A business automation processing method based on large language model and related devices

Abstract

The present invention provides a business automation processing method based on a large language model and related devices. The present invention relates to the field of business processing. The method includes: establishing a business knowledge graph based on business data stored in a business system; constructing a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, training the large language model, and obtaining a semantic recognition model; splicing user request information with historical conversations, and inputting the semantic recognition model and a graph neural network based on a business knowledge graph respectively to obtain semantic features and knowledge graph reasoning results; fusing the semantic features and knowledge graph reasoning results, classifying the fused data, and obtaining intent types; matching the intent types with the functional types of various business processes in the business system; if the intent type and the functional type are successfully matched, executing the business task corresponding to the user request information according to the business process corresponding to the functional type.

Description

Technical Field

[0001] The present invention relates to the field of business processing, and in particular to a business automation processing method based on a large language model and related devices. Background Art

[0002] In the application scenario where large language models are combined with business systems, the system registration information and knowledge reserves cannot be efficiently utilized when the business systems interact, resulting in inefficient business processing, inaccurate user intent recognition, and poor user experience, making it difficult to meet the needs of enterprise intelligent business development. Summary of the Invention

[0003] In order to address the deficiencies in the prior art, the present invention provides a business automation processing method based on a large language model to solve the problems of low business processing efficiency and inaccurate user intent recognition in the prior art.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions.

[0005] In a first aspect, the present application provides a method for automated business processing based on a large language model, comprising the following steps:

[0006] Through knowledge extraction technology, basic data is extracted from the business data stored in the business system, and a business knowledge graph is established based on the basic data. The business knowledge graph uses each link, each participating role and each data object in each business process in the business system as a node, and the relationship between each link, participating role and data object as an edge;

[0007] Building a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and training a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model;

[0008] Splicing user request information with historical conversations to obtain input data, and inputting the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0009] Fusing the semantic features with the knowledge graph reasoning results, and classifying the fused data through a multi-layer perception network to obtain the intent type corresponding to the user request information;

[0010] The intention type is matched with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

[0011] Optionally, the method further includes:

[0012] If the matching result indicates that the intent type and the function type fail to match, then based on the user request information and the business knowledge graph, a large language model is used to generate follow-up question information, and the follow-up question information is spliced ​​with the historical conversation to obtain follow-up question input data;

[0013] Inputting the question input data into the semantic recognition model and the graph neural network based on the business knowledge graph respectively, obtaining the question semantic features output by the semantic recognition model and the question reasoning results output by the graph neural network;

[0014] fusing the question semantic features and the question reasoning results to obtain question fusion data, and classifying the question fusion data through a multi-layer perception network to obtain a question intention type corresponding to the question information;

[0015] Matching the inquiry intention type with the function type of each business process in the business system; if the inquiry intention type successfully matches the function type, executing the business task corresponding to the inquiry information according to the business process corresponding to the function type;

[0016] If the question intention type fails to match the function type, the next round of question information is generated through a large language model based on the question information and the business knowledge graph until the question intention type successfully matches the function type.

[0017] Optionally, the step of generating follow-up information using a large language model based on the user request information and the business knowledge graph includes:

[0018] Performing entity extraction on the user request information to obtain entity information;

[0019] Determining, in the business knowledge graph, a node corresponding to the entity information and an association relationship corresponding to the entity information;

[0020] According to the node corresponding to the entity information and the association relationship corresponding to the entity information, follow-up question information is generated through a large language model.

[0021] Optionally, before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type, the method further includes:

[0022] Using the function type that successfully matches the intent type as the first information, using the entity information as the second information, and generating information to be confirmed based on the first information and the second information;

[0023] Displaying the information to be confirmed and obtaining user confirmation information for the information to be confirmed;

[0024] If the user confirmation information provides positive feedback for the information to be confirmed, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type;

[0025] If the user confirmation information provides negative feedback for the information to be confirmed, the matching result is modified to indicate that the intent type and the function type fail to match.

[0026] Optionally, the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type includes:

[0027] Classify the business tasks using a preset classification model to obtain a classification result;

[0028] If the classification result of the business task is a simple task, the FunctionCall mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type;

[0029] If the classification result of the business task is a complex task, the MCP mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type.

