Application method for realizing travel-related enterprise assistant based on large model

Abstract

The invention discloses an application method for realizing a travel-related enterprise assistant based on a large model, and the method comprises the steps: obtaining the input data of a user and the depth information of the user through a front-end interface, and storing the depth information of the user in a custom variable; retrieving a static database based on a workflow arrangement technology according to the input data to obtain a static output result; establishing a logic address mapping relation between input data and dynamic database content according to the big language inference model, outputting a dynamic output result, combining the dynamic output result with a static data result to form an application output result, establishing a problem classifier, and outputting a problem classification result; and selecting an output mode of the application output result according to the classification condition of the question classifier, and displaying the selected output mode in a front-end interface. According to the method, a reinforcement learning mechanism based on the user staying duration and the secondary questioning rate enables the model to be adaptively optimized, the user satisfaction degree is continuously improved, a problem classifier dynamically selects an output form according to user preferences, and the information transmission efficiency is improved by 40%.

Description

Technical Field

[0001] The present invention relates to the field of data search technology, and in particular to an application method for implementing a travel-related enterprise assistant based on a large model. Background Art

[0002] There are some problems in the field of data search technology: 1. Processing power limitations: Traditional enterprise systems often rely on rule-driven algorithms that have limited capabilities when handling complex tasks such as natural language understanding, image recognition, or predictive analytics. They cannot learn patterns and features from large amounts of data like large models.

[0003] 2. Lack of contextual understanding: Traditional systems are usually unable to understand complex contextual information, resulting in the inability to provide accurate and in-depth answers or suggestions when processing user queries or automated decisions. Summary of the invention

[0004] In view of the problems existing in the existing application methods of implementing travel-related enterprise assistants based on large models, the present invention is proposed.

[0005] Therefore, the problem to be solved by the present invention is to provide a new type of enterprise assistant application module, which integrates a large pre-trained model to enhance the processing capability of complex tasks.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: an application method for realizing a travel-related enterprise assistant based on a large model, which includes: Obtain user input data and user depth information through the front-end interface, and save the user depth information into custom variables. The input data includes natural language, uploaded files and pictures. The user depth information includes the semantic complexity of user input, the user interface stay time and the page secondary inquiry rate. According to the input data, the static database is searched based on the workflow arrangement technology to obtain the static output result; Establishing a large language reasoning model according to the custom variable configuration model parameters; Establishing a logical address mapping relationship between input data and dynamic database content according to the large language reasoning model, and outputting a dynamic output result; Combine dynamic output results with static data results to form application output results; Establish a question classifier, and select an output mode for the application output result according to the classification of the question classifier. The question classifier refers to the output mode selection of the application output result by the user, and the output mode selection includes text output, generating charts and exporting reports; The application output result is output in a classification manner according to the output of the question classifier and displayed in the front-end interface.

[0007] As a preferred solution of the application method of realizing the travel-related enterprise assistant based on the big model of the present invention, the obtaining of user depth information through the front-end interface includes: Obtain the semantic complexity of user front-end input, user stay time and secondary inquiry rate; The semantic complexity of obtaining user front-end input includes: Determine the depth of dependency syntactic analysis and the density of associated words in semantic complexity. When the depth of dependency syntactic analysis exceeds the threshold or the keyword density is lower than the threshold, the system starts a three-level retrieval mechanism: the first layer uses BERT-based semantic embedding matching, the middle layer uses the improved TF-IDF weighted algorithm, and the bottom layer triggers cross-domain knowledge graph association. After the user searches, the model parameters are dynamically adjusted according to the user's stay time and the secondary inquiry rate, forming a reinforcement learning mechanism with a time decay factor. The dynamic adjustment of model parameters by user stay time and secondary inquiry rate includes forming a reinforcement learning mechanism with a time decay factor to dynamically reduce the model parameters in the large language inference model when the user stay time exceeds a stay time threshold and the secondary inquiry rate exceeds a secondary inquiry rate threshold.

