A computing power resource configuration method and system based on network intention analysis

Abstract

The application provides a kind of computing power resource configuration method and system based on network intention analysis, comprising: obtaining user natural language input and obtaining standardized intention text by text preprocessing;By large language model, semantic understanding is carried out, business type is identified and explicit intention primitive is extracted;Based on business knowledge graph, explicit intention primitive is inferred and enhanced, and complete intention requirement set is obtained by completing implicit demand;After demand standardization processing and resource semantic mapping, resource description framework is generated;The resource description framework is matched with the computing power identifier of the computing power node, and the candidate computing power node set is screened in combination with real-time network state;Based on resource description framework, candidate node, network state and business constraint, adaptive prompt word is constructed, and candidate configuration scheme is generated by large language model, and the final configuration scheme is output after verification and optimization.The application can improve the integrity and accuracy of intention analysis, realize intelligent generation and adaptive optimization from natural language business intention to computing power configuration scheme.

Description

Technical Field

[0001] This invention relates to the fields of communication networks and artificial intelligence technology, and in particular to a method and system for allocating computing resources based on network intent parsing. Background Technology

[0002] With the development of cloud computing, edge computing, and computing-network convergence technologies, computing resources in networks are gradually exhibiting characteristics of multi-node, heterogeneity, and dynamic changes. Different computing nodes show significant differences in computing power, storage capacity, network bandwidth, transmission latency, and service capabilities. Meanwhile, for business scenarios such as video processing, intelligent analysis, and real-time interaction, users' demands for computing resources are becoming increasingly complex. They often focus not only on resource scale but also on factors such as response speed, deployment location, service quality, and business constraints. Therefore, how to accurately allocate computing resources according to user business needs has become a crucial technical issue in computing networks.

[0003] In existing technologies, computing resource configuration typically relies on preset rules, business templates, or structured parameter input. Users explicitly fill in information such as resource type, resource scale, latency requirements, and bandwidth requirements, and the system then performs node matching and resource allocation. While this method can complete basic configuration, it usually requires users to have strong professional skills and is difficult to adapt to business expressions in natural language. To improve interactivity, some solutions introduce natural language processing technology to segment, extract keywords, identify intent, and classify user input. However, these methods mostly remain at the level of explicit requirement identification and are unable to further explore the implicit resource requirements and performance constraints in the business scenario, nor can they reliably transform vague business descriptions into calculable configuration indicators.

[0004] Furthermore, existing technologies still do not adequately consider the combined real-time network status, service constraints, and resource capabilities. The generated configuration results are prone to problems such as insufficient availability, unsatisfactory performance, or resource conflicts. Moreover, they lack automatic verification and feedback optimization mechanisms for candidate solutions, making it difficult to form a complete closed loop.

[0005] Therefore, there is an urgent need in this field for a computing resource allocation scheme based on network intent parsing, in order to achieve accurate understanding of users' natural language business needs, completion of implicit needs, standardized expression of resource needs, and intelligent generation and optimization of candidate configuration schemes. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method and system for configuring computing resources based on network intent parsing, in order to solve the problems of insufficient deep semantic understanding of natural language business intent in the prior art and the difficulty in adaptively matching dynamic network states with the generated resource configuration scheme.

[0007] One aspect of the present invention provides a method for allocating computing resources based on network intent parsing, the method comprising the following steps: Obtain user natural language input and perform text preprocessing to obtain standardized intent text; The standardized intent text is input into a large language model to perform semantic understanding on the standardized intent text, identify the business type, and extract a set of explicit intent primitives that directly represent user needs. Based on the business knowledge graph, reasoning enhancement is performed on the explicit intent primitive set to complete the implicit requirements corresponding to the business type and the explicit intent primitive set, resulting in a complete intent requirement set; wherein, the complete intent requirement set includes the explicit intent primitive set and the implicit requirements; the business knowledge graph includes the business type, resource type, and performance index corresponding to the network service, as well as the association between the business type, the resource type, and the performance index; Perform requirement normalization processing and resource semantic mapping on the complete intent requirement set to generate a resource description framework, wherein the resource description framework includes resource type, data volume and service performance constraints; The resource description framework is matched with the computing power identifiers of each computing power node in the network, and the matching results are filtered in combination with real-time network status information to obtain a set of candidate computing power nodes. The computing power identifier is used to represent the identity information and capability information of the corresponding computing power node. Based on the resource description framework, the candidate computing node set, the real-time network status information, and business constraints, adaptive prompt words are constructed, and the adaptive prompt words are input into the large language model to generate candidate configuration schemes. The candidate configuration schemes are verified for resource availability, performance metrics, and conflicts. If the verification passes, the final configuration scheme is output. If the verification fails, the adaptive prompts are updated based on the verification results, and the candidate configuration schemes are regenerated based on the updated adaptive prompts until the final configuration scheme is obtained.

[0008] In some embodiments of the present invention, reasoning enhancement is performed on the set of explicit intent primitives based on a business knowledge graph, specifically including: Each explicit intent primitive in the set of explicit intent primitives is linked to the business type entity, resource type entity, and performance indicator entity in the business knowledge graph. Based on the entity linking results, perform relational reasoning along the associations in the business knowledge graph; based on the relational reasoning results, supplement the implicit requirements corresponding to the business type and the set of explicit intent primitives; The association relationship includes at least one of the following: The relationship between the business type entity and the resource type entity in terms of demand, the relationship between the resource type entity and the performance indicator entity in terms of influence, and the equivalence relationship between different natural language expressions.

[0009] In some embodiments of the present invention, after obtaining the complete set of intent requirements, the method further includes: A completeness check is performed on the complete intent requirement set to determine whether the complete intent requirement set simultaneously includes the business type, the resource requirement, and the performance requirement; If the complete set of intent requirements does not meet the completeness requirement, the implicit requirements are supplemented based on the business knowledge graph.

[0010] In some embodiments of the present invention, performing requirement normalization processing and resource semantic mapping on the complete intent requirement set specifically includes: Identify the non-quantitative fuzzy descriptions that characterize user needs within the complete set of intent requirements; Based on a pre-built demand mapping rule base, semantic matching is performed on the non-quantitative fuzzy description, and the semantic matching results are mapped to the values, value ranges or thresholds of the corresponding technical indicators to obtain measurable technical indicators. The technical indicators are mapped into structured output according to preset field rules to form the resource description framework; The structured output includes at least a resource type field, a data volume field, and a service performance constraint field.

[0011] In some embodiments of the present invention, an adaptive prompt word is constructed, and the adaptive prompt word is input into the large language model to generate a candidate configuration scheme, specifically including: Generate configuration task descriptions and requirement descriptions based on the resource description framework; A candidate node description is generated based on the candidate computing power node set, a network status description is generated based on the real-time network status information, and a constraint description is generated based on the business constraints. The configuration task description, the requirement description, the candidate node description, the network status description, and the constraint description are filled into a preset prompt template to form the adaptive prompt words; Based on the adaptive prompt words, the large language model is invoked to generate configuration scheme description information; The configuration scheme description information is parsed to obtain a structured candidate configuration scheme.