[0030] Optionally, the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type further includes:

[0031] Determining, according to the business process, execution parameter type parameter data and user information type parameter data required for executing the business task corresponding to the user request information, and calling the execution parameter type parameter data to obtain call feedback data;

[0032] If the call feedback data indicates a call failure, generating an alarm message based on the call feedback data, and sending the alarm message to a preset device;

[0033] If the call feedback data indicates that the call execution parameter type parameter data is successfully executed, obtaining the user identity information corresponding to the user request information, and searching the user information type parameter data in the business data stored in the business system based on the user identity information to obtain the search feedback data;

[0034] If the search feedback data indicates that user information type parameter data corresponding to the user identity information is found, then executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user information type parameter data corresponding to the user identity information;

[0035] If the search feedback data indicates that the user information type parameter data corresponding to the user identity information is not found, generating data request information based on the user information type parameter data, and displaying the data request information;

[0036] User supplementary information for the data request information is obtained, and a business task corresponding to the user request information is executed according to a business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information.

[0037] Optionally, before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information, the step further includes:

[0038] Acquire historical data of the user information type parameter data, and acquire first characteristic data of the historical data;

[0039] Acquire second characteristic data of the user request information, and match the first characteristic data with the second characteristic data;

[0040] If the first characteristic data matches the second characteristic data, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the call execution parameter type parameter data and the user supplementary information;

[0041] If the first characteristic data does not match the second characteristic data, the step of generating data request information based on the user information type parameter data and displaying the data request information is continued.

[0042] On the other hand, the present application provides a business automation processing system based on a large language model, the system comprising:

[0043] A graph generation module is used to extract basic data from the business data stored in the business system through knowledge extraction technology, and to establish a business knowledge graph based on the basic data. The business knowledge graph uses each link, each participating role, and each data object in each business process in the business system as a node, and the relationship between each link, participating role, and data object as an edge;

[0044] A model training module is configured to construct a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and to train a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model;

[0045] A computing module is configured to combine user request information with historical conversations to obtain input data, and input the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0046] An intent recognition module is used to fuse the semantic features with the knowledge graph reasoning results, classify the fused data through a multi-layer perception network, and obtain the intent type corresponding to the user request information;

[0047] An execution module is used to match the intention type with the function type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the function type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the function type.

[0048] In a third aspect, the present application also provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method for automated business handling based on a large language model as described above is implemented.

[0049] In a fourth aspect, the present application also provides a computer-readable storage medium, which stores a computer program, and is characterized in that when the computer program is executed by a processor, it implements the above-mentioned business automation processing method based on the large language model.

[0050] Beneficial effects:

[0051] This application uses knowledge extraction technology to extract basic data from the business data stored in the business system, and builds a business knowledge graph based on the basic data. The business knowledge graph uses each link, each participating role and each data object of each business process in the business system as a node, and the association relationship between each link, participating role and data object as an edge; constructs a training set based on the historical request information input into the business system and the semantic features corresponding to the historical request information, and trains the large language model according to the training set through the transfer learning method and the comparative learning method to obtain a semantic recognition model; splices the user request information with the historical conversation to obtain the input data, and converts the input The data are respectively input into the semantic recognition model and the graph neural network based on the business knowledge graph to obtain the semantic features output by the semantic recognition model and the knowledge graph reasoning results output by the graph neural network; the semantic features and the knowledge graph reasoning results are fused, and the fused data are classified through a multi-layer perception network to obtain the intent type corresponding to the user request information; the intent type is matched with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intent type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type. The business data stored in the business system is used to construct a knowledge graph, and the graph neural network is used to implement targeted business intent recognition, accurately execute various business processes, and improve business processing efficiency and user interaction experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is an operational flow chart of a method for automated business processing based on a large language model in an embodiment of the present invention;

[0053] Figure 2 Schematic diagram of the structure of a business automation processing system based on a large language model in an embodiment of the present invention;

[0054] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0055] Figure 4 It is a structural diagram of a computer-readable storage medium provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0057] The embodiments described in this application are only part of the embodiments of the present invention, not all of the embodiments. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0058] See also Figure 1 As shown, the present invention provides a method for automated business processing based on a large language model, comprising the following steps:

[0059] S110. Extract basic data from the business data stored in the business system using knowledge extraction technology, and establish a business knowledge graph based on the basic data. The business knowledge graph uses each link, each participating role, and each data object in each business process in the business system as a node, and the relationship between each link, participating role, and data object as an edge.