[0008] As a preferred solution of the application method of realizing the travel-related enterprise assistant based on the big model of the present invention, when the dependency syntax analysis depth exceeds the threshold or the keyword density is lower than the threshold, the system starts the three-level search mechanism including: The first layer uses BERT-based semantic embedding matching, the middle layer uses an improved TF-IDF weighted algorithm, introduces a word vector cosine similarity correction factor α=0.65, and the bottom layer triggers cross-domain knowledge graph association; The dependency syntactic analysis depth exceeds a threshold or the keyword density is lower than a threshold, including: When the dependency syntax analysis depth is ≥5 layers or the keyword density is <30%, the retrieval mechanism is triggered.

[0009] As a preferred solution of the application method of the travel-related enterprise assistant based on the big model of the present invention, the method of retrieving the static database based on the input data and obtaining the static output result based on the workflow orchestration technology includes: When the application module receives a data flow exceeding the input data threshold, the fast sorting channel is automatically started, wherein the data flow refers to the number of access requests from users; Rapidly extract key features based on hardware acceleration technology, wherein the key features include data fluctuation patterns and time sensitivity; A coarse classification is performed based on multi-layer filtering of key features. The first layer of the coarse classification screens out obviously irrelevant data, the second layer pre-groups by domain labels, and the third layer performs priority sorting based on historical pattern matching.

[0010] As a preferred solution of the application method of realizing the travel-related enterprise assistant based on the big model described in the present invention, the content within the logical address range of the database is output to the front-end interface, and the content output includes replying text, generating charts and exporting reports, and the content output method is freely selected by the user during the content input process; The database includes a static database and a dynamic database generated in real time; The content output method includes taking the application output result formed by combining the dynamic output result with the static data result as the content output option.

[0011] As a preferred solution of the application method of the travel-related enterprise assistant based on the big model of the present invention, the reply text formation process includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the output result of the application, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report; if it is determined to display text, the text is directly replied to the front-end interface; if it is determined to generate a chart or export a report, the output is stopped and the problem classification stage is returned.

[0012] As a preferred solution of the application method of realizing the travel-related enterprise assistant based on the big model of the present invention, the process of generating the chart includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the output result of the application, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report. If it is determined to generate a chart, the x-axis data type and y-axis data type of the chart are obtained from sys.query through the parameter extractor, the application output result is input into the x-axis and y-axis, the x-axis and y-axis data of the chart are generated, and the bar chart tool node is called to generate the chart to the front-end interface; if it is determined to reply text or export a report, the output is stopped and the problem classification stage is returned.

[0013] As a preferred solution of the application method of the travel-related enterprise assistant based on the big model of the present invention, the process of generating the export report includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the application output result, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report. If it is determined to export a report, the horizontal data type and vertical data type of the report are obtained from sys.query through the parameter extractor, and the report tool is called to generate a report to input the application output result into the horizontal data type and vertical data type, and the generated report is displayed in the front-end interface; if it is determined to reply text or generate a chart, the output is stopped and the problem classification stage is returned.

[0014] The present invention provides the following technical solution: an electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement an AI-based method for identifying illegal business system data operation behaviors.

[0015] The present invention provides the following technical solution: an electronic device, comprising: A computer-readable storage medium stores executable instructions, which, when executed by a processor, enable the processor to implement an AI-based method for identifying illegal data operation behaviors in a business system.

[0016] Beneficial effects of the present invention 1. Enhanced processing capabilities: Provides a new type of enterprise assistant application module that integrates large pre-trained models to enhance the processing capabilities of complex tasks, especially in natural language understanding, image recognition and predictive analysis, so as to learn deep patterns and features from large amounts of data.

[0017] 2. Improve contextual understanding capabilities: Develop an enterprise assistant application module that can understand and process complex contextual information to ensure that more accurate and in-depth answers or suggestions can be provided when processing user queries or automated decisions.

[0018] 3. Optimize data utilization: Design a mechanism to use large pre-trained models to fully tap the potential value of a large amount of underutilized data within the enterprise to improve the effectiveness and efficiency of data-driven decision-making.