[0012] In some embodiments of the present invention, in response to a failed verification, the adaptive prompt word is updated according to the verification result, and the candidate configuration scheme is regenerated based on the updated adaptive prompt word, specifically including: Identify at least one of the following: resource non-compliance, performance non-compliance, and constraint conflict that causes the verification to fail; Based on the recognition results, update the candidate node constraints, resource allocation constraints, or optimization objectives in the adaptive prompts; The updated adaptive prompts are then input into the large language model again to regenerate the candidate configuration scheme.

[0013] In some embodiments of the present invention, acquiring user natural language input and performing text preprocessing to obtain standardized intent text specifically includes: The user's natural language input is sequentially processed through text cleaning, word segmentation, part-of-speech tagging, and named entity recognition to obtain a normalized word sequence; and the normalized intent text is generated based on the normalized word sequence.

[0014] In some embodiments of the present invention, the computing power identifier includes node identity information and node capability information. The node identity information includes at least a node identifier and a node geographical location, and the node capability information includes at least a resource type, a resource scale, and a service capability level. The real-time network status information includes at least one of link bandwidth, link latency, node load, and network congestion status. The step of filtering the matching results by combining real-time network status information includes: removing computing power nodes that do not meet the service performance constraints or the business constraints, and retaining computing power nodes that meet the resource description framework requirements as the candidate computing power node set.

[0015] On the other hand, the present invention also provides a computing resource allocation system based on network intent parsing, including a processor, a memory, and a computer program or instructions stored in the memory, wherein the processor is used to execute the computer program or instructions, and when the computer program or instructions are executed, the system implements the steps of the method described in any of the above.

[0016] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the method as described in any of the preceding claims.

[0017] The computing resource allocation method and system based on network intent parsing provided by this invention integrates the semantic understanding capabilities of a large language model with the domain reasoning capabilities of a business knowledge graph. Based on the extraction of explicit intent primitives from the user's natural language input, it automatically completes the implicit requirements for business operation and transforms the complete intent into a structured resource description framework. Combining computing power identifier matching and real-time network status filtering, it determines the set of candidate computing power nodes. Then, through adaptive prompt words, it drives the large language model to generate and optimize the configuration scheme, realizing end-to-end intelligent generation and adaptive optimization from natural language business intent to computing resource allocation scheme.

[0018] Furthermore, when constructing adaptive prompts, the system generates corresponding descriptive content based on the resource description framework, the candidate computing node set, real-time network status information, and business constraints, and combines and fills them according to the preset prompt template. This enables the large language model to comprehensively perceive business needs, node capabilities, network status, and optimization constraints when generating configuration schemes, thereby outputting candidate configuration schemes that are compatible with the current environment.

[0019] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the description, or may be learned by practice of the invention. The objects and other advantages of the invention can be realized and obtained by means of the structures specifically pointed out in the description and drawings.

[0020] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description

[0021] The accompanying drawings, which are provided to further illustrate the invention and form part of this application, are not intended to limit the scope of the invention.

[0022] Figure 1 This is a flowchart illustrating the computing resource allocation method based on network intent parsing according to an embodiment of the present invention.

[0023] Figure 2 This is an overall architecture diagram of a computing resource allocation system based on network intent parsing according to an embodiment of the present invention.

[0024] Figure 3 This is a schematic diagram of the core process of intent parsing in the computing resource allocation method based on network intent parsing according to an embodiment of the present invention.

[0025] Figure 4This is a schematic diagram of the business knowledge graph structure in the computing resource allocation method based on network intent parsing according to an embodiment of the present invention.

[0026] Figure 5 This is a schematic diagram of the adaptive prompt word template structure in the computing resource allocation method based on network intent parsing according to an embodiment of the present invention.

[0027] Figure 6 This is a schematic diagram of the semantic embedding process for knowledge graph enhancement in the computing resource allocation method based on network intent parsing according to an embodiment of the present invention.

[0028] Figure 7 This is a scenario diagram illustrating the implementation of the computing resource allocation method based on network intent parsing according to an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the embodiments and accompanying drawings. Here, the illustrative embodiments and descriptions of this invention are used to explain the invention, but are not intended to limit the invention.

[0030] It should also be noted that, in order to avoid obscuring the invention with unnecessary details, only the structures and / or processing steps closely related to the solution according to the invention are shown in the accompanying drawings, while other details that are not closely related to the invention are omitted.

[0031] Currently, existing technologies mainly suffer from the following prominent problems: 1) Intent matching based on predefined keyword libraries and rule engines cannot infer and complete implicit needs in natural language input; 2) When using traditional machine learning classification models, the output is discrete business category labels, which cannot directly generate structured resource descriptions containing resource types, data volume, and service performance constraints; 3) The configuration generation process does not incorporate real-time network status information and lacks a verification feedback mechanism, resulting in deviations between the generated configuration scheme and the current availability of network resources.

[0032] In view of this, one aspect of the present invention provides a method for allocating computing resources based on network intent parsing, such as... Figure 1 As shown, the method includes the following steps S101~S107: Step S101: Obtain user natural language input and perform text preprocessing to obtain standardized intent text; Step S102: Input the standardized intent text into the large language model, perform semantic understanding on the standardized intent text, identify the business type, and extract the set of explicit intent primitives that directly represent user needs; Step S103: Based on the business knowledge graph, perform reasoning enhancement on the explicit intent primitive set to complete the implicit requirements corresponding to the business type and the explicit intent primitive set, thus obtaining a complete intent requirement set; wherein, the complete intent requirement set includes the explicit intent primitive set and implicit requirements; the business knowledge graph includes the business type, resource type, and performance index corresponding to the network business, as well as the relationship between the business type, resource type, and performance index; Step S104: Perform requirement normalization processing and resource semantic mapping on the complete intent requirement set to generate a resource description framework, wherein the resource description framework includes resource type, data volume and service performance constraints; Step S105: Match the resource description framework with the computing power identifiers of each computing power node in the network, and filter the matching results in combination with real-time network status information to obtain a set of candidate computing power nodes. The computing power identifier is used to represent the identity information and capability information of the corresponding computing power node. Step S106: Based on the resource description framework, candidate computing power node set, real-time network status information and business constraints, construct adaptive prompt words, and input the adaptive prompt words into the large language model to generate candidate configuration schemes; Step S107: Perform resource availability verification, performance indicator verification, and conflict detection on the candidate configuration schemes; in response to successful verification, output the final configuration scheme; in response to unsuccessful verification, update the adaptive prompt words according to the verification results, and regenerate the candidate configuration schemes based on the updated adaptive prompt words until the final configuration scheme is obtained.

[0033] In step S101, the system acquires the user's natural language input and performs text preprocessing to obtain standardized intent text.