[0060] S120: constructing a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and training a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model;

[0061] S130: Concatenate user request information with historical conversations to obtain input data, and input the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0062] S140: Fusing the semantic features with the knowledge graph reasoning results, and classifying the fused data through a multi-layer perception network to obtain an intent type corresponding to the user request information;

[0063] S150. Match the intention type with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

[0064] In a possible implementation, the method further includes:

[0065] If the matching result indicates that the intent type and the function type fail to match, then based on the user request information and the business knowledge graph, a large language model is used to generate follow-up question information, and the follow-up question information is spliced ​​with the historical conversation to obtain follow-up question input data;

[0066] Inputting the question input data into the semantic recognition model and the graph neural network based on the business knowledge graph respectively, obtaining the question semantic features output by the semantic recognition model and the question reasoning results output by the graph neural network;

[0067] fusing the question semantic features and the question reasoning results to obtain question fusion data, and classifying the question fusion data through a multi-layer perception network to obtain a question intention type corresponding to the question information;

[0068] Matching the inquiry intention type with the function type of each business process in the business system; if the inquiry intention type successfully matches the function type, executing the business task corresponding to the inquiry information according to the business process corresponding to the function type;

[0069] If the question intention type fails to match the function type, the next round of question information is generated through a large language model based on the question information and the business knowledge graph until the question intention type successfully matches the function type.

[0070] Exemplarily, follow-up information is generated until the question intention type is successfully matched with the function type, so that the user needs can be clarified, that is, the user intention is accurately identified.

[0071] In one possible implementation, the step of generating follow-up information using a large language model based on the user request information and the business knowledge graph includes:

[0072] Performing entity extraction on the user request information to obtain entity information;

[0073] Determining, in the business knowledge graph, a node corresponding to the entity information and an association relationship corresponding to the entity information;

[0074] According to the node corresponding to the entity information and the association relationship corresponding to the entity information, follow-up question information is generated through a large language model.

[0075] For example, when the user inputs "My campus card is lost" (user request information), the entity is extracted as the campus card, and the node corresponding to the campus card is determined in the business knowledge graph, and the association relationship corresponding to the entity information is formed to form a guiding question and clarify the user needs.

[0076] In a possible implementation manner, before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type, the method further includes:

[0077] Using the function type that successfully matches the intent type as the first information, using the entity information as the second information, and generating information to be confirmed based on the first information and the second information;

[0078] Displaying the information to be confirmed and obtaining user confirmation information for the information to be confirmed;

[0079] If the user confirmation information provides positive feedback for the information to be confirmed, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type;

[0080] If the user confirmation information provides negative feedback for the information to be confirmed, the matching result is modified to indicate that the intent type and the function type fail to match.

[0081] Exemplarily, in order to ensure the reliability of business task execution, before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type, the user is required to confirm the function type that successfully matches the intent type. For example, when the user enters "My campus card is lost", "campus card" is used as the entity information, and the function type that successfully matches the intent type can be understood as the processing / reissue function type, then the pending confirmation information "Do you apply for a campus card?" is generated. The user replies with confirmation words such as yes / okay as user confirmation information, and the user confirmation information provides positive feedback for the pending confirmation information; conversely, the user replies with negative words such as no / no as user confirmation information, and the user confirmation information provides negative feedback for the pending confirmation information.

[0082] In a possible implementation, the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type includes:

[0083] Classify the business tasks using a preset classification model to obtain a classification result;

[0084] If the classification result of the business task is a simple task, the FunctionCall mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type;

[0085] If the classification result of the business task is a complex task, the MCP mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type.

[0086] For example, after a business process is triggered, the ReAct mechanism is used to reason and decompose the business task within the large language model. For simple tasks, the FunctionCall mechanism is directly used. The large language model generates function instructions that comply with tool call specifications, clearly defining input parameters and expected outputs for fast and accurate calls. For complex tasks, the Multi-Call Processing (MCP) mechanism is activated. The large language model uses dynamic prompt templates generated by MCP to perform multi-step reasoning and multi-tool combination call planning.

[0087] Exemplarily, a training set is composed of different business tasks and classification results of different business tasks, and the initial classification model is trained using the training set to obtain a preset classification model.

