Business standard content generation method and device, equipment and storage medium

Abstract

The invention belongs to the technical field of computer software, is applied to the field of medical health and the field of financial science and technology, and discloses a business standard content generation method, device and equipment and a storage medium. Analyzing the business related content to obtain a generation intention of a user for the business standard content; matching a target template corresponding to the generation intention in a preset template library; if the target template exists in the preset template library, generating the business standard content according to the target template and the generation intention; and if the target template does not exist in the preset template library, inputting the generation intention into a preset large language model to obtain output business standard content. The technical problem that in the prior art, when the business standard is generated, the generation efficiency and the generation accuracy are low is solved.

Description

Technical Field

[0001] The present invention belongs to the field of computer software technology and is applied to the fields of medical health and financial technology. In particular, it relates to a method, device, equipment and storage medium for generating business standard content. Background Art

[0002] In various industries, business standards are often defined to ensure the quality and speed of business operations. For example, in the insurance industry, various clauses (such as the reimbursement ratio for critical illness insurance in the healthcare industry or the compensation ratio for commercial insurance in the financial insurance sector) need to be stipulated to ensure the normal operation of insurance business.

[0003] However, managing and updating business standards is a time-consuming and complex task, often requiring manual compilation and verification. However, the volume of data involved in business standards is enormous, making manual processing extremely inefficient.

[0004] In this regard, in order to improve the efficiency of business standard management and updating in the existing technology, when generating business standards, the generation requirements corresponding to the business standards are first obtained, and the generation requirements are decomposed into keywords, and the keywords are filled in the format template corresponding to the business standards.

[0005] However, when generating requirements for business standards, the providers of these requirements vary widely, and some may be users who lack business knowledge. Directly extracting keywords from these user-provided requirements and incorporating them into a format template results in the generated business standards being unsuitable for a broad range of businesses and only catering to specific customer needs. This approach significantly reduces the efficiency of generating a large number of business standards. Furthermore, when using format templates to generate business standards, they essentially only change the format of the user-provided requirements and do not perform any additional operations on them. This clearly cannot guarantee the accuracy of the generated business standards.

[0006] Therefore, when generating business standards in the prior art, both the generation efficiency and the generation accuracy are low. Summary of the Invention

[0007] The present invention provides a method, device, equipment and storage medium for generating business standard content, so as to solve the technical problem of low generation efficiency and low generation accuracy in the prior art when generating business standards.

[0008] In a first aspect, the present invention provides a method for generating business standard content, comprising:

[0009] Obtain business-related content entered by the user;

[0010] Analyze the business-related content to obtain the user's generation intention for business standard content;

[0011] Matching a target template corresponding to the generation intention in a preset template library;

[0012] If the target template exists in the preset template library, generating the business standard content according to the target template and the generation intention;

[0013] If the target template does not exist in the preset template library, the generation intention is input into the preset large language model to obtain the output business standard content.

[0014] In a second aspect, the present invention provides a device for generating business standard content, comprising:

[0015] The acquisition module is used to obtain business-related content input by the user;

[0016] An analysis module, configured to analyze the business-related content and obtain the user's generation intention for the business standard content;

[0017] A matching module, configured to match a target template corresponding to the generation intention in a preset template library;

[0018] A first generating module is configured to generate the business standard content according to the target template and the generating intention if the target template exists in the preset template library;

[0019] The second generation module is used to input the generation intention into a preset large language model to obtain output business standard content if the target template does not exist in the preset template library.

[0020] In a third aspect, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method for generating the above-mentioned business standard content when executing the computer program.

[0021] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for generating the above-mentioned business standard content are implemented.

[0022] The solution implemented by the above-mentioned method, device, equipment and storage medium for generating business standard content, when obtaining business standard content, if the generation intent corresponding to the business standard content is relatively complex, the corresponding template may not be found in the preset template library. For this reason, the generation intent can be input into the large language model to obtain the business standard content. Obviously, compared with the prior art of directly inputting complex generation intent into the format template and changing the format of the generation intent to directly obtain the business standard content, this can effectively improve the accuracy of the generated business standard content.

