Generative intelligent large screen construction method and system based on natural language questions and answers

Abstract

The invention discloses a generation type intelligent large screen construction method and system based on natural language questions and answers, and belongs to the technical field of computer software, and the method comprises the steps: data preprocessing and knowledge base construction: constructing an interface metadata description system based on a data interface, generating a high-dimensional vector through large-scale pre-training, and constructing a knowledge base; dense distributed representation of text semantics is realized; performing question segmentation and extraction: performing automatic segmentation on an intelligent large screen generation question proposed by a user by adopting a generative pre-training model, and matching an interface corresponding to an optimal interface description based on a model reasoning capability; index data acquisition: performing iterative processing by adopting a polling mode, and acquiring index data according to the interface metadata information; and large screen code generation and analysis to realize code generation of an intelligent large screen. According to the method, the problems that traditional large screen development needs to be participated by professional technicians and the response period is long are solved, automatic analysis of multi-source heterogeneous data and visual code generation are achieved, and finally an intelligent large screen application capable of being directly deployed is generated.

Description

Technical Field

[0001] The present invention relates to the field of computer software technology, and more specifically to a method and system for constructing a generative intelligent large-screen based on natural language question and answer. Background Art

[0002] The current mainstream approach to developing large-scale data visualization screens requires professional technicians to write code or configure visualization tools. This approach suffers from the following drawbacks: 1. Long development cycles, requiring multiple steps such as requirements analysis, interface integration, and chart design; 2. High technical barriers to entry, making it difficult for non-professionals to operate; 3. Poor dynamic adaptability, making it difficult to quickly respond to changing business needs. Existing generative technologies primarily focus on text or image generation, lacking systematic solutions for large-scale data visualization screens. Summary of the Invention

[0003] The technical task of the present invention is to address the above shortcomings and provide a generative smart large-screen construction method and system based on natural language question and answer, which can solve the problem that traditional large-screen development requires the participation of professional and technical personnel and has a long response cycle, realize automatic parsing and visual code generation of multi-source heterogeneous data, and finally generate a smart large-screen application that can be directly deployed.

[0004] The technical solution adopted by the present invention to solve its technical problem is:

[0005] A method for constructing a generative intelligent large screen based on natural language question answering, the implementation of which includes:

[0006] Data preprocessing and knowledge base construction: Data preprocessing organizes the required indicator data for large-screen display in the form of interfaces, and builds an interface metadata description system based on the data interfaces. Knowledge base construction adopts vectorization, generating high-dimensional vectors (such as 3072 dimensions) through large-scale pre-training to achieve dense distributed representation of text semantics.

[0007] Question segmentation and extraction: The smart large screen generation questions raised by users are automatically segmented using a generative pre-trained model to generate sub-questions containing question identifiers and question descriptions, which are then formatted and processed to form a sub-question queue. The sub-question extraction method adopts a polling method. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-question from the knowledge base (the number of retrieved interface descriptions can be manually configured). The sub-questions and the matching interface description queue are then passed to the generative pre-trained model. Based on the model's reasoning capabilities, the interface corresponding to the optimal interface description is matched.

[0008] Indicator data acquisition: It uses polling to iterate and obtain indicator data based on interface metadata information; it supports calling the interface to obtain indicator data based on "interface input parameters" and obtains the optimal value of "interface input parameters" based on the generative pre-training model;

[0009] Large screen code generation and parsing: Supports passing the acquired indicator data, interface return value, interface description, and chart type fields in the interface metadata to the generative pre-trained model, and realizes code generation for the smart large screen according to the prompt word project; supports uploading the code to the server after formatting, and generates the smart large screen access address after configuring the proxy and port, realizing the automatic generation of the smart large screen.

[0010] Furthermore, the data preprocessing and knowledge base construction,

[0011] Build an interface metadata description system based on the data interface. The corresponding fields include: interface name, interface description, URL address, interface input parameter, interface return value, and corresponding chart type;

[0012] The knowledge base uses a vectorized database, supports efficient approximate nearest neighbor search, and can balance retrieval speed and accuracy; and supports manual configuration of retrieval boundary conditions and behavior patterns through thresholds.

