Network service generation method, device and equipment and readable storage medium

Abstract

The invention discloses a network service generation method, device and equipment and a readable storage medium, and is applied to the technical field of computers, and the method comprises the steps: carrying out the multi-granularity intention recognition of a network service generation intention, and obtaining the intention recognition results of different granularities; analyzing the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention; constructing a scene graph corresponding to each sub-intention based on the sub-intentions and the parameter information corresponding to the sub-intentions; the scene graph is a topological graph formed based on network nodes; and executing all the scene graphs to generate a network service to obtain a target network service. According to the method, the scene graph corresponding to each sub-intention is constructed based on the sub-intention and the parameter information corresponding to the sub-intention, so that all the scene graphs are executed to generate the network service, and the target network service is obtained. The target network service can be directly generated based on the scene graph generated by the network service intention, so that the accuracy of generating the target network service can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a network service generation method, device, equipment and readable storage medium. Background Art

[0002] Before the widespread application of artificial intelligence technology, traditional intent recognition and service generation were based on manual methods, that is, users put forward specific network service requirements, network managers understand user intentions, and formulate corresponding network service plans based on user intentions. On this basis, manual methods are used to complete the corresponding network configuration and provide services to users. Manual methods have problems such as low efficiency and high labor costs.

[0003] It can be seen that how to improve the efficiency of network service generation is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a network service generation method, device, equipment and readable storage medium, which solves the technical problem of low efficiency of network service generation in the prior art.

[0005] In order to solve the above technical problems, the present invention provides a network service generation method, comprising:

[0006] Perform multi-granularity intent recognition on the network service generation intent to obtain intent recognition results of different granularities;

[0007] Analyze the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention;

[0008] Based on the sub-intentions and the parameter information corresponding to the sub-intentions, a scene graph corresponding to each sub-intention is constructed; wherein the scene graph is a topological graph composed of network nodes;

[0009] Execute all scene graphs to generate network services and obtain the target network services.

[0010] Optionally, the intent recognition results of different granularities include three granularities: language level, intent level and parameter level. The language level is the intent determined based on the semantic information of the intent generated by the network service, the intent level is the structural information related to the network topology structure, and the parameter level is the parameter information related to the network topology structure.

[0011] Optionally, the intent recognition results of different granularities are analyzed to obtain each sub-intent and parameter information corresponding to the sub-intent, including:

[0012] Determine the intent list corresponding to the intent recognition result of each granularity;

[0013] Intent fusion is performed based on the intent list corresponding to the intent recognition result of each granularity to obtain each sub-intent and parameter information corresponding to the sub-intent.

[0014] Optionally, the intent list corresponding to the intent recognition result based on each granularity is subjected to intent fusion to obtain each sub-intent and parameter information corresponding to the sub-intent, including:

[0015] Determine the number of occurrences of each sub-intent based on the intent list corresponding to the intent recognition result of each granularity;

[0016] The sub-intents are filtered based on the number of occurrences and the set number of occurrences threshold to obtain each sub-intent and parameter information corresponding to the sub-intent.

[0017] Optionally, execute all scene graphs to generate network services and obtain target network services, including:

[0018] The scene graphs are executed in sequence to generate interactive network services to obtain the target network service; wherein the interactive network service generation refers to generating a network service based on information input by a user during the process of generating the network service.

[0019] Optionally, executing the scene graph in sequence to generate an interactive network service to obtain the target network service includes:

[0020] Input the scene graph into a start node to obtain a verified scene graph; wherein the start node is used to determine the integrity and validity of the scene graph;

[0021] updating the verified scene graph based on the supplementary information and / or modified information input by the user to obtain an updated scene graph;

[0022] The updated scene graph is executed to call a network interface for service generation.

[0023] Optionally, updating the verified scene graph based on the supplementary information and / or modified information input by the user to obtain an updated scene graph includes:

[0024] Using the large network intent model, multi-granular intent recognition is performed on the supplementary information and / or modified information input by the user to obtain an intent update recognition result;

[0025] The verified scene graph is updated based on the intention update recognition result to obtain the updated scene graph.

[0026] The present invention also provides a network service generation device, comprising:

[0027] A multi-granularity recognition module is used to perform multi-granularity intent recognition on the network service generation intent and obtain intent recognition results of different granularities;

[0028] An analysis module, used to analyze the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention;

[0029] A scene graph construction module, used to construct a scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention; wherein the scene graph is a topological graph composed of network nodes;

[0030] The network service generation module is used to execute all scene graphs to generate network services and obtain target network services.

[0031] The present invention also provides a network service generation device, comprising:

[0032] Memory for storing computer programs;

[0033] The processor is used to execute the computer program to implement the steps of the network service generation method as described above.

[0034] The present invention also provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned network service generation method are implemented.

[0035] The present invention also provides a computer program product, comprising a computer program / instruction, wherein the computer program / instruction implements the steps of the above-mentioned network service generation method when executed by a processor.

