Support device

Abstract

A support device 1 comprises: a lower-level information conversion unit 12 that converts each of different setting methods for devices having the same role, which are included in a plurality of first documents, into a vector corresponding to a setting purpose, and associates similar setting methods with each other on the basis of inter-vector similarity; and a one-to-one conversion unit 13 that searches for a second setting method associated with a first setting method using the association information, and generates the first setting method and the second setting method as a setting method conversion rule.

Description

Support device

[0001] The present disclosure relates to a support device.

[0002] There is a network composed of a plurality of device nodes (transfer devices, transmission devices). When developing a control device that provides and operates network services in this network, the following cases occur.

[0003] Case 1; When using device nodes with the same role from different vendors, or when the setting methods are different even for the same vendor due to differences in the types of device nodes and OS versions, it is necessary to newly create device configurations and control commands for the new vendor.

[0004] Case 2; When defining a new role for a device node in the network, it is necessary to investigate and analyze existing documents in order to achieve consistency based on the existing method, the development history of existing devices, information on opposing devices, etc.

[0005] Case 3; It is necessary to newly create device configurations and control commands for device nodes different from the existing setting method for which a new role has been defined.

[0006] The developer of the control device needs to read a huge amount of documents in any of the above cases.

[0007] Therefore, for Case 1 and Case 3, a method of utilizing prompt engineering can be considered. By pre-inputting the conversion rules of device configurations into an LLM (Large Language Models) and inputting the device configuration to be converted into the LLM, a device configuration for the new vendor can be obtained (Non-Patent Document 1).

[0008] For Case 2, a method of utilizing RAG (Retrieval Augmented-Generation) can be considered. By giving an instruction to the LLM in advance to search the database of existing documents when generating an answer to a question, an answer that is consistent with the existing method can be obtained (Non-Patent Document 2).

[0009] "Large-scale language models, prompts", Wikipedia, the free encyclopedia, [online], [Retrieved January 11, 2025], <URL: https: / / ja.wikipedia.org / wiki / %E5%A4%A7%E8%A6%8F%E6%A8%A1%E8%A8%80%E8%AA%9E%E3%83%A2%E3%83%87%E3%83%AB#%E3%83%97%E3%83%AD%E3%83%B3%E3%83%97%E3%83%88> "Large-scale language models, tools", Wikipedia, the free encyclopedia, [online], [Retrieved January 11, 2025], <URL: https: / / ja.wikipedia.org / wiki / %E5%A4%A7%E8%A6%8F%E6%A8%A1%E8%A8%80%E8%AA%9E%E3%83%A2%E3%83%87%E3%83%AB#%E3%83%84%E3%83%BC%E3%83%AB%E3%81%AE%E4%BD%BF%E7%94%A8>

[0010] However, when using prompt engineering, if the conversion rules are not detailed enough, the accuracy of the conversion of device configurations and control commands will decrease. While it is possible to have the LLM create the conversion rules, if there is a shortage of tokens that the LLM can handle, the accuracy of the conversion rules created by the LLM will decrease.

[0011] When using RAG, the search terms for the document database are crucial. However, even if the information is related across documents, the context and keywords often differ, making it difficult to specify appropriate search terms and thus preventing the proper retrieval of necessary information from the document database.

[0012] This disclosure is made in view of the above circumstances, and the purpose of this disclosure is to provide technology that can support the development of control devices.

[0013] A support device according to one aspect of the present disclosure includes: a first conversion unit that converts different setting methods for devices having the same role included in a plurality of first documents into vectors according to the purpose of setting, and links similar setting methods together based on the similarity between the vectors; and a first generation unit that searches for a second setting method linked to a first setting method using the linked linking information, and generates the first setting method and the second setting method as a conversion rule for setting methods.

[0014] This disclosure can support the development of control devices.