[0013] Furthermore, the problem segmentation and extraction,

[0014] The generative pre-trained model used for automatic segmentation of original questions uses the Deepseek V3 instruction model; the configured prompt word project includes core tasks, output rules, atomization requirements, and output instance information;

[0015] The generative pre-trained model used to match the optimal interface description uses the DeepSeek R1 inference model; the configured prompt word project includes context data, processing steps, comprehensive matching degree, and output format information.

[0016] Furthermore, the data formatting of the sub-problem uses regular expressions.

[0017] Based on the configuration of data cleaning rules, abnormal data removal including title symbols, bold and italic symbols, link symbols, list symbols, reference symbols, etc. is achieved;

[0018] Furthermore, the hybrid retrieval method adopts a hybrid form of full-text retrieval and vector retrieval, supports the application of a re-ranking step, selects the best result matching the sub-question from the two types of query results, and supports the option of setting weights or configuring a re-ranking model.

[0019] Furthermore, the indicator data acquisition is specifically implemented as follows:

[0020] First, determine whether the "Interface Input Parameters" field in the interface metadata is empty. For interfaces whose "Interface Input Parameters" field is not empty, support passing the sub-problem corresponding to the interface and the JSON data corresponding to the "Interface Input Parameters" field to the generative pre-trained model, and match the optimal value corresponding to the "Interface Input Parameters" based on the model's reasoning ability. After the parameters are obtained, support calling the interface according to the URL address and interface input parameters in the interface metadata to obtain the indicator data required for the chart display on the large screen. For interfaces whose "Interface Input Parameters" field is empty, support calling the corresponding interface according to the URL address in the interface metadata information to obtain indicator data.

[0021] The generative pre-training model for obtaining the optimal value of the "interface input parameter" can adopt the DeepSeek R1 inference model; the configured prompt word project includes context data, extraction rules, value range verification, output format and other information.

[0022] Furthermore, the generative pre-training model for large-screen code adopts the DeepSeek V3 instruction model; the configured prompt word project includes large-screen title generation rules, chart type mapping relationships, code syntax rules, data format, legends and other information.

[0023] The present invention also claims protection for a generative intelligent large-screen construction system based on natural language question answering, comprising:

[0024] Data preprocessing and knowledge base construction module, used to build an interface metadata description system based on the data interface and achieve dense distributed representation of text semantics;

[0025] The problem segmentation and extraction module is used to generate sub-problems containing problem identifiers and problem descriptions, format them to form a sub-problem queue, and match the interface corresponding to the optimal interface description based on the model's reasoning capabilities;

[0026] The indicator data acquisition module is used to implement iterative processing in a polling manner and call the corresponding interface to obtain indicator data;

[0027] Large screen code generation and parsing module, used to realize code generation for smart large screens and automatic generation of smart large screens;

[0028] The system specifically implements the construction of a generative intelligent large screen based on natural language question and answer through the above method.

[0029] The present invention also claims protection for a generative intelligent large-screen construction device based on natural language question and answer, comprising: at least one memory and at least one processor;

[0030] The at least one memory is configured to store a machine-readable program;

[0031] The at least one processor is configured to call the machine-readable program to implement the above method.

[0032] The present invention also claims protection for a computer-readable medium having computer instructions stored thereon, which are capable of implementing the above method when executed by a processor.

[0033] Compared with the prior art, the method and system for constructing a generative intelligent large screen based on natural language question answering of the present invention have the following beneficial effects:

[0034] This invention achieves fully automated generation from natural language input to large-scale visualization screens, eliminating the need for human intervention. Non-professionals with low technical skills can complete large-scale screen construction through natural language interaction. Dynamic interface matching and multi-round iterative optimization mechanisms significantly shorten development cycles and improve response efficiency. Supporting unified standard access for multi-source heterogeneous data enables rapid response to changing business needs and flexible adaptation. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 This is a flowchart of a method for constructing a generative smart large screen based on natural language question answering provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] This embodiment of the present invention provides a generative intelligent large-screen construction method based on natural language question-answering, which realizes the automatic generation of intelligent large-screen through question-answering. The implementation of this method includes four major steps: data preprocessing and knowledge base construction, question segmentation and extraction, indicator data acquisition, and large-screen code generation and parsing.

