Analyzer, system and program

Abstract

To efficiently generate detailed output data.SOLUTION: An analyzer comprises: an intermediate data acquisition unit which inputs input data to a first learning model generated by machine learning using first learning data, and acquires intermediate data outputted by the first learning model; and an output data acquisition unit which inputs the intermediate data to a second learning model generated by the machine learning using second learning data specialized in a specific field, than the first learning data, and acquires output data outputted by the second learning model.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to an analysis device, a system, and a program. [Background technology]

[0002] Patent Document 1 describes a creation server that "implements a claim generation AI engine that generates and updates claims by having AI learn information from past patent applications (paragraph 0016)." Patent Document 2 describes a patent document creation support device (claim 1) that "accepts a selection from the extracted multiple sentence candidates based on a user's instruction via the acceptance unit, and adds the sentence candidate that has been selected to document information related to the invention." Patent Document 1: JP 2020-095653 A Patent Document 2: JP 2023-115837 A Summary of the Invention

[0003] A first aspect of the present invention provides an analysis device that generates output data for input data. The analysis device may include an intermediate data acquisition unit that inputs the input data to a first learning model generated by machine learning using first learning data, and acquires intermediate data output from the first learning model. Any of the analysis devices may include an output data acquisition unit that inputs the intermediate data to a second learning model generated by machine learning using second learning data that is more specialized in a specific field than the first learning data, and acquires the output data output from the second learning model.

[0004] In any of the above analysis devices, the second learning model may be a model generated by performing additional learning on the first learning model.

[0005] In any one of the above analysis devices, the first learning model may generate a plurality of the intermediate data. In any one of the above analysis devices, the output data acquisition unit may acquire the output data for at least one of the intermediate data.

[0006] Any of the above analysis devices may include a reaction acquisition unit that acquires a user's reaction to the intermediate data. In any of the above analysis devices, the output data acquisition unit may input one or more of the intermediate data selected based on the user's reaction to the second learning model.

[0007] Any of the above analysis devices may include a reaction acquisition unit that acquires a user's reaction to the intermediate data. In any of the above analysis devices, the intermediate data acquisition unit may input the user's reaction to the first learning model to acquire new intermediate data.

[0008] Any of the above analysis devices may include a similarity calculation unit that calculates a similarity between the new intermediate data acquired by the intermediate data acquisition unit and the past intermediate data. In any of the above analysis devices, the output data acquisition unit may input at least one of the intermediate data to the second learning model when the similarity calculated by the similarity calculation unit is equal to or greater than a set reference value.

[0009] Any of the above analysis devices may include a field determination unit that determines the field corresponding to the intermediate data. In any of the above analysis devices, the second learning model may be prepared for each field in the output data acquisition unit. In any of the above analysis devices, the output data acquisition unit may input the intermediate data to the second learning model corresponding to the field determined by the field determination unit.

[0010] In any of the above analysis devices, the output data acquisition unit may input the input data and the intermediate data to the second learning model.

[0011] In any of the above analysis devices, the second learning model may generate the output data including content that explains the relationship between the input data and the intermediate data.

[0012] In any of the above analysis devices, the intermediate data acquisition unit may input a configuration of the invention as the input data to the first learning model and acquire at least one of a problem to be solved by the invention and an effect of the invention as the intermediate data. In any of the above analysis devices, the output data acquisition unit may acquire the output data including content explaining a relationship between the configuration of the invention and at least one of the problem to be solved by the invention and an effect of the invention.

[0013] In any of the above analysis devices, the intermediate data acquisition unit may input at least one of a problem to be solved by the invention and an effect of the invention as the input data to the first learning model, and acquire a configuration of the invention as the intermediate data. In any of the above analysis devices, the output data acquisition unit may acquire a configuration of the invention that is more detailed than the intermediate data as the output data.

[0014] A second aspect of the present invention provides a system for generating output data for input data. The system may include an intermediate data acquisition unit that inputs the input data to a first learning model generated by machine learning using first learning data and acquires intermediate data output by the first learning model. Any of the above systems may include a second learning model generated by machine learning using second learning data that is more specialized in a specific field than the first learning data, and second inference means that inputs the intermediate data to the second learning model and infers the output data.

[0015] In a third aspect of the present invention, there is provided a program for causing a computer to function as the analysis device of the first aspect. In a fourth aspect of the present invention, there is provided a program for causing a computer to function as the system of the second aspect.

[0016] The above summary of the invention does not list all of the features of the present invention, and subcombinations of these features may also constitute inventions. [Brief explanation of the drawings]

[0017] [Figure 1] 1 is a block diagram illustrating an example of a system 10 according to one embodiment of the present invention. [Figure 2] FIG. 10 is a diagram showing an example of input data and a first prompt. [Figure 3] FIG. 10 is a diagram showing an example of intermediate data output by the first learning model 210 in response to a first prompt. [Figure 4] FIG. 10 is a diagram showing an example of a second prompt and output data. [Figure 5] FIG. 10 is a diagram illustrating another example of the operation of the system 10. [Figure 6] FIG. 10 is a diagram illustrating an example of the process of S1003. [Figure 7] FIG. 10 is a diagram illustrating an example of the processing in steps S1004 and S1005. [Figure 8] FIG. 10 is a diagram illustrating an example of the processing from S1006 to S1008. [Figure 9] FIG. 10 is a diagram illustrating an example of the process of step S1009. [Figure 10] 10 is a diagram illustrating another example of the processing of the system 10. FIG. [Figure 11] FIG. 11 is a diagram showing an example of intermediate data output by the first learning model 210 in response to the first prompt shown in FIG. [Figure 12] 12 is a diagram illustrating an example of processing performed by the output data acquisition unit 120 and the second learning model 220 on the intermediate data shown in FIG. 11. FIG. [Figure 13] FIG. 1 is a diagram illustrating another example of the system 10. [Figure 14] FIG. 1 is a diagram illustrating another example of the system 10. [Figure 15] FIG. 1 is a diagram illustrating another example of the system 10. [Figure 16] FIG. 1 is a diagram illustrating another example of the system 10. [Figure 17] 12 illustrates an example computer 1200 in which aspects of the present invention may be embodied, in whole or in part. DETAILED DESCRIPTION OF THE INVENTION

[0018] The present invention will be described below through embodiments of the invention, but the following embodiments do not limit the scope of the invention according to the claims. Furthermore, not all of the combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0019] FIG. 1 is a block diagram showing an example of a system 10 according to an embodiment of the present invention. The system 10 generates output data in response to input data. The input data is data input by a user or the like. The input data includes at least one of text information, numerical information, program information, image information, audio information, video information, etc. The output data may also include at least one of text information, numerical information, program information, image information, audio information, video information, etc.