[0019] 4. Improve response speed and accuracy: By integrating large models, it is possible to quickly and accurately respond to users' complex or ambiguous queries, improving user experience and service quality, especially in the areas of customer service and data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1This is a flow chart of an implementation method of the application method of realizing a travel-related enterprise assistant based on a large model in Example 1; Figure 2 This is an enterprise assistant architecture diagram for implementing the application method of the travel-related enterprise assistant based on the big model in Example 2; Figure 3 This is a schematic diagram of the connections between modules of the application method for implementing a travel-related enterprise assistant based on a large model in Example 2; Figure 4 This is a schematic diagram of the electronic structure of the application method of implementing a travel-related enterprise assistant based on a large model in Example 4. DETAILED DESCRIPTION

[0021] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0022] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0023] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0024] Example 1 Reference Figure 1 , which is the first embodiment of the present invention, and provides an application method for implementing a travel-related enterprise assistant based on a large model, which includes: Obtain user input data and user depth information through the front-end interface, and save the user depth information into custom variables. The input data includes natural language, uploaded files and pictures. The user depth information includes the semantic complexity of user input, the user interface stay time and the page secondary inquiry rate. According to the input data, the static database is searched based on the workflow arrangement technology to obtain the static output result; Establishing a large language reasoning model according to the custom variable configuration model parameters; Establishing a logical address mapping relationship between input data and dynamic database content according to the large language reasoning model, and outputting a dynamic output result; Combine dynamic output results with static data results to form application output results; Establish a question classifier, and select an output mode for the application output result according to the classification of the question classifier. The question classifier refers to the output mode selection of the application output result by the user, and the output mode selection includes text output, generating charts and exporting reports; The application output results are output in the classification method of the problem classifier and displayed in the front-end interface.

[0025] The system judges the depth of dependency syntactic analysis and the density of associated words in semantic complexity. When the depth of dependency syntactic analysis is ≥5 layers or the keyword density is <30%, the system starts a three-level retrieval mechanism: the first layer adopts BERT-based semantic embedding matching, the middle layer uses the improved TF-IDF weighted algorithm, and introduces the word vector cosine similarity correction factor α=0.65. The bottom layer triggers cross-domain knowledge graph association. When the user's stay time exceeds the stay time threshold of 8s and the secondary question rate exceeds the secondary question rate threshold of 40%, a reinforcement learning mechanism with a time attenuation factor is formed to dynamically reduce the model parameters in the large language inference model.

[0026] Dynamically adjusting model parameters through user stay time and secondary inquiry rate includes forming a reinforcement learning mechanism with a time decay factor to dynamically reduce model parameters in the large language inference model when the user stay time exceeds a stay time threshold and the secondary inquiry rate exceeds a secondary inquiry rate threshold.

[0027] Retrieving a static database based on input data based on workflow orchestration technology to obtain a static output result includes automatically starting a fast sorting channel when the application module receives a data flow exceeding the input data threshold, wherein the data flow refers to the number of access requests from users; Rapidly extract key features based on hardware acceleration technology, wherein the key features include data fluctuation patterns and time sensitivity; A coarse classification is performed based on multi-layer filtering of key features. The first layer of the coarse classification screens out obviously irrelevant data, the second layer pre-groups by domain labels, and the third layer performs priority sorting based on historical pattern matching.

[0028] Output the content within the logical address range of the database to the front-end interface. The content output includes reply text, generated charts and exported reports. The content output method is freely selected by the user during the content input process. The database includes a static database and a dynamic database that is generated on-the-fly.

[0029] The process of forming the reply text is: The user inputs data, and the user input is obtained through the sys.query variable. sys.query is input as a parameter into the static database for retrieval, and the retrieval results are input into the large language reasoning model. According to the large language reasoning model prompt words and the large language reasoning model type, the large language reasoning model is called to retrieve the dynamic database output results. Then, according to the content input by the user, the static database retrieval content and the dynamic database retrieval content are reorganized, and the question classifier is used to determine whether to reply to text, generate charts or export reports; if it is determined to display text, the text is directly replied to the front-end interface; if it is determined to generate charts or export reports, the output is stopped and the question classification stage is returned.