[0036] It can be seen that the present invention obtains intent recognition results of different granularities by performing multi-granularity intent recognition on the network service generation intention; analyzes the intent recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention; constructs a scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention; wherein the scene graph is a topological graph composed of network nodes; executes all scene graphs to generate network services to obtain the target network service.

[0037] The beneficial effect of the present invention is that compared with the current manual network service generation, the present application obtains intent recognition results of different granularities through multi-granularity intent recognition, analyzes the intent recognition results of different granularities, obtains each sub-intent and the parameter information corresponding to the sub-intent, and constructs a scene graph corresponding to each sub-intent based on the sub-intent and the parameter information corresponding to the sub-intent, thereby executing all scene graphs to generate network services and obtain target network services. Since the target network service can be directly generated based on the scene graph generated by the network service intent, the accuracy of the target network service generation is improved.

[0038] In addition, the present invention also provides a network service generating device, equipment and readable storage medium, which also have the above-mentioned beneficial effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0040] Figure 1 A flowchart of a network service generation method provided by an embodiment of the present invention;

[0041] Figure 2 A flowchart of a network service generation method provided by an embodiment of the present invention;

[0042] Figure 3 A structural framework diagram of a network service generation method provided by an embodiment of the present invention;

[0043] Figure 4 A flowchart of a method for generating a network intent training data set provided by an embodiment of the present invention;

[0044] Figure 5 A flowchart of a scene graph generation method provided by an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of the structure of a network service generating device provided by an embodiment of the present invention;

[0046] Figure 7 A schematic diagram of the structure of a network service generating device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] Please refer to Figure 1 , Figure 1 A flowchart of a network service generation method provided by an embodiment of the present invention. The method may include:

[0049] S101, performing multi-granularity intent recognition on the network service generation intent to obtain intent recognition results of different granularities.

[0050] The execution subject of this embodiment is an electronic device. This embodiment does not limit the specific electronic device. For example, the electronic device in this embodiment can be a computer, a mobile phone, etc. This embodiment does not limit the specific network service intention. For example, the network service generation intention in this embodiment can be to generate a VPN (virtual private network) channel between A and B; or the network service generation intention in this embodiment can be to build a set of Internet of Things data transmission and processing architecture for a specific industry; or the network service generation intention in this embodiment can be to build a virtual network environment that supports large-scale online collaborative office. Multi-granularity recognition in this embodiment refers to analyzing and identifying data at different levels or scales to obtain more comprehensive and accurate information. This embodiment does not limit specific multi-granularity recognition. For example, multi-granularity recognition in this embodiment is to perform semantic recognition and parameter recognition. Or multi-granularity recognition in this embodiment is to perform overall semantic recognition, decomposition intention recognition and parameter recognition.

[0051] It should be noted that, based on any of the above embodiments, the above-mentioned intent recognition results of different granularities include three granularities: language level, intent level and parameter level. The language level is the intent determined based on the semantic information of the network service generating intent, the intent level is the structural information related to the network topology, and the parameter level is the parameter information related to the network topology. This embodiment can identify the intent generated by the network service based on a multi-granularity network intent recognition model. The intent input by the user is combined with prompts of different granularities, and then the network intent large model API (interface) is called separately to generate network intent recognition results of different granularities. The different granularities here include three granularities: language level, intent level and parameter level. Among them, 1) language level is to first semantically divide the network intent input by the user, and decompose it into language sentences corresponding to different intents (for example: the user intent "I need a VPN dedicated line from Beijing to Nanjing, which needs to support srv6 and open a tunnel" is decomposed into two sentences: "I need a VPN dedicated line from Beijing to Nanjing, which needs to support srv6," and "To open a tunnel", which correspond to the sub-intents of "VPN dedicated line opening" and "Tunnel opening" respectively). Then, each decomposed language sentence is combined with the single intent recognition prompt, and input into the network intent large model for single intent recognition, to obtain a single sub-intent, and finally combined into a sub-intent list ["VPN dedicated line opening", "Tunnel opening" ]; 2) The intent level is to classify and identify the network intent input by the user and generate an intent list (for example: the user intent "I need a VPN dedicated line from Beijing to Nanjing, which needs to support SRV6 and open a tunnel" is identified as: ["VPN dedicated line opened", "Tunnel opened"]); 3) The parameter level is to directly classify and identify the network intent input by the user, generate an intent list, and extract the existing parameters (for example: the user intent "I need a VPN dedicated line from Beijing to Nanjing, which needs to support SRV6 and open a tunnel" is identified as: [{"VPN dedicated line opened": {"Starting point": "Beijing", "End point": "Nanjing", "Type": "SRV6"}, {"Tunnel opened": {}}]).

[0052] S102, analyzing the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention.