[0037] 1. Data preprocessing and knowledge base construction: Data preprocessing organizes the required indicator data for large-screen display in the form of an interface, and builds an interface metadata description system based on the data interface, corresponding to fields such as interface name, interface description, URL address, interface input parameter, interface return value, and corresponding chart type. The knowledge base is constructed using a vectorized approach, generating high-dimensional vectors (such as 3072 dimensions) through large-scale pre-training to achieve a dense distributed representation of text semantics. The knowledge base uses a vectorized database to support efficient approximate nearest neighbor search, which can balance retrieval speed and accuracy. The boundary conditions and behavior patterns that support retrieval are manually configured through thresholds.

[0038] 2. Question segmentation and extraction: The smart screen generation questions raised by users are automatically segmented using a generative pre-trained model. Sub-questions containing question identifiers and question descriptions are generated and formatted to form a sub-question queue. The sub-question extraction method adopts a polling method. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-question from the knowledge base (the number of retrieved interface descriptions can be manually configured). Then, the sub-question and the matching interface description queue are passed to the generative pre-trained model. Based on the model's reasoning capabilities, the interface corresponding to the optimal interface description is matched.

[0039] The data formatting processing method uses regular expressions and is configured based on data cleaning rules to remove abnormal data such as title symbols, bold and italic symbols, link symbols, list symbols, and reference symbols.

[0040] The hybrid retrieval method adopts a hybrid form of full-text retrieval and vector retrieval, supports applying a re-ranking step, selects the best result matching the sub-question from the two types of query results, and supports the option of setting weights or configuring a re-ranking model.

[0041] 3. Indicator data acquisition: Iterative processing is performed in a polling manner. Indicator data is acquired based on the interface metadata information, and the interface is called according to the "interface input parameters" to acquire indicator data, and the optimal value of the "interface input parameters" is acquired based on the generative pre-trained model. First, a judge is used to determine whether the "interface input parameters" field in the interface metadata is empty. For interfaces whose "interface input parameters" field is not empty, it supports passing the sub-problem corresponding to the interface and the json data corresponding to the "interface input parameters" field to the generative pre-trained model, and matching the optimal value corresponding to the "interface input parameters" based on the model's reasoning ability. After the parameters are acquired, the method supports calling the interface according to the url address and interface input parameters in the interface metadata to acquire the indicator data required for the chart display on the large screen. For interfaces whose "interface input parameters" field is empty, the method supports calling the corresponding interface according to the url address in the interface metadata information to acquire indicator data.

[0042] 4. Large screen code generation and parsing: Based on the acquired indicator data, the interface metadata, including the interface return value, interface description, and chart type fields, is passed to the generative pre-trained model. Based on the prompt word project, code generation for the smart large screen is achieved. The code is formatted and uploaded to the server. After configuring the proxy and port, the access address for the smart large screen is generated, enabling automatic generation of the smart large screen.

[0043] This method achieves fully automated generation from natural language queries to large-scale visualization screens through technologies such as data preprocessing, scenario knowledge base construction, dynamic prompt word generation, intelligent sub-problem segmentation, multi-round iterative optimization, and code generation. It employs a collaborative reasoning mechanism between domain knowledge graphs and large models, combined with a dynamic interface matching algorithm, to address the long response times and the need for specialized technical personnel in traditional large-scale screen development. By establishing a problem-interface mapping library, it enables automated parsing of heterogeneous multi-source data and visualization code generation, ultimately resulting in a deployable intelligent large-scale screen application.

[0044] The detailed steps to implement this method are as follows:

[0045] Step 1: Data preprocessing and knowledge base construction.

[0046] To display data on smart screens, unified data standards are used to organize indicator data in the form of interfaces. A metadata description system is constructed based on the data interfaces, including information such as interface name, interface description, URL, interface input parameters, interface return value, and corresponding chart type. The configured interface metadata is mapped to a vector database using a vectorized approach, and its semantic associations are quantified using cosine similarity to form a scenario knowledge base. This establishes a network of associations between indicators, interfaces, and charts.

