Knowledge graph-based Hash government affair service vertical domain large model construction method and system

Abstract

The invention provides a knowledge graph-based Hash government affair service vertical domain large model construction method and system, and belongs to the technical field of artificial intelligence and government affair services. The method comprises the following steps: S100, analyzing an original query demand of a user to obtain a plurality of query keywords; s200, determining a reference large model; s300, outputting a plurality of extended query requirements by the reference large model, wherein the extended query requirements are visual knowledge maps taking query keywords as root nodes; s400, determining at least one vertical domain corpus in response to the selection of the user for the expansion query demand, the vertical domain corpus comprising a plurality of government affair service corpora of different modalities; and S500, loading the vertical domain corpus, and training the reference large model to obtain a trained vertical domain large model. The system comprises a plurality of candidate reference large models and a plurality of candidate vertical domain corpora, and is used for realizing the method. According to the method, the government affair service vertical domain large model can be quickly and effectively constructed and is easy to use.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence and government service technology, and in particular, relates to a method and system for constructing a vertical domain big model of Xiyan government service based on a knowledge graph, a computer-readable storage medium, a computer program product and an electronic device for implementing the method. Background Art

[0002] Vertical domain big model, also known as industry big model, professional / field big model, is the abbreviation of vertical field big model. It is relative to general big model, and is a big model application dedicated to a certain field or industry. General big models are usually trained on a wide range of public data sets, which enables them to learn and solve various common problems, but these data sets cannot fully cover all the professional knowledge and technical details in certain specific fields. This results in that although general big models have a wide range of general knowledge, they lack sufficient knowledge depth to meet the complex needs of certain specific fields. Therefore, it becomes particularly important to build vertical field big models according to the needs of specific industries.

[0003] After searching, it is known that Chinese invention patent CN112100344B discloses a knowledge question-answering method in the financial field based on knowledge graph. Through neural network training model and similarity comparison adaptive parameter adjustment, the accuracy of the question-answering system is effectively improved; related literature ("Customization of vertical domain big model: theoretical basis and key technology", Chen Haolong et al., Data Collection and Processing, 2024, 39(03)) also has some introduction and reports on related theories and technologies. However, reports on the specific use of vertical domain big models for government services are relatively rare.

[0004] After searching, the relevant literature ("Application Status and Opportunities and Challenges of Government Big Models", Li Wei et al. Communication Enterprise Management, 2024, (06): 42-44.) also introduces the basic status and principles of general big models for government services. There is no clear technical solution for how to train and construct vertical domain big models when they are used for government services. Summary of the invention

[0005] In response to the above technical problems, the present invention proposes a method and system for constructing a vertical domain big model of Xiyan government services based on knowledge graph, a computer-readable storage medium, a computer program product and an electronic device for implementing the method.

[0006] In the first aspect of the present invention, a method for constructing a vertical domain big model of Xiyan government service based on a knowledge graph is proposed, and the method comprises the following steps:

[0007] S100: Analyze the user's original query requirements to obtain multiple query keywords;

[0008] The original query requirements include text query requirements, voice query requirements, and image query requirements. S200: Determine a benchmark large model based on the query keywords;

[0009] S300: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements;

[0010] S400: In response to a user's selection of the extended query requirement, determining at least one vertical domain corpus;

[0011] S500: Load the vertical domain corpus, train the benchmark large model, and obtain a trained vertical domain large model;

[0012] The vertical domain corpus includes a plurality of government service corpora in different modalities, wherein the different modalities include one of text, static images, video frames, and speech sequences, or any combination thereof;

[0013] The multiple extended query requirements are a visualized knowledge graph with the query keyword as the root node.

[0014] When the method is actually applied, after step S500, the method further includes:

[0015] S600: Input the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs the government service results in at least one modal form according to the current attributes of the user terminal.

[0016] The benchmark big model is a general industry big model, which includes a text-to-text model, a prompt word-class text-to-graph model, and a graph-to-graph model.