[0053] This embodiment does not limit the specific method of analyzing different intent recognition structures. For example, this embodiment can summarize the intent recognition results of different granularities, and use the intents that exceed the preset number of existences as each sub-intention; or this embodiment can directly use all existing intents as sub-intentions, as well as the parameter information corresponding to the sub-intentions. This embodiment does not limit the specific sub-intentions. For example, in this embodiment, the sub-intention can be a line from Beijing to Nanjing, or a line from Nanjing to Guangzhou; this embodiment does not limit the specific parameter information, for example, the line from Beijing to Nanjing is a VPN (virtual private network) dedicated line, which needs to support srv6 (a technology for carrying VPN services in IPv6 networks through SRv6 tunnels).

[0054] It should be further explained that, in order to improve the accuracy of sub-intention recognition, the above analysis of the intention recognition results of different granularities to obtain the parameter information corresponding to each sub-intention and the sub-intention may include: determining the intention list corresponding to the intention recognition result of each granularity; performing intention fusion based on the intention list corresponding to the intention recognition result of each granularity to obtain the parameter information corresponding to each sub-intention and the sub-intention. This embodiment fuses the intention recognition results of different granularities. Since the intention recognition of different granularities can obtain the intention list, the fusion strategy is to fuse the intention list. This embodiment does not limit the specific fusion method. For example, if the intention recognition results of three granularities appear 3 times, it means that the sub-intention exists. At the same time, the sub-intention that appears 2 times is recorded and marked as a suspicious intention. It can be confirmed to the user in the subsequent conversation, and the sub-intention that appears 1 time is considered not to be the user's true intention. Or this embodiment can directly indicate that the sub-intention exists if the number of appearances is greater than or equal to 2, otherwise the intention does not exist. This embodiment provides a specific method for determining each sub-intention, which improves the accuracy of sub-intention determination.

[0055] It should be further explained that, based on the above-mentioned embodiment, the above-mentioned intent fusion based on the intent list corresponding to the intent recognition result of each granularity to obtain the parameter information corresponding to each sub-intention and the sub-intention may include: determining the number of occurrences corresponding to each sub-intention based on the intent list corresponding to the intent recognition result of each granularity; filtering the sub-intentions based on the number of occurrences and the set number of occurrences threshold to obtain the parameter information corresponding to each sub-intention and the sub-intention. This embodiment can determine the number of occurrences of each sub-intention based on a list. This embodiment does not specify a specific number of occurrences threshold. For example, the number of occurrences threshold in this embodiment may be 2; or the number of occurrences threshold in this embodiment may be 3. This embodiment only regards the intents whose number of occurrences is greater than the set number threshold as true sub-intentions, so the accuracy of sub-intention determination can be improved.

[0056] S103, based on the sub-intentions and parameter information corresponding to the sub-intentions, construct a scene graph corresponding to each sub-intention; wherein the scene graph is a topological graph composed of network nodes.

[0057] In this embodiment, for each sub-intention, a corresponding scene graph is constructed. The scene graph consists of a series of nodes related to the sub-intention and the association relationships between the nodes, which is used to constrain the execution process of the sub-intention.

[0058] S104, executing all scene graphs to generate network services and obtain target network services.

[0059] This embodiment does not limit the specific process of obtaining the target network service. For example, this embodiment can execute each scene graph in sequence, and can interact with the user in natural language through the user interaction interface during the execution process, dynamically perceive the user's latest needs and the real-time status of the network, and can present the relevant results to the user in a timely manner, and finally generate relevant network services for the user according to the intention. Or when there is no latest demand, the network service can be generated directly according to the scene graph. This embodiment can automatically generate and deploy corresponding network services directly based on the scene graph.

[0060] It should be noted that, based on any of the above embodiments, the above execution of all scene graphs to generate a network service to obtain a target network service may include: executing the scene graphs in sequence to generate an interactive network service to obtain a target network service; wherein the interactive network service generation refers to generating a network service based on information input by a user during the process of generating the network service. This embodiment can generate the target network service based on real-time interactive information with the user during the process of generating the target network service, thereby improving the accuracy of generating the target network service and improving the user experience.

[0061] It should be further explained that, based on the above embodiment, the above-mentioned sequentially executing the scene graph to generate an interactive network service to obtain a target network service may include:

[0062] S1041, input the scene graph into the start node to obtain a verified scene graph; wherein the start node is used to determine the integrity and validity of the scene graph.

[0063] The scene graph of this embodiment is a data structure for describing objects and their relationships in a scene. In computer vision and graphics, a scene graph usually contains nodes and edges, where nodes represent objects in a scene and edges represent relationships between objects. In this embodiment, the start node is the entry point of the scene graph processing flow, and is mainly used to determine the integrity and validity of the scene graph. Specifically, the start node performs the following tasks: Integrity check: Ensure that all nodes and edges in the scene graph are intact. Validity verification: Verify whether the nodes and edges in the scene graph comply with predefined rules and constraints. The user or system inputs the scene graph into the start node. The scene graph can be a file (such as JSON, XML, etc.) or a data structure. The start node traverses all nodes and edges in the scene graph to ensure that there are no missing parts. Check whether each node has the necessary attributes and relationships. Check whether each edge is connected to a valid node. The start node verifies each node and edge in the scene graph according to predefined rules and constraints. For example, verify whether the type of the node is correct and whether the relationship between the edges is reasonable. Verify whether the scene graph complies with specific business logic or application scenarios. If the scene graph is complete and valid, the start node outputs the verified scene graph. If the scene graph is incomplete or invalid, the start node returns an error message to prompt the user or system to make corrections.