[0034] Specifically, in some embodiments, users submit service request descriptions in natural language form through an interactive interface. The user's natural language input can be text input or text content converted from speech information. After receiving the user's natural language input, the system first performs text cleaning to remove redundant spaces, meaningless special characters, irrelevant punctuation, and duplicate fields, and performs preliminary rectification of the expression in the text.

[0035] In some embodiments, the system performs word segmentation on the cleaned text, dividing the continuous text into word units with independent semantics; then, it performs part-of-speech tagging on the word segmentation results to identify the grammatical attributes corresponding to each word unit; based on this, it further performs named entity recognition to identify entity information related to user needs from the text.

[0036] In some embodiments, the identified entity information includes quantity entities, time entities, business entities, and performance description entities. The system can extract and annotate content such as the number of users, business duration, business type, and performance requirements from user natural language input using named entity recognition. After text cleaning, word segmentation, part-of-speech tagging, and named entity recognition, the system obtains a normalized word sequence and generates standardized intent text based on this sequence.

[0037] In step S101, the system converts the original user natural language input into standardized intent text with a unified format, providing an input basis for subsequent business type identification and requirement extraction.

[0038] In step S102, the system inputs the standardized intent text obtained in step S101 into the large language model, which performs semantic understanding on the standardized intent text, identifies the business type, and extracts a set of explicit intent primitives that directly represent user needs.

[0039] Specifically, in some embodiments, the system invokes the text analysis capabilities of the large language model to perform contextual semantic analysis on the standardized intent text in order to identify semantic content in the text that is related to business scenarios, resource requirements, and performance requirements.

[0040] In some embodiments, the large language model, based on the business description, behavioral object, and demand expression in the standardized intent text, combined with the semantic representation capabilities formed through pre-training, classifies the business scenario corresponding to the user's current intent and outputs the corresponding business type. Business types include video conferencing, file transfer, real-time interaction, and data analysis.

[0041] In some embodiments, the large language model directly extracts the user's explicitly expressed requirements from standardized intent text. These requirements include at least one of the following: a description of business scale, a description of service duration, a description of performance requirements, and a description of constraints. The extracted requirements are written as explicit intent primitives into the explicit intent primitive set, and together with the business type, serve as the output of step S102.

[0042] In some embodiments, the system first constructs a corresponding semantic representation based on standardized intent text, and then performs business type identification and explicit intent primitive extraction based on the semantic representation. The large language model outputs both the business type identification result and a set of explicit intent primitives corresponding to the explicit expressions in the user's natural language input.

[0043] In some embodiments, the large language model can also identify the attribute tendencies of business intent based on the demand characteristics reflected in the standardized intent text. Attribute tendencies include at least one of latency-sensitive, bandwidth-intensive, reliability-required, cost-constrained, and real-time-response types. Attribute tendencies can serve as auxiliary semantic basis for business type identification and explicit intent primitive extraction.

[0044] In step S102, the system obtains the business type and explicit intent primitive set from the standardized intent text, providing input for subsequent implicit requirement completion based on the business knowledge graph.

[0045] In step S103, the system, based on a pre-built business knowledge graph, performs reasoning enhancement on the explicit intent primitive set obtained in step S102, supplementing the implicit requirements corresponding to the business type and the explicit intent primitive set, to obtain a complete intent requirement set. The complete intent requirement set includes the explicit intent primitive set and the implicit requirements.

[0046] Specifically, in some embodiments, the system takes the business type and each explicit intent primitive output in step S102 as input and performs entity linking in the business knowledge graph. The business knowledge graph pre-sets business type entities, resource type entities, and performance indicator entities, and establishes associations between these entities. Through semantic matching, the system links business types to corresponding business type entities and links each explicit intent primitive to its corresponding business type entity, resource type entity, or performance indicator entity.

[0047] In some embodiments, after completing entity linking, the system performs relational reasoning based on the entity linking results, following the associations in the business knowledge graph, to identify requirement content related to the current business type and explicit intent primitives but not directly given in the user's natural language input, and identifies this requirement content as implicit requirements. Associations include at least one of the following: requirement associations between business type entities and resource type entities, influence relationships between resource type entities and performance indicator entities, and equivalence relationships between different natural language expressions. Based on the relational reasoning results, the system supplements the set of explicit intent primitives to obtain a complete set of intent requirements.

[0048] In some embodiments, the demand relationship between business type entities and resource type entities is used to characterize the dependency of a certain type of business on specific computing or network resources; the influence relationship between resource type entities and performance indicator entities is used to characterize the impact of different resource configurations on service performance; and the equivalence relationship between different natural language expressions is used to unify the semantics of different expressions in user input. Through the joint reasoning of the above relationships, the system can complete resource type requirements, performance indicator requirements, and other implicit requirements related to business scenarios from explicit expressions.

[0049] In some embodiments, after obtaining the complete intent requirement set, the system further performs a completeness judgment on the complete intent requirement set to determine whether the complete intent requirement set simultaneously includes business type, resource requirements, and performance requirements; if the complete intent requirement set does not meet the completeness requirements, the system continues to supplement the implicit requirements corresponding to the missing parts based on the business knowledge graph until a complete intent requirement set that meets the completeness requirements is obtained.

[0050] In some embodiments, for user-proposed business scale requirements, service quality requirements, and business duration requirements, the system can first map them to corresponding entities through entity links, and then combine them with the relationships in the business knowledge graph to supplement the resource type requirements and performance indicator requirements that are suitable for the business type, and together with the set of explicit intent primitives, form a complete set of intent requirements.

[0051] Through step S103, the system supplements the implicit requirements related to business configuration that are not directly expressed by the user based on the set of explicit intent primitives, thereby improving the completeness and accuracy of intent parsing results.

[0052] In step S104, the system performs requirement normalization processing and resource semantic mapping on the complete intent requirement set obtained in step S103 to generate a resource description framework.

[0053] Specifically, in some embodiments, the system first scans and parses each requirement in the complete intent requirement set, identifying non-quantitative fuzzy descriptions that represent user needs. Non-quantitative fuzzy descriptions refer to natural language requirement expressions that do not directly correspond to specific numerical values, value ranges, or thresholds. The system treats the identified non-quantitative fuzzy descriptions as objects to be normalized.

[0054] In some embodiments, the system invokes a pre-built demand mapping rule base to perform semantic matching on non-quantitative fuzzy descriptions, and maps the matching results to the values, ranges, or thresholds of corresponding technical indicators to obtain measurable technical indicators. The demand mapping rule base stores the correspondence between different demand expressions and technical indicators; this correspondence is used to convert fuzzy demand expressions in natural language into computable and comparable parameter constraints. The system can match non-quantitative fuzzy descriptions with rule items in the rule base and determine the corresponding technical indicators and their constraint forms based on the matching results.

[0055] In some embodiments, the technical indicators include at least one of resource scale indicators, data volume indicators, and service performance indicators; the constraint forms include at least one of definite values, value ranges, and thresholds. For requirement items in the complete intent requirement set that already contain explicit values ​​or ranges, the system can directly retain their corresponding technical indicators without performing fuzzy description mapping.

