Multi-scenario, multi-user, and multi-service intention translation method for satellite communication networks

Abstract

The present invention discloses a method for translating the intent of multiple scenarios, multiple users and multiple services in a satellite communication network. The method solves the problem of low resource utilization due to large restrictions on user input and a wide variety of types. The implementation includes: constructing an intent recognition module; inputting and preprocessing the service intent of the front-end user; identifying the scenario, user and service to obtain a classification result; extracting the intent information extraction result; storing the original service intent, and dynamically learning the model; implementing the intention translation of the front-end user; and implementing the intention translation of multiple scenarios, multiple users and multiple services in a satellite communication network. The present invention adds an intent classification model to the intent recognition module, and can provide network strategies that meet different service quality requirements for front-end users through coarse-grained and fine-grained network strategies, thereby improving user satisfaction and experience. The coarse-grained fills the default value of the fine-grained strategy to ensure the stability and reliability of the intent translation. It is used for the intent translation of multiple scenarios, multiple users and multiple services in a satellite communication network.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite communication networks, and mainly relates to user intention translation. It specifically discloses a method for translating the intention of a satellite communication network in multiple scenarios, multiple users, and multiple services, which is used in the field of intent-driven satellite network communications. Background Art

[0002] Satellite networks are networked systems that utilize space platforms (such as geosynchronous satellites, medium- and low-orbit satellites, stratospheric balloons, and aircraft) and integrate ground-based network nodes to complete satellite acquisition, preprocessing, transmission, and reprocessing tasks. These systems support deep space exploration and Earth observation. As a key national infrastructure development, satellites not only play a key role in the military but also have a profound impact on human production and lifestyles. With an increasing number of users and increasingly complex services connected to satellite networks, satellite networks have entered an era of reliable transmission and diversified information. Satellite network communications effectively compensate for the shortcomings of terrestrial communications and play a vital role in television broadcasting, global communications, maritime rescue, and telemedicine.

[0003] However, satellite networks serving the global market need to provide various types of services to different users. Each type of service has different requirements for network resources. At the same time, the corresponding network management is becoming more and more complicated. Traditional network management has a high degree of manual participation and a high error rate.

[0004] Secondly, existing satellite networks, built for specific mission requirements, use isolated and independent network systems to meet different mission needs. For example, Earth observation satellite systems and long-distance communication satellite systems cannot use each other. Even for the same type of mission, they are implemented by multiple isolated subnets (such as environmental satellite systems and resource satellite systems). At the same time, my country's satellites currently generally use overhead transmission, which makes real-time data transmission difficult. In summary, the closed segmentation of network resources, the repeated deployment of network facilities, and the overhead transmission of data make it difficult to efficiently share limited network resources.

[0005] In summary, the challenges of existing satellite network resource management are:

[0006] (1) Satellite network architectures vary, network management is complex, and traditional manual network management methods have a high error rate;

[0007] (2) Compared with the limited satellite network resources, the business information traffic has increased dramatically, and the scarcity of satellite network resources has made it difficult to meet the current mission's demand for resource diversity.

[0008] The emergence of intent-driven networking (IDN) offers a new approach to addressing these challenges. It is a programmable and customizable automated network that integrates deep application intent mining, global network status awareness, and real-time network configuration optimization capabilities. Intent is a declarative description of system state. It abstracts network objects and capabilities from a demand perspective and can be translated into high-level policies. Intent-driven networks can automatically convert, verify, deploy, configure, and optimize according to operator intent to achieve the target network state. Furthermore, relying on holistic network perception and a closed-loop feedback optimization loop, anomalies can be automatically resolved to ensure network reliability. Intent-driven networks play a crucial role in the evolution of satellite networks.

[0009] Currently, satellite network resource management relies primarily on operators specifying network policies using low-level languages. This makes it difficult for ordinary users without network skills to effectively control network status. The intent translation system provides network operators and ordinary users with a method for converting mission intent expressed in natural language into standard intent expressions recognizable by the network. This system is the first step in realizing intent-driven satellite networks.