[0023] Furthermore, for less complex generation intents, directly generating business-standard content using target templates from the preset template library is still sufficiently accurate. Furthermore, using templates from the preset template library to generate business-standard content can offload some of the generation of business-standard content, making it easier to generate business-standard content for large language models, thereby reducing their complexity and consuming fewer resources.

[0024] Furthermore, in this application, after obtaining the business-related content input by the user, the user's generation intent is analyzed, and then business standard content is generated based on the generation intent. It is understandable that multiple business-related contents may have the same generation intent, and the generation intent can represent the key content of a category of business-related content. Therefore, when generating business standard content based on the generation intent, it is possible to avoid generating corresponding business standard content for each business-related content, thereby effectively improving the generation efficiency of business standard content.

[0025] In summary, this solution can solve the problem of low generation efficiency and generation accuracy in the existing technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments of the present invention. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0027] Figure 1 1 is a flow chart of a method for generating business standard content in one embodiment of the present invention;

[0028] Figure 2 is another flowchart of a method for generating business standard content in one embodiment of the present invention;

[0029] Figure 3 yes Figure 2 A flow chart of step S155;

[0030] Figure 4 yes Figure 1 A flow chart of step S160;

[0031] Figure 5 It is a structural diagram of a device for generating business standard content in one embodiment of the present invention;

[0032] Figure 6 is a structural diagram of a computer device in one embodiment of the present invention;

[0033] Figure 7 FIG. 2 is another structural diagram of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0035] Figure 1 The flow chart of the method for generating business standard content provided by the embodiment of the present invention is as follows: Figure 1 As shown, the method for generating service standard content provided by the embodiment of the present invention includes the following steps.

[0036] Step S110: obtaining the business-related content input by the user.

[0037] Specifically, the business-related content in this step is any content related to the business field entered by the user. For example, if the business field is insurance, the business-related content entered by the user may include inquiries about the type of insurance, coverage, or claim conditions. As a specific example, in the field of medical and health insurance, the business-related content entered by the user may include the claim conditions for critical illness insurance. As another specific example, in the field of financial insurance, the business-related content entered by the user may include inquiries about the types of auto insurance.

[0038] Step S120: analyzing the business-related content to obtain the user's intention to generate business-standard content.

[0039] Specifically, in this step, business standard content refers to business-related content that complies with business standard rules. The user's generation intent for business standard content in this step represents the core intent expressed by the user. Multiple different business-related content may correspond to the same generation intent for business standard content. The so-called generation intent for business standard content refers to the form of the generated intent content complying with business standard rules.

[0040] For example, in the field of medical insurance and health, business-related content can be "How can I make a claim if I have stomach cancer?", and business-related content can also be "What are the claim conditions for stomach cancer?". It can be understood that the above two business-related contents both want to know the claim conditions for cancer. In the business standards of the insurance industry, cancer is a critical illness insurance, and the claim conditions for cancer are the claim conditions for critical illness insurance in the business standard rules. Therefore, the generation intention of the above two business-related contents for the business standard content is "the claim conditions for critical illness insurance."

[0041] Step S130: Match a target template corresponding to the generation intention in a preset template library.

[0042] Specifically, in this step, the templates in the preset template library are pre-set and verified templates, which can be set according to business standard rules. For example, in the field of financial insurance, for the business standard rule of "Claim conditions for motor vehicle loss insurance", the corresponding template can include two parts: insurance type and claim conditions. These two parts can be set with corresponding placeholders in the template. In this way, after obtaining the user's generation intention for the business standard content, the generation intention is analyzed to see whether it includes insurance type and claim conditions. If included, the above-mentioned template can be its corresponding target template.