[0015] Figure 1 shows an example of the support system configuration. Figure 2 shows an example of the function and operation of the higher-level information conversion unit. Figure 3 shows an image of the upper-level document. Figure 4 shows an example of vector information of the upper-level text. Figure 5 shows an image of the graph structure of the upper-level text. Figure 6 shows an example of the function and operation of the lower-level information conversion unit. Figure 7 shows an image of the lower-level configuration. Figure 8 shows an image of the lower-level command tree structure. Figure 9 shows an example of the hierarchical structure of the lower-level configuration. Figure 10 shows an example of vector information of the purpose of the lower-level command and the purpose of the arguments. Figure 11 shows an image of the graph structure of the lower-level command. Figure 12 shows an example of the function and operation of the one-to-one conversion unit. Figure 13 shows an image of the generation of conversion rules. Figure 14 shows an example of conversion rules. Figure 15 shows an embodiment of the second case. Figure 16 shows an example of the function and operation of the conversion support unit. Figure 17 shows an example of a method for rearranging multiple commands. Figure 18 shows an example of the hardware configuration of the support device.

[0016] Embodiments of this disclosure will be described below with reference to the drawings. In the drawings, the same parts are denoted by the same reference numerals and their descriptions are omitted.

[0017] [Summary of this disclosure] This disclosure provides a method for supporting information retrieval and information transformation by linking information within documents across documents. This disclosure relates to a technology for applying natural language processing technology to the configuration creation process of device nodes.

[0018] Specifically, the information within the document is converted into vectors based on its semantic content, a graph structure is created by linking information with similar vector magnitudes, and then this information is further transformed according to its purpose.

[0019] In the first case, different device configurations and control commands contained in configuration manuals from multiple vendors are converted into semantic vectors, linked based on the similarity of these semantic vectors, and rule-based conversion rules are created that use this linked information to convert device configurations and control commands to a one-to-one correspondence.

[0020] In the second case, specifications and other documents from multiple vendors are divided into chapters, etc., converted into semantic vectors, linked between documents and within documents based on the similarity of these semantic vectors, and the linked information is stored in a database so that it can be accessed by third parties. Linking information regarding device configurations and control commands is also stored in the database so that it can be accessed by third parties. By referring to each linking information, the LLM can appropriately retrieve relevant information from the document database.

[0021] For the third case as well, we will create rule-based conversion rules. We will focus on the fact that even if the configuration system and setting strings differ, the purpose of the function to be implemented will be described and mean similarly in the manual. Therefore, we will use the two types of linking information mentioned above to create rule-based templates for device configurations and control commands.

[0022] [Example of Support System Configuration] Figure 1 shows an example of the configuration of the support system according to this embodiment.

[0023] The support system comprises a support device 1, a plurality of user terminals 2, and an AI (Artificial Intelligence) device 3.

[0024] Support device 1 is a server device that assists in the development of control devices. User terminal 2 is a client terminal used by the developers of the control devices. AI device 3 is a server device on which LLM operates. Support device 1 is connected to multiple user terminals 2 and AI device 3 in a manner that allows for mutual communication between them.

[0025] A control device is a server device that provides and operates network services in a network composed of multiple device nodes. A device node is a transfer device or transmission device.

[0026] [Example of the configuration of the support device 1] As shown in Figure 1, the support device 1 comprises a higher-level information conversion unit 11, a lower-level information conversion unit 12, a one-to-one conversion unit 13, and a conversion support unit 14.

[0027] The higher-level information conversion unit (second conversion unit) 11 has the function of converting documents of the higher layer (second documents related to the first document) into vectors and structuring (linking) them. Documents of the higher layer include, for example, system specifications and meeting documents.

[0028] Specifically, the higher-level information conversion unit 11 converts each paragraph's text or each semantic unit's text within a paragraph contained in documents from multiple different vendors or each higher-level layer of the same vendor into semantic vectors corresponding to the content of the text, and has a function to link similar texts together based on the similarity between semantic vectors (for example, cosine similarity).

[0029] The lower-level information conversion unit (first conversion unit) 12 has the function of converting lower-layer documents (first documents) into vectors and structuring (linking) them. Lower-layer documents include, for example, configuration manuals and interface specifications.