[0020] The system 10 may be configured with pre-set request information indicating what output data is required or what processing is required for the input data. The system 10 generates output data according to the request information. In another example, the request information may be included in the input data.

[0021] The system 10 includes one or more information processing devices. Each information processing device may be a computer terminal managed by a user receiving a service, a server managed by a service provider, or another computer. Each information processing device may be connected via a general-purpose network such as the Internet, or may be connected via a dedicated line. Each information processing device may be connected via a wireless line or a wired line.

[0022] The system 10 of this example includes an analysis device 100, a first inference means 200, and a second inference means 202. The means (e.g., software) for realizing the analysis device 100, the first inference means 200, and the second inference means 202 may be provided by the same business operator, or some of the means may be provided by different business operators. The analysis device 100, the first inference means 200, and the second inference means 202 may be realized by a single information processing device, or may be realized by multiple information processing devices.

[0023] The first inference means 200 has a first learning model 210 generated by machine learning using first learning data. The first inference means 200 may include one or more information processing devices. In the first inference means 200, the multiple information processing devices may cooperate to realize processing by the first learning model 210. The first learning model 210 may use a publicly available general-purpose generative artificial intelligence (also referred to as generative AI).

[0024] The second inference means 202 has a second learning model 220 generated by machine learning using second learning data that is more specialized in a specific field than the first learning data. The second inference means 202 may include one or more information processing devices. In the second inference means 202, the multiple information processing devices may cooperate to realize processing by the second learning model 220. The second learning model 220 may be generative artificial intelligence.

[0025] Generally, a technique for generating a learning model using structured learning data including explanatory variables and corresponding objective variables is known. Such a learning model learns the relationship between explanatory variables and objective variables. When an explanatory variable of the same or similar type as the learned explanatory variable is input to a trained model, an estimation result of the corresponding objective variable is obtained.

[0026] The generative artificial intelligence in this example is generated by deep learning. The generative artificial intelligence can generate outputs not only for a certain type of explanatory variables but also for a variety of requests or questions. The first learning model 210 and the second learning model 220 may be large-scale language models. The first learning model 210 and the second learning model 220 may be GPT (registered trademark) (Generative Pre-trained Transformer), BERT (Bidirectional Encoder Representations from Transformers), or other language models.

[0027] As described above, the second learning model 220 is generated by machine learning using second learning data that is more specialized for a specific field than the first learning data. A field may refer to each field when human activities, etc., are divided into multiple domains. The domains may overlap. Examples of fields include academic fields, entertainment fields, business fields, and art fields. Each field may have further sub-fields. For example, academic fields may be subdivided into natural science fields, medical fields, literature fields, economics fields, etc. Each field of natural science may be further subdivided into electrical fields, communications fields, chemistry fields, biology fields, etc. Fields may be defined by commonly known methods such as patent classification, or may be defined by a user of the system 10 or a provider of the system 10.

[0028] The first training data may be data with no field restriction, data related to some selective fields, or data excluding some fields. The second training data is data with a field restriction greater than that of the first training data. The field restricted in the second training data is referred to as a restricted field. The second training data may include more data related to the restricted field than the first training data. The first training data may include more data related to fields other than the restricted field than the second training data. Furthermore, the proportion of data related to the restricted field contained in the second training data to the entire second training data may be higher than the proportion of data related to the restricted field contained in the first training data to the entire first training data.

[0029] The second training data may be data that matches search conditions set by a user or provider of the system 10. More specifically, it may be information containing keywords specified by the user or provider, information contained in a subject specified by the user (e.g., a specific journal, a specific trade paper, etc.), information classified by a known classification method such as a patent classification, or information collected by a general-purpose training model such as the first training model based on search conditions. This information may be text data.

[0030] The second learning model 220 may be a model that has been machine-learned using the second learning data from the beginning. In another example, the second learning model 220 may be a model that has been machine-learned using learning data from a wide range of fields, such as the first learning data, and then additionally trained using the second learning data. The second learning model 220 may be a model that has been additionally trained using the second learning data based on the first learning model 210. Additional training includes, for example, fine tuning.

[0031] Analysis device 100 generates output data for input data in cooperation with first inference means 200 and second inference means 202. The above-mentioned request information may be set in analysis device 100 in advance, or the request information may be included in the input data.

[0032] The analysis device 100 includes an intermediate data acquisition unit 110 and an output data acquisition unit 120. The intermediate data acquisition unit 110 inputs a first prompt including input data to a first learning model 210 and acquires intermediate data output by the first learning model 210. In this example, the intermediate data acquisition unit 110 generates a first prompt including the input data and request information.

[0033] The first learning model 210 uses input data included in the first prompt to execute the processing indicated by the request information included in the first prompt. For example, the input data includes technical content (or the configuration of an invention), and the request information is a request to infer the effect achieved by the technical content or the problem solved by the technical content. In this specification, the effect achieved by the technical content or the problem solved by the technical content may be simply referred to as the effect of the technical content or the problem solved by the technical content. However, the input data and the request information are not limited to these contents. The first learning model 210 outputs intermediate data generated in response to the first prompt.

[0034] The intermediate data acquisition unit 110 outputs the acquired intermediate data to the output data acquisition unit 120. The output data acquisition unit 120 inputs a second prompt including the intermediate data to the second learning model 220 and acquires output data output by the second learning model 220. The output data acquisition unit 120 in this example generates a second prompt including the intermediate data and request information.

[0035] The second learning model 220 uses the intermediate data included in the second prompt to execute the process indicated by the request information included in the second prompt. For example, the intermediate data includes an inference result of the effect of the technical content or the problem to be solved, and the request information is a request to infer the problem to be solved in more detail. The request information may also be a request to infer the relationship between the technical content and the effect or the problem to be solved in more detail. The second learning model 220 in this example is a model that has been trained in advance using second learning data in a field corresponding to the technical content. For example, the second learning data is data that associates input data with output data. An example of input data is the technical content, and an example of output data is the effect or the problem to be solved.