[0056] In some embodiments, after completing the requirements standardization process, the system maps technical indicators into structured output according to preset field rules to form a resource description framework. The structured output includes at least a resource type field, a data volume field, and a service performance constraint field. The resource type field characterizes the types of resources required for business operation, the data volume field characterizes the scale of data involved in the business, and the service performance constraint field characterizes the performance requirements related to business operation.

[0057] In some embodiments, the system categorizes, populates, and formats the standardized technical indicators, uniformly writing demand information from different sources into the corresponding fields to obtain a complete and uniformly formatted resource description framework. In some embodiments, the resource description framework can be encoded and output using a structured data format to facilitate subsequent execution of computing node matching and configuration scheme generation.

[0058] In step S104, the system transforms the natural language requirements in the complete set of intent requirements into a structured and measurable resource description framework, thereby providing a unified input for subsequent resource matching and configuration scheme generation.

[0059] In step S105, the system matches the resource description framework generated in step S104 with the computing power identifiers of each computing power node in the network, and filters the matching results in conjunction with real-time network status information to obtain a set of candidate computing power nodes.

[0060] Specifically, in some embodiments, each computing node in the network is assigned a corresponding computing power identifier, which represents the node's identity and capability information. The node identity information includes at least a node identifier and a node's geographical location, while the node capability information includes at least resource type, resource size, and service capability level. The system reads the resource type field, data volume field, and service performance constraint field from the resource description framework and obtains the computing power identifier information corresponding to each computing node in the network.

[0061] In some embodiments, the system first matches the resource description framework with the computing power identifier of each computing power node. Specifically, the system compares the resource type requirements in the resource description framework with the resource types of each computing power node to filter out computing power nodes whose resource types meet the requirements; then, it compares the data volume requirements and service performance constraints in the resource description framework with the node capability information to further filter out computing power nodes whose resource scale and service capability level meet the requirements, thereby obtaining the initial matching result.

[0062] In some embodiments, the service capability level is used to characterize the ability of computing power nodes to support different business needs. The system can further constrain the initial matching results based on the service capability level to improve the compatibility between candidate computing power nodes and current business needs.

[0063] In some embodiments, after completing the above matching, the system filters the matching results by combining real-time network status information. Real-time network status information includes at least one of link bandwidth, link latency, node load, and network congestion status. The system obtains the real-time network status information corresponding to each candidate computing node and compares the real-time network status information with the service performance constraints and business constraints in the resource description framework.

[0064] In some embodiments, the matching results are filtered by combining real-time network status information, including: removing computing power nodes that do not meet service performance constraints or business constraints, and retaining computing power nodes that meet the resource description framework requirements as a candidate computing power node set. Specifically, for business requirements with latency constraints, the system checks whether the current latency of the link corresponding to the candidate computing power node meets the constraint requirements; for business requirements with bandwidth constraints, the system checks whether the current bandwidth of the link corresponding to the candidate computing power node meets the constraint requirements; for node load and network congestion status, the system removes computing power nodes in a high-load or congested state according to preset filtering conditions.

[0065] In some embodiments, the system generates a set of candidate computing power nodes by first static matching and then dynamic filtering. Static matching is used to filter computing power nodes that meet the requirements of the resource description framework based on computing power identifiers, while dynamic filtering is used to eliminate computing power nodes that do not currently meet the service performance constraints or business constraints by combining real-time network status information.

[0066] In step S105, the system selects a set of candidate computing nodes from the computing nodes in the network that meet the requirements of the resource description framework and are suitable for the current network state, providing a candidate node basis for the subsequent configuration scheme generation.

[0067] In step S106, the system constructs adaptive prompt words based on the resource description framework, the candidate computing node set, real-time network status information, and business constraints, and inputs the adaptive prompt words into the large language model to generate candidate configuration schemes.

[0068] Specifically, in some embodiments, the system first generates a configuration task description and a requirement description based on the resource description framework. The configuration task description is used to characterize the target content of the current resource configuration task, and the requirement description is used to characterize the resource type requirements, data volume requirements, and service performance constraint requirements corresponding to the business type.

[0069] In some embodiments, the system generates candidate node descriptions based on a set of candidate computing power nodes. The candidate node descriptions include the node identifier, geographical location, available resource type, remaining resource scale, and service capability level of each candidate computing power node. The system also generates network status descriptions based on real-time network status information, which characterize link bandwidth, link latency, node load, and network congestion status related to the candidate computing power nodes. The system further generates constraint descriptions based on business constraints, which include at least one of cost constraints, latency constraints, deployment location constraints, load balancing requirements, and reliability requirements.

[0070] In some embodiments, the system fills the preset prompt template with configuration task description, requirement description, candidate node description, network status description, and constraint description to form adaptive prompt words. Since the candidate computing node set and real-time network status information are updated as the network environment changes, the system updates and fills the prompt template based on the current candidate computing node set and real-time network status information each time a candidate configuration scheme is generated, so that the adaptive prompt words are consistent with the current configuration environment.

[0071] In some embodiments, the system inputs adaptive prompts into a large language model, which then performs comprehensive reasoning based on resource requirements, candidate node status, network status, and business constraints to output configuration scheme description information. The configuration scheme description information includes the selected computing nodes, the resource allocation results for each node, and deployment strategy information. The resource allocation results include computing resource allocation results, storage resource allocation results, and network resource allocation results; the deployment strategy information includes collaborative deployment strategies and load balancing strategies.

[0072] In some embodiments, the system parses the configuration scheme description information output by the large language model, extracts node identifiers, resource allocation parameters and deployment strategy parameters, and converts them into structured data according to preset field rules to obtain structured candidate configuration schemes.

[0073] In some embodiments, the process of constructing adaptive prompt words and generating candidate configuration schemes includes: generating configuration task descriptions and requirement descriptions based on a resource description framework; generating candidate node descriptions based on a candidate computing power node set, generating network status descriptions based on real-time network status information, and generating constraint descriptions based on business constraints; filling the configuration task description, requirement description, candidate node description, network status description, and constraint description into a preset prompt template to form adaptive prompt words; calling a large language model based on the adaptive prompt words to generate configuration scheme description information; and parsing the configuration scheme description information to obtain structured candidate configuration schemes.

[0074] In some embodiments, the system may use a dynamic template selection method to construct prompt templates. The system selects the corresponding scenario template based on the business type in the resource description framework, selects the corresponding constraint template based on the optimization objective in the business constraints, and combines the selected scenario template, constraint template, and basic template. Then, it fills in the candidate computing power node information and real-time network status information to form adaptive prompt words.

[0075] In step S106, the system combines the resource description framework, the candidate computing node set, real-time network status information, and business constraints to generate a structured candidate configuration scheme.