[0030] Specifically, the lower-level information conversion unit 12 converts different configuration methods (e.g., configuration, commands) for device nodes with the same role, contained in documents from multiple different vendors or each lower layer of the same vendor, into semantic vectors corresponding to the purpose of the configuration and the purpose of the arguments, and has a function to link similar configuration methods together based on the similarity between the semantic vectors (e.g., cosine similarity).

[0031] The one-to-one conversion unit (first generation unit) 13 has the function of generating conversion rules for setting methods between different setting methods.

[0032] Specifically, the one-to-one conversion unit 13 has the function of using the linking information linked by the lower-level information conversion unit 12 to search for other setting methods (second setting methods) linked to a predetermined setting method (first setting method), and generating a conversion rule for setting methods between the predetermined setting method and the other setting method.

[0033] Furthermore, the conversion rules for configuration methods are not limited to those used between different vendors. Even within the same vendor, the conversion rules may apply to configuration methods based on differences in device node type or OS version. In other words, regardless of whether the vendors are different or not, it is sufficient if the configuration system or method for setting configuration information to the device nodes differs.

[0034] The conversion support unit (second generation unit) 14 has a function to support the process of converting higher-level information of a higher-level document into lower-level information of a lower-level document.

[0035] Specifically, the conversion support unit 14 receives information for setting a desired role on a predetermined device node, and uses the linking information between texts and the linking information between setting methods to search for texts associated with the desired role, search for setting methods similar to those texts, and has the function of using those setting methods as templates for setting methods of application-specific device nodes that perform the desired role.

[0036] [Correspondence between each functional part of the support device 1 and each case] In this embodiment, various pieces of information within a document are automatically linked and structured, and the information is processed for various purposes to correspond to the first to third cases.

[0037] The first case is a method that enables support for configuration conversion. The first case is processed by the lower-level information conversion unit 12 and the one-to-one conversion unit 13.

[0038] The second case is a method for enabling cross-document searching of the contents of documents written in natural language. The second case is processed by a higher-level information conversion unit 11 and a lower-level information conversion unit 12.

[0039] The third case is a method that enables the content of a document written in natural language to be converted into a lower layer such as a device configuration by associating it with higher-level information such as a system specification. The third case is processed by a higher-level information conversion unit 11, a lower-level information conversion unit 12, and a conversion support unit 14.

[0040] [Examples of Functions and Operations of the Higher-Level Information Conversion Unit] Figure 2 shows examples of functions and operations of the higher-level information conversion unit 11.

[0041] The higher-level information conversion unit 11 includes a higher-level information processing unit 111 that vectorizes and structures the information of the upper-level layer document, and a higher-level information structured vector DB 112 that stores the structured vectors.

[0042] Step S101; The higher-level information processing unit 111 receives documents from multiple different vendors or from each higher-level layer of the same vendor, and uses existing natural language processing technology to divide each of the documents from each higher-level layer into paragraph-based text. For example, the higher-level information processing unit 111 divides the document from the higher-level layer shown in Figure 3 into "aaa\naaa" and "iii\niii". "\n" means a newline.

[0043] Step S102; The upper information processing unit 111 uses existing natural language processing techniques to divide the text for each divided paragraph into texts for each meaning within the paragraph. For example, the upper information processing unit 111 divides "あああ￥nあああ" into "あああ" and "あああ".

[0044] Step S103; The upper information processing unit 111 uses an existing encoder to convert the text for each paragraph and / or for each meaning within the paragraph into a meaning vector according to the content of the text. The existing encoder is, for example, BERT (Bidirectional Encoder Representations from Transformers).

[0045] Step S104; The upper information processing unit 111 associates the meaning vector for each paragraph and / or for each meaning within the paragraph with the text of the meaning vector and stores it in the vector information storage DB. FIG. 4(a) is an example of the vector information of the text for each paragraph. FIG. 4(b) is an example of the vector information of the text for each meaning within the paragraph.