[0036] The output data acquisition unit 120 provides the user with the output data acquired from the second learning model 220. The output data acquisition unit 120 may display the output data or transmit it to the user's terminal.

[0037] According to the system 10, by using the first learning model 210, intermediate data can be generated according to the learning results using general-purpose first learning data. Therefore, comprehensive intermediate data based on a wide range of knowledge can be generated. For example, when the input data is technical content and the requirement information is an inference of an effect or a problem to be solved, the first learning model 210 can comprehensively estimate the effect or the problem to be solved based on a wide range of information not limited to the field of the technical content. Therefore, it is expected that the effect or the problem to be solved can be estimated using knowledge in fields other than the technical content.

[0038] By inputting the intermediate data into the second learning model 220 specialized for that field, processing can be performed on the comprehensively generated intermediate data according to knowledge specialized for that field. For example, output data that provides a more detailed explanation of the problem to be solved indicated in the intermediate data can be obtained. Furthermore, output data that more precisely estimates the relationship between the technical content indicated in the input data and the effect or problem to be solved indicated in the intermediate data can be obtained. This allows for both processing that utilizes comprehensive knowledge from a wide range of fields and processing that utilizes detailed knowledge from a specific field. Furthermore, if one learning model were to utilize comprehensive knowledge from a wide range of fields and also detailed knowledge from a specific field, the learning model would have to be constructed using detailed learning data from all fields. In this case, a huge amount of learning data would be required. By combining the first learning model 210 and the second learning model 220, a learning model can be efficiently generated using a relatively small amount of learning data.

[0039] 2 is a diagram showing an example of input data and a first prompt. The input data in this example is a sentence showing the configuration of the invention. The input data may be a list of the configuration of the invention, or may be written as a single sentence.

[0040] The intermediate data acquisition unit 110 inputs a first prompt including the configuration of the invention as input data to the first learning model 210. In this example, the intermediate data acquisition unit 110 adds request information specifying the content of the output that the first learning model 210 should generate based on the input data to the first prompt. In this example, the intermediate data acquisition unit 110 requests the first learning model 210 to comprehensively output the problem to be solved by the invention.

[0041] FIG. 3 is a diagram showing an example of intermediate data output by the first learning model 210 in response to the first prompt. In this example, the first learning model 210 outputs at least one of the problem to be solved and the effect of the invention indicated in the first prompt as intermediate data. The first learning model 210 preferably outputs multiple problems to be solved or effects as intermediate data. The intermediate data acquisition unit 110 may specify the minimum number of problems to be solved or effects output by the first learning model 210 in the first prompt. The intermediate data in the example of FIG. 3 includes information indicating five problems to be solved. In this specification, each item included in the intermediate data may be treated as a single piece of intermediate data. In other words, intermediate data including multiple items may be treated as multiple pieces of intermediate data.

[0042] FIG. 4 is a diagram showing an example of a second prompt and output data. The output data acquisition unit 120 generates the second prompt based on the intermediate data shown in FIG. 3. The output data acquisition unit 120 generates the second prompt including an extracted portion extracted from the intermediate data. The output data acquisition unit 120 may include one or more items selected from the multiple items included in the intermediate data (in this example, multiple problems to be solved or effects) in the second prompt. In the example of FIG. 4, the fifth problem to be solved shown in FIG. 3 is included in the second prompt.

[0043] The output data acquisition unit 120 may extract an item specified by the user from among multiple items included in the intermediate data and include it in the second prompt. In another example, the output data acquisition unit 120 may extract an item from among multiple items included in the intermediate data that best meets conditions preset by the user or the like and include it in the second prompt. For example, the output data acquisition unit 120 may be configured to prioritize multiple keywords in advance by the user or the like. In another example, the output data acquisition unit 120 may compare a problem described in a publication of a patent application in the field of the invention with multiple solved problems included in the intermediate data and extract one of the solved problems based on the comparison results. For example, the output data acquisition unit 120 may calculate the similarity between each solved problem included in the intermediate data and a group of problems described in the publication, and extract one of the solved problems based on the similarity. The similarity to one solved problem may be the average similarity between the problem described in each publication and the solved problem. The output data acquisition unit 120 may extract the solved problem with the lowest similarity. The similarity may be calculated, for example, based on the similarity of included terms. In this case, issues with a relatively high degree of novelty can be extracted.

[0044] The output data acquisition unit 120 may add request information that is set in advance or specified in the input data to the second prompt. The output data acquisition unit 120 may add request information that requests a more detailed explanation of the relationship between the content of the input data and the items extracted from the intermediate data than the items themselves. The request information in the example of FIG. 4 is information that requests an explanation of the relationship between the configuration of the invention shown in the input data and at least one of the problem to be solved and the effect extracted from the intermediate data. The output data acquisition unit 120 may include the content of the input data in the second prompt.

[0045] The second prompt in the example of Figure 4 includes the content indicated by the input data as a "claim" and the extracted portion extracted from the intermediate data as a "problem in the past" or an "effect." The second prompt also includes, as request information, a request to further specify the extracted "problem in the past" in order to prepare a patent specification.

[0046] The second learning model 220 generates output data in response to the second prompt. The second learning model 220 of this example generates output data including content explaining the relationship between the input data and the intermediate data. The output data acquisition unit 120 acquires the output data. The second learning model 220 of this example can generate more detailed output about the field of content included in the input data (in this example, the field related to hollow fiber membrane blood purifiers or a field encompassing that field) than the first learning model 210. Therefore, the output data acquisition unit 120 can acquire more detailed information than the content written in the second prompt. The second learning model 220 of this example outputs a problem-to-solve that describes in more detail the content of the problem-to-solve contained in the intermediate data. The output data output by the second learning model 220 of this example includes additional information, such as an explanation of the background or premise of the intermediate data, a detailed explanation of each term in the intermediate data, more detailed causal relationships between the components included in the intermediate data, and additional information related to the content of the intermediate data.

[0047] The output data acquisition unit 120 acquires output data for at least one item included in the intermediate data. In this example, the output data acquisition unit 120 selects one of the items included in the intermediate data and includes it in the second prompt. In this case, the output data acquisition unit 120 acquires output data for that item. In another example, the output data acquisition unit 120 may select multiple items included in the intermediate data and include them in the second prompt. In this case, the output data acquisition unit 120 may generate a second prompt requesting a detailed relationship between each item of the intermediate data and the input data. The output data acquisition unit 120 may acquire output data such as that shown in FIG. 4 for each item included in the second prompt.