[0076] In step S107, the system performs resource availability verification, performance indicator verification, and conflict detection on the candidate configuration schemes generated in step S106. If the verification passes, the final configuration scheme is output; if the verification fails, the adaptive prompts are updated based on the verification results, and candidate configuration schemes are regenerated based on the updated adaptive prompts until the final configuration scheme is obtained.

[0077] Specifically, in some embodiments, the system first performs resource availability verification on the candidate configuration schemes. The system reads the resource allocation results corresponding to each computing node in the candidate configuration schemes and compares them with the currently available computing resources, storage resources, and network resources of each computing node to determine whether the resource allocation in the candidate configuration schemes exceeds the actual available resource range of the corresponding node. If there is a situation where the resource allocation result exceeds the available resources of the node, the resource availability verification is determined to have failed, and the corresponding resource non-compliance items are identified.

[0078] In some embodiments, the system performs performance metric verification on candidate configuration schemes. Based on the node selection results, resource allocation results, and deployment strategies in the candidate configuration schemes, the system analyzes the performance metrics achievable by the candidate configuration schemes and compares the analysis results with the service performance constraints in the resource description framework to determine whether the candidate configuration schemes meet the corresponding service performance requirements. If a candidate configuration scheme fails to meet the service performance constraints, the performance metric verification is deemed unsuccessful, and the corresponding performance non-compliance items are identified.

[0079] In some embodiments, the system performs conflict detection on candidate configuration schemes to identify whether resource allocation conflicts or constraint conflicts exist among the candidate configuration schemes. Resource allocation conflicts include resource over-utilization caused by multiple service requests competing for the same computing power node resources, while constraint conflicts include inconsistencies between different service constraints or different optimization objectives. If a conflict is detected, the conflict detection is deemed to have failed, and the corresponding constraint conflict item is identified.

[0080] In some embodiments, when a candidate configuration scheme passes resource availability verification, performance metric verification, and conflict detection, the system determines the candidate configuration scheme as the final configuration scheme and outputs it. The final configuration scheme includes the selected computing nodes, the resource allocation results corresponding to each computing node, and deployment strategy information.

[0081] In some embodiments, in response to a verification failure, the system identifies at least one of the following: resource non-compliance, performance non-compliance, and constraint conflict. Based on the identification result, the system updates the candidate node constraints, resource allocation constraints, or optimization objectives in the adaptive prompts. For resource non-compliance, the system adjusts the candidate node constraints or resource allocation constraints in the prompts to limit the resource allocation range of the corresponding node. For performance non-compliance, the system adjusts the performance-related constraints in the prompts to guide the large language model to generate candidate configuration schemes that better meet service performance constraints. For constraint conflict, the system modifies the optimization objectives or constraints in the prompts to reduce the probability of conflict.

[0082] In some embodiments, the system re-inputs the updated adaptive prompts into the large language model, which then regenerates candidate configuration schemes. The system then repeatedly performs resource availability verification, performance metric verification, and conflict detection on the regenerated candidate configuration schemes until a final configuration scheme that passes verification is obtained.

[0083] In some embodiments, the system may also store and record the final configuration scheme that has been verified, and write the prompt word construction method and configuration parameters corresponding to the final configuration scheme into the knowledge base for reference in subsequent similar business scenarios.

[0084] Through step S107, the system verifies and iteratively optimizes the candidate configuration schemes, thereby improving the reliability of the final configuration scheme in terms of resource availability, performance satisfaction, and constraint consistency.

[0085] On the other hand, the present invention also provides a computing resource allocation system based on network intent parsing, including a processor, a memory, and a computer program or instructions stored in the memory, wherein the processor is used to execute the computer program or instructions, and when the computer program or instructions are executed, the system implements the steps of the method described in any of the above.

[0086] On the other hand, the present invention also provides a computer-readable storage medium having a computer program or instructions stored thereon, which, when executed by a processor, implement the steps of the method as described in any of the preceding claims.

[0087] The present invention will now be described with reference to a specific embodiment: This embodiment will elaborate in detail on the complete technical solution of the computing resource allocation method and system based on network intent parsing proposed in this invention.

[0088] This invention provides a method and system for allocating computing resources based on network intent parsing. By introducing the semantic understanding capabilities of a large language model, combined with business knowledge graph enhancement and adaptive prompt word mechanism, it achieves intelligent conversion from user natural language input to a structured resource description framework.

[0089] I. System Architecture for Computing Resource Allocation Based on Network Intent Parsing This invention proposes a computing resource allocation system architecture based on network intent parsing. For example... Figure 2 As shown, this architecture integrates the semantic understanding capabilities of a large language model with network domain knowledge, and consists of three layers: an input interface layer, an intent parsing core layer, and a knowledge support layer, enabling intelligent conversion from user natural language input to candidate configuration schemes. A detailed description of each layer is as follows.

[0090] Input Interface Layer: This layer is responsible for receiving and preprocessing user input, supporting various input methods including text and speech-to-text conversion. Through modules such as format normalization, text cleaning, and word segmentation, it converts raw unstructured input into standardized intent text and passes it to lower layers via the northbound interface. This layer serves as the entry point for user interaction with the system, ensuring that input of various forms and expressions can be correctly recognized and processed by subsequent modules.

[0091] The core layer for intent parsing is the heart of the system's intelligent intent parsing, primarily composed of three key components: the intent parsing component, the requirement translation component, and the configuration generation component. The intent parsing component utilizes a large language model for deep semantic understanding, combined with business knowledge graph enhancement technology, to accurately identify the user's business type and explicit intent primitives, and complete implicit requirements. The requirement translation component transforms abstract explicit intent primitives into concrete, measurable Resource Description Frameworks (RDFs) through requirement reasoning and normalization. The configuration generation component, based on an adaptive prompting mechanism, automatically generates executable resource configuration schemes by calling the large language model and performs verification and optimization. These three components work collaboratively to achieve end-to-end intent parsing and configuration generation.

[0092] Knowledge Support Layer: This layer provides necessary domain knowledge and real-time status information support for intent parsing, mainly composed of three functional modules: a business knowledge graph, a network status information base, and a requirement mapping rule base. The business knowledge graph stores entities and their relationships, such as business types, resource types, and performance metrics, supporting semantic reasoning and implicit requirement completion. The network status information base collects and updates information such as resource availability, network load, and link status of edge nodes and cloud nodes in real time, providing real-time status awareness for configuration generation. The requirement mapping rule base defines conversion rules from fuzzy descriptions to specific metrics, such as mapping "no lag" to "latency less than 100 milliseconds." Through the southbound interface, this layer provides knowledge query and status query services to various components of the intent parsing core layer.

[0093] Furthermore, to achieve precise mapping from intent to resources, a Computing Power Identifier (CPI) is defined for each computing power node in this system. This CPI comprises two parts: node identity information and node capability information. The node identity information consists of the node's geographical location and IP address, enabling rapid node authentication and location. The node capability information consists of resource type and service capability level. Resource type includes descriptions of resources such as GPU, CPU, storage, and network. The service capability level uses a scoring system to describe the node's service capabilities for different business types, such as latency-sensitive, bandwidth-intensive, compute-intensive, and security-sensitive services, with scores ranging from 0 to 100. By matching the user's required Resource Description Framework (RDF) with the node's Computing Power Identifier (CPI), a set of candidate computing power nodes matching the business intent can be quickly selected.