[0043] Step S140: If the target template exists in the preset template library, the business standard content is generated according to the target template and the generation intention.

[0044] As a specific example, according to the example of step S130, the specific contents of the insurance types and claim conditions included in the generation intention can be filled into the corresponding placeholders in the template to form a complete template.

[0045] In some embodiments of the present invention, Figure 2 As shown, before step S160, the following steps are also included.

[0046] Step S151: Acquire sample data of the business domain corresponding to the business-related content.

[0047] Specifically, in this step, the sample data may include business specifications, business standard content, business-related content, the generation intent corresponding to the business-related content, and the corresponding relationship between the generation intent and the business standard content. In other words, the sample data is all the learning data used by the large language model in the process of outputting the business standard content after learning the input generation intent. For example, if the business domain corresponding to the business-related content is medical health insurance, the corresponding sample data to be acquired is sample data in the medical health insurance field.

[0048] Step S152: input the sample data into a preset teacher model for training to obtain a trained teacher model.

[0049] In this step, the sample data is fed into a pre-set teacher model for training. This is the process by which the teacher model learns from the sample data. The teacher model is typically a high-performance and complex model that can fully learn the connections between the sample data.

[0050] Step S153: Migrate the content of the trained teacher model to the preset student model.

[0051] Specifically, the learning model is a model with low complexity and low resource requirements. In this way, the teacher model's ample resources can be used to fully learn the sample data, thereby achieving accurate prediction of the input. It is understandable that the knowledge learned by the teacher model is relatively accurate knowledge. At this time, the knowledge of the teacher model is transferred to the preset student model. Naturally, the knowledge in the student model is also relatively accurate knowledge. In this way, when the student model is used to predict business standard content, the predicted results are also more accurate. Moreover, because the student model is less complex and requires fewer resources than the teacher model, it can predict faster when making predictions.

[0052] Step S154: input the sample data into the preset student model to obtain the business standard content corresponding to the sample data.

[0053] Specifically, in this step, the sample data of the preset student model is input as the generation intention corresponding to the business-related content, and then the business standard content corresponding to the generation intention predicted by the student model can be output through the student model.

[0054] Step S155: Calculate the prediction error of the preset student model according to the business standard content.

[0055] Specifically, in this step, the method for calculating the prediction error of the student model can be any feasible method. For example, an error function can be set, and the prediction error of the student model can be calculated based on the error function and the predicted business standard content.

[0056] Step S156: When the prediction error is less than a preset error value, the preset student model is determined as the preset large language model.

[0057] In some embodiments of the present invention, Figure 3 As shown, step S155 includes the following steps.

[0058] Step S1551, calculating a first prediction error value between the business standard content output by the student model and the business standard content output by the teacher model;

[0059] Step S1552 , calculating a second prediction error value between the business standard content output by the student model and the historical business standard content in the sample data.

[0060] The prediction error includes the first prediction error value and the second prediction error value.

[0061] Specifically, in step S1551, the business standard content output by the student model and the business standard content output by the teacher model are obtained for the same input generation intention. In this way, the calculated first prediction error value can determine whether the student model has fully transferred the knowledge learned by the teacher model. Since the knowledge learned by the teacher model is the basis for the student model's prediction, the more comprehensive the knowledge of the teacher model acquired by the student model, the stronger its prediction accuracy for the output. The second prediction error value is the error between the business standard content output by the student model and the historical business standard content, that is, the error between the business standard content obtained by the student model for the generation intention prediction and the actual business standard content corresponding to the generation intention. It can be understood that the smaller the second prediction error value, the stronger the prediction accuracy of the student model. In summary, this embodiment calculates two prediction error values. When both prediction error values are small, it can ensure that the student model can fully learn the knowledge learned by the teacher model and can also make the student model's prediction accuracy stronger.

[0062] In some embodiments of the present invention, before step S160, the method further includes:

[0063] The hyperparameters of the preset large language model are updated based on a preset Bayesian optimization algorithm.