[0047] Vectorization utilizes a deep learning-based text embedding model. Large-scale pre-training generates high-dimensional vectors (e.g., 3072 dimensions) to achieve a dense, distributed representation of text semantics, better supporting interface metadata retrieval. The vector library utilizes the Chroma database, supporting efficient approximate nearest neighbor search and effectively balancing retrieval speed and accuracy. By setting different thresholds, precise matching of sub-questions and interface descriptions is achieved.

[0048] Step 2: Question segmentation and extraction.

[0049] For the generation of smart large screen questions raised by users, a generative pre-trained model is used to automatically segment the original question, generate sub-questions containing question identifiers and question descriptions, and format them into the sub-question queue. For each question in the question queue, polling is used for iterative processing. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-question from the knowledge base (the number of retrieved interface descriptions can be manually configured), and then the sub-question and the matching interface description list are passed to the generative pre-trained model, and the interface corresponding to the optimal interface description is matched based on the model's reasoning ability.

[0050] The generative pre-trained model used for automatic segmentation of original questions uses the Deepseek V3 instruction model. This requires configuring a prompt word project, including information such as core tasks, output rules, atomization requirements, and output examples.

[0051] The generative pre-trained model used to match optimal interface descriptions uses the DeepSeek R1 inference model. A prompt word project must be configured, including context data, processing steps, overall matching degree, and output format.

[0052] Data formatting uses regular expressions and is configured based on data cleaning rules to remove abnormal data such as title symbols, bold and italic symbols, link symbols, list symbols, and quote symbols.

[0053] The hybrid retrieval method uses full-text retrieval and vector retrieval simultaneously, and applies a re-ranking step to select the best result matching the sub-question from the two types of query results. In this method, you can choose to set weights or configure the re-ranking model.

[0054] Step 3: Obtain indicator data.

[0055] Based on the interface list returned in step 2, iterative processing is performed in a polling manner. First, determine whether the "Interface Input Parameters" field in the interface metadata is empty. For interfaces whose "Interface Input Parameters" field is not empty, pass the sub-problem corresponding to the interface in step 2 and the json data corresponding to the "Interface Input Parameters" field to the generative pre-training model, and match the optimal value corresponding to the "Interface Input Parameters" based on the model's reasoning ability. Then, call the corresponding interface based on the url address and interface input parameter information in the interface metadata to obtain the indicator data required for the chart display on the large screen. For interfaces whose "Interface Input Parameters" field is empty, directly call the corresponding interface based on the url address in the interface metadata information to obtain the indicator data.

[0056] The generative pre-trained model used to obtain optimal values for interface input parameters uses the DeepSeek R1 inference model. This requires configuring a prompt word project, including context data, extraction rules, value range validation, and output format.

[0057] Step 4: Generate and parse large screen code.

[0058] Based on the indicator data obtained through polling in step 3, combined with the interface return value, interface description, and chart type fields in the interface metadata, the model is fed into the generative pre-trained model. Based on the prompt word project, the code for the entire smart screen is generated. After formatting, the code is uploaded to the server. After configuring the proxy and port, the access address for the smart screen is returned, achieving automatic generation of the smart screen.

[0059] The generative pre-trained model for large-screen code generation uses the DeepSeekV3 instruction model. You need to configure a prompt project, including large-screen title generation rules, chart type mappings, code syntax rules, data formats, legends, and other information.

[0060] Based on the above detailed steps, the specific application case of this method is implemented as follows:

[0061] For indicator data that needs to be displayed on the large screen, unified data standards are encapsulated in the form of interfaces. An interface metadata description system is constructed based on the encapsulated data interface, including information such as interface name, interface description, URL address, interface input parameters, interface return value, and corresponding chart type. The configured interface metadata information is mapped to the Chroma vector database in a vectorized manner to achieve a dense distributed representation of text semantics. Its semantic associations are quantified through cosine similarity to form a scenario knowledge base, and a relationship network between indicators, interfaces, and charts is established. The vectorization process can achieve retrieval boundary conditions and behavioral pattern control of sub-problems and interface descriptions by setting different thresholds.