[0048] 5 is a diagram showing another example of the operation of the system 10. The system 10 of this example performs the processes of S1001 to S1009. The processes of S1001 to S1009 may include the processes of the system 10 described with reference to FIGS. 1 to 4.

[0049] In S1001, the intermediate data acquisition unit 110 acquires input data. In S1002, the intermediate data acquisition unit 110 creates a first prompt and inputs it to the first learning model 210. The processing of the intermediate data acquisition unit 110 and the first learning model 210 in S1001 and S1002 is the same as the processing described in FIG. 2.

[0050] 6 is a diagram illustrating an example of the processing of S1003. In this example, the intermediate data acquisition unit 110 acquires first intermediate data generated by the first learning model 210 in response to the first prompt. The outline of the first intermediate data is similar to that of the intermediate data shown in FIG. 3. The first intermediate data in this example includes more items (items 1 to 20 in the example of FIG. 6) than the intermediate data in FIG. 3.

[0051] FIG. 7 is a diagram illustrating an example of the processing of S1004 and S1005. In S1004, the intermediate data acquisition unit 110 of this example generates a third prompt for the acquired first intermediate data and inputs it to the first learning model 210. The third prompt includes request information for narrowing down one or more items that meet a predetermined condition from among the multiple items included in the first intermediate data. The predetermined condition may be set in advance by a user or provider of the system 10. In S1005, the first learning model 210 generates second intermediate data in response to the third prompt. The second intermediate data is data obtained by extracting items that meet a predetermined condition from the items included in the first intermediate data. The intermediate data acquisition unit 110 acquires the second intermediate data.

[0052] The intermediate data acquisition unit 110 may generate a third prompt if the number of items included in the first intermediate data is greater than a set reference value. If the number of items included in the first intermediate data is equal to or less than the reference value, the intermediate data acquisition unit 110 may treat the first intermediate data as second intermediate data and perform subsequent processing. In another example, the intermediate data acquisition unit 110 may include the request information indicated in the third prompt in the first prompt in advance. In this case, the first learning model 210 outputs second intermediate data including the narrowed-down items, as shown in FIG. 7, in response to the first prompt.

[0053] 8 is a diagram illustrating an example of the processing from S1006 to S1008. In S1006, the output data acquisition unit 120 generates a second prompt according to the second intermediate data and inputs it to the second learning model 220. In S1007, the second learning model 220 generates first output data according to the second prompt. The output data acquisition unit 120 acquires the first output data. The processing for generating the second prompt and the first output data is similar to the processing for generating the second prompt and output data described in FIG. 4.

[0054] In S1008, the output data acquisition unit 120 of this example generates a fourth prompt based on the first output data and inputs it to the second learning model 220. The fourth prompt includes request information requesting a detailed explanation of the reason and principle behind how the problem to be solved contained in the intermediate data can be obtained from the configuration of the invention contained in the input data. The request information to be included in the fourth prompt may be set in advance by the user or provider of the system 10.

[0055] FIG. 9 is a diagram illustrating an example of the processing of S1009. In S1009, the second learning model 220 generates second output data in response to the fourth prompt. The output data acquisition unit 120 acquires the second output data. The second output data in this example includes a detailed explanation of the reason why the problem to be solved contained in the second intermediate data can be obtained from the configuration of the invention contained in the input data, and a theoretical explanation.

[0056] According to this example, the first learning model 210 can be used to comprehensively infer the problem to be solved or the effect of the target invention. Then, the second learning model 220 can be used to infer the problem to be solved or the effect in detail. Furthermore, by further using the second learning model 220, the relationship between the problem to be solved or the effect and the configuration of the invention can be inferred in detail.

[0057] FIG. 10 is a diagram illustrating another example of processing by the system 10. In this example, the intermediate data acquisition unit 110 acquires at least one of the problem to be solved and the effect of the invention as input data. The intermediate data acquisition unit 110 generates a first prompt including the input data and inputs it into the first learning model 210. In this example, the first prompt includes the input data and request information that comprehensively requests configurations of the invention that can solve the problem or achieve the effect indicated in the input data.

[0058] Figure 11 is a diagram showing an example of intermediate data output by the first learning model 210 in response to the first prompt shown in Figure 10. The first learning model 210 in this example comprehensively shows technical elements related to the problem to be solved or the effect shown in the input data, and creates an explanation that simply describes the relationship between each element and the problem to be solved or the effect. The intermediate data acquisition unit 110 acquires intermediate data such as that shown in Figure 11.

[0059] FIG. 12 is a diagram illustrating an example of processing performed by the output data acquisition unit 120 and the second learning model 220 on the intermediate data shown in FIG. 11. The output data acquisition unit 120 generates a second prompt including one or more items (extracted portions) extracted from the intermediate data and inputs the second prompt to the second learning model 220. The output data acquisition unit 120 includes the items of the intermediate data, input data, and request information in the second prompt. In this example, the request information requests a more specific configuration of the invention than the content shown in the intermediate data, as well as a detailed explanation of the configuration and the problem to be solved or the effect. The request information may also request the content of the claims in a patent specification. The request information included in the second prompt may be preset by the user or provider of the system 10.

[0060] According to this example, the first learning model 210 can be used to comprehensively infer candidate configurations of an invention from the problem to be solved or the effect. Then, the second learning model 220 can be used to infer a more detailed configuration of an invention and to infer a detailed relationship between the configuration of the invention and the problem to be solved or the effect. The second learning model 220 in this example may be a model trained using technical information related to the user as second learning data. The technical information related to the user may include patent specifications, papers, technical reports, etc. created by the user or the organization to which the user belongs. This increases the degree of correlation between the detailed configuration of the invention output by the second learning model 220 and technologies related to the user. Therefore, the second learning model can output a configuration of an invention that is easy for the user to manufacture. The second inference means 202 may have a second learning model 220 for each user. The output data acquisition unit 120 may include information identifying the user in the second prompt. The user's identification information may be included in the input data. The second inference means 202 may input the second prompt to the second learning model 220 corresponding to the user indicated in the second prompt. In another example, the output data acquisition unit 120 may include information specifying the second learning model 220 corresponding to the user in the second prompt.