[0094] II. Intent parsing mechanism of large language model based on knowledge graph enhancement To achieve accurate intent understanding and implicit requirement completion in network intent parsing, this invention categorizes business intent types from two aspects. First, it classifies intents based on the quality of service (QoS) provided by the network, such as latency-sensitive intents, bandwidth-intensive intents, reliability-requirement intents, and accuracy-requirement intents. These intents often indicate user requirements related to QoS metrics such as network latency, bandwidth capacity, service reliability, and computational accuracy. Second, it classifies intents based on the user's subjective business needs, such as security-sensitive intents, cost-constrained intents, real-time response intents, and high-throughput intents. These intents often indicate the user's subjective expectations regarding security, cost-effectiveness, response speed, and processing capacity.

[0095] To abstract service intents in the network into resource requirements and further map them to configuration policies, this invention employs the following approach: Figure 3 The diagram illustrates an intent parsing process based on the fusion of a large language model and a business knowledge graph. Specifically, the input interface layer receives and preprocesses the user's natural language input. The normalized text is then fed into the core intent parsing layer. Deep semantic understanding based on the large language model identifies the business type and explicit intent primitives. Implicit requirements are then completed using business knowledge graph enhancement. Next, the explicit intent primitives are transformed into a resource description framework, and fuzzy descriptions are normalized into measurable technical indicators. Finally, a structured resource description framework containing resource requirements and performance constraints is generated and input into the configuration generation component to generate an executable configuration scheme. During configuration deployment, the feasibility and performance of the configuration scheme must be verified to allow for adjustments, optimizations, or regeneration of the configuration scheme in case of resource conflicts or performance issues.

[0096] In the intent parsing phase, after receiving and preprocessing user input, the system performs deep semantic understanding based on a pre-trained large language model. By encoding the text into high-dimensional semantic vectors, it captures the deep semantic information and contextual relationships of the text, and identifies the user's business type and explicit intent primitives through semantic analysis. Business types include video conferencing, file transfer, and real-time monitoring, while explicit intent primitives refer to key phrases in the user input that directly reflect the user's needs, such as "no lag" or "1000 people online." This phase expresses the recognition results using a combination of business type and explicit intent primitive set.

[0097] Further, after completing intent parsing and requirement translation, the system enters the configuration generation phase. The configuration generation component performs Prompt construction based on the resource description framework, real-time network status information, business constraints, and candidate computing node information, and generates adaptive prompts according to a preset prompt template. These adaptive prompts can include task definitions, requirement descriptions, network status descriptions, and constraint descriptions, providing the large language model with the task objectives, resource requirements, and optimization directions needed for configuration generation. The large language model generates configuration scheme description information containing node selection, resource allocation, and deployment strategies based on the adaptive prompts. The system parses the configuration scheme description information to obtain structured candidate configuration schemes. Subsequently, the system verifies the candidate configuration schemes; if the verification fails, the adaptive prompts are adjusted according to the verification results, and the large language model is called again to generate new candidate configuration schemes. During multiple iterations, multiple adaptive prompts generated in the order of generation can form a Prompt sequence, thereby achieving iterative optimization of the configuration scheme.

[0098] After obtaining the business type and explicit intent primitives, business knowledge graph enhancement is needed to complete the implicit requirements. This invention pre-constructs a business knowledge graph containing entities such as business types, resource types, performance metrics, and their relationships, such as... Figure 4 As shown in the diagram, this knowledge graph defines "demand association" relationships, indicating that business types require specific resource types; it defines "impact" relationships, indicating that resource configuration affects performance metrics; and it defines "equivalence" relationships, indicating the semantic equivalence between different expressions. Based on this business knowledge graph, the system links the identified business types and explicit intent primitives to the corresponding entities in the graph and performs inference through the relationships defined in the graph. Starting from the "high-definition video conferencing" business type, implicit requirements such as GPU computing resources, high-bandwidth networks, and low-latency transmission can be inferred through demand association relationships; starting from the explicit intent primitive "1000 people online," implicit requirements such as load balancing capabilities and high-concurrency processing capabilities can be inferred. These inferred implicit requirements are then added to the explicit intent primitives to form a complete set of intent requirements.

[0099] In the requirement translation phase, the identified and completed requirements need to be converted into a structured resource description framework. First, requirement standardization is performed, converting vague descriptions in the text, such as non-quantitative descriptions like "no lag," "smooth," and "high definition," into specific measurable indicators. This invention predefines mapping rules in a requirement mapping rule base, such as mapping "no lag" to "latency less than 100 milliseconds," "1000 people online" to "concurrent users greater than or equal to 1000," and "high definition quality" to "video resolution greater than or equal to 1080p." Then, the translated requirements are output in a resource description framework format, which consists of three parts: a resource type field, a data volume field, and a service performance constraint field. The resource type field describes the types of computing resources required by the user's business, including GPU resources, CPU resources, storage resources, bandwidth resources, etc.; the data volume field describes the data scale of the business; and the service performance constraint field describes the business's requirements for specific service quality, including latency less than 100 milliseconds, concurrency greater than or equal to 1000, and resolution greater than or equal to 1080p.

[0100] Based on the resource description framework and the computing power identifiers of computing power nodes, the mapping from intent to policy can be completed, and a set of candidate computing power nodes can be selected. First, the resource description framework and computing power identifiers are matched using methods such as similarity calculation and vector matching. Then, computing power nodes with a matching degree reaching a preset threshold are set as candidate computing power nodes. Finally, based on the service performance constraints in the resource description framework and the current network state, the optimization objectives for configuration generation are determined. Optimization objectives include load balancing, cost optimization, and latency optimization, and the candidate computing power node information and optimization objectives are passed to the configuration generation component. Typical representative business scenarios and configuration schemes will be stored in the system knowledge base for reference in subsequent similar tasks.

[0101] III. Configuration Generation and Optimization Mechanism Based on Adaptive Prompt Words like Figure 5 As shown, a configuration generation process based on an adaptive prompting word mechanism is presented. This process fully utilizes the generation capabilities of a large language model, combined with real-time network conditions and business constraints, to automatically generate an executable resource configuration scheme.