[0064] Specifically, hyperparameters can be continuous parameters, discrete parameters, and categorical parameters. For example, a continuous parameter can be a learning rate, a discrete parameter can be a batch, and a categorical parameter can be an activation function.

[0065] Step S160: If the target template does not exist in the preset template library, the generation intention is input into a preset large language model to obtain output business standard content.

[0066] Specifically, for more complex generation intents, it is not possible to directly fill the target template with the relevant content from the generation intent to obtain complete business standard content. In this case, to ensure sufficiently accurate business standard content is output, the generation intent can be input into a preset large language model to output sufficiently accurate business standard content.

[0067] In this way, when the embodiment of the present invention obtains the business standard content, if the generation intent corresponding to the business standard content is relatively complex, the corresponding template may not be found in the preset template library. For this reason, the generation intent can be input into the large language model to obtain the business standard content. Obviously, compared with the prior art of directly inputting the complex generation intent into the format template and changing the format of the generation intent to directly obtain the business standard content, this can effectively improve the accuracy of the generated business standard content.

[0068] Furthermore, for less complex generation intents, directly generating business-standard content using target templates from the preset template library is still sufficiently accurate. Furthermore, using templates from the preset template library to generate business-standard content can offload some of the generation of business-standard content, making it easier to generate business-standard content for large language models, thereby reducing their complexity and consuming fewer resources.

[0069] Furthermore, in this application, after obtaining the business-related content input by the user, the user's generation intent is analyzed, and then business standard content is generated based on the generation intent. It is understandable that multiple business-related contents may have the same generation intent, and the generation intent can represent the key content of a category of business-related content. Therefore, when generating business standard content based on the generation intent, it is possible to avoid generating corresponding business standard content for each business-related content, thereby effectively improving the generation efficiency of business standard content.

[0070] In summary, this solution can solve the problem of low generation efficiency and generation accuracy in the existing technology.

[0071] In some embodiments of the present invention, Figure 4 As shown, step S160 includes the following steps.

[0072] Step S161: input the generation intention into the input layer of a preset large language model to obtain preprocessed generation intention data.

[0073] Specifically, in this step, the generated intent can be input into the input layer of the large language model. This input layer then preprocesses the generated intent to obtain a feature vector corresponding to the generated intent. For example, preprocessing for the generated intent can include parsing and cleaning the input data, such as using regular expressions to remove special symbols, identifying illegal words using a domain dictionary, or extracting key elements through named entity recognition. The preprocessed generated intent data can then be converted into a feature vector.

[0074] Step S162: input the preprocessed generation intention data into the encoding layer of the preset large language model to obtain the encoding vector corresponding to the generation intention data.

[0075] Among them, the encoding layer integrates the business domain knowledge corresponding to the generated intention data.

[0076] Specifically, in this step, the business domain knowledge corresponding to the generated intent data, such as a knowledge graph, can be introduced at the coding layer. As a specific example, in the field of medical health insurance, the medical knowledge graph and the insurance knowledge graph can be injected into the coding layer. It can be understood that incorporating business domain knowledge into the coding layer can enable the model to have a better understanding of the relevant business domain, so that it can have a deeper understanding of the semantics and purpose corresponding to the generated intent data, thereby making the business standard content corresponding to the generated intent data obtained in the final prediction more accurate.

[0077] Step S163: input the encoding vector into the decoding layer of the preset large language model to obtain the decoded content of the encoding vector.

[0078] Wherein, the coding layer is provided with a business standard constraint layer of the business field corresponding to the generation intention data.

[0079] Specifically, in this step, the business standard constraint layer is used to constrain the final generated business standard content to meet the final generation purpose.