[0061] FIG. 13 is a diagram showing another example of the system 10. In addition to the configuration shown in FIG. 1, the analysis device 100 of this example further includes a reaction acquisition unit 130. The reaction acquisition unit 130 presents intermediate data to a user and acquires the user's reaction to the intermediate data. The reaction acquisition unit 130 may display the intermediate data on a display unit of the analysis device 100, or may transmit the intermediate data to the user's terminal.

[0062] The reaction acquisition unit 130 acquires the user's reaction to each item included in the intermediate data. For example, the reaction acquisition unit 130 may acquire a reaction indicating which item the user selected from among multiple items included in the intermediate data. The user may select one or multiple items. The reaction acquisition unit 130 may acquire the contents of comments, corrections, or changes made by the user to any of the items included in the intermediate data. The user may add, correct, or change the contents of the intermediate data.

[0063] The reaction acquisition unit 130 selects one or more items from the multiple items included in the intermediate data based on the user's reaction and inputs them as intermediate data to the output data acquisition unit 120. The reaction acquisition unit 130 may extract the intermediate data items selected by the user and input them to the output data acquisition unit 120, or may extract items to which the user has added comments, corrections, etc. and input them to the output data acquisition unit 120. The first intermediate data shown in FIG. 6 is an example of intermediate data input to the reaction acquisition unit 130, and the second intermediate data shown in FIG. 7 is an example of intermediate data output by the reaction acquisition unit 130.

[0064] The output data acquisition unit 120 inputs a second prompt, including one or more intermediate data items selected based on the user's response, to the second learning model 220. The second prompt is similar to the example shown in FIG. 8, etc. The second learning model 220 generates output data in response to the second prompt. The second learning model 220 may generate the first output data shown in FIG. 8. The output data acquisition unit 120 may output the first output data to the user. In another example, the output data acquisition unit 120 may further input a fourth prompt such as that shown in FIG. 8 to the second learning model 220 and output the resulting second output data to the user.

[0065] 14 is a diagram showing another example of the system 10. The analysis device 100 of this example inputs a modified version of the first prompt, which reflects the user's reaction to the intermediate data, into the first learning model 210.

[0066] The reaction acquisition unit 130 acquires the user's reaction to the intermediate data. The reaction acquisition unit 130 notifies the intermediate data acquisition unit 110 of the acquired user's reaction. The intermediate data acquisition unit 110 inputs a modified version of the first prompt that reflects the acquired user's reaction into the first learning model 210, and acquires new intermediate data. The reaction acquisition unit 130 may present the new intermediate data to the user, or may input it to the output data acquisition unit 120.

[0067] The reaction acquisition unit 130 may acquire a user's reaction to the new intermediate data and notify the intermediate data acquisition unit 110 again. The intermediate data acquisition unit 110 inputs a first prompt that reflects the re-notified user's reaction into the first learning model 210. The process of acquiring the user's reaction and generating a first prompt that reflects the user's reaction may be repeated until the user's reaction satisfies a predetermined condition. When the user's reaction satisfies the predetermined condition, the reaction acquisition unit 130 may input the latest intermediate data to the output data acquisition unit 120. For example, when receiving an instruction from the user to generate output data using the current intermediate data, the reaction acquisition unit 130 inputs the intermediate data to the output data acquisition unit 120.

[0068] The user's response in this example may include a request for intermediate data. The response acquisition unit 130 may acquire a response requesting narrowing down of the items in the intermediate data, as shown in the third prompt of FIG. 7 . Based on the response, the intermediate data acquisition unit 110 may generate a modified first prompt similar to the third prompt of FIG. 7 and input it to the first learning model 210. The user's response may include corrections, additions, or changes to the input data. The intermediate data acquisition unit 110 may generate a modified first prompt using the corrected input data and input it to the first learning model 210.

[0069] FIG. 15 is a diagram showing another example of the system 10. The analysis device 100 of this example further includes a similarity calculation unit 140 in addition to the configuration shown in FIG. 14. As described in FIG. 14, the analysis device 100 of this example updates the intermediate data based on the user's reaction to the intermediate data. The similarity calculation unit 140 calculates the similarity between new intermediate data acquired by the intermediate data acquisition unit 110 and past intermediate data. The past intermediate data may be the intermediate data immediately before. The similarity calculation unit 140 may calculate the similarity of sentences or words included in each intermediate data. The similarity calculation unit 140 may calculate the similarity of sentences between intermediate data using a known method. For example, the similarity calculation unit 140 may vectorize the sentences included in the intermediate data to calculate the similarity.

[0070] The similarity calculation unit 140 may compare the calculated similarity with a set reference value. When the calculated similarity is equal to or greater than the reference value, the similarity calculation unit 140 may notify the output data acquisition unit 120 of this fact. When the similarity is equal to or greater than the reference value, it can be determined that no new content can be obtained even if the intermediate data is updated any further. Note that the method for determining that no new content can be obtained is not limited to the above, and other methods may also be used.

[0071] In this example, when the similarity is equal to or greater than a reference value, the output data acquisition unit 120 generates a second prompt according to at least one intermediate data that has already been obtained. The output data acquisition unit 120 may generate the second prompt according to the latest intermediate data. The output data acquisition unit 120 inputs the second prompt into the second learning model 220 to acquire output data.

[0072] This process allows the first prompt to be optimized according to the user's response. Also, when the changes in the updated intermediate data become small, the process can be shifted to processing using the second learning model 220. This improves the efficiency of the processing in the analysis device 100.

[0073] The first inference means 200 may further train the first learning model 210 using the user's reaction acquired by the reaction acquisition unit 130. The first inference means 200 may learn the content of corrections made by the user.

[0074] When the output data acquiring unit 120 outputs multiple pieces of output data, the reaction acquiring unit 130 may acquire the reaction of the user as to which output data they selected. Furthermore, when the user modifies and uses the output data, the reaction acquiring unit 130 may acquire the modification content of the output data. The second inference means 202 may further train the second learning model 220 using the user's reaction to the output data.

[0075] FIG. 16 is a diagram showing another example of the system 10. The analysis device 100 of this example further includes a field determination unit 150. The second inference means 202 has a plurality of second learning models 220. The second learning models 220 are prepared for each of a plurality of fields. Each second learning model 220 is created using second learning data for that field. The other structures are similar to those of the analysis device 100 of any of the aspects described in FIGS. 1 to 15.

