Method for constructing, processing and verifying service intention data of intention network

Abstract

The invention belongs to the technical field of Internet, and particularly discloses a method for constructing, processing and verifying business intention data of an intention network, which comprises the following steps of: S1, collecting original business intention data from various sources to form an initial data set; s2, performing preprocessing operation on data in the initial data set, and labeling the preprocessed business intention data; and S3, detecting the correlation, repeatability and integrity of the data by using a third-party detection tool, and verifying that the data can be applied in an intention recognition scene through a BERT-based named entity recognition natural language processing method. By adopting the method for constructing, processing and verifying the business intention data for the intention network, valuable data driving support can be provided for development of an intention-driven network management system, so that improvement of the network management efficiency is promoted, and the network management cost is reduced.

Description

Technical Field

[0001] The present invention relates to the field of Internet technology, and in particular to a method for constructing, processing and verifying business intent data for an intent network. Background Art

[0002] With the rapid development of emerging technologies such as the Internet of Things, cloud computing, and big data, future networks will face more complex and diverse demands and challenges. In this context, network management systems urgently need to achieve automation and intelligence, and ultimately achieve closed-loop autonomy in order to have the ability to self-learn and evolve. As an emerging technology in the field of network management, intent-based networking (IBN) realizes automated network management and configuration under the condition of customizing user intent. Accurately and autonomously extracting user intent and converting it into a language that the network management system can understand is the key. However, it is worth noting that without data support, intent-based closed-loop autonomous systems will find it difficult to make effective decisions and adjustments. As the cornerstone of system operation, data provides real-time network status information and historical operation records, and is the key to achieving system self-optimization and self-protection. With the advent of the big data era, how to study future networks in a data-driven way has become a new trend.

[0003] In the process of network management based on business intent, users or administrators usually express their specific requirements for network performance, service quality, and resource configuration through natural language. These requirements are first parsed by the business intent recognition system and converted into machine-understandable business intent data. The network management system then orchestrates resources and generates corresponding network slices to flexibly respond to different business needs. In this process, the network management system relies on business intent data and the associated data between business intent and network slices to achieve full-process management from demand to service.

[0004] Most of the data constructed by existing research only describes certain characteristics and is difficult to adapt to the current rapidly changing network scenarios. Some studies rely on using public network data for model training, but public network data sets are unreliable or even outdated due to the rapid changes in network scenarios and patterns. In addition, the current network slice generation still mainly relies on a standardized service model based on network communication functions, that is, selecting appropriate slices from existing network slices. In order to achieve more flexible and business-driven network slice orchestration, it is urgent to build a method for building intent data that can dynamically generate and optimize slice-related intent data according to business needs. Summary of the invention

[0005] The purpose of the present invention is to provide a method for constructing, processing and verifying business intent data for an intent-based network, which provides important data support for an intent-based network closed-loop autonomous system while ensuring data quality, thereby promoting the automation process of the network management system.

[0006] To achieve the above object, the present invention provides a method for constructing, processing and verifying service intent data for an intent-based network, comprising the following steps: S1. Collect original business intent data from multiple sources to form an initial data set; S2. Preprocess the data in the initial data set and annotate the preprocessed business intent data; S3. Use third-party detection tools to detect the relevance, duplication, and completeness of the data, and verify that the data can be used in intent recognition scenarios through natural language processing methods based on BERT-based named entity recognition; In S3, the original input data is converted into Encoded as The vector form of Represents the input sequence Quantity, vector Represents the hidden state vector after BERT encoding; Then perform sequence modeling to obtain a new sequence form; Given a predefined set of entity categories , where m represents the number of entity categories contained in the set E; The goal is to maximize the conditional probability of the true label sequence among the combined scores of all possible label sequences, that is: (1); in, represents the score of each input sequence corresponding to each entity category, represents the actual annotation sequence, represents all possible annotation sequences; Decode the function S to get the corresponding business intent entity: (2); The harmonic mean F1 value of precision and recall is used as the evaluation indicator of the model.

[0007] Preferably, in S1, the source of the original business intent data includes recording corresponding operation data from intent-based network operations; expressing and converting user business intents according to predefined reference templates and in combination with speech-to-text APIs; crawling relevant academic papers, industry standards, and public network documents from official websites of enterprises, campuses, and operators, and obtaining original business intent data by parsing relevant documents or academic papers.

[0008] Preferably, S1 comprises the following steps: S101. Collecting real data from the intent-based network operation process: extracting relevant data indicators of network operation, including service name, service type, service start and end time, network performance requirements, and network functions, based on the network requirements and performance of the system in different service scenarios, and converting these indicator data into corresponding network intent expression paradigms; S102, interacting and expressing network requirements according to predefined task scenarios to generate user intent data, where the task scenarios include network configuration modification, service request, and performance optimization; S103. Obtain network problems, intent examples and their solutions raised by relevant academic papers, industry standards and websites of operators, educational institutions and enterprises, and obtain the business intent of the corresponding network by parsing the data.