[0064] S1042: Update the verified scene graph based on the supplementary information and / or modification information input by the user to obtain an updated scene graph.

[0065] The supplementary information in this embodiment refers to adding edges, nodes, parameter information, etc. in the scene graph. The modified information in this embodiment refers to modifying nodes, edges, parameter information, etc. in the scene graph. When determining the supplementary information and / or modified information, this embodiment can identify the user's intention based on the network intention big model to obtain updated information (supplementary information and / or modified information).

[0066] S1043, executing the updated scene graph to call the network interface to generate services.

[0067] In this embodiment, the scene graph is a network topology graph composed of multiple nodes for each sub-intention and the parameter information corresponding to each sub-intention. Each node represents a state of the scene corresponding to the intent, and the nodes are migrated by supplementing information. State nodes may include: start, end, interrupt, supplement, modify, execute and other nodes, which can be expanded according to the scene corresponding to the intent. Each state node represents the stage of the current network service generating the intent execution. At the same time, each state node can design a corresponding prompt template. By combining the current scene graph state and network state information with the corresponding prompt, the current execution state is output through the network intent model and interacts with the user. When the sub-intention starts to execute, 1) first enter the start node, and then call the network intent model to extract the single intent parameters (need to use the single intent parameter extraction prompt); 2) check each extracted parameter to verify the legitimacy of the parameter (for example: whether the provided location is in the network topology node, whether the provided bandwidth exceeds the network's own capacity, etc.); 3) when all parameter information is obtained, enter the modification node and call the network intent model (need to use the ask to modify parameter prompt) to ask the user whether to adjust the parameters; 4) if the user answers to confirm the parameter-related information, the network intent model can sense the user's confirmation status and enter the execution phase; if the user enters 5) If the user proposes to abandon or terminate the intention, the network intention model can also perceive the user's current intention, enter the interrupt state, and save the session state information to facilitate the user's subsequent request to continue to execute the sub-intention; 6) The process of intention execution is a natural interactive process. The network intention model can perceive the network status and, combined with the current scene graph execution status, feedback relevant information to the user in the form of chat, and conduct human-computer interaction. When all sub-intentions are executed, the user is informed of the relevant information of the network service that has been generated for the user. This embodiment gives the specific process of network service generation. The accuracy of network service generation is improved.

[0068] The network service generation method provided by the embodiment of the present invention may include: S101, performing multi-granularity intent recognition on the network service generation intention to obtain intent recognition results of different granularities; S102, analyzing the intent recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention; S103, based on the sub-intention and the parameter information corresponding to the sub-intention, constructing a scene graph corresponding to each sub-intention; wherein the scene graph is a topological graph composed of network nodes; S104, executing all scene graphs to generate network services to obtain the target network service. Compared with the current manual network service generation, the present application obtains intent recognition results of different granularities through multi-granularity intent recognition, analyzes the intent recognition results of different granularities, obtains each sub-intention and the parameter information corresponding to the sub-intention, and constructs the scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention, thereby executing all scene graphs to generate network services to obtain the target network service. Since the target network service generation can be directly performed based on the scene graph generated by the network service intention, the accuracy of the target network service generation is improved.

[0069] In order to make the present invention easier to understand, please refer to Figure 2 , Figure 2 A flowchart of a network service generation method provided by an embodiment of the present invention may specifically include:

[0070] S201. Call the network intent big model to perform multi-granularity intent recognition on the user's network intent to obtain intent recognition results of different granularities.