[0010] The intent translation method solves the problem of multiple user input intents in multi-user and multi-service scenarios. It is applicable to both users with network-related professional knowledge and non-professional users. Currently, there are related studies on intent translation methods in operator networks. However, they are all targeted at specific intent types in specific scenarios, specific users, and specific services, and have not been studied for multi-scenario, multi-user, and multi-service scenarios. Existing research on intent translation methods is as follows:

[0011] Existing technology captures user intent through an interactive interface and uses named entity recognition in natural language processing to directly obtain relevant information about network bandwidth, latency, endpoints, and other aspects of the user intent. However, this method lacks the ability to treat all users and services equally, failing to consider the diverse needs of different users and services and failing to flexibly address the specific service requirements of specific users.

[0012] The second existing technology is to provide users with a set of templates to specify their user intent based on the network scope and intent through a graphical user interface. The template guides users to fill in information by showing which network attributes are required, which are conditional, and which are optional. The disadvantage is that the use of graphical interfaces and template tools may limit users to specify some possible options that are not provided, reducing flexibility and being time-consuming.

[0013] The third existing technology identifies user intent by deploying a set of APIs (applications) in the northbound interface of an SDN (Software Defined Network) environment. This intent primarily emphasizes connection-related intent and supports point-to-point and point-to-multipoint connections. However, this approach is primarily targeted at application developers, with a narrow audience and limited intent coverage.

[0014] Existing technologies have many restrictions on user input, treating all users and businesses equally without considering the differentiated needs of different users and businesses. They cannot flexibly handle the specific business needs of specific users in specific scenarios, nor can they achieve reasonable allocation of resources. Summary of the Invention

[0015] In response to the problems existing in the prior art, the present invention provides a method for translating intentions in multiple scenarios, multiple users, and multiple services in a satellite communication network, which has fewer restrictions on the form of user input intentions and finer divisions of scenarios and services.

[0016] The present invention is a multi-scenario, multi-user, and multi-service intention translation method for a satellite communication network. The method is characterized in that a user's original service intention is received at the front end through a web page interface, and the user's original service intention is processed at the back end through multiple modules; the multiple modules include a text preprocessing module, an intention recognition module, and a strategy mapping module, and ultimately obtain a network strategy, which is then used to translate the user's intention in the satellite communication network. The multi-scenario, multi-user, and multi-service intention translation method for a satellite communication network includes the following steps:

[0017] Step 1: Construct an intent recognition module. The constructed intent recognition module receives the original business intent input by the front-end user and inputs it into the intent classification model and intent extraction model in sequence. The recognition results of the above two models are used as a whole output, and the correct recognition result is output. The front-end user business intent input into the model needs to meet the following specifications: the intent text is truncated or padded to a specified length and converted into a digital index. The intent classification model is constructed based on the text classification algorithm in natural language processing and is used to identify the front-end user's scenario, user and business type in multiple scenarios, multiple users and multiple businesses. The intent extraction model is constructed based on the entity recognition algorithm in natural language processing and is used to extract key network parameter information from the front-end user's business intent.

[0018] Step 2: The front-end user inputs the business intent and preprocesses it. The front-end user inputs the original business intent in natural language through the web interface. The original business intent text input by the front-end user is preprocessed, including error detection, error correction, padding or truncation operations, and conversion into a digital index. This ensures that the business intent text originally input by the front-end user meets the input specifications of the intent recognition module and is then input into the intent recognition module.

[0019] Step 3: Identify the scenario, user, and business to obtain a classification result. Use the intent classification model in the intent recognition module to simultaneously identify the scenario, user, and specific business type of the front-end user's business intent to obtain an identification result. Then, by judging the identification result, specific business processing for a specific user in a specific scenario is implemented. Any identification result that is not a specific scenario, user, or business is processed according to the default method or user-specified method.

[0020] Step 4 extracts the front-end user's service intent information and obtains the intent information extraction results. After classifying all front-end user service intents, the intent extraction model is used to extract key network parameter information, including bandwidth, latency, duration, spatial location, jitter, and network transmission speed parameters. The service intent classification results and intent information extraction results are fed back to the web page interface for the front-end user to confirm or modify, and the correct identification result is obtained after the front-end user confirms it.

[0021] Step 5: Store the original business intent of the front-end user so that the intent classification and intent extraction models can learn dynamically. The original business intent input by the front-end user and the correct recognition result after the user's confirmation are stored in the database so that the intent classification and extraction models can learn dynamically.