[0076] The field determination unit 150 determines the field corresponding to the intermediate data acquired by the intermediate data acquisition unit 110. The field determination unit 150 may determine the field of the intermediate data based on words, etc. included in the intermediate data. The field determination unit 150 may determine the field of the intermediate data using a known method. The field determination unit 150 may determine to which second learning model 220 the field of the intermediate data corresponds. The field determination unit 150 may register multiple fields corresponding to multiple available second learning models 220. When the second inference means 202 creates a new second learning model 220, it may register the corresponding field in the field determination unit 150. The field determination unit 150 may input the intermediate data and field data indicating the field of the intermediate data to the output data acquisition unit 120.

[0077] The output data acquisition unit 120 inputs the intermediate data into the second learning model 220 corresponding to the field determined by the field determination unit 150. The output data acquisition unit 120 may add information specifying the corresponding field to the second prompt, or may add information specifying the second learning model 220 to be used. The second learning model 220 generates output data in response to the second prompt. The output data acquisition unit 120 acquires the output data.

[0078] According to this example, the second learning model 220 to be used can be selected depending on the content of the input data or intermediate data. Therefore, output data with appropriate and detailed explanations can be generated for a variety of input data or intermediate data.

[0079] 1 to 16, examples have been described in which input data and output data represent the configuration, problem to be solved, or effect of the invention. The data handled by system 10 is not limited to these. System 10 is applicable to a variety of input data and a variety of output data.

[0080] In one embodiment, the input data may be a patient's symptoms or biometric information, and the output data may be a detailed treatment method. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise treatment methods. The second learning model 220 may generate output data including more detailed treatment methods. The second learning data may be data that associates at least the patient's symptoms or biometric information with detailed treatment information.

[0081] In one embodiment, the input data may be a student's answer sheet, and the output data may be a detailed study method. In this case, the first learning model 210 may generate intermediate data including comprehensive and concise study methods. The second learning model 220 may generate output data including more detailed study methods. The second learning data may be data that associates at least the student's answer sheet with more detailed study methods (by subject, by purpose, etc.).

[0082] In one embodiment, the input data may be characteristics of a finished product, and the output data may be detailed materials and manufacturing methods for the finished product. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise information about materials and manufacturing methods. The second learning model 220 may generate output data including more detailed material compositions and manufacturing methods. The second learning data may be data that associates at least the characteristics of the finished product with more detailed materials and manufacturing methods for the finished product.

[0083] In one embodiment, the input data may be an overview of the building's exterior, and the output data may be a detailed internal structure. In this case, the first learning model 210 may generate intermediate data including a comprehensive and simplified internal structure. The second learning model 220 may generate output data including a more detailed internal structure. The second learning data may be data that associates at least the building's exterior with a more detailed internal structure (such as required materials).

[0084] In one embodiment, the input data may be technical materials, and the output data may be promotional materials tailored to a specific purpose. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise materials. The second learning model 220 may generate output data including more detailed promotional materials (e.g., by purpose, by format such as text or video). The second learning data may be data that associates the technical materials with the promotional materials.

[0085] In one embodiment, the input data may be crop information, and the output data may be an optimal cultivation method specific to a region. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise cultivation methods. The second learning model 220 may generate output data including the optimal cultivation method specific to a region. The second learning data may be data that associates at least the crop information with the optimal cultivation method specific to a region.

[0086] In one embodiment, the input data may be information about the growing environment, and the output data may be crops and their cultivation methods that are optimal for that environment. The output data may also be livestock and their raising methods that are optimal for that environment. In this case, the first learning model 210 may generate intermediate data that includes comprehensive yet simple information about candidate crops that can be grown in the environment and their cultivation methods. The first learning model 210 may generate intermediate data that includes comprehensive yet simple information about livestock that can be grown in the environment and their raising methods. The second learning model 220 may generate output data that includes crops and their cultivation methods that are optimal for the environment. The second learning model 220 may generate output data that includes livestock and their raising methods that are optimal for the environment. The second learning data may be data that associates at least information about the growing environment with crops and their cultivation methods that are optimal for that environment. The second learning data may also be data that associates livestock and their raising methods that are optimal for the environment.

[0087] In one embodiment, the input data may be a final product in organic synthesis, and the output data may be an optimal synthetic route for a specific purpose. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise synthetic routes. The second learning model 220 may generate output data including detailed and optimal synthetic routes. The second learning data may be data that associates at least information about the final product with the synthetic route.

[0088] In the above example, the intermediate data and the output data of the second learning model are data of the same category, such as the "problem to be solved" in one embodiment of the present invention, but this is not limited to this. For example, the intermediate data and the output data may be data of different categories. The following describes an example in which the intermediate data and the output data are data of different categories.

[0089] In one embodiment, the input data may be electronic medical record information, and the output data may be test methods required for making a definitive diagnosis. In this case, the first learning model 210 may generate intermediate data including disease name candidates that are expected to be presented in a comprehensive and simplified differential diagnosis result. The second learning model 220 may generate output data including test methods required for making a definitive diagnosis. The second learning data may be data that associates at least one of the electronic medical record information and disease name candidates with test methods required for making a definitive diagnosis.

[0090] In one embodiment, the input data may be product information, and the output data may be standards and test information to be applied to the product. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise product categories. The second learning model 220 may generate output data including at least one of the standards and test information to be applied to the product. The second learning data may be data that associates at least one of the product information and product categories with the standards and test information to be applied to the product.

[0091] In one embodiment, the input data may be sounds emitted from a device, and the output data may be the device's status and, if there is a problem, how to deal with it. In this case, the first learning model 210 may generate intermediate data including comprehensive yet concise sound categories (such as normal or abnormal sounds and the location where the sound is generated). The second learning model 220 may generate output data including a concrete description of the situation and how to deal with it. The second learning data may be data that associates at least one of the sounds emitted from the device and the sound categories with the device's status and, if there is a problem, how to deal with it.

[0092] In one embodiment, the input data may be product information, and the output data may be market information on new applications to which the product can be applied. In this case, the first learning model 210 may generate intermediate data that comprehensively and simply includes new applications to which the product can be applied. The second learning model 220 may generate output data that includes market information on new applications to which the product can be applied. The second learning data may be data that associates at least one of the product information and new applications to which the product can be applied with the new applications to which the product can be applied.

[0093] In one embodiment, the input data may be manufacturing information, and the output data may be an alternative process plan for complying with legal regulations. In this case, the first learning model 210 may generate intermediate data that includes comprehensive yet concise information about issues in the manufacturing information. The second learning model 220 may generate output data that includes an alternative process plan for the manufacturing information that satisfies legal regulations. The second learning data may be data that associates at least one of the manufacturing information and the issues in the manufacturing information with an existing process that satisfies legal regulations.