[0102] Step 1: During the requirements translation phase, generate the resource description framework for the business and transfer it to the configuration generation component; Step 2: Obtain the current network resource status information from the network status information database, including the real-time status of each node's CPU and GPU resource availability, storage capacity, network bandwidth, link latency, etc., and update it to the configuration generation component; Step 3: Match the resource requirements in the Resource Description Framework (RDF) with the Computing Power Identifier (CPI) of the computing power nodes to filter out a set of candidate computing power nodes that meet the business intent; Step 4: Combine the resource description framework, candidate node information, and network status to construct adaptive prompts. These prompts include: task definition, clearly defining the generated goals and output format; requirement information, injecting the business type, performance requirements, and resource requirements from the resource description framework; network status, injecting real-time resource availability and network status; and constraints, clearly defining optimization goals and constraints, including prioritizing edge resources, ensuring load balancing, and minimizing costs. Step 5: Call the large language model, input the constructed adaptive prompt words, and generate a configuration scheme that includes computing resources, storage resources, network resources, and other dimensions. Step 6: Verify and optimize the generated configuration scheme, including resource availability checks, performance indicator verification, conflict detection, and optimize the configuration or return to regenerate based on the verification results, until a feasible and optimized final configuration scheme is obtained.

[0103] In the aforementioned intent parsing and configuration generation process based on a large language model, the adaptive prompt word mechanism is one of the key technologies. To improve the accuracy and applicability of configurations generated by the large language model, this invention designs an adaptive prompt word mechanism for different business scenarios and network conditions. The core of this mechanism lies in dynamically selecting and combining prompt word templates based on the current business requirement type, network condition, and optimization goals to form prompt words suitable for the current scenario.

[0104] Furthermore, to enhance the support capability of the preceding intent parsing results for subsequent configuration generation, such as... Figure 6 As shown, in the semantic embedding process enhanced by the knowledge graph, after completing entity linking and relation reasoning, the system fuses the original semantic vectors generated by the large language model with the domain knowledge vectors obtained from the business knowledge graph to form the enhanced semantic embedding. This fusion process can be represented as: V-fusion = α·V-input + β·V-knowledge graph; Wherein, Vinput represents the high-dimensional semantic vector obtained by encoding the user input text using the large language model; Vknowledge graph represents the knowledge vector encoded based on the entity links and relation reasoning results in the business knowledge graph; α represents the fusion weight of the semantic vector; and β represents the fusion weight of the knowledge vector, used to adjust the contribution ratio between the original semantic understanding and domain knowledge enhancement. Through the above semantic embedding fusion, the system incorporates implicit requirement information reasoned from the knowledge graph into the original semantic representation, including GPU computing resource requirements, low-latency transmission requirements, and high-bandwidth network requirements, enabling subsequent requirement translation and configuration generation to be carried out on a richer and more accurate semantic basis.

[0105] During configuration generation, the system pre-designs a prompt word template library, which includes basic templates, scenario templates, and constraint templates. The basic templates define task objectives and output formats; the scenario templates adapt to different business scenarios such as video conferencing, file transfer, and online gaming; and the constraint templates define the expression of different optimization objectives such as cost optimization, latency optimization, and load balancing. During configuration generation, the system selects the corresponding scenario template based on the business type in the resource description framework, and the corresponding constraint template based on the business constraints and the current network status. The selected templates are then combined with the basic templates, and specific parameter values ​​are filled in, including candidate computing node information, real-time resource availability, and specific requirement indicators from the resource description framework, forming a complete adaptive prompt word. After this prompt word is input into a large language model, the model can generate candidate configuration schemes optimized for the current scenario based on the task guidance and constraint information provided in the prompt word.

[0106] During configuration generation, the system employs a dynamic template selection and combination strategy. Based on the service type in the resource description framework, the system automatically selects the corresponding scenario template. For video conferencing services, a video conferencing scenario template containing guidance information such as low-latency transmission, stable bandwidth guarantees, and GPU-accelerated encoding / decoding can be selected. Based on service constraints and the current network status, the system selects the corresponding constraint template. If current edge resources are sufficient and users require low latency, the constraint template prioritizing edge resources is selected; if the current network load is unbalanced, the constraint template prioritizing load balancing is selected. The system combines the selected basic template, scenario template, and constraint template, and fills in specific parameter values ​​to form a complete adaptive prompt.

[0107] For verified and feasible configuration schemes, the system establishes a continuous optimization mechanism. After the configuration scheme is issued and executed, the system monitors the execution effect in real time, collecting indicators such as actual latency, resource utilization, and service quality. If the execution effect is unsatisfactory, such as actual latency exceeding requirements, resource utilization being too low, or service quality failing to meet standards, the system will use the feedback information to update the requirement mapping rule base and prompt word template base. By continuously accumulating configuration experience and optimizing templates, the system can continuously improve the quality and accuracy of generated configuration schemes. For typical and representative business scenarios and configuration schemes, the system can also store prompt word combination patterns and configuration parameters in a knowledge base, providing reference templates for subsequent similar tasks.

[0108] Through the aforementioned adaptive prompting mechanism, the system can dynamically generate configuration schemes suitable for the current situation based on different business scenarios and network conditions, thereby achieving intelligent and adaptive configuration generation.

[0109] The present invention will now be described with reference to another specific embodiment: This embodiment uses a typical video conferencing service deployment request as an example scenario to fully demonstrate the end-to-end processing of the present invention's method from user natural language input to the generation of the final computing resource configuration scheme. Specifically, this embodiment showcases the complete workflow of text preprocessing, large language model semantic understanding, business knowledge graph reasoning and completion, requirement translation and resource description framework generation, computing power identifier matching and candidate computing power node selection, adaptive prompt word construction, large language model configuration scheme generation, and configuration verification.

[0110] like Figure 7 The following is a specific implementation case of intent parsing. A user inputs via natural language: "I need to deploy a high-definition video conferencing system that supports 1000 simultaneous online users, requiring smooth, uninterrupted video playback, and a meeting duration of approximately 2 hours." The system first preprocesses the input text, identifying key information such as the quantity entity "1000 people," the time entity "2 hours," and the business entity "video conferencing." Then, through a large language model for semantic understanding, it identifies the business type as "high-definition video conferencing," with explicit intent primitives including "supports 1000 online users," "smooth video playback," "uninterrupted playback," and "duration of 2 hours." Next, through business knowledge graph enhancement, it infers implicit requirements such as GPU resources, low-latency networks, and high bandwidth from the "high-definition video conferencing" business type, and specific performance requirements such as network latency of less than 100 milliseconds and packet loss rate of less than 1% from the "uninterrupted playback" intent primitive. The requirements were then translated, converting the explicit requirement of "1000 online users" into "concurrent users greater than or equal to 1000," and the implicit requirements into measurable metrics such as "bandwidth greater than or equal to 1Gbps" and "availability greater than or equal to 99.9%." This resulted in a structured resource description framework: the resource type field was GPU and high-bandwidth network; the data volume field was video traffic; and the service performance constraints included latency less than 100 milliseconds, concurrency greater than or equal to 1000, and duration of 2 hours. This resource description framework was then matched with the computing power identifiers of candidate nodes to filter out eligible edge and cloud nodes. Finally, adaptive prompts were constructed and a large language model was invoked to generate candidate configuration schemes: allocating 3 edge GPU nodes and 2 cloud GPU nodes, each node configured with 8 CPU cores, 64GB of memory, 500GB of cache, and a 1Gbps dedicated channel, employing an edge-cloud collaborative deployment strategy and load balancing mechanism. After configuration verification, this scheme met the resource availability and performance requirements and was deployed to the network layer for execution as the final configuration scheme.