[0071] For easier understanding, please refer to Figure 3 , Figure 3A structural framework diagram of a network service generation method provided in an embodiment of the present invention, and the framework diagram corresponds to the network service generation method provided in an embodiment of the present invention. In this embodiment, the user can input the user's network intent in natural language through an interactive interface to the network intent big model for multi-granularity intent recognition. In this embodiment, the user's network intent refers to the network configuration, network optimization and other related requirements proposed by the user in natural language, for example: "I need a VPN dedicated line from Beijing to Nanjing, which needs to support srv6 and open a tunnel." In this embodiment, the network intent big model API interface can be called to perform multi-granularity intent recognition on the network intent input by the user to obtain intent recognition results of different granularities. In this embodiment, the different granularities include three granularities: language level, intent level and parameter level. In this embodiment, the network intent big model builds a network intent data set based on an open source big language model (for example: LLaMA series, GLM4, Qwen, etc.), and obtains a special big model for network intent recognition and service generation through instruction fine-tuning. The instruction fine-tuning method is a general large language model fine-tuning method, which can be one of the fine-tuning methods such as full parameter fine-tuning, Lora (a method for efficient parameter fine-tuning of large language models), P-Tuning (a method for efficient parameter fine-tuning), P-Tuningv2, etc. Among them, the network intent dataset is a dataset specially designed and constructed for training large network intent models. The construction method is as follows: 1) define a set of multi-granularity (three granularities: language level, intent level and parameter level) intent recognition prompts, and edit them according to the three granularities to form corresponding single-round dialogue data samples; 2) define a set of intent scene graph execution prompt templates to guide the large model to generate corresponding intent execution status-related information based on node status, network status and user feedback, and construct single-round dialogue data samples according to different nodes, different network status and different user feedback; 3) use ChatGPT4 to expand the data samples, and guide ChatGPT4 to generate a large number of high-quality expanded data sets as required by constructing data expansion prompts; 4) review and screen the generated data, remove invalid and duplicate samples, select high-quality samples as reference samples, and continue to guide ChatGPT4 to generate until the number of samples that meet the training requirements can be obtained; 5) based on the generated samples and the pre-built multi-granularity intent recognition prompts, combine and generate a network intent data set, that is, each data sample must add a specific prompt information as a special instruction to the large model, which can stimulate the potential of the large model during training. By setting appropriate training fine-tuning parameters, a large network intent model is trained based on the generated training dataset.

[0072] For easier understanding, please refer to Figure 4 , Figure 4The following is a flowchart of a method for generating a network intent training dataset provided by an embodiment of the present invention. Figure 4 As shown, the main steps include:

[0073] 1. Design a multi-granularity prompt for network intent recognition, which includes three granularities: language level, intent level, and parameter level. Each granularity has a different depth of natural language understanding. The language level prompt prompts the model to decompose the user intent into clauses containing a single independent sub-intent, and then perform single intent recognition on this basis; the intent level prompt is used to require the large model to directly decompose the user intent and form a corresponding sub-intent list; the parameter level prompt guides the large model to extract parameter information that may exist in the user input information in addition to extracting the sub-intent list.

[0074] 2. Design the intent scenario graph prompt, including a series of prompt templates, involving parameter extraction, intent confirmation, parameter confirmation, information modification, intent termination, etc. Each type of prompt gives the big model a specific role and guides the big model to have the ability to perform specific tasks.

[0075] 3. Design examples for prompts. The examples are in the form of dialogue, i.e., questions + answers. 1) For multi-granular intent recognition, the questions of each granularity are the same, but the corresponding answers are different. For example: the question is "I need a VPN from Beijing to Nanjing, which needs to support srv6 and a tunnel needs to be opened", the language-level answer is: ["I need a VPN from Beijing to Nanjing, which needs to support srv6", "A tunnel needs to be opened"], the intent-level answer is: ["VPN line opened", "Tunnel opened"]), the parameter-level answer is: [{"VPN line opened": {"Starting point": "Beijing", "End point": "Nanjing", "Type": "srv6"}, {"Tunnel opened": {}}]; 2) For intent scenario execution, it is necessary to simulate the state of each node of the scene graph execution and the state of the current network, and ask questions based on the relevant information and set answers.

[0076] 4. The original sample generation prompt design is a prompt template designed for large models (can be any large model, domestic or foreign) to generate a large number of samples. In this prompt, the role of the large model needs to be specified and the requirements for generating samples need to be put forward.

[0077] 5. Combining prompts and examples combines the original sample generation prompt and manually designed examples, which provides reference examples for the large model to generate similar samples.

[0078] 6. Call the big model to generate a large number of samples, then screen the samples, and combine the screened samples into prompt as samples, and call the big model iteratively to generate a sufficient number of samples;

[0079] 7. Concatenate the designed multi-granularity prompt with the corresponding original sample, and generate a network intent training dataset according to the format requirements of instruction fine-tuning.

[0080] S202. Fusing the intent recognition results of different granularities to obtain a fused network intent recognition result; wherein the network intent recognition result includes sub-intents and parameter information associated with each sub-intent.

[0081] In this embodiment, the intention recognition results of different granularities can be fused through the intention fusion module to obtain the fused network intention recognition result, which contains sub-intentions and parameter information that may be associated with each sub-intention. In this embodiment, the intention recognition results of different granularities can be input into the intention fusion module for fusion and confirmation. The specific method is that if the sub-intention appears three times in the intention recognition results of the three granularities, it means that the sub-intention exists. At the same time, the sub-intention that appears twice is recorded and marked as a suspicious intention, which can be confirmed to the user in the subsequent dialogue, and the sub-intention that appears once is considered not to be the user's true intention. When confirming with the user, firstly, the prompt requesting the user to confirm is input into the network intention big model, and the network intention big model generates a user intention confirmation request. Then, after the intention fusion module receives the user's intention confirmation reply, the reply result and the corresponding prompt are input into the network intention big model again, and the confirmation information analysis result is obtained through the network intention big model for the intention fusion module to make a judgment.