[0062] For the generation problem of smart large screen raised by users, Deepseek V3 generative pre-trained model is used to automatically segment the original problem according to the configured prompt word project including core tasks, output rules, atomization requirements, output instances and other information, generate sub-problems containing problem identification and problem description, and form a sub-problem list after formatting some abnormal data through regular expressions. For each question in the question list, polling is used for iterative processing. First, a hybrid retrieval method including full-text retrieval and vector retrieval is used to retrieve the interface description with the highest matching degree with the sub-problem from the knowledge base (the number of retrieved interface descriptions can be configured in advance), and then the sub-problem and the matching interface description list are passed to Deepseek R1 generative pre-trained model, and the interface corresponding to the optimal interface description is matched based on the model reasoning capability.

[0063] For the returned interface list, polling is used for iterative processing. First, determine whether the "Interface Input Parameters" field in the interface metadata is empty. For interfaces whose "Interface Input Parameters" field is not empty, pass the sub-problem corresponding to the interface and the json data corresponding to the "Interface Input Parameters" field to the DeepSeekR1 generative pre-training model, and match the optimal value corresponding to the "Interface Input Parameters" based on the model's reasoning ability. Then, call the corresponding interface according to the url address and interface input parameter information in the interface metadata to obtain the indicator data required for the chart display on the large screen. For interfaces whose "Interface Input Parameters" field is empty, directly call the corresponding interface according to the url address in the interface metadata to obtain the indicator data.

[0064] The indicator data list returned after polling is combined with the interface return value, interface description, and chart type fields in the interface metadata and passed to the DeepSeekV3 generative pre-trained model. Based on the large-screen title generation rules configured by the prompt word project, chart type mapping relationships, code syntax rules, data format, legend, and other information, the code for the entire smart large-screen is generated. After formatting, the code is uploaded to the server, and after configuring the proxy and port, the access address of the smart large-screen is returned, realizing the automatic generation of the smart large-screen.

[0065] The embodiment of the present invention further provides a generative intelligent large-screen construction system based on natural language question answering, including:

[0066] 1. Data preprocessing and knowledge base construction module, used to implement the construction of an interface metadata description system based on the data interface, and to achieve dense distributed representation of text semantics. Data preprocessing organizes the required indicator data for large-screen display in the form of an interface, and builds an interface metadata description system based on the data interface, corresponding to the interface name, interface description, URL address, interface input parameter, interface return value, corresponding chart type and other fields. The knowledge base is constructed in a vectorized manner, and high-dimensional vectors (such as 3072 dimensions) are generated through large-scale pre-training to achieve dense distributed representation of text semantics. The knowledge base uses a vectorized database, supports efficient approximate nearest neighbor search, and can balance retrieval speed and accuracy. The boundary conditions and behavior patterns that support retrieval are manually configured through thresholds.

[0067] 2. Problem segmentation and extraction module, which is used to generate sub-problems including problem identification and problem description, format them to form a sub-problem queue, and match the interface corresponding to the optimal interface description based on the model reasoning ability. The smart large-screen generation problem proposed by the user is automatically segmented using a generative pre-trained model to generate sub-problems including problem identification and problem description, and format them to form a sub-problem queue. The sub-problem extraction method adopts a polling method. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-problem from the knowledge base (the number of retrieved interface descriptions supports manual configuration), and then the sub-problem and the matching interface description queue are passed to the generative pre-trained model, and the interface corresponding to the optimal interface description is matched based on the model reasoning ability.

[0068] The data formatting processing method uses regular expressions and is configured based on data cleaning rules to remove abnormal data such as title symbols, bold and italic symbols, link symbols, list symbols, and reference symbols.

[0069] The hybrid retrieval method adopts a hybrid form of full-text retrieval and vector retrieval, supports applying a re-ranking step, selects the best result matching the sub-question from the two types of query results, and supports the option of setting weights or configuring a re-ranking model.