[0046] Step S105; The upper information processing unit 111 reads out the meaning vectors of all the texts from the vector information storage DB, and associates (graph-structurizes) the texts with high cosine similarity between the meaning vectors. Whether the cosine similarity is high or not is determined using, for example, a threshold value. Thereafter, the upper information processing unit 11l stores the association information of the texts in the upper information structured vector DB112.

[0047] An image of the graph structure of the associated texts is shown in FIG. 5. As methods for associating texts, for example, (1) a method of associating only the texts for each paragraph across documents, (2) a method of associating only the texts for each meaning within a paragraph across the same document or across documents, and (3) a method of associating all texts regardless of the division unit can be considered. These methods can be used appropriately according to the purpose.

[0048] [Function Example and Operation Example of the Lower Information Conversion Unit] FIG. 6 is a diagram showing a function example and an operation example of the lower information conversion unit 12.

[0049] The lower-level information conversion unit 12 includes a lower-level information processing unit 121 that vectorizes and structures the information of the lower-layer document, and a lower-level information structured vector DB 122 that stores the structured vectors.

[0050] Step S201; The lower-level information processing unit 121 receives documents from multiple different vendors or from each lower layer of the same vendor, and uses existing natural language processing technology to create a tree structure of commands within the configurations contained in each lower-level document according to their purpose.

[0051] For example, the lower-level information processing unit 121 structures the commands for setting IP addresses in the configuration manual (Vendor A, Device X, OS version xx.yy.zz) shown in Figure 7 into a tree structure.

[0052] Commands for setting IP addresses include, for example, "ip-address {<IPv6 Address_own> |<IPv6 Prefix> It is defined as "ip-address" meaning the purpose of the command setting, and "{<IPv6 Address_own> |<IPv6 Prefix> The '}' indicates the purpose of the argument.

[0053] In the case of this command, "ip-address<IPv6 Address_own> " and "ip-address<IPv6 Prefix> By entering these two commands into device X, the objective of setting an IP address for device X can be achieved.

[0054] In this case, the lower-level information processing unit 121 sets the "ip-address" to the parent node,<IPv6 Address_own> "and"<IPv6 Prefix> These are set as child nodes. In the example tree structure shown in Figure 8, the expression "OR" is used to link the two child nodes to the parent node in parallel.

[0055] Furthermore, the lower-level information processing unit 121 may use a sample configuration (Figure 6) as supplementary information to the configuration manual in order to improve the accuracy of the tree structure. The same applies to step S202.

[0056] Step S202; The lower-level information processing unit 121 uses existing natural language processing technology to create a tree structure of the entire command system contained in the lower-layer documents. For example, as shown in Figure 9, the configuration has a hierarchical structure such as Layer 1 - Layer 2 - Layer 3, so the lower-level information processing unit 121 links the information of the upper layers and the information of the lower layers together to create a tree structure. In doing so, it starts from the highest level and creates a tree structure for each depth of all layers.

[0057] Step S203; The lower-level information processing unit 121 uses an existing encoder to convert the tree-structured command into a semantic vector corresponding to the purpose of the setting and the purpose of the argument. In the above example, the lower-level information processing unit 121 converts "ip-address" (purpose) - "<IPv6 Address_own> It generates a semantic vector of "" (first argument)" and ""ip-address" (purpose) - "<IPv6 Prefix> This generates a semantic vector for "(second argument)".

[0058] Furthermore, the lower-level information processing unit 121 also generates semantic vectors subdivided according to the number of arguments for a given command, such as all arguments, only the third argument, or only the fourth argument. This improves the accuracy of linking configurations with different command and argument combinations, which can occur between different vendors.

[0059] Step S204; The lower-level information processing unit 121 stores the purpose-specific semantic vectors in the vector information storage DB, linking them to the purpose of the command and the purpose of the arguments of the semantic vectors. Figure 10(a) is an example of command vector information when there are two arguments. Figure 10(b) is an example of command vector information when there is one argument.

[0060] Step S205; The lower-level information processing unit 121 reads the semantic vectors of all commands from the vector information storage DB and links commands with high cosine similarity between their semantic vectors (creates a graph structure). Whether or not the cosine similarity is high is determined, for example, using a threshold. After that, the lower-level information processing unit 121 stores the command linking information in the lower-level information structured vector DB 122.