[0088] In a possible implementation, the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type further includes:

[0089] Determining, according to the business process, execution parameter type parameter data and user information type parameter data required for executing the business task corresponding to the user request information, and calling the execution parameter type parameter data to obtain call feedback data;

[0090] If the call feedback data indicates a call failure, generating an alarm message based on the call feedback data, and sending the alarm message to a preset device;

[0091] If the call feedback data indicates that the call execution parameter type parameter data is successfully executed, obtaining the user identity information corresponding to the user request information, and searching the user information type parameter data in the business data stored in the business system based on the user identity information to obtain the search feedback data;

[0092] If the search feedback data indicates that user information type parameter data corresponding to the user identity information is found, then executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user information type parameter data corresponding to the user identity information;

[0093] If the search feedback data indicates that the user information type parameter data corresponding to the user identity information is not found, generating data request information based on the user information type parameter data, and displaying the data request information;

[0094] User supplementary information for the data request information is obtained, and a business task corresponding to the user request information is executed according to a business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information.

[0095] For example, parameter data of the execution parameter type can be understood as tool call parameters, such as the parameter instructions required to call a tool to create a new page, or the size parameters of the created page. Parameter data of the user information type can be understood as parameters corresponding to the user's personal identity information, such as name, ID number, etc.

[0096] Specifically, in a business process for reissuing a campus card, the execution parameter type parameter data includes but is not limited to parameters for calling a blank campus card creation tool, and the user information type parameter data includes but is not limited to parameters such as the user's name and class.

[0097] Exemplarily, if the call feedback data indicates a call failure, an alarm message is generated based on the call feedback data and sent to a preset device, which is a device of a technician (business system maintenance personnel) and is used to promptly remind the technician (business system maintenance personnel) of a failure, thereby facilitating timely maintenance.

[0098] Exemplarily, if the search feedback data indicates that the user information type parameter data corresponding to the user identity information is not found, data request information is generated based on the user information type parameter data and displayed, thereby reducing the number of user operations and providing a better user experience.

[0099] In a possible implementation, before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information, the method further includes:

[0100] Acquire historical data of the user information type parameter data, and acquire first characteristic data of the historical data;

[0101] Acquire second characteristic data of the user request information, and match the first characteristic data with the second characteristic data;

[0102] If the first characteristic data matches the second characteristic data, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the call execution parameter type parameter data and the user supplementary information;

[0103] If the first characteristic data does not match the second characteristic data, the step of generating data request information based on the user information type parameter data and displaying the data request information is continued.

[0104] For example, feature matching can be used to avoid irrelevant information gaps in the execution of business tasks. For example, in the process of applying for a campus card, the user enters "application reason XXX", which is related to the business process; if the user fills in "help me write a poem about spring", the model will determine that the content is irrelevant to the business process and will prompt the user that the input content is irrelevant to the application process.

[0105] In a possible implementation, the method further includes:

[0106] Acquire historical data of the user information type parameter data, and use the data format of the historical data as the target data format;

[0107] The data format of the user supplementary information is converted into a target data format, and the business task corresponding to the user request information is executed according to the business process corresponding to the function type based on the user supplementary information in the target data format.

[0108] For example, different data formats can easily lead to misidentification, which in turn can easily cause incorrect execution of business tasks. Operations with a unified format can facilitate the rapid and correct execution of business tasks.

[0109] In a possible implementation, the method further includes:

[0110] The user supplementary information is verified by using preset verification data, and the verified user supplementary information is used to execute the business task corresponding to the user request information according to the business process corresponding to the function type.

[0111] For example, taking the ID card number as an example, the ID card number is numbered in the form of "area code + date of birth + code". Therefore, the preset verification data for the date of birth should be "not exceeding the current date (year, month, day), and the month is not more than 12, and the day is not more than 31", thereby ensuring the correctness of the user's supplementary information and avoiding the execution of incorrect tasks due to incorrect information.

[0112] On the other hand, Figure 2 As shown, the present application provides a business automation processing system based on a large language model, the system comprising:

[0113] A graph generation module 201 is used to extract basic data from the business data stored in the business system using knowledge extraction technology, and to establish a business knowledge graph based on the basic data. The business knowledge graph uses each link, each participating role, and each data object in each business process in the business system as a node, and the relationship between each link, participating role, and data object as an edge;

[0114] A model training module 202 is configured to construct a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and train a large language model based on the training set using a transfer learning method and a comparative learning method to obtain a semantic recognition model;

[0115] The computing module 203 is configured to combine user request information with historical conversations to obtain input data, and input the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0116] Intent recognition module 204 is used to fuse the semantic features with the knowledge graph reasoning results, classify the fused data through a multi-layer perception network, and obtain the intent type corresponding to the user request information;