[0017] The step S500 loads the vertical domain corpus and trains the benchmark large model to obtain a trained vertical domain large model, specifically including:

[0018] Adopting a retrieval enhancement generation method, the vertical domain corpus is used as a supplementary training library and plug-in query library for the benchmark large model;

[0019] Adjusting the training parameters of the benchmark large model based on the supplementary training library;

[0020] The trained vertical domain large model is only queried and retrieved in the external query library.

[0021] The step S600 inputs the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs at least one modal form of government service results according to the current attributes of the user terminal, specifically including:

[0022] S610: Generate multiple domain prompt words based on the original query requirement;

[0023] S620: Integrate the domain prompt word into the original query requirement as additional context information and use it together with the original query requirement as input of the vertical domain macro model;

[0024] S630: The vertical domain big model is searched in the vertical domain corpus;

[0025] S640: When the search results contain multiple modalities, output government service results in at least one modality according to the current attributes of the user terminal.

[0026] Corresponding to the method for constructing a vertical domain big model of Xiyan government service based on knowledge graph described in the first aspect, in the second aspect of the present invention, a system for constructing a vertical domain big model of Xiyan government service based on knowledge graph is proposed, the system comprising multiple candidate benchmark big models and multiple candidate vertical domain corpora, the multiple candidate benchmark big models comprising text-generated text models, prompt word class text-generated graph models, and graph-generated graph class models; the multiple candidate vertical domain corpora comprise government service corpora of multiple different modalities, the different modalities comprising one of text, static images, video frames, and speech sequences, or any combination thereof;

[0027] The system further comprises:

[0028] Analysis unit: analyzes the user's original query requirements to obtain multiple query keywords;

[0029] A benchmark large model determining unit: determining a benchmark large model based on the query keyword;

[0030] A knowledge graph expansion unit: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements; the plurality of extended query requirements are visual knowledge graphs with the query keyword as the root node;

[0031] A vertical domain corpus determining unit: in response to a user's selection of the extended query requirement, determining at least one vertical domain corpus;

[0032] Training unit: loads the vertical domain corpus, trains the benchmark large model, and obtains a trained vertical domain large model.

[0033] In a specific application, the system further includes:

[0034] Input unit: The user inputs the original query requirement through the input unit, and the original query requirement includes a text query requirement, a voice query requirement, and an image query requirement.

[0035] Result output unit: inputs the user's original query requirements into the vertical domain big model, and the vertical domain big model outputs government service results in at least one modal form according to the current attributes of the user terminal.

[0036] The aforementioned method for constructing the vertical domain big model of Xiyan government service based on knowledge graph can be automatically implemented through various forms of electronic devices through computer program instructions; the computer program instructions can be stored in different forms of storage media and loaded into computer electronic devices for execution.

[0037] Therefore, in the third aspect of the present invention, a computer-readable storage medium is also provided for storing computer instructions. When the computer instructions are executed on an electronic device, the electronic device executes all or part of the steps of the aforementioned method for building a vertical domain big model of Xiyan government services based on knowledge graph.

[0038] In the fourth aspect of the present invention, a computer device is also proposed, which includes a processor and a memory, the memory is used to store instructions, and the processor is used to call the instructions in the memory, so that the computer device executes the aforementioned knowledge graph-based Xiyan government service vertical domain big model construction method.

[0039] In the fifth aspect of the present invention, a computer program product is also proposed, which includes a computer program. When the computer program is executed, all or part of the steps of the aforementioned method for building a vertical domain big model of Xiyan government services based on knowledge graph are implemented.

[0040] The present invention first determines a benchmark big model through query keywords, and then outputs multiple extended query requirements through the benchmark big model; then, in response to the user's selection of the extended query requirements, at least one vertical domain corpus is determined, and then the vertical domain corpus is loaded to train the benchmark big model to obtain a trained vertical domain big model, which can quickly and effectively construct a vertical domain big model for government services and is easy to use.

[0041] Further advantages of the present invention will be further reflected in detail in the specific embodiments section in conjunction with the drawings of the specification. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. 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 creative work.