[0022] Step 6 implements the front-end user's intention translation; the correct recognition result confirmed by the front-end user is fed back to the front-end web interface for the front-end user to save, and the correct recognition result is input into the strategy mapping module built based on the finite state machine to generate the network strategy, completing the network strategy mapping; the network strategy is used for the front-end user's intention translation in the satellite communication network, realizing the front-end user's original business intention translation in the satellite communication network with multiple scenarios, multiple users and multiple services.

[0023] The present invention is also a method for translating the intention of multiple scenarios, multiple users and multiple services in a satellite communication network as described in any one of claims 1 to 5, characterized in that, in response to the problems of scarce satellite communication network resources, weak dynamic resource adjustment capabilities, diverse user types, and mismatch and insufficiency between service demands and network resources, the method for translating the intention of multiple users and multiple services in a satellite communication network can identify the communication scenario, user type and service type to which the service belongs, so as to meet the needs of different users for different services in different scenarios and realize efficient utilization of limited satellite network resources.

[0024] The present invention solves the problem that the existing technology has significant limitations on user input and is suitable for both professional and non-professional users. Secondly, the present invention proposes a method that combines coarse and fine granularity strategies to solve the problem of low resource utilization in satellite communication networks due to the large number of scenarios, users and business types, thereby improving resource utilization and enhancing user experience.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] Fewer restrictions on user input: The intent translation method proposed in the present invention adds an intent classification model to the intent recognition part for the first time. Therefore, the front-end user no longer needs to describe the business intent in detail when inputting, that is, there is no need to reflect service quality indicator parameters such as bandwidth, latency, and duration. The intent recognition model can identify the scenario, user, and business type, and realize specific business processing for specific users in specific scenarios, and the policy mapping module can map and generate coarse-grained policies.

[0027] The method is more portable and can be quickly migrated to other types of networks: The intent classification and intent extraction models in the intent recognition module are fine-tuned based on the pre-trained language model Alber based on transfer learning. The method is more portable and can be quickly migrated to other types of networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following is a brief introduction to the drawings required for use in the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0029] Figure 1 It is a flowchart of a multi-scenario, multi-user, and multi-service intention translation method for a satellite communication network according to the present invention;

[0030] Figure 2 It is a schematic diagram of the system framework for multi-scenario, multi-user, and multi-service intention translation in a satellite communication network according to the present invention.

[0031] Figure 3 This is a block diagram of the multi-task module construction adopted by the intent classification model of the present invention. DETAILED DESCRIPTION

[0032] Example 1

[0033] To make the objectives, technical solutions, and advantages of the present invention more clearly understood, the following describes in detail the multi-scenario, multi-user, and multi-service intent translation method for a satellite communication network proposed by the present invention. It should be understood that the specific contents described herein are merely for the purpose of explaining the present invention and are not intended to limit the present invention.

[0034] Existing technologies have many restrictions on user input, treating all users and businesses equally without considering the differentiated needs of different users and businesses. They cannot flexibly handle the specific business needs of specific users in specific scenarios, nor can they achieve reasonable allocation of resources.

[0035] In view of the problems existing in the prior art, the present invention provides a method for translating the intent of multiple users and multiple services in a satellite communication network. The present invention is described in detail below with reference to the accompanying drawings.

[0036] The present invention is a multi-scenario, multi-user, and multi-service intention translation method for satellite communication networks. Figure 1 , Figure 1 This is a flowchart of the multi-scenario, multi-user, multi-service intent translation method for a satellite communication network according to the present invention. The translation system receives the user's original service intent through a web interface at the front end, and processes the user's original service intent through multiple modules at the back end. These modules include a text preprocessing module, an intent recognition module, and a policy mapping module. Ultimately, the network policy is obtained and used to translate user intent in the satellite communication network. The multi-scenario, multi-user, multi-service intent translation method for a satellite communication network according to the present invention includes the following steps:

[0037] Step 1: Construct an intent recognition module. The constructed intent recognition module receives the original business intent input by the front-end user and inputs it into the intent classification model and the intent extraction model in sequence. The recognition results of the above two models are output as a whole, and the correct recognition result is output. The front-end user business intent input into the model needs to meet the following specifications: the intent text is truncated or padded to a specified length and converted into a digital index. The intent classification model is constructed based on the text classification algorithm in natural language processing, and is used to identify the scenarios, users and business types of front-end users in multiple scenarios, multiple users and multiple businesses. The intent extraction model is constructed based on the entity recognition algorithm in natural language processing, and is used to extract key network parameter information in the front-end user business intent.

