array directly using a generative AI, but by using the results of numerical modeling using a multi-perspective model, it becomes possible to achieve things that could not be achieved using generative AI alone. (2) By vectorizing and visualizing the output of the generative AI, it becomes possible to visualize non-numerical data that could not be handled with multi-perspective models. (3) These technologies use multi-perspective models and generative AI in a complementary manner, enabling them to perform tasks that the other lacks, such as modeling, summarizing, and quantifying data. This creates a synergistic effect rather than simply combining them. (4) Multi-perspective models can integrate multiple data. By treating the output of generative AI as a type of data, it is also possible to integrate non-numerical data. This is not simply a combination, but is achieved by integrating the processes of converting non-numerical data to numerical data and integrating multiple data.

[0064] In the embodiment of the present invention, an example of application to furniture marketing analysis has been described, but the present invention can be applied in a variety of ways, and can be applied to marketing analysis, policy survey analysis, and other areas within the scope of the technical concept of the present invention. [Explanation of symbols]

[0065] 1. Data visualization system 2 Integrated Map Section 20 Integrated map generation unit 21 Initialization processing section 22 Individual map information generation unit 23 Placement position calculation section 24 Integrated map information generation unit 25 Integrated map control unit 3 Generation AI section 30 Generative AI Model Section 31 Large-scale language models 32 Image generation model 33 Generation AI control unit 4. Aggregated data storage unit 5 Instruction section 6 Display section 7. Internet information (big data)

Claims

1. A data visualization system that combines an integrated map with AI generation such as a large-scale language model, an integrated map unit that generates an integrated map by mapping a model estimated based on a plurality of different multidimensional viewpoints of nonlinear data from a plurality of different viewpoints to a lower dimension; a generation AI unit that generates at least one of a summary sentence or an image image from nonlinear data from a plurality of different viewpoints using a generation AI such as a large language model; a display unit that displays the integrated map and at least one of the summary text and the image; an instruction unit into which an instruction from a user is input, When a user specifies any location on the integrated map using the specification unit, the data visualization system displays the integrated map data corresponding to that location and at least one of the summary text or image generated by the generation AI unit.

2. A data visualization system that combines an integrated map with AI generation such as a large-scale language model, an integrated map unit that generates an integrated map by mapping a model estimated based on a plurality of different multidimensional viewpoints of nonlinear data from a plurality of different viewpoints to a lower dimension; a generation AI unit that generates at least one of a summary sentence or an image image from nonlinear data from a plurality of different viewpoints using a generation AI such as a large language model; a display unit that displays the integrated map and at least one of the summary text and the image; an instruction unit into which an instruction from a user is input, When a user inputs words or text data from the instruction unit, the display unit displays at least one of the summary sentence or image generated by the generation AI unit, and the data visualization system links the corresponding locations of the summary sentence or image displayed on the integrated map so that they can be visually understood.

3. The integrated map unit converts the numerical data of each unit of the created integrated map into text and sends it to the generation AI unit, The generation AI unit converts numerical data of nonlinear data from a plurality of different viewpoints into text and learns the text, and also learns by linking it to the numerical data of each unit converted into text sent from the integrated map unit. The data visualization system according to claim 1 or 2.

4. The generation AI unit vectorizes non-digitized data such as text data and sends the vectorized data to the integrated map unit, The integrated map unit learns the vectorized summary and communicates to the generation AI unit which unit on the integrated map the summary corresponds to, The summary is linked to the units on the integrated map and learned. The data visualization system according to claim 1 or 2.

5. At least a portion of the non-linear data of the plurality of different viewpoints uses text data summarized by the large-scale language model. The data visualization system according to claim 1 or 2.

6. A data visualization method that combines an integrated map with AI generation of large-scale language models, etc. an integrated map process for generating an integrated map by mapping nonlinear data from a plurality of different viewpoints onto a low-dimensional model estimated based on a plurality of different multidimensional viewpoints; A generation AI process for generating at least one of summary text and image images from non-linear data from multiple different perspectives using a generation AI such as a large language model; a display step of displaying the integrated map and at least one of the summary text and the image; an instruction step of inputting an instruction from a user, When a user specifies a location on the integrated map in the specifying step, the data visualization method displays the integrated map data corresponding to the location and at least one of the summary text or image generated by the generation AI unit.

7. A data visualization method that combines an integrated map with AI generation of large-scale language models, etc. an integrated map process for generating an integrated map by mapping nonlinear data from a plurality of different viewpoints onto a low-dimensional model estimated based on a plurality of different multidimensional viewpoints; A generation AI process for generating at least one of summary text and image images from non-linear data from multiple different perspectives using a generation AI such as a large language model; a display step of displaying the integrated map and at least one of the summary text and the image; an instruction step of inputting an instruction from a user, When a user inputs words or text data in the instruction step, at least one of the summary sentence or image generated by the generation AI unit is displayed in the display step, and the corresponding locations of the summary sentence or image displayed on the integrated map are linked and displayed in a visually understandable manner.

8. One or more programs that cause an electronic device such as a computer to execute the data visualization method according to claim 6 or 7.

Images