[0080] For example, each business domain has its own set of illegal terms. To address this, the business standard constraint layer can construct a domain-allowed vocabulary to limit the output space of business standard content, preventing illegal terms from appearing in it. Another example is the introduction of a reward function to measure the compliance of generated content. For example, in the healthcare and insurance sector, the accuracy of medical terminology in decoded content must be sufficient, while in the financial and insurance sector, the data must conform to the conventional logic of financial data.

[0081] Step S164: input the decoded content of the encoding vector into the output layer of the preset large language model to obtain the business standard content corresponding to the generated intention data.

[0082] It is understood that in the embodiments of the present invention, the integration of the business domain knowledge corresponding to the generated intent data at the encoding layer can enable the language model to have a deeper understanding of the relevant business domain, thereby enabling a more accurate understanding of the generated intent data, and naturally making the business standard content ultimately obtained based on this more accurate. In addition, introducing a business standard constraint layer for the business domain corresponding to the generated intent data at the decoding layer can ensure that the content ultimately output by the decoding layer conforms to the standards of the business domain corresponding to the generated intent data, thereby making the ultimately output business standard content more accurate.

[0083] In some embodiments of the present invention, the fine-tuning process of the preset large language model includes the following steps:

[0084] Obtain historical business standard content;

[0085] Analyze the historical business standard content to obtain the actual business scenario corresponding to the historical business standard content;

[0086] Annotate the actual business scenarios corresponding to the historical business standard content on the historical business standard content to form model training samples;

[0087] Freeze the basic language capability layer parameters in the preset large language model to obtain an updated large language model.

[0088] Specifically, in this embodiment, the large language model can include multiple layers, with the basic language capability layer responsible for processing general language knowledge, such as grammar and vocabulary. Once the large language model is established, the parameters of these layers have been pre-trained on a large amount of text data, and already possess good language understanding and generation capabilities.

[0089] Specifically, freezing the parameters of the basic language capability layer can be achieved by setting the trainable attribute of the parameter. In this way, during the model training process, the trainable attribute of the parameter is set to non-trainable, so that the parameter will not be updated during model training.

[0090] It should be noted that, excluding the basic language capability layer parameters, other capability layer parameters of the large language model are level parameters related to specific services.

[0091] Therefore, it is understandable that the basic language capability layer of the large language model has been pre-trained on a large amount of general text and has already acquired mature basic capabilities such as grammatical understanding, semantic representation, and logical reasoning. Freezing the parameters of this layer can prevent these core capabilities from being damaged during fine-tuning, ensuring that the model maintains fluent language expression and basic logic when handling business tasks. Fine-tuning the business-related layers can significantly reduce the amount of computation and time without updating the entire model parameters. Fine-tuning the business-related layers can enable the large language model to accurately predict business-related business standard content.

[0092] In some embodiments of the present invention, the number of frozen basic language capability layers in the large language model is proportional to the number of model training samples.

[0093] Understandably, when a large number of domain-specific training samples are available (e.g., tens of thousands of historical business standard content), the model already has sufficient data to learn the rules, terminology, and logic of the specific domain. Freezing more foundational language capability layers at this time can prevent degradation of these capabilities and prevent the model from over-modifying underlying language understanding capabilities (e.g., grammar and semantic representation) during fine-tuning, thereby ensuring the fundamental fluency and logic of the generated content. Fine-tuning can focus on high-level parameters to quickly learn domain-specific business standards, format constraints, and scenario logic (e.g., compliance statements in financial reports, terminology standards in medical documents). If domain samples are relatively small (e.g., only a few thousand pieces of data), the model will struggle to learn complex domain knowledge from just a small amount of data. Unfreezing more foundational layers at this time allows the large model to leverage foundational capabilities such as semantic understanding and contextual associations learned from massive amounts of general text to compensate for the lack of domain data. By adjusting more underlying parameters, the model can flexibly combine general language capabilities with limited domain data, avoiding the rigidity of the model caused by excessive parameter freezing and its inability to capture domain characteristics.