[0038] Step 2: The front-end user's business intent is input and preprocessed; the front-end user inputs the original business intent in the form of natural language through the web interface, and the original business intent text input by the front-end user is preprocessed, including error detection, error correction, padding or truncation operations on the original business intent text and converting it into a digital index, so that the business intent text originally input by the front-end user meets the input specifications of the intent recognition module and is input into the intent recognition module.

[0039] Step 3 uses the constructed intent classification model to identify scenarios, users, and businesses to obtain classification results; uses the intent classification model in the intent recognition module to simultaneously identify the scenarios, users, and specific business types of the front-end user's business intent to obtain recognition results; then, by judging the recognition results, specific business processing for specific users in specific scenarios is realized; all recognition results that are non-specific scenarios, users, and businesses are processed in the default manner or user-specified manner; obtain the intent classification results of the scenarios, users, and specific business types to which the user has been processed by specific businesses or in a specified manner.

[0040] Step 4 uses the constructed intent extraction model to extract the front-end user's business intent information and obtain the intent information extraction result; after classifying all front-end user business intentions, use the intent extraction model to extract key network parameter information, including bandwidth, delay, duration, spatial position, jitter, and network transmission speed parameters, and feed back the business intent classification results and intent information extraction results to the web page interface for the front-end user to confirm or modify, and obtain the correct identification result after confirmation by the front-end user.

[0041] Step 5 stores the original business intent of the front-end user so that the intent classification and intent extraction models can learn dynamically; the original business intent input by the front-end user and the correct recognition results after user confirmation are stored in the database so that the intent classification and extraction models can learn dynamically and improve the recognition accuracy of the intent classification and intent extraction models.

[0042] Step 6 implements the front-end user's intention translation; the correct recognition result confirmed by the front-end user is fed back to the front-end web interface for the front-end user to save, and the correct recognition result is input into the strategy mapping module built based on the finite state machine to generate the network strategy, completing the network strategy mapping; the network strategy is used for the front-end user's intention translation in the satellite communication network, realizing the front-end user's original business intention translation in the satellite communication network with multiple scenarios, multiple users and multiple services.

[0043] The idea of ​​the present invention is as follows: first, an intent recognition module is constructed, which consists of an intent classification model and an intent extraction model, and is used to identify the scenarios, users and business types of front-end users, and extract key network parameter information from business intents; secondly, the original business intent input by the front-end user is preprocessed, including text error detection, error correction, padding or truncation, and converted into a digital index to make it conform to the input specification of the intent recognition module; after the preprocessing is completed, the intent classification model is used to identify the scenarios, users and business types of the front-end user business intent, and specific business processing for specific users in specific scenarios is performed based on the recognition results; after the intent is classified, the intent extraction model is used to extract key network parameter information for all classified business intents, and the classification results and information extraction results are fed back to the front-end user for confirmation or modification; after the intent extraction is completed, the front-end user's original business intent and the correct recognition results are stored so that the intent classification and intent extraction models can be dynamically learned to improve the recognition accuracy; finally, the front-end user's intent is translated, and the correct recognition results are input into the policy mapping module to generate a network policy for translating the front-end user's original business intent in the satellite communication network.

[0044] The technical solution and approach of this invention: The intent recognition module is constructed using the open-source deep learning framework PyTorch, and is based on text classification and entity extraction algorithms used in natural language processing. The system framework is constructed using the DJango framework, with the front-end built using HTML and the back-end using MySQL to store user intent data.

[0045] The technical effect of the present invention is to achieve automatic identification of the business intentions of front-end users. By constructing an intent recognition module, the original business intention text input by the front-end user is converted into a digital index, and the user's scenario, user type and business type are identified through the intent classification model and the intent extraction model, thereby achieving automatic identification of the front-end user's business intentions; and achieving automatic translation of the front-end user's business intentions. By constructing a policy mapping module, the business intention classification results output by the intent recognition module are mapped to the corresponding network policy, thereby achieving automatic translation of the front-end user's business intentions, improving the communication efficiency and user experience in the satellite communication network.