[0043] Figure 1This is a main flow chart of a method for constructing a vertical domain big model of Xiyan government service based on knowledge graph in one embodiment of the present invention.

[0044] Figure 2 This is a main flow chart of a method for constructing a vertical domain big model of Xiyan government service based on a knowledge graph according to another embodiment of the present invention.

[0045] Figure 3 This is a data analysis diagram of the present invention analyzing the user's original query requirements to obtain multiple query keywords

[0046] Figure 4 This is a schematic diagram of the knowledge graph of the present invention outputting multiple extended query requirements

[0047] Figure 5 This is a schematic diagram of a visualized knowledge graph with a query keyword as the root node in a specific application example of the present invention.

[0048] Figure 6 This is a structural diagram of a system for building a vertical domain big model of Xiyan government service based on a knowledge graph according to an embodiment of the present invention.

[0049] Figure 7 This is a schematic diagram of the partial interaction structure between the input unit and the output unit of the Xiyan government service vertical domain big model construction system based on the knowledge graph DETAILED DESCRIPTION

[0050] The embodiments of this section adopt a step-by-step approach. First, the application status and implementation effect of the Xiyan government service vertical domain big model proposed by the applicant are introduced, and then the basic construction principle of the vertical domain big model is introduced, that is, the relevant method embodiments and the relevant schematic principles of applying the method; finally, the relevant system embodiments corresponding to the method embodiments are introduced.

[0051] It can be understood that the method embodiment and the system embodiment should correspond to each other in terms of execution process and execution principle, and the corresponding functional modules and method steps should correspond to each other. Therefore, after the method embodiment is introduced in detail, the system embodiment does not need to be introduced repeatedly.

[0052] The technical solution of the present invention comes from the key R&D project undertaken by the applicant's related units - research and demonstration of technology based on the construction of a large language model for government services. The "Xiyan Big Model" developed based on the project has successfully passed the "double filing" of generative artificial intelligence (large language model) and deep synthesis service algorithm. It integrates advanced AI capabilities such as multi-round dialogue, planning suggestions, content generation, etc., and provides customized solutions for government services, housing provident funds, education and other fields.

[0053] Corresponding to the above-mentioned industrial model, the present invention proposes the following relevant technical solutions for the corresponding vertical large-scale model construction method and system.

[0054] See first Figure 1 , Figure 1 It is the main flow chart of a method for constructing a vertical domain big model of Xiyan government service based on knowledge graph in one embodiment of the present invention.

[0055] exist Figure 1 The method includes steps S100-S500, and each step is specifically implemented as follows:

[0056] S100: Analyze the user's original query requirements to obtain multiple query keywords;

[0057] S200: Determine a benchmark macro model based on the query keyword;

[0058] S300: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements;

[0059] S400: In response to a user's selection of the extended query requirement, determining at least one vertical domain corpus;

[0060] S500: Load the vertical domain corpus, train the benchmark large model, and obtain a trained vertical domain large model.

[0061] After the vertical domain big model is constructed in step S500, the method can be specifically applied to the current query, that is, the vertical domain big model of this embodiment realizes dynamic construction of query-ready and each query and each call.

[0062] For details, see Figure 2 In an embodiment, after step S500, the method further includes:

[0063] S600: Input the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs the government service results in at least one modal form according to the current attributes of the user terminal.

[0064] In the above embodiment, the vertical domain corpus includes a plurality of government service corpora in different modalities, and the different modalities include one of text, static images, video frames, and speech sequences, or any combination thereof;

[0065] Next, combine Figure 3-Figure 5 ,right Figure 1 or Figure 2 Each step of the above method embodiment is specifically introduced.

[0066] Step S100: Analyze the user's original query requirements to obtain multiple query keywords.

[0067] Figure 3It is a data analysis schematic diagram of the present invention analyzing the original query demand of the user to obtain multiple query keywords.

[0068] In a specific application, the original query requirements include text query requirements, voice query requirements, and image query requirements.