[0030] The formation process of generating the chart is: The user inputs data, and the user input is obtained through the sys.query variable. The sys.query is input as a parameter into the static database for retrieval, and the retrieval results are input into the large language reasoning model. According to the large language reasoning model prompt words and the large language reasoning model type, the large language reasoning model is called to retrieve the dynamic database output results, and then according to the content input by the user, the static database retrieval content is combined with the dynamic database retrieval content to generate the application output results. Use the question classifier to determine whether to reply to text, generate a chart, or export a report; if it is determined to generate a chart, the x-axis data type and y-axis data type of the chart are obtained from sys.query through the parameter extractor, and the application output results are input into the x-axis and y-axis to generate the x and y-axis data of the chart, and the bar chart tool node is called to generate the chart to the front-end interface; if it is determined to reply to text or export a report, the output is stopped and returned to the question classification stage.

[0031] The process of generating the export report is: The user inputs data, and the user input is obtained through the sys.query variable. The sys.query is input as a parameter into the static database for retrieval, and the retrieval results are input into the large language reasoning model. According to the large language reasoning model prompt words and the large language reasoning model type, the large language reasoning model is called to retrieve the dynamic database output results, and then according to the content input by the user, the static database retrieval content is combined with the dynamic database retrieval content to generate the application output results. Use the question classifier to determine whether to reply to text, generate charts or export reports; if it is determined to export reports, the horizontal data type and vertical data type of the report are obtained from sys.query through the parameter extractor, and the report tool is called to generate a report. The application output results are input into the horizontal data type and vertical data type, and the generated report is displayed in the front-end interface; if it is determined to reply to text or generate charts, the output is stopped and the question classification stage is returned.

[0032] The application output results are output in the classification method of the problem classifier and displayed in the front-end interface.

[0033] Example 2 The second embodiment of the present invention is different from the first embodiment in that: in the previous embodiment, the application method of implementing the travel-related enterprise assistant based on the big model includes: On the basis of building smart tourism, travel-related enterprises implement the application method of travel-related enterprise assistant based on big model, provide a chat window, and let enterprise users obtain the desired data through the front-end function, including historical basic data, real-time dynamic data, statistical data through analysis, etc.

[0034] The application includes functions such as enterprise assistant, intelligent customer service and financial reports, and can be expanded according to needs in the future.

[0035] Establish application architecture: Implement the application method architecture of travel-related enterprise assistants based on the big model, which is divided into application front-end, intelligent agent, knowledge base, database and big model.

[0036] Application front-end: A program that provides an operational interface for enterprise employees to operate. Through the front-end interface, employees can interact with the intelligent agent of the travel enterprise assistant to obtain the desired data. Intelligent agent: Through workflow orchestration, static knowledge base data and calling dynamic database data, combined with the emergence ability of large models, to realize enterprise assistant functions; Knowledge base / database: Provide users with comprehensive enterprise data by importing static data into the knowledge base and calling dynamic data generated in real time; Big model: Call the general big model provided by the platform and the locally deployed open source big model.

[0037] Application Architecture Figure 2 As shown below.

[0038] When the relationship between modules is called, the user accesses the application front end and enters content in the chat window of the front end. The intelligent agent calls the knowledge base and database to achieve real-time access to local data, and then submits the obtained content to the large model for processing or prediction, and finally outputs the results to the user. The calling relationship between modules is as follows: Figure 3 shown.

[0039] Processing flow: Users input the required questions in the front-end interface, and classify the questions according to the big model through the "Question Classification" node. The questions are divided into three categories: reply text, generate charts, and export reports. The categories can be expanded according to business needs, and different categories have special processing flows.