[0094] It should be understood that the order of execution of the steps in the above embodiments does not necessarily imply a specific order of execution. The order of execution of each process should be determined by its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The software tools or components not provided by our company that appear in the embodiments of this application are merely examples and do not represent actual use.

[0095] In one embodiment, a device for generating business standard content is provided, which corresponds to the method for generating business standard content in the above embodiment. Figure 5 As shown, the generating device includes an acquisition module 510, an analysis module 520, a matching module 530, a first generating module 540, and a second generating module 550. The functional modules are described in detail as follows:

[0096] Acquisition module 510, used to acquire business-related content input by the user;

[0097] An analysis module 520 is configured to analyze the business-related content to obtain the user's intention to generate business-standard content;

[0098] A matching module 530 is configured to match a target template corresponding to the generation intention in a preset template library;

[0099] A first generating module 540 is configured to generate the business standard content according to the target template and the generating intention if the target template exists in the preset template library;

[0100] The second generation module 550 is configured to input the generation intention into a preset large language model to obtain output business standard content if the target template does not exist in the preset template library.

[0101] In one embodiment, the fine-tuning process of the preset large language model includes the following steps:

[0102] Obtain historical business standard content;

[0103] Analyze the historical business standard content to obtain the actual business scenario corresponding to the historical business standard content;

[0104] Annotate the actual business scenarios corresponding to the historical business standard content on the historical business standard content to form model training samples;

[0105] Freeze the basic language capability layer parameters in the preset large language model to obtain an updated large language model;

[0106] Inputting the model training samples into the updated large language model to obtain the predicted business scenario corresponding to the historical business standard content;

[0107] Calculating a scenario error value between the predicted business scenario and the actual business scenario;

[0108] When the scene error value exceeds a preset error range, the parameters of the preset large language model are adjusted until the scene error value is within the preset error range.

[0109] In one embodiment, the number of frozen basic language capability layers in the large language model is proportional to the number of model training samples.

[0110] In one embodiment, the second generating module 550 is specifically configured to:

[0111] Inputting the generation intention into the input layer of a preset large language model to obtain preprocessed generation intention data;

[0112] Input the preprocessed generation intent data into the encoding layer of a preset large language model to obtain an encoding vector corresponding to the generation intent data, wherein the encoding layer incorporates the business domain knowledge corresponding to the generation intent data;

[0113] Inputting the encoding vector into a decoding layer of a preset large language model to obtain decoded content of the encoding vector, wherein the encoding layer is provided with a business standard constraint layer for the business field corresponding to the generation intention data;

[0114] The decoded content of the encoding vector is input into the output layer of the preset large language model to obtain the business standard content corresponding to the generated intention data.

[0115] In one embodiment, the second generating module 550 is further configured to:

[0116] The hyperparameters of the preset large language model are updated based on a preset Bayesian optimization algorithm.

[0117] In one embodiment, the second generating module 550 is further configured to:

[0118] Obtain sample data of the business domain corresponding to the business-related content;

[0119] Input the sample data into the preset teacher model for training to obtain the trained teacher model;

[0120] Migrating the content of the trained teacher model to the preset student model;

[0121] Inputting the sample data into the preset student model to obtain the business standard content corresponding to the sample data;

[0122] Calculating the prediction error of the preset student model according to the business standard content;

[0123] When the prediction error is less than a preset error value, the preset student model is determined as the preset large language model.

[0124] In one embodiment, the second generating module 550 is further configured to:

[0125] Calculating a first prediction error value between the business standard content output by the student model and the business standard content output by the teacher model;

[0126] Calculate a second prediction error value between the business standard content output by the student model and the historical business standard content in the sample data, wherein the prediction error includes the first prediction error value and the second prediction error value.

[0127] The present invention provides a device for generating business standard content. When obtaining business standard content, if the generation intent corresponding to the business standard content is relatively complex, the corresponding template may not be found in the preset template library. For this reason, the generation intent can be input into a large language model to obtain the business standard content. Obviously, compared with the prior art of directly inputting complex generation intent into the format template and changing the format of the generation intent to directly obtain the business standard content, this can effectively improve the accuracy of the generated business standard content.