[0069] like Figure 3 Illustrative examples of original query requirements include:

[0070] Receive user's voice input: "Process and certification materials for handling XXX";

[0071] Receive scanned copies of documents uploaded by users, that is, users upload scanned copies of the completed XXX forms through terminal devices, and users need to be informed of the next submission window;

[0072] Receive text description words directly entered by the user: the process and supporting materials for handling XXX.

[0073] After receiving the user's original query demand, semantic analysis is performed on it, including voice semantic analysis, text semantic analysis, and image semantic analysis, etc., to obtain multiple query keywords.

[0074] Figure 3 Only one keyword XXX obtained through analysis is shown as an example. It can be understood that other keywords, such as "text, picture, window, proof", etc., can be obtained based on the above example. Figure 3 Only one XXX is shown as an example.

[0075] Next, step S200 is performed: determining a benchmark macro model based on the query keyword;

[0076] The benchmark big model is an industry-wide big model, and the industry-wide big model includes a text-generated-text industry-wide big model, a prompt-word text-generated-image industry-wide big model, and a picture-generated-image industry-wide big model.

[0077] The industry-wide big model can be a generative AI engine that is well known to those skilled in the art. Generative AI can use techniques such as neural networks and deep learning to identify patterns and structures in existing data to generate new original content. One of the breakthroughs of generative AI models is the ability to train using different learning methods, including unsupervised or semi-supervised learning. This makes it easier and faster for organizations to create base models using large amounts of unlabeled data. As the name suggests, base models can serve as the foundation for AI systems that can perform multiple tasks. Examples of base models include GPT-3 and stable diffusion, which allow users to harness the power of language. For example, popular applications such as ChatGPT (which draws on GPT-3) allow users to generate an article based on a short text request. On the other hand, stable diffusion allows users to generate realistic images based on text input.

[0078] A typical example of a common large model in the literary and artistic industry is the artificial intelligence engine supported by GPT (including GPT1-GPT3) technology. The most typical example is ChatGPT (released by OpenAI on November 3, 2022, and is constantly evolving).

[0079] The general large model of the prompt word category text image industry uses the text actively input by the user as the prompt word and intelligently generates the image output result. The most typical example is the Midjourney tool released by David Holz in March 2022. In addition, the DALL E2 tool released by OpenAI in January 2022 also has similar functions.

[0080] The general large model of the image generation industry uses the partial image description text entered by the user and a picture as the prompt word input to intelligently generate preliminary design pictures. The most typical examples are the Stabile Diffusion tool released by Stabilty AI in August 2022 and the FUgenerator tool released by Tongji CAUP in March 2023.

[0081] It should be noted that the above three different types of industry-wide big models are constantly evolving, and the later-stage industry-wide big models usually have the functions of the previous-stage industry-wide big models, but they all have different focuses. For example, the natural language-driven text-based industry-wide big model (such as the evolved ChatGPT) is mainly good at text-based text, but also has the primary text-based graph function; the prompt word text-based graph industry-wide big model has both text-based text and text-based graph functions; the graph-based graph industry-wide big model can also have text-based text, text-based graph, and graph-based graph functions at the same time.

[0082] In various embodiments of the present invention, when it comes to the general big model of the relevant industry, its improved functional parts are mainly used. For example, for the general big model of the prompt word category text-generated graph industry, its text-generated graph function is mainly used, and for the general big model of the graph-generated graph industry, its partial functions of the graph-generated graph are mainly used, that is, the advantages of each different type of industry general big model are fully utilized.

[0083] In step S200, at least one industry macro model may be determined as a benchmark macro model based on the query keyword;

[0084] For example, if the query keywords involve image input or output, the prompt word text-to-image model or the image-to-image model is preferably used as the benchmark large model; if the query keywords only involve or the user terminal attributes only support text input and output, then only the natural language driven text-to-text model can be used as the benchmark large model.

[0085] The benchmark big model can be an open source big model or a commercial paid big model. Preferably, the present embodiment uses an open source big model to quickly start the subsequent training process. Of course, the use of the open source big model also needs to comply with the corresponding open source rules.