[0070] 3. An indicator data acquisition module is used to implement iterative processing in a polling manner and call the corresponding interface to obtain indicator data. Iterative processing is performed in a polling manner. Indicator data is obtained according to the interface metadata information, and it supports calling the interface to obtain indicator data according to the "interface input parameters", and obtaining the optimal value of the "interface input parameters" based on the generative pre-training model. First, a judge is used to determine whether the "interface input parameters" field in the interface metadata is empty. For interfaces whose "interface input parameters" field is not empty, it supports passing the sub-problem corresponding to the interface and the json data corresponding to the "interface input parameters" field to the generative pre-training model, and matching the optimal value corresponding to the "interface input parameters" based on the model reasoning ability. After the parameters are obtained, the method supports calling the interface according to the url address and interface input parameters in the interface metadata to obtain the indicator data required for the chart display in the large screen. For interfaces whose "interface input parameters" field is empty, the method supports calling the corresponding interface according to the url address in the interface metadata information to obtain indicator data.

[0071] 4. The large-screen code generation and parsing module is used to implement code generation for smart large-screen displays and automatically generate them. This module supports passing acquired indicator data, combined with the interface return value, interface description, and chart type fields in the interface metadata, to a generative pre-trained model. Based on the prompt word project, code generation for the smart large-screen is achieved. The module supports formatting the code and uploading it to the server. After configuring the proxy and port, the access address for the smart large-screen is generated, enabling automatic generation of the smart large-screen.

[0072] The system specifically implements the generative smart large screen construction based on natural language question answering through the generative smart large screen construction method based on natural language question answering described in the above embodiment. The specific implementation is as follows:

[0073] Step 1: Data preprocessing and knowledge base construction.

[0074] To display data on smart screens, unified data standards are used to organize indicator data in the form of interfaces. A metadata description system is constructed based on the data interfaces, including information such as interface name, interface description, URL, interface input parameters, interface return value, and corresponding chart type. The configured interface metadata is mapped to a vector database using a vectorized approach, and its semantic associations are quantified using cosine similarity to form a scenario knowledge base. This establishes a network of associations between indicators, interfaces, and charts.

[0075] Vectorization utilizes a deep learning-based text embedding model. Large-scale pre-training generates high-dimensional vectors (e.g., 3072 dimensions) to achieve a dense, distributed representation of text semantics, better supporting interface metadata retrieval. The vector library utilizes the Chroma database, supporting efficient approximate nearest neighbor search and effectively balancing retrieval speed and accuracy. By setting different thresholds, precise matching of sub-questions and interface descriptions is achieved.

[0076] Step 2: Question segmentation and extraction.

[0077] For the generation of smart large screen questions raised by users, a generative pre-trained model is used to automatically segment the original question, generate sub-questions containing question identifiers and question descriptions, and format them into the sub-question queue. For each question in the question queue, polling is used for iterative processing. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-question from the knowledge base (the number of retrieved interface descriptions can be manually configured), and then the sub-question and the matching interface description list are passed to the generative pre-trained model, and the interface corresponding to the optimal interface description is matched based on the model's reasoning ability.

[0078] The generative pre-trained model used for automatic segmentation of original questions uses the Deepseek V3 instruction model. This requires configuring a prompt word project, including information such as core tasks, output rules, atomization requirements, and output examples.

[0079] The generative pre-trained model used to match optimal interface descriptions uses the DeepSeek R1 inference model. A prompt word project must be configured, including context data, processing steps, overall matching degree, and output format.

[0080] Data formatting uses regular expressions and is configured based on data cleaning rules to remove abnormal data such as title symbols, bold and italic symbols, link symbols, list symbols, and quote symbols.

[0081] The hybrid retrieval method uses full-text retrieval and vector retrieval simultaneously, and applies a re-ranking step to select the best result matching the sub-question from the two types of query results. In this method, you can choose to set weights or configure the re-ranking model.

[0082] Step 3: Obtain indicator data.

[0083] Based on the interface list returned in step 2, iterative processing is performed in a polling manner. First, determine whether the "Interface Input Parameters" field in the interface metadata is empty. For interfaces whose "Interface Input Parameters" field is not empty, pass the sub-problem corresponding to the interface in step 2 and the json data corresponding to the "Interface Input Parameters" field to the generative pre-training model, and match the optimal value corresponding to the "Interface Input Parameters" based on the model's reasoning ability. Then, call the corresponding interface based on the url address and interface input parameter information in the interface metadata to obtain the indicator data required for the chart display on the large screen. For interfaces whose "Interface Input Parameters" field is empty, directly call the corresponding interface based on the url address in the interface metadata information to obtain the indicator data.