[0128] Furthermore, for less complex generation intents, directly generating business-standard content using target templates from the preset template library is still sufficiently accurate. Furthermore, using templates from the preset template library to generate business-standard content can offload some of the generation of business-standard content, making it easier to generate business-standard content for large language models, thereby reducing their complexity and consuming fewer resources.

[0129] Furthermore, in this application, after obtaining the business-related content input by the user, the user's generation intent is analyzed, and then business standard content is generated based on the generation intent. It is understandable that multiple business-related contents may have the same generation intent, and the generation intent can represent the key content of a category of business-related content. Therefore, when generating business standard content based on the generation intent, it is possible to avoid generating corresponding business standard content for each business-related content, thereby effectively improving the generation efficiency of business standard content.

[0130] In summary, this solution can solve the problem of low generation efficiency and generation accuracy in the existing technology.

[0131] The specific definition of the business standard content generation device can be found in the definition of the business standard content generation method above and will not be repeated here. The various modules in the above-mentioned business standard content generation device can be implemented in whole or in part through software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor of the computer device in hardware form, or can be stored in the memory of the computer device in software form, so that the processor can call and execute the corresponding operations of each of the above modules.

[0132] Based on the above business standard content generation method, such as Figure 6 As shown, an embodiment of the present invention further provides a schematic structural diagram of a device for generating business standard content. The device includes a processor 61 and a memory 62 coupled to the processor 61. The memory 62 stores a computer program. When the computer program is executed by the processor 61, the processor 61 performs the steps of the method for generating business standard content in the above embodiment.

[0133] For other details about how the processor 61 in the above-mentioned service standard content generation device implements the above-mentioned technical solution, please refer to the description of the service standard content generation method provided in the above-mentioned invention embodiment, which will not be repeated here.

[0134] Among them, the processor 61 can also be called a CPU (Central Processing Unit), and the processor 61 may be an integrated circuit chip with signal processing capabilities; the processor 61 can also be a general-purpose processor, DSP (Digital Signal Process), ASIC (Application Specific Integrated Circuit), FPGA (Field Programmable Gate Array) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, among which the general-purpose processor can be a microprocessor or the processor 61 can also be any conventional processor, etc.

[0135] like Figure 7 As shown, an embodiment of the present invention further provides a schematic diagram of the structure of a computer-readable storage medium, on which a readable computer program 71 is stored; wherein, the computer program 71 can be stored in the above-mentioned storage medium in the form of a software product, including a number of instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) or a processor to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a magnetic disk or an optical disk, a ROM (Read-Only Memory), a RAM (Random Access Memory), and other media that can store program code, or a terminal device such as a computer, server, mobile phone, or tablet.

[0136] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is merely a logical function division. In actual implementation, there may be other division methods, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or module, which can be electrical, mechanical or other forms.

[0137] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules may be selected to achieve the purpose of the present embodiment according to actual needs.

[0138] In addition, the functional modules in the various embodiments of the present invention may be integrated into a single processing module, each module may exist physically separately, or two or more modules may be integrated into a single module. The integrated modules may be implemented in the form of hardware or software functional modules. If the integrated modules are implemented in the form of software functional modules and sold or used as independent products, they may be stored in a computer-readable storage medium.

[0139] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware, or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0140] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center via a wired (e.g., coaxial cable, optical fiber, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) method. The computer-readable storage medium can be any available medium that a computer can store or a data storage device such as a server or data center that includes one or more available media integrated. The available medium can be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium), or a semiconductor medium (e.g., an SSD (solid state disk)).