[0082] S203. Based on the fused network intent recognition results, construct a scene graph corresponding to each sub-intent; wherein the scene graph is composed of a series of nodes related to the sub-intent and the association relationships between the nodes, which are used to constrain the execution process of the sub-intent.

[0083] This embodiment constructs a corresponding scene graph for each sub-intention, and the scene graph consists of a series of nodes related to the sub-intention and the association relationship between the nodes, which is used to constrain the execution process of the sub-intention. In this embodiment, the sub-intention is input into the scene graph for calculation, including scene graph construction, parameter extraction, modification, verification, confirmation and other processes, and finally the service generation result is obtained after confirmation by the user. In the process, the network intention large model is called cyclically and the user is interacted with to achieve the purpose of parameter supplementation, information confirmation, etc., and the network resources can be accessed in real time during the scene graph calculation process to obtain network status information and autonomously call the network resource interface.

[0084] For easier understanding, please refer to Figure 5, Figure 5 A flow chart of a scene graph generation method provided by an embodiment of the present invention, first, according to a pre-set scene graph template and a user sub-intent, a scene graph for the sub-intent is constructed; secondly, 1) after the construction is completed, the scene graph first enters the start node, which can autonomously call the network intent big model to parse the parameter information provided by the user and judge its integrity and validity, 2) then ask the user to supplement or modify as needed, if there is a parameter missing, enter the supplement node, otherwise enter the modification node, and ask the user whether to supplement or modify the parameter; 3) then decide whether to execute or terminate according to the user's confirmation information and the current network status; 4) if entering the execution node, the scene graph autonomously calls the relevant network interface to generate the service, and feeds back the result to the user, and finally enters the end node; 5) at the termination node or the end node, the scene graph can automatically save all the information of the current session for the user to trace back and continue the operation. In the process of scene graph calculation, the scene graph drives the network intent big model to ask and answer questions, output parameter requests, confirmation requests, modification requests, service generation results and other information, and the user conducts corresponding dialogues according to the information output by the scene graph to realize an interactive network service generation mode.

[0085] S204, executing each scene graph in turn to generate a network service, and obtaining a target generation service.

[0086] This embodiment executes each scene graph in turn. During the execution, the network intent model can interact with the user in natural language through the user interaction interface, dynamically perceive the user's latest needs and the real-time status of the network, and can present the relevant results to the user in a timely manner, and finally generate relevant network services for the user according to the intention. After the scene graph calculation is completed, this embodiment calls the relevant interface to generate the network service, and inputs the generated network service related information and the corresponding prompt into the network intent model, and the network intent model returns the network service generation result to the user.

[0087] This invention combines a large model (large language model) with network intent, and improves the network intent recognition and service generation capabilities by training a dedicated large model for network intent recognition and service generation based on the open source base large model. Specifically, it solves the following technical problems:

[0088] (1) Design a method for generating network intent training data: To address the problems of misidentification and missing key information that may exist in the process of network intent decomposition and classification, a multi-granularity training data generation and instruction fine-tuning method is proposed. For each simulated user intent information, prompts of different granularities are constructed, and the intent recognition result is constructed based on the prompt. This method can be used to train and generate a large network intent model, which can greatly improve the large model's ability to understand semantic information related to network intent, thereby improving the accuracy of network intent recognition.

[0089] (2) A multi-granularity intent understanding and intent fusion method is proposed: In order to reduce the impact of hallucinations that may be caused by large language models, when using a large network intent model for intent recognition, prompts of different granularities are used to constrain and reason about user intent, and intent recognition results of different granularities are obtained. By fusing recognition results of different granularities, the accuracy and stability of the network intent model reasoning can be improved. The multi-granularity network intent training data generation method is used to generate training data and train the network intent model to have a stronger understanding of network intent. The proposed multi-granularity intent understanding and intent fusion method can achieve more accurate intent decomposition and classification capabilities.

[0090] (3) Design of a network service generation method based on scene graph: In order to optimize the user network intention execution process and provide users with more convenient and friendly network services, an interactive and perceptible network service generation method based on scene graph is designed. The user intention is converted into a scene graph and executed step by step according to the process. Each step executed in the scene graph can perceive the network status and feedback relevant information to the user. The user can perceive the service generation process and status, and can intervene (such as parameter modification) through natural language. Improve the convenience and friendliness of network service generation. The designed intention scene graph calculation method can realize interactive intention execution, and combined with the dynamic perception of network status and real-time interactive information of users, it provides more reasonable and reliable network services, greatly improving the convenience and friendliness of network service generation.

[0091] The following is an introduction to a network service generation device provided in an embodiment of the present invention. The network service generation device described below and the network service generation method described above can be referenced to each other.