[0040] Reply text: To enter the reply text process, first use the local knowledge base "Knowledge Retrieval" node to recall relevant content in the knowledge base, then enter the "Call Big Model" node, and let the big model infer and output the results of the knowledge base recalled by the user's question, and then use the "Question Classification" node to determine whether to display it as text or list, and call the "Reply" node to display the list or individual text on the front end.

[0041] Generate charts: Enter the chart generation process. First, classify the questions through the "Question Classification" node, which are divided into "Query Passenger Flow Data", "Query Ticketing Data", "Query Hotel Data", and "Query Scenic Spot Data". Then call the "Knowledge Retrieval" node at the same time to recall relevant knowledge and call the interface through the "Http Request" node to obtain data from the database. Input the data of both into the "Big Model" node, and let the big model perform reasoning and give replies. Finally, call the "Chart Generation Tool" to organize the data into charts, which are returned to the front-end display by the "Reply" node.

[0042] Export report: Enter the export report process, first classify the questions through the "Question Classification" node, and divide them into "Query passenger flow data", query ticketing data, query hotel data, and query attraction data". Then call the "Knowledge Retrieval" node at the same time to recall relevant knowledge and call the interface through the "Http Request" node to obtain data from the database. Input the data of both into the "Big Model" node, and let the big model perform reasoning and give reply content. Finally, call the "Export Report Tool" to organize the data into downloadable reports, which are returned to the front-end display by the "Reply" node.

[0043] The specific processing flow is as follows: Figure 4 shown.

[0044] Start node: Get user input and save data to sys.query variables. You can also customize variables.

[0045] Question classification node: By defining the classification description, the question classifier can use the large model to infer the matching classification based on the user input and output the classification result, providing more accurate information to the downstream nodes.

[0046] Configuration steps: Select input variables, which refers to the input content used for classification. Input file variables are supported. In the customer service Q&A scenario, it is generally the question entered by the user sys.query; Select the reasoning model. The question classifier is based on the natural language classification and reasoning capabilities of the large language model. Selecting a suitable model will help improve the classification effect. Write category labels / descriptions. You can manually add multiple categories and write category keywords or description statements to help the large language model better understand the classification basis.

[0047] Select the downstream node corresponding to the classification. After the problem classification node completes the classification, the subsequent process path can be selected based on the relationship between the classification and the downstream node.

[0048] Knowledge retrieval node: Retrieve text content related to user questions from the knowledge base, which can be used as context for downstream large model nodes.

[0049] Configuration process: Select query variables. Query variables usually represent questions entered by users. This variable can be used as an input item to retrieve relevant text segments in the knowledge base. In common dialog applications, sys.query of the start node is generally used as a query variable. The maximum query content that the knowledge base can accept is 200 characters; Select the knowledge base you want to query. The optional knowledge base needs to be created in advance in the knowledge base. Specify the recall mode: connect and configure downstream nodes, usually large model nodes; Large model node: calls the large language model to process the information (natural language, uploaded files or pictures) entered by the user in the "Start" node and gives effective response information.

[0050] Configuration steps: Select a model. It provides support for mainstream models around the world, including OpenAI's GPT series, Anthropic's Claude series, Google's Gemini series, etc. Selecting a model depends on factors such as its reasoning ability, cost, response speed, context window, etc. You need to select a suitable model based on scenario requirements and task types.

[0051] Configure model parameters. Model parameters are used to control the generation results of the model, such as temperature, TopP, maximum mark, response format, etc. For easy selection, the system also provides 3 sets of preset parameters: creativity, balance and precision. If you are not familiar with the above parameters, select the default settings. If you want the application to have image analysis capabilities, you can choose a model with visual capabilities.

[0052] Fill in the context (optional). The context can be understood as the background information provided to the large model, and is often used to fill in the output variables of knowledge retrieval.

[0053] To write prompt words, the large model node provides an easy-to-use prompt word arrangement page. Selecting the chat model or the completion model will display different prompt word arrangement structures. If you select the chat model, you can customize the system prompt words (SYSTEM) / user (USER) / assistant (ASSISTANT) in three parts.