[0094] In one embodiment, the input data may be information about sleep states, and the output data may be methods for improving the sleep environment or lifestyle habits. In this case, the first learning model 210 may generate intermediate data that comprehensively and simply includes sleep state issues. The second learning model 220 may generate output data that includes methods for improving the sleep environment or lifestyle habits necessary to solve the sleep state issues. The second learning data may be data that associates at least one of the sleep state information and sleep state issues with methods for improving the sleep environment or lifestyle habits linked to the sleep state issues.

[0095] In one embodiment, the input data may be an ingredient list, and the output data may be a recipe for a specific dish. In this case, the first learning model 210 may generate intermediate data including comprehensive and simplified dish candidates that can be created from the ingredient list. The second learning model 220 may generate output data including the recipe for the specific dish. The second learning data may be data that associates the recipe for the specific dish with at least one of the ingredient list and the dish candidates that can be created from the ingredient list.

[0096] The technical scope of the present invention is not limited to the above-described embodiments, and is applicable to the present invention as long as at least two of the input data, intermediate data, and output data have a correlation or association that can be inferred by a person skilled in the art. The second training data may be data including the input data and the output data, or may be data in which the input data and the output data are associated. The second training data may also be data in which at least one of the input data and the intermediate data is associated with the output data.

[0097] The system 10 described in Figures 1 to 16 may be realized by installing a program in one or more computers. The analysis device 100 described in Figures 1 to 16 may be realized by installing a program in one or more computers. The first inference means 200 described in Figures 1 to 16 may be realized by installing a program in one or more computers. The second inference means 202 described in Figures 1 to 16 may be realized by installing a program in one or more computers. These programs may be recorded on computer-readable media.

[0098] Various embodiments of the present invention may be described with reference to flowcharts and block diagrams, where the blocks may represent (1) stages of a process in which operations are performed or (2) sections of an apparatus responsible for performing the operations. Particular stages and sections may be implemented by dedicated circuitry, programmable circuitry provided with computer-readable instructions stored on a computer-readable medium, and / or a processor provided with computer-readable instructions stored on a computer-readable medium. Dedicated circuitry may include digital and / or analog hardware circuitry, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuitry may include reconfigurable hardware circuitry, including logical AND, OR, XOR, NAND, NOR, and other logical operations, flip-flops, registers, memory elements such as field programmable gate arrays (FPGAs), programmable logic arrays (PLAs), and the like.

[0099] A computer-readable medium may include any tangible device capable of storing instructions that are executed by an appropriate device, such that the computer-readable medium having instructions stored thereon comprises an article of manufacture containing instructions that can be executed to create means for performing the operations specified in the flowcharts or block diagrams. Examples of computer-readable media may include electronic, magnetic, optical, electromagnetic, and semiconductor storage media. More specific examples of computer-readable media may include floppy disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disc read-only memory (CD-ROM), digital versatile disc (DVD), Blu-ray (RTM) disc, memory stick, integrated circuit card, and the like.

[0100] The computer readable instructions may include either assembler instructions, Instruction Set Architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, or source or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, JAVA®, C++, etc., and conventional procedural programming languages ​​such as the “C” programming language or similar programming languages.

[0101] The computer-readable instructions may be provided to a processor or programmable circuit of a programmable data processing device, such as a general-purpose computer, a special-purpose computer, or another computer, either locally or via a wide-area network (WAN) such as a local area network (LAN) or the Internet, which executes the computer-readable instructions to create means for performing the operations specified in the flowcharts or block diagrams. Here, the computer may be a personal computer (PC), a tablet computer, a smartphone, a workstation, a server computer, a general-purpose computer, a special-purpose computer, or the like, or may be a computer system in which multiple computers are connected. Such a computer system in which multiple computers are connected is also called a distributed computing system, and is a broad definition of computers. In a distributed computing system, multiple computers collectively execute a program by each executing a portion of the program and passing data between computers as needed during program execution.

[0102] Examples of processors include a computer processor, a central processing unit (CPU), a processing unit, a microprocessor, a digital signal processor, a controller, a microcontroller, etc. A computer may have one or more processors. In a multiprocessor system with multiple processors, each processor executes a portion of a program and passes data between processors as needed during program execution, allowing the multiple processors to collectively execute a program. For example, in multitasking, each of the multiple processors may execute a portion of each task in small chunks by switching tasks at time slice intervals. In this case, which portion of a program each processor executes changes dynamically. Which portion of a program each of the multiple processors executes may also be statically determined by multiprocessor-aware programming.

[0103] 17 illustrates an example of a computer 1200 in which aspects of the present invention may be embodied, in whole or in part. Programs installed on the computer 1200 may cause the computer 1200 to function as or perform operations associated with an apparatus or one or more sections of the apparatus according to embodiments of the present invention, and / or to perform a process or steps of a process according to embodiments of the present invention. Such programs may be executed by the CPU 1212 to cause the computer 1200 to perform specific operations associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0104] A computer 1200 according to this embodiment includes a CPU 1212, a RAM 1214, a graphics controller 1216, and a display device 1218, which are interconnected by a host controller 1210. The computer 1200 also includes input / output units such as a communication interface 1222, a storage device 1224 such as a hard disk drive, a DVD-ROM drive 1226, and an IC card drive, which are connected to the host controller 1210 via an input / output controller 1220. The computer also includes legacy input / output units such as a ROM 1230 and a keyboard 1242, which are connected to the input / output controller 1220 via an input / output chip 1240.

[0105] The CPU 1212 operates according to programs stored in the ROM 1230 and the RAM 1214, thereby controlling each unit. The graphics controller 1216 acquires image data generated by the CPU 1212 into a frame buffer or the like provided in the RAM 1214 or into the graphics controller 1216 itself, and causes the image data to be displayed on the display device 1218.

[0106] The communication interface 1222 communicates with other electronic devices via a network. The storage device 1224 stores programs and data used by the CPU 1212 in the computer 1200. The DVD-ROM drive 1226 reads programs or data from a DVD-ROM 1227 and provides the programs or data to the storage device 1224 via the RAM 1214. The IC card drive reads programs and data from an IC card and / or writes programs and data to an IC card.