[0061] Figure 11 shows a graph structure image of the linked commands. For example, the IP address setting command for vendor A is linked to the IP address setting command for vendor B and the vendor-independent OpenConfig IP address setting command.

[0062] [Examples of Functions and Operations of the One-to-One Conversion Unit] Figure 12 shows examples of functions and operations of the one-to-one conversion unit 13. The one-to-one conversion unit 13 includes a conversion rule generation unit 131 that generates conversion rules, and a conversion rule management DB 132 that stores and manages the generated conversion rules.

[0063] Step S301; The conversion rule generation unit 131 accepts the specified command for the desired purpose of the source and the vendor to be converted.

[0064] Step S302; The conversion rule generation unit 131 obtains commands from the vector information storage DB of the lower-level information processing unit 121 that match the desired command of the specified source and the vendor to be converted. The conversion rule generation unit 131 searches the lower-level information structured vector DB 122 for commands associated with the obtained commands and obtains them from the vector information storage DB of the lower-level information processing unit 121.

[0065] Step S303: The conversion rule generation unit 131 generates a conversion rule that associates the vendor command specified in step S301 with the vendor command obtained in step S302, and stores the conversion rule in the conversion rule DB 132.

[0066] Step S304; The user terminal 2 uses the conversion rules stored in the conversion rule DB 132 to convert a command from a predetermined vendor to a command from another vendor. For example, it generates commands and configurations from vendor A for vendor B.

[0067] Let's explain a specific example of step S303.

[0068] When the conversion rule generation unit 131 receives the IP address setting command shown in the upper part (input) of Figure 13 and vendor A as input, it obtains the IP address setting command of vendor B shown in the lower part (output) of the same figure as a command associated with the command of vendor A.

[0069] The conversion rule generation unit 131 generates a conversion rule that converts a command for setting an IP address for vendor A to a command for vendor B, as shown in Figure 14. In other words, the conversion rule generation unit 131 uses the source information to search for an element that matches "original" and obtains the "converts" portion as the destination, thereby generating a conversion rule that performs a one-to-one conversion. Information about the source and destination is stored in "meta_info". This is used to facilitate modifications to the conversion rule if any changes are made.

[0070] The arguments should be in the form of substituting values ​​from the source information. If the matching result of the possible values ​​(thresholds) for the arguments indicates that there are constraints on the possible values, a warning message will be sent to the user.

[0071] Furthermore, since vector information for not only the entire command and its arguments, but also parts of the command and its arguments are stored in the vector information storage database, it is possible to support vendors with different command counts. It is also possible to support vendors with different hierarchies and configuration systems.

[0072] [First Case and Effects] To address the first case, in this embodiment, the lower-level information conversion unit 12 converts each different command to a device node having the same role contained in documents of multiple lower layers into a semantic vector corresponding to the purpose of the command and the purpose of the argument, links similar commands together based on the cosine similarity between the semantic vectors, and the one-to-one conversion unit 13 searches for other commands linked to a predetermined command using the linking information linked by the lower-level information conversion unit 12, and generates the predetermined command and the other commands as a conversion rule for the setting method. This makes it possible to address the first case and supports the development of a control device.

[0073] [Second Case and Effects] Figure 15 shows an example of the second case.

[0074] The higher-level information conversion unit 11 has a function to store the linking information between texts in the higher-level information structured vector DB 112 so that a third party can refer to it.

[0075] The lower-level information conversion unit 12 has a function to store information linking setting methods (e.g., configuration, commands) in the lower-level information structure vector DB 122 so that a third party can refer to it.

[0076] Step S401; When the AI ​​device 3 is given an instruction to search for documents in higher layers or lower layers, it searches for information associated with the question (e.g., text, configuration, command) from the higher-level information structure vector DB 112 and the lower-level information structure vector DB 122.

[0077] Subsequently, the AI ​​device 3 uses the retrieved information as search terms to obtain the necessary information from the document database.