[0054] HTTP request node: allows sending server requests via HTTP protocol, suitable for obtaining external data, webhooks, generating images, downloading files, etc. It can send customized HTTP requests to the specified network address to achieve interconnection with various external services.

[0055] This node supports common HTTP request methods: GET: Used to request the server to send a resource.

[0056] POST: Used to submit data to the server, usually used to submit a form or upload a file.

[0057] HEAD: Similar to a GET request, but the server does not return the requested resource body, only the response header.

[0058] PATCH: Used to obtain the transmission path at each node in the request-response chain.

[0059] PUT: used to upload resources to the server, usually used to update existing resources or create new resources.

[0060] DELETE: used to request the server to delete the specified resource.

[0061] You can configure the HTTP request including URL, request header, query parameters, request body content, and authentication information.

[0062] Tool node: The "Tool" node can provide powerful third-party capability support for workflows.

[0063] Configuration steps: Authorize tools / create custom tools / publish workflows as tools; Configure tool inputs and parameters.

[0064] Direct reply: Define the reply content in a process.

[0065] You can freely define the response format in the text editor, including customizing a fixed text content, using the output variable in the previous step as the response content, or combining custom text with variables for the response.

[0066] You can add nodes at any time to stream content to the conversation reply, support WYSIWYG configuration mode and support mixed text and image layout, such as: Output the reply content of the large model node; Output generated image; Outputs plain text.

[0067] Example 3 The third embodiment of the present invention is a method for implementing an application of a travel-related enterprise assistant based on a large model, including: The test scenario is an intelligent customer service system of a large enterprise, which is used to handle technical support requests from customers. The system front-end interface supports natural language input, file upload (such as log files) and picture upload (such as photos of equipment failures). The test data comes from the company's historical work order records, including 1,000 real user requests, covering three categories: simple queries, complex technical issues and cross-domain issues. The static database is the company's internal knowledge base, containing 100,000 technical documents; the dynamic database is a real-time updated work order processing record, containing 50,000 data.

[0068] Implementation process (1) Data input and state capture: Users input questions through the front-end interface, and the system captures user in-depth information in real time, including semantic complexity (calculated through dependency syntax analysis), interface dwell time (time from page loading to submission), and secondary question rate (number of times users ask questions about the initial answer). Input data and status are saved in custom variables for subsequent processing.

[0069] (2) Static data retrieval: When the input data flow exceeds the threshold (500 requests per second), the system starts the fast sorting channel. Through FPGA hardware acceleration technology, key features (such as data fluctuation patterns and time sensitivity) are extracted, and three layers of filtering are used for rough classification: the first layer filters out obviously irrelevant data (such as advertisements or invalid inputs), the second layer pre-groups by domain labels (such as network, hardware, software), and the third layer prioritizes based on historical pattern matching.

[0070] (3) Dynamic reasoning and result generation: Configure the parameters of the large language inference model according to the user's depth information. When the semantic complexity is ≥5 layers or the keyword density is <30%, the three-level retrieval mechanism is activated: the first layer uses BERT semantic embedding matching, the middle layer uses the improved TF-IDF weighted algorithm (α=0.65), and the bottom layer triggers cross-domain knowledge graph association. The dynamic output results are combined with the static retrieval results to form the final application output results.

[0071] (4) Output method selection and display: The problem classifier selects the output method based on the user's historical preferences and problem type. For example, technical parameter comparison is output in charts, troubleshooting steps are output in text, and statistical analysis results are exported in reports. The final results are displayed through the front-end interface.

[0072] (5) Dynamic adjustment of model parameters: When the user's stay time is greater than 8 seconds and the secondary inquiry rate is greater than 40%, the system activates the reinforcement learning mechanism and dynamically adjusts the model parameters through the time decay factor to ensure continuous optimization of the model.

[0073] This embodiment conducted six groups of experiments, namely, simple query scenario performance data, complex technical scenario performance data, cross-domain problem scenario performance data, high concurrency scenario performance data, dynamic parameter adjustment test performance data, and traditional system comparison performance data.