[0092] Please refer to Figure 6 , Figure 6 A schematic diagram of the structure of a network service generation device provided by an embodiment of the present invention may include:

[0093] The multi-granularity recognition module 100 is used to perform multi-granularity intent recognition on the network service generation intent to obtain intent recognition results of different granularities;

[0094] An analysis module 200 is used to analyze the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention;

[0095] A scene graph construction module 300, for constructing a scene graph corresponding to each sub-intent based on the sub-intent and the parameter information corresponding to the sub-intent; wherein the scene graph is a topological graph composed of network nodes;

[0096] The network service generation module 400 is used to execute all scene graphs to generate network services and obtain target network services.

[0097] Further, based on the above embodiments, the intent recognition results of different granularities may include three granularities: language level, intent level and parameter level. The language level is the intent determined based on the semantic information of the intent generated by the network service, the intent level is the structural information related to the network topology structure, and the parameter level is the parameter information related to the network topology structure.

[0098] Further, based on any of the above embodiments, the above analysis module 200 may include:

[0099] An intent list determination unit, used to determine an intent list corresponding to an intent recognition result of each granularity;

[0100] The sub-intention determination unit is used to perform intent fusion based on the intent list corresponding to the intent recognition result of each granularity, and obtain each sub-intention and parameter information corresponding to the sub-intention.

[0101] Further, based on any of the above embodiments, the sub-intention determination unit may include:

[0102] An occurrence count determination subunit, used to determine the occurrence count corresponding to each sub-intention based on the intent list corresponding to the intent recognition result of each granularity;

[0103] A filtering and fusion unit is used to filter the sub-intentions based on the number of occurrences and the set number of occurrences threshold to obtain each sub-intention and parameter information corresponding to the sub-intention.

[0104] Further, based on any of the above embodiments, the network service generation module 400 may include:

[0105] The target network service generation unit is used to execute the scene graph in sequence to generate an interactive network service to obtain the target network service; wherein the interactive network service generation refers to generating a network service based on information input by a user during the process of generating the network service.

[0106] Further, based on the above embodiment, the above target network service generation unit may include:

[0107] A verification subunit, used to input the scene graph into a start node to obtain a verified scene graph; wherein the start node is used to determine the integrity and validity of the scene graph;

[0108] An updated scene graph determination subunit is used to update the verified scene graph based on the supplementary information and / or modification information input by the user to obtain an updated scene graph;

[0109] The service generation subunit is used to execute the updated scene graph to call the network interface to generate services.

[0110] Further, based on the above embodiment, the updated scene graph determination subunit may include:

[0111] An updating subunit, used to perform multi-granularity intent recognition on the supplementary information and / or modified information input by the user using a large network intent model, and obtain an intent update recognition result;

[0112] A scene graph updating unit is used to update the verified scene graph based on the intention update recognition result to obtain the updated scene graph.

[0113] It should be noted that the order of the modules and units in the above network service generation device can be changed without affecting the logic.

[0114] The network service generation device provided by the embodiment of the present invention may include: a multi-granularity recognition module 100, which is used to perform multi-granularity intent recognition on the network service generation intention to obtain intent recognition results of different granularities; an analysis module 200, which is used to analyze the intent recognition results of different granularities to obtain each sub-intention and the parameter information corresponding to the sub-intention; a scene graph construction module 300, which is used to construct a scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention; wherein the scene graph is a topological graph composed of network nodes; a network service generation module 400, which is used to execute all scene graphs to generate network services to obtain the target network service. Compared with the current manual network service generation, the present application obtains intent recognition results of different granularities through multi-granularity intent recognition, analyzes the intent recognition results of different granularities, obtains each sub-intention and the parameter information corresponding to the sub-intention, and constructs the scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention, thereby executing all scene graphs to generate network services to obtain the target network service. Since the target network service can be generated directly based on the scene graph generated by the network service intention, the accuracy of the target network service generation is improved.

[0115] A network service generation device provided by an embodiment of the present invention is introduced below. The network service generation device described below and the network service generation method described above can refer to each other.

[0116] Please refer to Figure 7 , Figure 7 A schematic diagram of the structure of a network service generating device provided by an embodiment of the present invention may include:

[0117] A memory 10, used for storing computer programs;

[0118] The processor 20 is used to execute a computer program to implement the above-mentioned network service generation method.

[0119] The memory 10 , the processor 20 , and the communication interface 30 all communicate with each other via a communication bus 40 .

[0120] In the embodiment of the present invention, the memory 10 is used to store one or more programs, and the program may include program code, and the program code includes computer operation instructions. In the embodiment of the present invention, the memory 10 may store programs for implementing the following functions:

[0121] Perform multi-granularity intent recognition on the network service generation intent to obtain intent recognition results of different granularities;

[0122] Analyze the intention recognition results of different granularities to obtain each sub-intention and the corresponding parameter information of the sub-intention;

[0123] Based on the sub-intentions and the parameter information corresponding to the sub-intentions, a scene graph corresponding to each sub-intention is constructed; wherein the scene graph is a topological graph composed of network nodes;

[0124] Execute all scene graphs to generate network services and obtain the target network services.