[0046] Implementation Column 2

[0047] The method for translating intent for multiple users and multiple services in a satellite communication network is the same as in Example 1. Step 3 of the present invention identifies the scene, user, and service to obtain a classification result. The intent classification model in the intent recognition module uses a shared-layer multi-task model based on the large-scale pre-trained model Albert and the one-dimensional convolutional model TextCNN to simultaneously identify the scene, user, and service type to obtain a classification result. The present invention improves recognition accuracy by using the large-scale pre-trained language model Albert. Albert is a powerful natural language processing model with a large amount of training data and a deep neural network structure, which can provide more accurate prediction results. TextCNN can extract key features through the convolutional layer, which is conducive to identifying the scene, user, and service type. The reusability and scalability of the model are improved. Based on the implementation of the shared-layer multi-task model, the scene, user, and service type identification problems can be treated as a whole and processed simultaneously, reducing the model's parameters and computational complexity, improving the model's reusability and scalability, and making the model suitable for different application scenarios such as multiple scenes, multiple users, and multiple services.

[0048] Implementation 3

[0049] The method for translating intent for multiple users and multiple services in a satellite communication network is the same as that in Examples 1-2. The intent extraction model described in step 4 of the present invention is constructed based on the large-scale pre-trained language model Albert and the Conditional Random Field (CRF). During the extraction process, the <domain, operation, object, result> quadruple is used to characterize user intent. The <domain> includes: communication scenario, user type, and service type; the <operation> includes: upload, download, delete, and add; the <object> includes: network nodes, service flows, and network resources; and the <result> includes: performance indicators, expected states, and spatiotemporal constraints. The present invention utilizes the large-scale pre-trained language model Albert and the Conditional Random Field (CRF) to make intent extraction more accurate and precise. Furthermore, the <domain, operation, object, result> quadruple is used to characterize user intent, enabling a better understanding of user needs and intent. This improves the accuracy and efficiency of intent translation, enhances user experience and satisfaction, and enables more intelligent intent translation for multiple users and multiple services in satellite communication networks.

[0050] Implementation 4

[0051] The multi-user, multi-service intent translation method for a satellite communication network is similar to that of Examples 1-3. Step 4 of the present invention extracts front-end user service intent information to obtain an intent information extraction result. During intent extraction, the user intent text is annotated using a named entity recognition algorithm based on the BIOES annotation method to accurately extract key information about the user service intent. The present invention utilizes the BIOES annotation method combined with the named entity recognition algorithm to more accurately annotate the user intent text, thereby improving the accuracy of intent extraction.

[0052] Implementation 5

[0053] The method for translating intent for multiple users and multiple services in a satellite communication network is similar to that of Examples 1-4. In step 6 of the present invention, in implementing the front-end user's intent translation, the policy mapping module generates a network policy. Specifically, the intent classification results are used to map and generate a coarse-grained network policy. The coarse-grained network policy is a set of network policies provided by the system to meet the basic service quality requirements of the front-end user. The intent extraction results are used to map and generate a fine-grained network policy, which is generated based on the front-end user's original service intent. When a certain indicator in the fine-grained network policy has a default value, the corresponding item in the coarse-grained network policy is filled in to complete the translation of the user's original service intent in the satellite communication network for multiple scenarios, multiple users, and multiple services. The present invention combines coarse-grained and fine-grained network policies to provide front-end users with network policies that meet different service quality requirements, thereby improving user satisfaction and experience. Secondly, using the coarse-grained network policy to fill the default value in the fine-grained network policy avoids the situation where the fine-grained network policy cannot be generated due to the lack of a certain indicator, ensuring the stability and reliability of intent translation.

[0054] Implementation List 6

[0055] The present invention relates to a multi-scenario, multi-user, and multi-service intent translation method for a satellite communication network. The multi-user, multi-service intent translation method for a satellite communication network is similar to that of Examples 1-5. This method addresses the problems of scarce satellite communication network resources, weak dynamic resource adjustment capabilities, diverse user types, and a mismatch or insufficiency between service demands and network resources. The multi-user, multi-service intent translation method for a satellite communication network can identify the communication scenario, user type, and service type to which a service belongs, thereby meeting the service demands of different users in different scenarios and achieving efficient utilization of limited satellite network resources. The multi-user, multi-service intent translation method for a satellite communication network can identify the communication scenario, user type, and service type to which a service belongs, enabling specific processing for specific scenarios, users, and specific services, thereby avoiding a "one-size-fits-all" approach. This method addresses the problems of scarce satellite communication network resources, weak dynamic adjustment capabilities, diverse user types, and a mismatch or insufficiency between service demands and network resources. By identifying the communication scenario, user type, and service type to which a service belongs, the multi-user, multi-service intent translation method can meet the service demands of different users in different scenarios and achieve efficient utilization of limited satellite network resources and enhance user experience.