[0074] Table 1: Simple query scenario performance data Table 2: Performance data for complex technical scenarios Table 3: Cross-domain problem scenario performance data Table 4: High concurrency scenario performance data Table 5: Dynamic parameter adjustment test performance data Table 6: Comparative performance data of traditional systems 1. Performance comparison analysis It can be seen from the table data that the system of the present invention is superior to the traditional system in various key indicators: Retrieval time: In a high-concurrency scenario, the retrieval time of the system of the present invention is 200 milliseconds, while that of the traditional system is 600 milliseconds, with an efficiency improvement of 67%.

[0075] Accuracy: For complex technical problems, the accuracy of the system of the present invention reaches 89%, while the accuracy of the traditional system is only 75%, an increase of 14 percentage points.

[0076] User satisfaction: In the cross-domain problem scenario, the user satisfaction of the present invention is 86%, while that of the traditional system is 70%, an increase of 16 percentage points.

[0077] 2. Analysis of technical advantages Three-level retrieval mechanism: Through BERT semantic embedding matching and cross-domain knowledge graph association, the system can handle complex semantic problems (such as nested compound sentences) with significantly improved accuracy.

[0078] Dynamic parameter adjustment: The reinforcement learning mechanism based on user stay time and secondary inquiry rate enables the model to be adaptively optimized and user satisfaction to continuously improve.

[0079] Output mode adaptation: The question classifier dynamically selects the output format (text, chart, report) based on user preferences, improving information communication efficiency by 40%.

[0080] 3. Summary of advantages High efficiency: Hardware acceleration and multi-layer filtering enable the system to maintain millisecond-level response in high-concurrency scenarios.

[0081] Accuracy: The three-level retrieval mechanism and dynamic parameter adjustment ensure the accuracy and stability of complex problem processing.

[0082] User experience: Output mode adaptation and dynamic optimization mechanism significantly improve user satisfaction and task completion rate.

[0083] Conclusion: Through innovative workflow orchestration technology and dynamic feedback mechanism, the present invention demonstrates significant advantages in retrieval efficiency, accuracy and user experience, and has clear technological advancement and industrial application value.

[0084] Example 4 Reference Figure 4 , which is the fourth embodiment of the present invention, and which is different from the first three embodiments in that: the application method of realizing the travel-related enterprise assistant based on the big model includes: Reference below Figure 4 , which shows a schematic diagram of the structure of an electronic device 300 suitable for implementing some embodiments of the present invention. The electronic devices in some embodiments of the present invention may include but are not limited to mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 2 The enterprise assistant architecture shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0085] like Figure 4 As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 to a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0086] Typically, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Figure 4 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 4 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0087] Furthermore, the storage medium of the embodiment of the present application stores program instructions that can implement all the above methods, wherein the program instructions can be stored in the above storage medium in the form of a software product, including several instructions for enabling 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 methods described in each embodiment of the present application. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, or terminal devices such as a computer, a server, a mobile phone, and a tablet.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An application method for realizing a travel-related enterprise assistant based on a large model, characterized in that , the method comprises: Obtain user input data and user depth information through the front-end interface, and save the user depth information into custom variables. The input data includes natural language, files and pictures. The user depth information includes the semantic complexity of the user input, the user interface stay time and the page secondary inquiry rate. Based on the input data, the static database is searched according to the workflow orchestration technology to obtain the static output result; Establishing a large language reasoning model according to the custom variable configuration model parameters; Establishing a logical address mapping relationship between input data and dynamic database content according to the large language reasoning model, and outputting a dynamic output result; Combine dynamic output results with static data result analysis to form application output results; Establish a question classifier, and select an output mode for the application output result according to the classification of the question classifier. The question classifier refers to the output mode selection of the application output result by the user, and the output mode selection includes text output, generating charts and exporting reports; The application output result is output in a classification manner according to the output of the question classifier and displayed in the front-end interface.

2. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 1 is characterized in that: The obtaining of user depth information through the front-end interface includes: Obtain the semantic complexity of user front-end input, user stay time and secondary inquiry rate; The semantic complexity of obtaining user front-end input includes: Determine the depth of dependency syntactic analysis and the density of associated words in semantic complexity. When the depth of dependency syntactic analysis exceeds the threshold or the keyword density is lower than the threshold, the system starts a three-level retrieval mechanism. The first layer uses BERT-based semantic embedding matching, the middle layer uses the improved TF-IDF weighted algorithm, and the bottom layer triggers cross-domain knowledge graph association. After the user searches, the model parameters are dynamically adjusted according to the user's stay time and the secondary inquiry rate, forming a reinforcement learning mechanism with a time decay factor. The method of dynamically adjusting model parameters by user stay time and secondary inquiry rate includes forming a reinforcement learning mechanism with a time decay factor to dynamically reduce model parameters in the large language inference model when the user stay time exceeds a stay time threshold of 8s and the secondary inquiry rate exceeds a secondary inquiry rate threshold of 40%.

3. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 2 is characterized in that: When the dependency syntax analysis depth exceeds a threshold or the keyword density is lower than a threshold, the system starts a three-level search mechanism including: The first layer uses BERT-based semantic embedding matching, the middle layer uses an improved TF-IDF weighted algorithm, introduces a word vector cosine similarity correction factor α=0.65, and the bottom layer triggers cross-domain knowledge graph association; The dependency syntactic analysis depth exceeds a threshold or the keyword density is lower than a threshold, including: When the dependency syntax analysis depth is ≥5 layers or the keyword density is <30%, the retrieval mechanism is triggered.

4. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 1 is characterized in that: The step of retrieving a static database based on input data and workflow orchestration technology to obtain a static output result includes: When the application module receives a data flow exceeding the input data threshold, the fast sorting channel is automatically started, wherein the data flow refers to the number of access requests from users; Rapidly extract key features based on hardware acceleration technology, wherein the key features include data fluctuation patterns and time sensitivity; A coarse classification is performed based on multi-layer filtering of key features. The first layer of the coarse classification screens out obviously irrelevant data, the second layer pre-groups by domain labels, and the third layer performs priority sorting based on historical pattern matching.

5. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 1 is characterized in that: The output of the application results is outputted in a classification manner according to the output of the problem classifier and displayed on the front-end interface, including: The content output includes replying text, generating charts and exporting reports. The content output method is freely selected by the user during the content input process; The content output method includes taking the application output result formed by combining the dynamic output result with the static data result as the content output option.

6. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 5 is characterized in that: The process of forming the reply text includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the output result of the application, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report; if it is determined to display text, the text is directly replied to the front-end interface; if it is determined to generate a chart or export a report, the output is stopped and the problem classification stage is returned.

7. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 5 is characterized in that: The formation process of generating the graph includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the application output result, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report; if it is determined to generate a chart, the x-axis data type and y-axis data type of the chart are obtained from sys.query through the parameter extractor, and the application output result is input into the x-axis and y-axis to generate the x and y-axis data of the chart, and the bar chart tool node is called to generate the chart to the front-end interface; if it is determined to reply text or export a report, the output is stopped and the problem classification stage is returned.

8. The application method for realizing a travel-related enterprise assistant based on a large model as claimed in claim 5 is characterized in that: The process of forming the export report includes: The user inputs data, and the application interface obtains the user input through the sys.query variable. After obtaining the application output result, the question classifier is used to determine whether the content output method is to reply text, generate a chart, or export a report; if it is determined to export a report, the horizontal data type and the vertical data type of the report are obtained from sys.query through the parameter extractor, and the report tool is called to generate a report to input the application output result into the horizontal data type and the vertical data type, and the generated report is displayed in the front-end interface; if it is determined to reply text or generate a chart, the output is stopped and the problem classification stage is returned.

9. An electronic device, comprising: one or more processors; a storage device having one or more programs stored thereon; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 8.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enables the processor to implement the method according to any one of claims 1 to 8.

Images