[0086] Step S300: input the query keyword into the benchmark large model, and the benchmark large model outputs multiple extended query requirements.

[0087] The multiple extended query requirements are a visualized knowledge graph with the query keyword as the root node.

[0088] For details, see Figure 4 , Figure 4 It is a schematic diagram of the knowledge graph outputting multiple extended query requirements of the present invention.

[0089] Taking the natural language driven text model as the benchmark big model as an example, the query keyword xxx is input into the benchmark big model, and the benchmark big model outputs multiple extended query requirements, including:

[0090] XXX-AXXX;XXX-BXXX;XXX-CXXX;XXX-DXXX-dXX1;

[0091] To describe the above process more intuitively, jump to Figure 5 , Figure 5 A schematic diagram of a visualized knowledge graph with a query keyword as the root node in a specific application example of the present invention.

[0092] exist Figure 5 In the example, the user's query keyword includes "education". After entering it into the benchmark model, multiple extended query requirements are obtained, including:

[0093] Education-Preschool Education-Private Schools;

[0094] Education-Private education;

[0095] Education - junior high school education;

[0096] Education-Preschool-Kindergarten;

[0097] …

[0098] It can be seen that multiple extended query requirements are visualized knowledge graphs with the query keywords as root nodes.

[0099] Next, step S400 determines at least one vertical domain corpus in response to the user's selection of the expanded query requirement.

[0100] The vertical domain corpus includes a variety of government service corpora in a variety of different modalities, and the different modalities include one of text, static images, video frames, and voice sequences, or any combination thereof.

[0101] Continue to see Figure 5 The vertical domain corpus composed of various government service corpora includes:

[0102] Special column corpus, service corpus, convenient query corpus, interactive knowledge base, policy and regulation database, etc.

[0103] exist Figure 5 In the example, assuming that the extended query requirement selected by the user is the path "education-preschool education-kindergarten", at least one vertical domain database determined is the "interactive knowledge base".

[0104] Next, step S500 loads the vertical domain corpus, trains the benchmark large model, and obtains a trained vertical domain large model.

[0105] It can be understood that after determining the training corpus and query database, the training process of the vertical domain large model is basically similar to the training process of the ordinary (general) large model. Therefore, the specific training process can be referred to the prior art, and this embodiment will not be further elaborated on this.

[0106] As a part different from the prior art, this embodiment is further preferably as follows:

[0107] Specifically, the step S500 loads the vertical domain corpus, trains the benchmark large model, and obtains the trained vertical domain large model, which specifically includes:

[0108] Adopting a retrieval enhancement generation method, the vertical domain corpus is used as a supplementary training library and plug-in query library for the benchmark large model;

[0109] Adjusting the training parameters of the benchmark large model based on the supplementary training library;

[0110] The trained vertical domain large model is only queried and retrieved in the external query library.

[0111] Correspondingly, the step S600 inputs the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs at least one modal form of government service results according to the current attributes of the user terminal, specifically including:

[0112] S610: Generate multiple domain prompt words based on the original query requirement;

[0113] For example, the domain prompt words for preschool education can also be kindergarten, childcare, etc.;

[0114] S620: Integrate the domain prompt word into the original query requirement as additional context information and use it together with the original query requirement as input of the vertical domain macro model;

[0115] S630: The vertical domain big model is searched in the vertical domain corpus;

[0116] S640: When the search results contain multiple modalities, output government service results in at least one modality according to the current attributes of the user terminal.

[0117] The retrieval results described in step S640 include multiple modalities, which means that there are retrieval results with multiple optional outputs, for example, output results in video mode, output results in picture mode, and output results in text mode.

[0118] Nevertheless, the current properties of the user terminal may not support all modes. For example, the graphics card resources and GPU resources currently available to the user terminal are insufficient, resulting in the inability to play videos. Therefore, if the data results of the video mode are sent to the user terminal, it may cause freezes and bring a bad human-computer interaction experience to the user. In this case, the output results of this mode can be output according to the current properties of the user terminal; and when the relevant graphics card resources and GPU resources are sufficient, the output results of the video mode and the output results of the picture mode can be re-pushed.