[0125] In a possible implementation, the memory 10 may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required for at least one function, etc.; the data storage area may store data created during use.

[0126] In addition, the memory 10 may include a read-only memory and a random access memory, and provide instructions and data to the processor. A portion of the memory may also include an NVRAM. The memory stores an operating system and operating instructions, executable modules or data structures, or a subset thereof, or an extended set thereof, wherein the operating instructions may include various operating instructions for implementing various operations. The operating system may include various system programs for implementing various basic tasks and processing hardware-based tasks.

[0127] The processor 20 may be a central processing unit (CPU), an application specific integrated circuit, a digital signal processor, a field programmable gate array or other programmable logic device, a microprocessor or any conventional processor, etc. The processor 20 may call a program stored in the memory 10 .

[0128] The communication interface 30 may be an interface of a communication module, and is used to connect to other devices or systems.

[0129] Of course, it should be noted that Figure 7 The structure shown does not constitute a limitation on the network service generation device in the embodiment of the present invention. In actual applications, the network service generation device may include Figure 7 More or fewer components than shown, or combinations of certain components.

[0130] The following is an introduction to a readable storage medium provided in an embodiment of the present invention. The readable storage medium described below and the network service generating method described above can be referenced to each other.

[0131] The present invention also provides a readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned network service generation method are implemented.

[0132] The readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0133] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0134] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0135] Finally, it should be noted that, in this article, relationships such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprises" or any other variations are intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0136] The above is a detailed introduction to a network service generation method, device, equipment and readable storage medium provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for a general technician in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A network service generation method, characterized in that: include: Perform multi-granularity intent recognition on the network service generation intent to obtain intent recognition results of different granularities; Analyze the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention; Based on the sub-intentions and the parameter information corresponding to the sub-intentions, a scene graph corresponding to each sub-intention is constructed; wherein the scene graph is a topological graph composed of network nodes; Execute all scene graphs to generate network services and obtain the target network services.

2. The network service generation method according to claim 1, characterized in that: The intention recognition results of different granularities include three granularities: language level, intention level and parameter level. The language level is the intention determined based on the semantic information of the intent generated by the network service, the intention level is the structural information related to the network topology structure, and the parameter level is the parameter information related to the network topology structure.

3. The network service generation method according to claim 1, characterized in that: The intent recognition results of different granularities are analyzed to obtain each sub-intent and parameter information corresponding to the sub-intent, including: Determine the intent list corresponding to the intent recognition result of each granularity; Intent fusion is performed based on the intent list corresponding to the intent recognition result of each granularity to obtain each sub-intent and parameter information corresponding to the sub-intent.

4. The network service generation method according to claim 3, characterized in that: The intent list corresponding to the intent recognition result based on each granularity is subjected to intent fusion to obtain each sub-intent and parameter information corresponding to the sub-intent, including: Determine the number of occurrences of each sub-intent based on the intent list corresponding to the intent recognition result of each granularity; The sub-intents are filtered based on the number of occurrences and the set number of occurrences threshold to obtain each sub-intent and parameter information corresponding to the sub-intent.

5. The network service generation method according to any one of claims 1 to 4, characterized in that: Execute all scene graphs to generate network services and obtain target network services, including: The scene graphs are executed in sequence to generate interactive network services to obtain the target network service; wherein the interactive network service generation refers to generating a network service based on information input by a user during the process of generating the network service.

6. The network service generation method according to claim 5, characterized in that: Executing the scene graph in sequence to generate an interactive network service to obtain the target network service includes: Input the scene graph into a start node to obtain a verified scene graph; wherein the start node is used to determine the integrity and validity of the scene graph; updating the verified scene graph based on the supplementary information and / or modified information input by the user to obtain an updated scene graph; The updated scene graph is executed to call a network interface for service generation.

7. The network service generation method according to claim 6, characterized in that: The updating of the verified scene graph based on the supplementary information and / or modification information input by the user to obtain an updated scene graph includes: Using the large network intent model, multi-granular intent recognition is performed on the supplementary information and / or modified information input by the user to obtain an intent update recognition result; The verified scene graph is updated based on the intention update recognition result to obtain the updated scene graph.

8. A network service generating device, characterized in that: include: A multi-granularity recognition module is used to perform multi-granularity intent recognition on the network service generation intent and obtain intent recognition results of different granularities; An analysis module, used to analyze the intention recognition results of different granularities to obtain each sub-intention and parameter information corresponding to the sub-intention; A scene graph construction module, used to construct a scene graph corresponding to each sub-intention based on the sub-intention and the parameter information corresponding to the sub-intention; wherein the scene graph is a topological graph composed of network nodes; The network service generation module is used to execute all scene graphs to generate network services and obtain target network services.

9. A network service generating device, characterized in that: include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the steps of the network service generation method according to any one of claims 1 to 7.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the network service generation method according to any one of claims 1 to 7 are implemented.

11. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the network service generation method according to any one of claims 1 to 7.

Images