[0111] In summary, the computing resource allocation method and system based on network intent parsing provided by this invention integrates the semantic understanding capabilities of a large language model with the domain reasoning capabilities of a business knowledge graph. Based on extracting explicit intent primitives from user natural language input, it automatically completes the implicit requirements for business operation and transforms the complete intent into a structured resource description framework. Combining computing power identifier matching and real-time network status filtering, it determines a set of candidate computing power nodes. Furthermore, through adaptive prompts, it drives the large language model to generate and optimize the configuration scheme, achieving end-to-end intelligent generation and adaptive optimization from natural language business intent to computing resource allocation schemes.

[0112] Furthermore, when constructing adaptive prompts, the system generates corresponding descriptive content based on the resource description framework, the candidate computing node set, real-time network status information, and business constraints, and combines and fills them according to the preset prompt template. This enables the large language model to comprehensively perceive business needs, node capabilities, network status, and optimization constraints when generating configuration schemes, thereby outputting candidate configuration schemes that are compatible with the current environment.

[0113] Those skilled in the art will understand that the exemplary components, systems, and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software, or a combination of both. Whether implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. When implemented in hardware, it can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this invention are programs or code segments used to perform the desired tasks. The programs or code segments can be stored in a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried in a carrier wave.

[0114] It should be clarified that the present invention is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of the present invention.

[0115] In this invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or in place of features of other embodiments.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations of the embodiments of the present invention are possible. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for allocating computing resources based on network intent parsing, characterized in that, The method includes: Obtain user natural language input and perform text preprocessing to obtain standardized intent text; The standardized intent text is input into a large language model to perform semantic understanding on the standardized intent text, identify the business type, and extract a set of explicit intent primitives that directly represent user needs. Based on the business knowledge graph, reasoning enhancement is performed on the explicit intent primitive set to complete the implicit requirements corresponding to the business type and the explicit intent primitive set, resulting in a complete intent requirement set; wherein, the complete intent requirement set includes the explicit intent primitive set and the implicit requirements; the business knowledge graph includes the business type, resource type, and performance index corresponding to the network service, as well as the association between the business type, the resource type, and the performance index; Perform requirement normalization processing and resource semantic mapping on the complete intent requirement set to generate a resource description framework, wherein the resource description framework includes resource type, data volume and service performance constraints; The resource description framework is matched with the computing power identifiers of each computing power node in the network, and the matching results are filtered in combination with real-time network status information to obtain a set of candidate computing power nodes. The computing power identifier is used to represent the identity information and capability information of the corresponding computing power node. Based on the resource description framework, the candidate computing node set, the real-time network status information, and business constraints, adaptive prompt words are constructed, and the adaptive prompt words are input into the large language model to generate candidate configuration schemes. The candidate configuration schemes are verified for resource availability, performance metrics, and conflicts. If the verification passes, the final configuration scheme is output. If the verification fails, the adaptive prompts are updated based on the verification results, and the candidate configuration schemes are regenerated based on the updated adaptive prompts until the final configuration scheme is obtained.

2. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, Based on the business knowledge graph, reasoning enhancement is performed on the set of explicit intent primitives, specifically including: Each explicit intent primitive in the set of explicit intent primitives is linked to the business type entity, resource type entity, and performance indicator entity in the business knowledge graph. Based on the entity linking results, perform relational reasoning along the associations in the business knowledge graph; based on the relational reasoning results, supplement the implicit requirements corresponding to the business type and the set of explicit intent primitives; The association relationship includes at least one of the following: The relationship between the business type entity and the resource type entity in terms of demand, the relationship between the resource type entity and the performance indicator entity in terms of influence, and the equivalence relationship between different natural language expressions.

3. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, After obtaining the complete set of intent requirements, it also includes: A completeness check is performed on the complete intent requirement set to determine whether the complete intent requirement set simultaneously includes the business type, the resource requirement, and the performance requirement; If the complete set of intent requirements does not meet the completeness requirement, the implicit requirements are supplemented based on the business knowledge graph.

4. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, Perform requirement normalization and resource semantic mapping on the complete set of intent requirements, specifically including: Identify the non-quantitative fuzzy descriptions that characterize user needs within the complete set of intent requirements; Based on a pre-built demand mapping rule base, semantic matching is performed on the non-quantitative fuzzy description, and the semantic matching results are mapped to the values, value ranges or thresholds of the corresponding technical indicators to obtain measurable technical indicators. The technical indicators are mapped into structured output according to preset field rules to form the resource description framework; The structured output includes at least a resource type field, a data volume field, and a service performance constraint field.

5. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, Construct adaptive prompt words and input the adaptive prompt words into the large language model to generate candidate configuration schemes, specifically including: Generate configuration task descriptions and requirement descriptions based on the resource description framework; A candidate node description is generated based on the candidate computing power node set, a network status description is generated based on the real-time network status information, and a constraint description is generated based on the business constraints. The configuration task description, the requirement description, the candidate node description, the network status description, and the constraint description are filled into a preset prompt template to form the adaptive prompt words; Based on the adaptive prompt words, the large language model is invoked to generate configuration scheme description information; The configuration scheme description information is parsed to obtain a structured candidate configuration scheme.

6. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, In response to a failed verification, the adaptive prompt words are updated based on the verification result, and the candidate configuration scheme is regenerated based on the updated adaptive prompt words, specifically including: Identify at least one of the following: resource non-compliance, performance non-compliance, and constraint conflict that causes the verification to fail; Based on the recognition results, update the candidate node constraints, resource allocation constraints, or optimization objectives in the adaptive prompts; The updated adaptive prompts are then input into the large language model again to regenerate the candidate configuration scheme.

7. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, Obtain user natural language input and perform text preprocessing to obtain standardized intent text, specifically including: The user's natural language input is sequentially processed through text cleaning, word segmentation, part-of-speech tagging, and named entity recognition to obtain a normalized word sequence; and the normalized intent text is generated based on the normalized word sequence.

8. The computing resource allocation method based on network intent parsing according to claim 1, characterized in that, The computing power identifier includes node identity information and node capability information. The node identity information includes at least a node identifier and a node geographical location. The node capability information includes at least a resource type, a resource scale, and a service capability level. The real-time network status information includes at least one of link bandwidth, link latency, node load, and network congestion status. The step of filtering the matching results by combining real-time network status information includes: removing computing power nodes that do not meet the service performance constraints or the business constraints, and retaining computing power nodes that meet the resource description framework requirements as the candidate computing power node set.

9. A computing resource allocation system based on network intent parsing, comprising a processor, a memory, and computer programs or instructions stored in the memory, characterized in that, The processor is configured to execute the computer program or instructions, and when the computer program or instructions are executed, the system implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 8.

Images