[0119] Figure 6 It is a structural diagram of a system for building a vertical domain big model of Xiyan government service based on knowledge graph in one embodiment of the present invention.

[0120] exist Figure 6In the system, the system includes multiple candidate benchmark large models and multiple candidate vertical domain corpora, wherein the multiple candidate benchmark large models include natural language driven text-generated text models, prompt word-generated text-generated graph models, and graph-generated graph models; the multiple candidate vertical domain corpora include government service corpora of multiple different modalities, wherein the different modalities include one of text, static images, video frames, and speech sequences, or any combination thereof;

[0121] The system further comprises:

[0122] Analysis unit: analyzes the user's original query requirements to obtain multiple query keywords;

[0123] A benchmark large model determining unit: determining a benchmark large model based on the query keyword;

[0124] A knowledge graph expansion unit: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements; the plurality of extended query requirements are visual knowledge graphs with the query keyword as the root node;

[0125] A vertical domain corpus determining unit: in response to a user's selection of the extended query requirement, determining at least one vertical domain corpus;

[0126] Training unit: loads the vertical domain corpus, trains the benchmark large model, and obtains a trained vertical domain large model.

[0127] Figure 7 This is a schematic diagram of the partial interaction structure between the input unit and the output unit of the Xiyan government service vertical domain big model construction system based on the knowledge graph.

[0128] exist Figure 7 In the embodiment, it is further shown that the system further includes:

[0129] Input unit: The user inputs the original query requirement through the input unit, and the original query requirement includes a text query requirement, a voice query requirement, and an image query requirement.

[0130] Result output unit: inputs the user's original query requirements into the vertical domain big model, and the vertical domain big model outputs government service results in at least one modal form according to the current attributes of the user terminal.

[0131] Specifically, the result output unit is also connected to a user terminal attribute detection unit, which is used to detect the resource types currently available to the user terminal, and determine the output mode of the government service result based on the resource types currently available to the user terminal. The output mode includes one of text output, interactive animation output or picture / video output to meet user expectations and enhance user experience.

[0132] Other principles or implementation functions of the system embodiment can be found in the method embodiment and will not be described in detail.

[0133] For other technologies, principles, algorithms or models not elaborated in detail in this application, please refer to the prior art.

[0134] In the aforementioned embodiment section, the present invention provides multiple embodiments, each of which can constitute an independent technical solution and may contribute to the prior art and solve corresponding technical problems.

[0135] However, it should be pointed out that different embodiments can be combined with each other without violating logic; at the same time, each embodiment can solve at least one technical problem, but it is not required that each individual embodiment must solve multiple or all technical problems.

[0136] At the same time, in the specific implementation of this application, if user-related data is involved, when the embodiment of this application is applied to a specific product or technology, the user's permission or consent must be obtained, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0137] It can be seen that the present invention first determines the benchmark big model through query keywords, and then outputs multiple extended query requirements through the benchmark big model; then in response to the user's selection of the extended query requirements, at least one vertical domain corpus is determined, and then the vertical domain corpus is loaded to train the benchmark big model to obtain the trained vertical domain big model, which can quickly and effectively construct a vertical domain big model for government services and is easy to use.

[0138] The name of the large model of the technical solution of the present invention, "Xiyan", is taken from Laozi's "Tao Te Ching" "Xiyan is natural", which means that the words are simple and in line with nature. "Xiyan" is positioned as a vertical domain large model focusing on industry applications, which better understands user needs and government service knowledge, can better coordinate business, is more lightweight and flexible, and is more secure and reliable. At present, the "Xiyan large model" has been initially applied in government services, education, agriculture and other fields.

[0139] The knowledge graph-based Xiyan government service vertical domain big model proposed in this application is the first artificial intelligence big model in the applicant's region (autonomous region) (see the relevant report on People's Daily: http: / / nx.peop l e.com.cn / n2 / 2024 / 1104 / c192482-41030161.html). Practice has proved that this technical solution has obvious improvement effects and has received good social response.