[0078] Thus, in order to address the second case, in this embodiment, the higher-level information conversion unit 11 stores the linking information between texts in the higher-level information structured vector DB 112 so that a third party can refer to it, and the lower-level information conversion unit 12 stores the linking information between setting methods in the lower-level information structured vector DB 122 so that a third party can refer to it. This makes it possible to address the second case and supports the development of the control device.

[0079] By pre-structuring the contextual connections across multiple documents, the accuracy of context-aware vector searches is improved, ultimately contributing to improved RAG performance. It can handle both upper-layer and lower-layer documents.

[0080] [Examples of Functions and Operations of the Conversion Support Unit] Figure 16 shows examples of functions and operations of the conversion support unit 14. The conversion support unit 14 includes a conversion processing unit 141 that converts information from upper layer documents and information from lower layer documents into templates for setting up device nodes for specific applications, and a device node template management DB 142 that stores and manages these templates for setting up device nodes for specific applications.

[0081] Step S501; The conversion processing unit 141 receives request information for setting a desired role for a predetermined device node.

[0082] Request information is information specified by arbitrarily combining various setting methods stored in the vector information storage DB and the upper-level information structured vector DB 112 of the upper-level information processing unit 111, and the vector information storage DB and the lower-level information structured vector DB 122 of the lower-level information processing unit 121.

[0083] For example, the request information might be, "We want to define the role of PE (Provider Edge) for a device in a node-to-node communication method called SR-MPLS (Segment Routing - Multi Protocol Label Switching)." A keyword alone, such as "SR-MPLS PE," would also suffice.

[0084] Step S502; The conversion processing unit 141 uses an existing encoder to divide the above request information into semantic units and converts all the divided text into semantic vectors corresponding to the content of the text.

[0085] Step S503; The conversion processing unit 141 uses the semantic vector to search the higher-level information structure vector DB 112 for text associated with the text and extracts the top n (n is a natural number and can be changed) higher-layer documents. For example, the conversion processing unit 141 extracts higher-layer documents that are similar to the semantic vectors of the words "SR-MPLS" and "PE".

[0086] Specifically, in the higher-level information structured vector DB 112 shown in Figure 5, each text node is associated with metadata (document name, text information, page, etc.) in the JSON structure shown in Figure 4 and vector information. Therefore, the conversion processing unit 141 extracts a set of metadata (information indicating which page and which part of the document in the higher layer the text belongs to) associated with the semantic vector converted in step S502.

[0087] Step S504; The conversion processing unit 141 converts the extracted top n upper layer documents (metainformation) into semantic vectors corresponding to their semantic content, searches the lower information structure vector DB 122 for commands associated with those documents (metainformation), and extracts the command with the highest similarity.

[0088] Specifically, in the lower-level information structured vector DB122 shown in Figure 11, each command node is associated with metadata and vector information in the JSON structure shown in Figure 10. Therefore, the conversion processing unit 141 extracts a set of lower-level metadata (information indicating which page and part of the lower-level document the command belongs to) that is associated with the semantic vector of the upper-level metadata.

[0089] In this case, if the vendor, device type, OS type, and OS version to be converted are specified, the conversion processing unit 141 extracts commands from the lower-layer documents of the specified vendor to be converted. If these are not specified and a common format such as OpenConfig is assumed, the conversion processing unit 141 extracts commands from all lower-layer documents.

[0090] In this way, the similarity between the semantic vector of the upper-layer document (metadata) and the semantic vector of the lower-layer command is calculated, and the command with the highest similarity is extracted. This allows us to link what you want to do as described in the upper-layer document with what you can do as described in the lower-layer document.

[0091] Step S505; Depending on the vendor and OS, the number of commands and the order in which the commands are entered may be important when completing the specified settings. For example, when setting an IP address, "ip-address<IPv6 Address_own> " and "ip-address<IPv6 Prefix> The two commands, "[command 1]" and "[command 2]", must be entered in that order. Meanwhile, in step S504, multiple commands are selected in a random order.