[0056] The present invention belongs to the field of satellite communication network technology and discloses a method for translating intent for multiple scenarios, multiple users, and multiple services in a satellite communication network. A user inputs the original service intent through a front-end web interface, and the original service intent text is preprocessed and sent to the back-end for further processing. The service intent text is sent to an intent classification model for scenario, user, and service identification. The user's service intent is then input into an intent extraction model to extract key information from the user's intent. The intent classification and extraction results are then fed back to the user for confirmation or modification, and the user's confirmed correct identification result is stored in a database for subsequent dynamic model learning. Finally, the identification result is fed back to the web interface and transmitted to a policy mapping module in JSON file format for parameter mapping to generate a network policy. The present invention constructs scenario, user, and service models by deeply analyzing the air interface characteristics, service types, service priorities, and service quality requirements of users at all levels and various types of satellite stations on the network. This method proposes an efficient intent translation method, laying the foundation for intent-driven network technology in satellite communication networks.

[0057] Implementation List 7

[0058] The intention translation method for multi-user and multi-service satellite communication network is the same as that of embodiments 1-6. Figure 1 As shown, Figure 1 The flowchart of the multi-scenario, multi-user, multi-service intention translation method for a satellite communication network provided by the present invention includes the following steps:

[0059] S101: The front-end user inputs the business intent and pre-processes it. The front-end user inputs the original business intent in natural language through a web page interface. The original business intent text input by the front-end user is pre-processed, specifically including error detection, error correction, padding or truncation operations on the original business intent text, and conversion into a digital index. This ensures that the business intent text input by the front-end user meets the input specifications of the intent recognition module and is then input into the intent recognition module.

[0060] S102: Identify the scenario, user, and service to obtain a classification result; use the intent classification model in the intent recognition module to simultaneously identify the scenario, user, and specific service type of the front-end user's service intent to obtain an identification result; then, by judging the identification result, implement specific service processing for a specific user in a specific scenario; any identification result that is not specific to a specific scenario, user, or service is processed according to the default method or the user-specified method;

[0061] S103: Extracting front-end user service intent information and obtaining an intent information extraction result. After classifying all front-end user service intents, the intent extraction model is used to extract key network parameter information, including bandwidth, latency, duration, spatial location, jitter, and network transmission speed parameters. The service intent classification result and intent information extraction result are fed back to the web page interface for the front-end user to confirm or modify, and the correct recognition result is obtained after the front-end user confirms it.

[0062] S104: Storing the original business intent of the front-end user for dynamic learning of the intent classification and intent extraction model; storing the original business intent input by the front-end user and the correct recognition result after the user's confirmation in the database for dynamic learning of the intent classification and extraction model;

[0063] S105: Realize the intention translation of the front-end user; feed back the correct recognition result confirmed by the front-end user to the front-end web page interface for the front-end user to save, and at the same time input the correct recognition result into the strategy mapping module built based on the finite state machine to generate the network strategy, and complete the network strategy mapping; the network strategy is used for the intention translation of the front-end user in the satellite communication network, realizing the original business intention translation of the front-end user in the satellite communication network with multiple scenarios, multiple users and multiple services.

[0064] The application principles of each module of the present invention are further described in detail below with reference to the accompanying drawings.

[0065] like Figure 2 As shown, Figure 2 This is a schematic diagram of the system framework for multi-scenario, multi-user, and multi-service intent translation in a satellite communication network according to the present invention. The multi-scenario, multi-user, and multi-service intent translation method provided by the present invention includes:

[0066] The front-end user web interface is used to obtain the user's original task intention and transmit the intention to the back-end for processing. The interactive interface includes a user input example and an intention input text box. The user can enter the task intention in the text box according to the prompts of the input example in the interactive interface;

[0067] The data preprocessing module converts the user's intended text into a numerical index that can be recognized by the intent recognition model, and truncates text that exceeds the model's specified input length or pads text that is less than the model's specified input length.