[0117] The execution module 205 is used to match the intention type with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

[0118] In one possible implementation, Figure 3 As shown, an embodiment of the present application provides a terminal device 300, including: a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor 320. When the processor 320 executes the computer program 311, basic data is extracted from the business data stored in the business system through knowledge extraction technology, and a business knowledge graph is established based on the basic data. The business knowledge graph uses each link, each participating role, and each data object of each business process in the business system as a node, and the association relationship between each link, participating role, and data object as an edge;

[0119] Building a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and training a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model;

[0120] Splicing user request information with historical conversations to obtain input data, and inputting the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0121] Fusing the semantic features with the knowledge graph reasoning results, and classifying the fused data through a multi-layer perception network to obtain the intent type corresponding to the user request information;

[0122] The intention type is matched with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

[0123] In one possible implementation, Figure 4 As shown, an embodiment of the present application provides a computer-readable storage medium 400 having a computer program 411 stored thereon. When the computer program 411 is executed by a processor, basic data is extracted from the business data stored in the business system through knowledge extraction technology, and a business knowledge graph is established based on the basic data. The business knowledge graph has each link, each participating role, and each data object of each business process in the business system as a node, and the relationship between each link, participating role, and data object as an edge;

[0124] Building a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and training a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model;

[0125] Splicing user request information with historical conversations to obtain input data, and inputting the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network;

[0126] Fusing the semantic features with the knowledge graph reasoning results, and classifying the fused data through a multi-layer perception network to obtain the intent type corresponding to the user request information;

[0127] The intention type is matched with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

[0128] It should be noted that the information interaction, execution process, etc. between the above-mentioned devices / units are based on the same concept as the method embodiment of this application. Their specific functions and technical effects can be found in the method embodiment section and will not be repeated here.

[0129] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated into one processing unit, or each unit can exist physically alone, or two or more units can be integrated into one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0130] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by using a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. The computer-readable medium can include at least: any entity or device capable of carrying computer program code to a camera / terminal device, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, removable hard drives, magnetic disks, or optical disks. In some jurisdictions, based on legislation and patent practice, computer-readable media cannot be electric carrier signals or telecommunication signals.

[0131] In the above embodiments, the description of each embodiment has its own focus. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant description of other embodiments.

[0132] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0133] In the embodiments provided in this application, it should be understood that the disclosed devices / network equipment and methods can be implemented in other ways. For example, the device / network equipment embodiments described above are merely illustrative. For example, the division of the modules or units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0134] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0135] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present application, and should all be included in the scope of protection of the present application.

[0136] The above disclosure is merely a preferred embodiment of the present invention and certainly cannot be used to limit the scope of the present invention. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims

1. A business automation processing method based on a large language model, characterized in that: The following steps are involved: Through knowledge extraction technology, basic data is extracted from the business data stored in the business system, and a business knowledge graph is established based on the basic data. The business knowledge graph uses each link, each participating role and each data object in each business process in the business system as a node, and the relationship between each link, participating role and data object as an edge; Building a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and training a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model; Splicing user request information with historical conversations to obtain input data, and inputting the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network; Fusing the semantic features with the knowledge graph reasoning results, and classifying the fused data through a multi-layer perception network to obtain the intent type corresponding to the user request information; The intention type is matched with the functional type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the functional type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the functional type.

2. The method for automated business processing based on a large language model according to claim 1, characterized in that: The method further comprises: If the matching result indicates that the intent type and the function type fail to match, then based on the user request information and the business knowledge graph, a large language model is used to generate follow-up question information, and the follow-up question information is spliced ​​with the historical conversation to obtain follow-up question input data; Inputting the question input data into the semantic recognition model and the graph neural network based on the business knowledge graph respectively, obtaining the question semantic features output by the semantic recognition model and the question reasoning results output by the graph neural network; fusing the question semantic features and the question reasoning results to obtain question fusion data, and classifying the question fusion data through a multi-layer perception network to obtain a question intention type corresponding to the question information; Matching the inquiry intention type with the function type of each business process in the business system; if the inquiry intention type successfully matches the function type, executing the business task corresponding to the inquiry information according to the business process corresponding to the function type; If the question intention type fails to match the function type, the next round of question information is generated through a large language model based on the question information and the business knowledge graph until the question intention type successfully matches the function type.