[0141] The technical solution provided by the present invention is introduced in detail above. Specific examples are used in the present invention to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

[0142] It will be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0143] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0144] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0146] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. A method for generating business standard content, characterized in that: include: Obtain business-related content entered by the user; Analyze the business-related content to obtain the user's generation intention for business standard content; Matching a target template corresponding to the generation intention in a preset template library; If the target template exists in the preset template library, generating the business standard content according to the target template and the generation intention; If the target template does not exist in the preset template library, the generation intention is input into the preset large language model to obtain the output business standard content.

2. The method for generating business standard content according to claim 1, characterized in that: The fine-tuning process of the preset large language model includes the following steps: Obtain historical business standard content; Analyze the historical business standard content to obtain the actual business scenario corresponding to the historical business standard content; Annotate the actual business scenarios corresponding to the historical business standard content on the historical business standard content to form model training samples; Freeze the basic language capability layer parameters in the preset large language model to obtain an updated large language model; Inputting the model training samples into the updated large language model to obtain the predicted business scenario corresponding to the historical business standard content; Calculating a scenario error value between the predicted business scenario and the actual business scenario; When the scene error value exceeds a preset error range, the parameters of the preset large language model are adjusted until the scene error value is within the preset error range.

3. The method for generating business standard content according to claim 2, characterized in that: The number of frozen basic language capability layers in the large language model is proportional to the number of model training samples.

4. The method for generating business standard content according to claim 1, characterized in that: If the target template does not exist in the preset template library, the generation intention is input into the preset large language model to obtain output business standard content, including: Inputting the generation intention into the input layer of a preset large language model to obtain preprocessed generation intention data; Input the preprocessed generation intent data into the encoding layer of a preset large language model to obtain an encoding vector corresponding to the generation intent data, wherein the encoding layer incorporates the business domain knowledge corresponding to the generation intent data; Inputting the encoding vector into a decoding layer of a preset large language model to obtain decoded content of the encoding vector, wherein the encoding layer is provided with a business standard constraint layer for the business field corresponding to the generation intention data; The decoded content of the encoding vector is input into the output layer of the preset large language model to obtain the business standard content corresponding to the generated intention data.

5. The method for generating business standard content according to claim 1, characterized in that: Before inputting the generation intention into a preset large language model to obtain outputted business standard content if the target template does not exist in the preset template library, the method further includes: The hyperparameters of the preset large language model are updated based on a preset Bayesian optimization algorithm.

6. The method for generating business standard content according to claim 1, characterized in that: Before inputting the generation intention into a preset large language model to obtain outputted business standard content if the target template does not exist in the preset template library, the method further includes: Obtain sample data of the business domain corresponding to the business-related content; Input the sample data into the preset teacher model for training to obtain the trained teacher model; Migrating the content of the trained teacher model to the preset student model; Inputting the sample data into the preset student model to obtain the business standard content corresponding to the sample data; Calculating the prediction error of the preset student model according to the business standard content; When the prediction error is less than a preset error value, the preset student model is determined as the preset large language model.

7. The method for generating business standard content according to claim 6, characterized in that: The step of calculating the prediction error of the preset student model according to the business standard content includes: Calculating a first prediction error value between the business standard content output by the student model and the business standard content output by the teacher model; Calculate a second prediction error value between the business standard content output by the student model and the historical business standard content in the sample data, wherein the prediction error includes the first prediction error value and the second prediction error value.

8. A device for generating business standard content, characterized in that: include: The acquisition module is used to obtain business-related content input by the user; An analysis module, configured to analyze the business-related content and obtain the user's generation intention for the business standard content; A matching module, configured to match a target template corresponding to the generation intention in a preset template library; A first generating module is configured to generate the business standard content according to the target template and the generating intention if the target template exists in the preset template library; The second generation module is used to input the generation intention into a preset large language model to obtain output business standard content if the target template does not exist in the preset template library.

9. A computer 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 steps of the method for generating service standard content according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method for generating service standard content according to any one of claims 1 to 7 are implemented.

Images