[0084] The generative pre-trained model used to obtain optimal values for interface input parameters uses the DeepSeek R1 inference model. This requires configuring a prompt word project, including context data, extraction rules, value range validation, and output format.

[0085] Step 4: Generate and parse large screen code.

[0086] Based on the indicator data obtained through polling in step 3, combined with the interface return value, interface description, and chart type fields in the interface metadata, the model is fed into the generative pre-trained model. Based on the prompt word project, the code for the entire smart screen is generated. After formatting, the code is uploaded to the server. After configuring the proxy and port, the access address for the smart screen is returned, achieving automatic generation of the smart screen.

[0087] The generative pre-trained model for large-screen code generation uses the DeepSeekV3 instruction model. You need to configure a prompt project, including large-screen title generation rules, chart type mappings, code syntax rules, data formats, legends, and other information.

[0088] An embodiment of the present invention further provides a generative intelligent large-screen construction device based on natural language question answering, comprising: at least one memory and at least one processor;

[0089] The at least one memory is configured to store a machine-readable program;

[0090] The at least one processor is used to call the machine-readable program to implement the generative smart large-screen construction method based on natural language question and answer described in the above embodiment.

[0091] An embodiment of the present invention further provides a computer-readable medium having computer instructions stored thereon, which, when executed by a processor, implement the method for constructing a generative smart large screen based on natural language question and answer described in the above embodiment. Specifically, a system or device equipped with a storage medium can be provided, on which software program codes for implementing the functions of any of the above embodiments are stored, and a computer (or CPU or MPU) of the system or device is enabled to read and execute the program code stored in the storage medium.

[0092] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute part of the present invention.

[0093] Examples of storage media for providing program code include floppy disks, hard disks, magneto-optical disks, optical disks (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), magnetic tapes, non-volatile memory cards, and ROMs. Alternatively, the program code can be downloaded from a server computer via a communication network.

[0094] In addition, it should be clear that the functions of any of the above embodiments can be achieved not only by executing the program code read by the computer, but also by enabling the operating system operating on the computer to complete part or all of the actual operations based on the instructions of the program code.

[0095] In addition, it can be understood that the program code read from the storage medium is written into the memory provided in the expansion board inserted into the computer or into the memory provided in the expansion unit connected to the computer, and then based on the instructions of the program code, the CPU installed on the expansion board or expansion unit is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above embodiments.

[0096] The present invention has been shown and described in detail above through the accompanying drawings and preferred embodiments. However, the present invention is not limited to these disclosed embodiments. Based on the above multiple embodiments, those skilled in the art can know that the code review methods in the above different embodiments can be combined to obtain more embodiments of the present invention, and these embodiments are also within the scope of protection of the present invention.

Claims

1. A method for constructing a generative intelligent large screen based on natural language question answering, characterized in that: The implementation of this method includes: Data preprocessing and knowledge base construction: Data preprocessing organizes the required indicator data for large-screen display in the form of interfaces, and builds an interface metadata description system based on the data interfaces. Knowledge base construction adopts a vectorized approach, generating high-dimensional vectors through large-scale pre-training to achieve dense distributed representation of text semantics. Question segmentation and extraction: The smart screen generation questions raised by users are automatically segmented using a generative pre-trained model to generate sub-questions containing question identifiers and question descriptions, which are then formatted and processed to form a sub-question queue. The sub-question extraction method adopts a polling method. First, a hybrid retrieval method is used to retrieve the interface description with the highest matching degree with the sub-question from the knowledge base. The sub-question and the matching interface description queue are then passed to the generative pre-trained model. Based on the model's reasoning ability, the interface corresponding to the optimal interface description is matched. Indicator data acquisition: It uses polling to iterate and obtain indicator data based on interface metadata information. It supports calling the interface to obtain indicator data based on "interface input parameters" and obtains the optimal value of "interface input parameters" based on the generative pre-training model. Large screen code generation and parsing: Based on the obtained indicator data, combined with the interface return value, interface description, and chart type fields in the interface metadata, it is passed to the generative pre-trained model, and according to the prompt word project, the code generation of the smart large screen is realized; it supports uploading the code to the server after formatting, and generates the smart large screen access address after configuring the proxy and port, realizing the automatic generation of the smart large screen.