3. The method for automated business processing based on a large language model according to claim 2, characterized in that: The step of generating follow-up information by using a large language model based on the user request information and the business knowledge graph includes: Performing entity extraction on the user request information to obtain entity information; Determining, in the business knowledge graph, a node corresponding to the entity information and an association relationship corresponding to the entity information; Generate follow-up question information through a large language model according to the node corresponding to the entity information and the association relationship corresponding to the entity information.

4. The method for automated business processing based on a large language model according to claim 3, characterized in that: Before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type, the method further includes: Using the function type that successfully matches the intent type as the first information, using the entity information as the second information, and generating information to be confirmed based on the first information and the second information; Displaying the information to be confirmed and obtaining user confirmation information for the information to be confirmed; If the user confirmation information provides positive feedback for the information to be confirmed, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type; If the user confirmation information provides negative feedback for the information to be confirmed, the matching result is modified to indicate that the intent type and the function type fail to match.

5. The method for automated business processing based on a large language model according to claim 1, characterized in that: The step of executing the business task corresponding to the user request information according to the business process corresponding to the function type includes: Classify the business tasks using a preset classification model to obtain a classification result; If the classification result of the business task is a simple task, the FunctionCall mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type; If the classification result of the business task is a complex task, the MCP mechanism is adopted to execute the business task corresponding to the user request information according to the business process corresponding to the function type.

6. The method for automated business processing based on a large language model according to claim 1, characterized in that: The step of executing the business task corresponding to the user request information according to the business process corresponding to the function type also includes: Determining, according to the business process, execution parameter type parameter data and user information type parameter data required for executing the business task corresponding to the user request information, and calling the execution parameter type parameter data to obtain call feedback data; If the call feedback data indicates a call failure, generating an alarm message based on the call feedback data, and sending the alarm message to a preset device; If the call feedback data indicates that the call execution parameter type parameter data is successfully executed, obtaining the user identity information corresponding to the user request information, and searching the user information type parameter data in the business data stored in the business system based on the user identity information to obtain the search feedback data; If the search feedback data indicates that user information type parameter data corresponding to the user identity information is found, then executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user information type parameter data corresponding to the user identity information; If the search feedback data indicates that the user information type parameter data corresponding to the user identity information is not found, generating data request information based on the user information type parameter data, and displaying the data request information; User supplementary information for the data request information is obtained, and a business task corresponding to the user request information is executed according to a business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information.

7. The method for automated business processing based on a large language model according to claim 6, characterized in that: Before the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the called execution parameter type parameter data and the user supplementary information, the method further includes: Acquire historical data of the user information type parameter data, and acquire first characteristic data of the historical data; Acquire second characteristic data of the user request information, and match the first characteristic data with the second characteristic data; If the first characteristic data matches the second characteristic data, then continuing to execute the step of executing the business task corresponding to the user request information according to the business process corresponding to the function type based on the call execution parameter type parameter data and the user supplementary information; If the first characteristic data does not match the second characteristic data, the step of generating data request information based on the user information type parameter data and displaying the data request information is continued.

8. A business automation processing system based on a large language model, characterized in that: The system comprises: A graph generation module is used to extract basic data from the business data stored in the business system through knowledge extraction technology, and to establish a business knowledge graph based on the basic data. The business knowledge graph uses each link, each participating role, and each data object in each business process in the business system as a node, and the relationship between each link, participating role, and data object as an edge; A model training module is configured to construct a training set based on historical request information input into the business system and semantic features corresponding to the historical request information, and to train a large language model based on the training set using a transfer learning method and a contrastive learning method to obtain a semantic recognition model; A computing module is configured to combine user request information with historical conversations to obtain input data, and input the input data into the semantic recognition model and the graph neural network based on the business knowledge graph, respectively, to obtain semantic features output by the semantic recognition model and knowledge graph inference results output by the graph neural network; An intent recognition module is used to fuse the semantic features with the knowledge graph reasoning results, classify the fused data through a multi-layer perception network, and obtain the intent type corresponding to the user request information; An execution module is used to match the intention type with the function type of each business process in the business system to obtain a matching result. If the matching result indicates that the intention type and the function type are successfully matched, the business task corresponding to the user request information is executed according to the business process corresponding to the function type.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the business automation processing method based on the large language model is implemented as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for automated business processing based on a large language model as described in any one of claims 1 to 7 is implemented.

Images