[0068] The intent classification model is used to classify the input user text intent by scene, user and business through natural language processing technologies such as text classification algorithms to achieve specific business processing for specific users in specific scenarios. The intent classification model is based on the Albert multi-task classification model. The model framework is as follows Figure 3 As shown, Figure 3 This is a block diagram of the multi-task module construction used in the intent classification model of the present invention. In this example, the intent classification model in the intent recognition module adopts a hard parameter sharing mode, that is, the bottom layer of the model shares parameters, and the upper layer tasks are independent. It is usually suitable for processing highly related tasks. The communication scenario, user type, and business type to which the identification service belongs are three independent tasks. Compared with using a single-task model to process each task separately, the multi-task model will reduce data processing, model training time, and maintenance costs. At the same time, multiple task results can be obtained by requesting the model only once, which can reduce online reasoning time and enhance the model generalization ability.

[0069] The intent extraction model is used to extract the key information of the user intent text through natural language processing technologies such as naming recognition, and construct the intent quadruple <domain, operation, object, result> to represent the user intent. For example, in this invention, <domain> includes: communication scenarios, user types, business types, etc.; <operation> includes: upload and download, deletion and addition, etc.; <object> includes network nodes, business flows, network resources, etc.; <result> includes performance indicators, expected states, time and space constraints, etc. The named entity recognition here is mainly based on the entity extraction algorithm of the named entity recognition model of Albert-CRF. The model uses a method that combines Albert with Conditional Random Field (CRF). The intent classification results and intent extraction results are fed back to the user for confirmation or modification.

[0070] The intent storage module stores the user's original intent and the recognition results after the user's confirmation in the database, so that the recognition model can be dynamically learned later to improve the recognition accuracy.

[0071] The strategy mapping module uses the intent recognition results to guide the network to generate executable strategies, and sends the generated specific network configuration information to the network elements of the underlying network for execution.

[0072] In summary, the present invention belongs to the field of satellite communication network technology and discloses a method for translating intent for multiple scenarios, multiple users, and multiple services in a satellite communication network. This method solves the problem of existing technologies having significant limitations on user input and is suitable for both professional and non-professional users. Furthermore, the present invention proposes a method combining coarse and fine granularity strategies to address the problem of low resource utilization in satellite communication networks due to the wide variety of scenarios, users, and service types, thereby improving resource utilization and user experience. The user enters the original service intent through the front-end web interface, and the original service intent text is pre-processed and sent to the back-end for further processing. The service intent text is sent to an intent classification model for scenario, user, and service identification. The user's service intent is then input into an intent extraction model to extract key information from the user's intent. The intent classification and extraction results are then fed back to the user for confirmation or modification, and the user's confirmed correct recognition result is stored in a database for subsequent dynamic model learning. Finally, the recognition result is fed back to the web interface and transmitted to the policy mapping module in JSON file format for parameter mapping to generate a network policy. Briefly, the implementation of the present invention includes: constructing an intent recognition module; inputting and preprocessing the front-end user's service intent; identifying scenarios, users, and services to obtain classification results; extracting the front-end user's service intent information to obtain intent information extraction results; storing the front-end user's original service intent to facilitate dynamic learning of the intent classification and intent extraction models; implementing front-end user intent translation; and implementing intent translation for multiple scenarios, multiple users, and multiple services in a satellite communication network. This invention constructs scenario, user, and service models through in-depth analysis of the air interface characteristics, service types, service priorities, and service quality requirements of users at all levels and various types of satellite stations in the network. It then proposes an efficient intent translation method, laying the foundation for intent-driven network technology in satellite communication networks.

[0073] This invention adds an intent classification model to the intent recognition module. It forms a coarse-grained policy by identifying scenarios, users, and services. It uses an intent extraction model to extract key user information to form a fine-grained policy. Through coarse-grained and fine-grained network policies, it can provide front-end users with network policies that meet different service quality requirements, improving user satisfaction and experience. Secondly, using coarse-grained network policies to fill in default values ​​in fine-grained network policies avoids the inability to generate fine-grained network policies due to missing indicators, ensuring the stability and reliability of intent translation.