2. A method for constructing a generative intelligent large screen based on natural language question answering according to claim 1, characterized in that: The data preprocessing and knowledge base construction, Build an interface metadata description system based on the data interface. The corresponding fields include: interface name, interface description, URL address, interface input parameter, interface return value, and corresponding chart type; The knowledge base uses a vectorized database to support efficient approximate nearest neighbor search; the boundary conditions and behavior patterns supported by retrieval are manually configured through thresholds.

3. The method for constructing a generative intelligent large screen based on natural language question answering according to claim 1, characterized in that: The problem segmentation and extraction, The generative pre-trained model used for automatic segmentation of original questions uses the Deepseek V3 instruction model; the configured prompt word project includes core tasks, output rules, atomization requirements, and output instance information; The generative pre-trained model used to match the optimal interface description uses the DeepSeek R1 inference model; the configured prompt word project includes context data, processing steps, comprehensive matching degree, and output format information.

4. A method for constructing a generative intelligent large screen based on natural language question answering according to claim 1 or 3, characterized in that: The data formatting of the sub-problem uses regular expressions. Based on the data cleaning rule configuration, abnormal data removal including title symbols, bold and italic symbols, link symbols, list symbols, and reference symbols is achieved.

5. The method for constructing a generative intelligent large screen based on natural language question answering according to claim 1 or 3, characterized in that: The hybrid retrieval method adopts a hybrid form of full-text retrieval and vector retrieval, supports applying a re-ranking step, selects the best result matching the sub-question from the two types of query results, and supports the option of setting weights or configuring a re-ranking model.

6. The method for constructing a generative intelligent large screen based on natural language question answering according to claim 1, characterized in that: The indicator data acquisition is specifically implemented as follows: First, determine whether the "Interface Input Parameters" field in the interface metadata is empty. For interfaces whose "Interface Input Parameters" field is not empty, support passing the sub-problem corresponding to the interface and the JSON data corresponding to the "Interface Input Parameters" field to the generative pre-trained model. Based on the model's reasoning ability, the optimal value corresponding to the "Interface Input Parameters" is matched. After the parameters are obtained, support calling the interface according to the URL address and interface input parameters in the interface metadata to obtain the indicator data required for the chart display on the large screen. For interfaces whose "Interface Input Parameters" field is empty, support calling the corresponding interface according to the URL address in the interface metadata information to obtain indicator data. The generative pre-training model for obtaining the optimal value of the "interface input parameter" can adopt the DeepSeek R1 inference model; the configured prompt word project includes context data, extraction rules, value range verification, and output format information.

7. The method for constructing a generative intelligent large screen based on natural language question answering according to claim 1, characterized in that: The generative pre-training model for large-screen code uses the DeepSeek V3 instruction model; the configured prompt word project includes large-screen title generation rules, chart type mapping relationships, code syntax rules, data format, and legend information.

8. A generative intelligent large-screen construction system based on natural language question answering, characterized in that: include: Data preprocessing and knowledge base construction module, used to build an interface metadata description system based on the data interface and achieve dense distributed representation of text semantics; The problem segmentation and extraction module is used to generate sub-problems containing problem identifiers and problem descriptions, format them to form a sub-problem queue, and match the interface corresponding to the optimal interface description based on the model's reasoning capabilities; The indicator data acquisition module is used to implement iterative processing in a polling manner and call the corresponding interface to obtain indicator data; Large screen code generation and parsing module, used to realize code generation for smart large screens and automatic generation of smart large screens; The system specifically realizes the construction of a generative intelligent large screen based on natural language question and answer through the method described in any one of claims 1 to 7.

9. A generative intelligent large-screen construction device based on natural language question answering, characterized in that: include: at least one memory and at least one processor; The at least one memory is configured to store a machine-readable program; The at least one processor is configured to call the machine-readable program to implement the method according to any one of claims 1 to 7.

10. A computer-readable medium, characterized in that The computer readable medium stores computer instructions, which, when executed by a processor, can implement the method according to any one of claims 1 to 7.

Images