[0107] The ROM 1230 stores therein a boot program or the like that is executed by the computer 1200 upon activation, and / or programs that depend on the hardware of the computer 1200. The input / output chip 1240 may also connect various input / output units to the input / output controller 1220 via a parallel port, a serial port, a keyboard port, a mouse port, etc.

[0108] The programs are provided by a computer-readable medium such as a DVD-ROM 1227 or an IC card. The programs are read from the computer-readable medium, installed in the storage device 1224, RAM 1214, or ROM 1230, which are also examples of computer-readable media, and executed by the CPU 1212. Information processing described in these programs is read by the computer 1200, and causes cooperation between the programs and the various types of hardware resources described above. An apparatus or a method may be configured by implementing information manipulation or processing in accordance with the use of the computer 1200.

[0109] For example, when communication is performed between the computer 1200 and an external device, the CPU 1212 may execute a communication program loaded into the RAM 1214 and instruct the communication interface 1222 to perform communication processing based on the processing described in the communication program. Under the control of the CPU 1212, the communication interface 1222 reads transmission data stored in a transmission buffer processing area provided in the RAM 1214, the storage device 1224, the DVD-ROM 1227, or a recording medium such as an IC card, and transmits the read transmission data to the network, or writes reception data received from the network to a reception buffer processing area or the like provided on the recording medium.

[0110] The CPU 1212 may read all or a necessary portion of a file or database stored on an external recording medium such as the storage device 1224, the DVD-ROM drive 1226 (DVD-ROM 1227), an IC card, etc. into the RAM 1214, and perform various types of processing on the data on the RAM 1214. The CPU 1212 then writes back the processed data to the external recording medium.

[0111] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and may undergo information processing. CPU 1212 may perform various types of processing on data read from RAM 1214, including various types of operations, information processing, conditional judgment, conditional branching, unconditional branching, information search / replacement, etc., as described throughout this disclosure and specified by the instruction sequences of the programs, and write the results back to RAM 1214. CPU 1212 may also search for information in a file, database, etc. on the recording medium. For example, if multiple entries each having an attribute value of a first attribute associated with an attribute value of a second attribute are stored on the recording medium, CPU 1212 may search for an entry that matches a condition specified by the attribute value of the first attribute from among the multiple entries, read the attribute value of the second attribute stored in the entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0112] The above-described programs or software modules may be stored in a computer-readable medium on or near the computer 1200. A recording medium such as a hard disk or RAM provided in a server system connected to a dedicated communication network or the Internet can also be used as a computer-readable medium, thereby providing the programs to the computer 1200 via the network.

[0113] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications and improvements can be made to the above embodiments. It is clear from the claims that such modifications and improvements can also be included within the technical scope of the present invention.

[0114] It should be noted that the execution order of each process, such as operations, procedures, steps, and stages, in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not specifically stated as "before," "prior to," etc., and that the processes can be performed in any order unless the output of a previous process is used in a subsequent process. Even if the operational flow in the claims, specifications, and drawings is described using "first," "next," etc. for convenience, this does not mean that the processes must be performed in this order. [Explanation of symbols]

[0115] 10. System, 100. Analysis device, 110. Intermediate data acquisition unit, 120. Output data acquisition unit, 130. Response acquisition unit, 140. Similarity calculation unit, 150. Field determination unit, 200. First inference means, 202. Second inference means, 210. First learning model, 220. Second learning model

Claims

1. An analysis device that generates output data for input data, an intermediate data acquisition unit that inputs the input data into a first learning model generated by machine learning using first learning data, and acquires intermediate data output by the first learning model; an output data acquisition unit that inputs the intermediate data into a second learning model generated by machine learning using second learning data that is more specialized in a specific field than the first learning data, and acquires the output data output by the second learning model; An analysis device comprising:

2. The second learning model is a model generated by performing additional learning on the first learning model. The analysis device according to claim 1 .

3. The first learning model generates a plurality of the intermediate data; The output data acquisition unit acquires the output data for at least one of the intermediate data. The analysis device according to claim 1 .

4. a reaction acquisition unit that acquires a user's reaction to the intermediate data; The output data acquisition unit inputs one or more of the intermediate data selected based on the user's response into the second learning model. The analysis device according to claim 2 .

5. a reaction acquisition unit that acquires a user's reaction to the intermediate data; The intermediate data acquisition unit inputs the user's reaction into the first learning model to acquire new intermediate data. The analysis device according to claim 1 .

6. a similarity calculation unit that calculates a similarity between the new intermediate data acquired by the intermediate data acquisition unit and the past intermediate data, The output data acquisition unit inputs at least one of the intermediate data to the second learning model when the similarity calculated by the similarity calculation unit is equal to or greater than a set reference value. The analysis device according to claim 5 .

7. a field determination unit that determines the field corresponding to the intermediate data; The second learning model is prepared for each of the fields for the output data acquisition unit, The output data acquisition unit inputs the intermediate data into the second learning model corresponding to the field determined by the field determination unit. The analysis device according to claim 1 .

8. The output data acquisition unit inputs the input data and the intermediate data into the second learning model. The analysis device according to any one of claims 1 to 7.

9. The second learning model generates the output data including a description of the relationship between the input data and the intermediate data. The analysis device according to claim 8.

10. the intermediate data acquisition unit inputs a configuration of the invention as the input data into the first learning model, and acquires at least one of a problem to be solved by the invention and an effect of the invention as the intermediate data; The output data acquisition unit acquires the output data including a description of the relationship between the configuration of the invention and at least one of the problem to be solved by the invention and the effect of the invention. The analysis device according to claim 8.

11. the intermediate data acquisition unit inputs at least one of a problem to be solved by the invention and an effect of the invention as the input data into the first learning model, and acquires a configuration of the invention as the intermediate data; The output data acquisition unit acquires, as the output data, a configuration of the invention that is more detailed than the intermediate data. The analysis device according to claim 8.

12. A system for generating output data for input data, comprising: an intermediate data acquisition unit that inputs the input data into a first learning model generated by machine learning using first learning data, and acquires intermediate data output by the first learning model; a second inference means having a second learning model generated by machine learning using second learning data that is more specialized in a specific field than the first learning data, and inputting the intermediate data into the second learning model to infer the output data; A system comprising:

13. A program for causing a computer to function as the analysis device according to claim 1.

14. A program for causing a computer to function as the system according to claim 12.

Images