Claims

1. A multi-scenario, multi-user, and multi-service intent translation method for a satellite communication network, characterized in that: The user's original service intention is received through a web interface at the front end, and the user's original service intention is processed through multiple modules at the back end; the multiple modules include a text preprocessing module, an intention recognition module, and a policy mapping module, and finally a network policy is obtained, and the network policy is used to translate the user's intention in the satellite communication network. The satellite communication network multi-scenario multi-user multi-service intention translation method includes the following steps: Step 1: Construct an intent recognition module. The constructed intent recognition module receives the original business intent input by the front-end user and inputs it into the intent classification model and intent extraction model in sequence. The recognition results of the above two models are used as a whole output, and the correct recognition result is output. The front-end user business intent input into the model must meet the following specifications: the intent text is truncated or padded to a specified length and converted into a digital index. The intent classification model is built based on the text classification algorithm in natural language processing and is used to identify the front-end user's scenario, user, and business type in multiple scenarios, multiple users, and multiple businesses. The intent extraction model is built based on the entity recognition algorithm in natural language processing and is used to extract key network parameter information from the front-end user's business intent; Step 2: The front-end user's business intention is input and pre-processed; The front-end user inputs the original business intent in the form of natural language through the web interface. The original business intent text input by the front-end user is preprocessed, including error detection, error correction, padding or truncation operations on the original business intent text and conversion into a digital index, so that the business intent text originally input by the front-end user meets the input specifications of the intent recognition module and is input into the intent recognition module; Step 3: Identify the scenario, user, and business to obtain a classification result. Use the intent classification model in the intent recognition module to simultaneously identify the scenario, user, and specific business type of the front-end user's business intent to obtain an identification result. Then, by judging the identification result, specific business processing for a specific user in a specific scenario is implemented. Any identification result that is not a specific scenario, user, or business is processed according to the default method or user-specified method. Step 4 extracts the front-end user's service intent information and obtains the intent information extraction results. After classifying all front-end user service intents, the intent extraction model is used to extract key network parameter information, including bandwidth, latency, duration, spatial location, jitter, and network transmission speed parameters. The service intent classification results and intent information extraction results are fed back to the web page interface for the front-end user to confirm or modify, and the correct identification result is obtained after the front-end user confirms it. Step 5: Store the original business intent of the front-end user so that the intent classification and intent extraction models can be dynamically learned. The original business intent input by the front-end user and the correct recognition results after user confirmation are stored in the database so that the intent classification and extraction model can be dynamically learned; Step 6 implements the front-end user's intention translation; the correct recognition result confirmed by the front-end user is fed back to the front-end web interface for the front-end user to save, and the correct recognition result is input into the strategy mapping module built based on the finite state machine to generate the network strategy, completing the network strategy mapping; the network strategy is used for the front-end user's intention translation in the satellite communication network, realizing the front-end user's original business intention translation in the satellite communication network with multiple scenarios, multiple users and multiple services.

2. The satellite communication network multi-scenario multi-user multi-service intent translation method according to claim 1, characterized in that: In step 3, the scene, user, and business are identified to obtain the classification results. The intent classification model in the intent recognition module uses a shared layer multi-task model based on the large-scale pre-trained model Albert and the one-dimensional convolutional model TextCNN to simultaneously identify the scene, user, and business type to obtain the classification results.

3. The satellite communication network multi-scenario multi-user multi-service intent translation method according to claim 1, characterized in that: The intent extraction model described in step 4 is built based on the large-scale pre-trained language model Albert and conditional random fields (CRF). During the extraction process, the <domain, operation, object, result> quadruple is used to represent user intent. The <domain> includes: communication scenarios, user types, and business types; the <operations> include: upload, download, delete, and add; the <objects> include network nodes, business flows, and network resources; and the <results> include performance indicators, expected states, and time and space constraints.

4. The satellite communication network multi-scenario multi-user multi-service intent translation method according to claim 1, characterized in that: Step 4 extracts the front-end user business intention information to obtain the intention information extraction result, wherein the user intention text is annotated using the named entity recognition algorithm based on the BIOES annotation method in the intention extraction.

5. The satellite communication network multi-scenario multi-user multi-service intent translation method according to claim 1, characterized in that: In the implementation of the front-end user's intention translation described in step 6, the strategy mapping module generates a network strategy. Specifically, the intent classification result is used to map and generate a coarse-grained network strategy. The coarse-grained network strategy is a set of network strategies provided by the system to meet the basic business service quality requirements of the front-end user; the intent extraction result is used to map and generate a fine-grained network strategy, which is generated based on the original business intent of the front-end user. When a default value appears in a certain indicator of the fine-grained network strategy, it is filled in by the corresponding item of the coarse-grained network strategy.

Images