[0140] The foregoing has shown and described the method embodiments and system of the present invention, but it is understandable to those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph, characterized in that: The method comprises the following steps: S100: Analyze the user's original query requirements to obtain multiple query keywords; S200: Determine a benchmark macro model based on the query keyword; S300: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements; S400: In response to a user's selection of the extended query requirement, determining at least one vertical domain corpus; S500: Load the vertical domain corpus, train the benchmark large model, and obtain a trained vertical domain large model; The vertical domain corpus includes a plurality of government service corpora in different modalities, wherein the different modalities include one of text, static images, video frames, and speech sequences, or any combination thereof; The multiple extended query requirements are a visualized knowledge graph with the query keyword as the root node.

2. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph as claimed in claim 1, characterized in that: After step S500, the method further includes: S600: Input the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs the government service results in at least one modal form according to the current attributes of the user terminal.

3. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph as claimed in claim 1, characterized in that: The benchmark big model is a general industry big model, which includes a text-to-text model, a prompt word-class text-to-graph model, and a graph-to-graph model.

4. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph as claimed in claim 1, characterized in that: The step S500 loads the vertical domain corpus and trains the benchmark large model to obtain a trained vertical domain large model, specifically including: Adopting a retrieval enhancement generation method, the vertical domain corpus is used as a supplementary training library and plug-in query library for the benchmark large model; Adjusting the training parameters of the benchmark large model based on the supplementary training library; The trained vertical domain large model is only queried and retrieved in the external query library.

5. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph as described in claim 2, characterized in that: The step S600 inputs the original query demand of the user into the vertical domain big model, and the vertical domain big model outputs at least one modal form of government service results according to the current attributes of the user terminal, specifically including: S610: Generate multiple domain prompt words based on the original query requirement; S620: Integrate the domain prompt word into the original query requirement as additional context information and use it together with the original query requirement as input of the vertical domain macro model; S630: The vertical domain big model is searched in the vertical domain corpus; S640: When the search results contain multiple modalities, output government service results in at least one modality according to the current attributes of the user terminal.

6. A method for constructing a vertical domain big model of Xiyan government service based on knowledge graph as described in any one of claims 1 to 5, characterized in that: The original query requirements include text query requirements, voice query requirements, and image query requirements.

7. A system for constructing a vertical domain big model of Xiyan government service based on knowledge graph, the system comprising multiple candidate benchmark big models and multiple candidate vertical domain corpora, the multiple candidate benchmark big models comprising text-generated text models, prompt word class text-generated graph models, and graph-generated graph class models; the multiple candidate vertical domain corpora comprise government service corpora of multiple different modalities, the different modalities comprising one of text, static images, video frames, and speech sequences, or any combination thereof; It is characterized in that The system further comprises: Analysis unit: analyzes the user's original query requirements to obtain multiple query keywords; A benchmark large model determining unit: determining a benchmark large model based on the query keyword; A knowledge graph expansion unit: inputting the query keyword into the benchmark large model, and the benchmark large model outputs a plurality of extended query requirements; the plurality of extended query requirements are visual knowledge graphs with the query keyword as the root node; A vertical domain corpus determining unit: in response to a user's selection of the extended query requirement, determining at least one vertical domain corpus; Training unit: loads the vertical domain corpus, trains the benchmark large model, and obtains a trained vertical domain large model.

8. A knowledge graph-based Xiyan government service vertical domain big model construction system as claimed in claim 7, characterized in that: The system further comprises: Result output unit: inputs the user's original query requirements into the vertical domain big model, and the vertical domain big model outputs government service results in at least one modal form according to the current attributes of the user terminal.

9. A knowledge graph-based Xiyan government service vertical domain big model construction system as described in claim 7, characterized in that: The system further comprises: Input unit: The user inputs the original query requirement through the input unit, and the original query requirement includes a text query requirement, a voice query requirement, and an image query requirement.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method for constructing a vertical domain big model of Xiyan government service based on a knowledge graph as described in any one of claims 1 to 6.

Images