[0092] Therefore, the conversion processing unit 141 obtains a tree structure describing the configuration system between the vendor to be converted and the OS, and rearranges the extracted results of multiple commands. In other words, the conversion processing unit 141 rearranges the commands, which are subtrees, based on the overall tree of the overall command system (the command grammar tree) that was calculated in step S202.

[0093] Specifically, a one-dimensional array list is created by arranging the entire tree shown in Figure 17 in a depth-first order from the top level, and then subtrees 1 to 3 of each command are rearranged in order of the top-level node of each subtree based on this one-dimensional array list.

[0094] Subsequently, the conversion processing unit 141 stores the rearranged commands in the application-specific device node template management DB 142 as command templates for the application-specific device nodes.

[0095] Step S506; The user terminal 2 uses templates stored in the application-specific device node template management DB 142 to generate commands for a predetermined vendor of a device node for a predetermined application. For example, it generates commands and configurations for vendor A of a device node for application X.

[0096] [Third Case and Effects] To address the third case, in this embodiment, the conversion support unit 14 receives information for setting a desired role in a predetermined device, searches for text associated with the desired role using the text linking information and the setting method linking information, searches for a command similar to the text, and uses the command as a template for the command in the application-specific device node. This makes it possible to address the third case and to support the development of the control device.

[0097] [Other] This disclosure is not limited to the embodiments described above. This disclosure can be modified in numerous ways within the scope of the gist of this disclosure.

[0098] The support device 1 of this embodiment described above can be realized using a general-purpose computer system, for example, as shown in Figure 18, which includes a CPU 901, memory 902, storage 903, communication device 904, input device 905, and output device 906.

[0099] Memory 902 and storage 903 are storage devices. In this computer system, each function of the support device 1 is realized when the CPU 901 executes a predetermined program loaded onto memory 902.

[0100] Support device 1 may be implemented on a single computer. Support device 1 may be implemented on multiple computers. Support device 1 may be a virtual machine implemented on a computer.

[0101] The program for support device 1 can be stored on a computer-readable recording medium such as an HDD, SSD, USB memory, CD, or DVD. A computer-readable recording medium is, for example, a non-transitory recording medium. The program for support device 1 can also be distributed via a communication network.

[0102] 1. Support Device 11. Higher-level information conversion unit 111. Higher-level information processing unit 112. Higher-level information structured vector DB 12. Lower-level information conversion unit 121. Lower-level information processing unit 122. Lower-level information structured vector DB 13. One-to-one conversion unit 131. Conversion rule generation unit 132. Conversion rule management DB 14. Conversion support unit 141. Conversion processing unit 142. Application-specific device node template management DB 2. User terminal 3. AI device 901. CPU 902. Memory 903. Storage 904. Communication device 905. Input device 906. Output device

Claims

1. A support device comprising: a first conversion unit that converts different setting methods for devices having the same role contained in multiple first documents into vectors according to the purpose of setting, and links similar setting methods together based on the similarity between the vectors; and a first generation unit that searches for a second setting method linked to the first setting method using the linked linking information, and generates the first setting method and the second setting method as a setting method conversion rule.

2. The support device according to claim 1, further comprising a second conversion unit that converts each paragraph text or each semantic unit text within a paragraph contained in a plurality of second documents related to each of the plurality of first documents into vectors corresponding to the content of the text, and links similar texts together based on the similarity between the vectors, wherein the first conversion unit stores the linking information between the setting methods in a first storage unit so that a third party can refer to it, and the second conversion unit stores the linking information between the texts in a second storage unit so that a third party can refer to it.

3. The support device according to claim 2, further comprising a second generation unit that receives information for setting a desired role in a predetermined device, searches for text associated with the desired role using the text linkage information and the setting method linkage information, searches for a setting method similar to the text, and uses the setting method as a template for a setting method for a device of a specific application that performs the desired role.

4. The support device according to claim 1, wherein the first conversion unit converts each of the setting methods into vectors corresponding to the purpose of the setting and the purpose of the argument.

Images