[0009] Preferably, S102 includes the following steps: S1021. Build a user interface using HTML, CSS, and JavaScript, develop a voice assistant system in combination with the voice recognition API, and collect user business intent expression data through the voice assistant system; S1022. Construct a reference model for intent expression to standardize intent expression and reduce the impact of different intent expression paradigms on the accuracy of the intent recognition system and subsequent intent strategy generation.

[0010] Preferably, S103 includes the following steps: S1031. Collect data from relevant documents or academic papers using web crawlers, keyword searches and snowballing methods; S1032. Clean the data, remove noise and irrelevant data, and retain relevant information; S1033. Combine automated tools and manual review to parse the cleaned data, extract information related to business intent and convert it into corresponding network intent.

[0011] Preferably, in S2, the preprocessing operation includes removing semantic repetitions, non-ASCII encoded characters, checking for spelling errors, and removing expressions irrelevant to the intent, and performing entity labeling, relationship labeling, slice type labeling, and BIO labeling on the preprocessed business intent data.

[0012] Preferably, in S2, entity labeling, relationship labeling, slice type labeling and BIO labeling are performed on the preprocessed data, including the following steps: S2021. Label the type of entity involved in each piece of data according to predefined data categories; S2022. Annotate the relationships between entities in each piece of data, where each piece of data contains multiple relationships. S2023. Mark the slice type corresponding to each piece of data. The predefined slice types include enhanced mobile broadband (eMBB), ultra-high reliability and low latency communication (URLLC), and massive machine type communication (mMTC). S2024, by assigning a BIO tag to each word in the data, marking the entity type of each entity in the corresponding text, B represents the beginning of the entity, I represents the middle part of the entity, and O represents a non-entity word; S2025. Store the data, store the collected original business intent data in csv format, and store the labeled structured data in json format.

[0013] Preferably, S3 comprises the following steps: S301. Use DataProfiler, a third-party automated analysis tool based on Python, to analyze the relevance, repeatability, and integrity of data and evaluate the reliability of data in network intent identification. Data relevance, i.e., consistency of input data format; The repeatability of the data is expressed as the ratio of the number of repeated data to the total number of data; The completeness of the data, that is, the proportion of rows consisting entirely of non-null values ​​to the total number of rows; S302. Use a BERT-based named entity recognition method to verify whether the data can be applied in the intent recognition scenario, evaluate the performance of the intent recognition model, and identify the user's network intent by using BERT-based named entity recognition and extracting key entities and related parameters in unstructured text.

[0014] Therefore, the present invention adopts the above-mentioned method for constructing, processing and verifying service intent data for the intent network, and the beneficial effects are as follows: (1) This invention constructs business intent and network slice association data from three different methods, and ensures the reliability and applicability of the data through preprocessing, annotation and data quality verification. This provides important data support for the intent-based network closed-loop autonomous system, thereby promoting the intelligent and automated development of the network.

[0015] (2) The present invention can provide valuable data-driven support for the development of intent-driven network management systems, thereby promoting the improvement of network management efficiency and reducing network management costs.

[0016] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is an overall flow chart of an embodiment of a method for constructing, processing and verifying service intent data for an intent-based network according to the present invention; Figure 2 1. The present invention is a method for constructing, processing and verifying business intent data of an intent network in accordance with an embodiment of the present invention. The method for acquiring data and data annotation are schematic diagrams, wherein (a) is a schematic diagram for converting real data collected from network operation, (b) is a schematic diagram for network intent of zone-related websites and files, (c) is a schematic diagram for expressing and generating data through a voice assistant system, and (d) is a schematic diagram for data annotation and storage. Figure 3 It is a voice assistant system and conversion example of an embodiment of a method for constructing, processing and verifying business intent data for an intent network of the present invention; Figure 4 It is a reference template for intent expression in an embodiment of a method for constructing, processing and verifying service intent data for an intent network of the present invention; Figure 5 It is the convergence of the model during the training process of the embodiment of the method for constructing, processing and verifying business intent data of the intent network of the present invention, wherein (a) is the change of the training loss of the intent recognition model, and (b) is the change of the F1 value of the intent recognition model. DETAILED DESCRIPTION

[0018] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0019] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0020] like Figure 1 As shown, the method for constructing, processing and verifying business intent data for an intent-based network includes the following steps: S1. In view of the general lack of data support for the extraction of network intent, original business intent data from multiple sources are collected to form an initial data set. The sources of original business intent data in this embodiment include three types: first, the corresponding operation data is recorded from the actual intent-based network operation, and then the various network indicators involved are converted into corresponding business intents; according to the predefined reference template and combined with the speech-to-text API, the user's business intent is expressed and converted; relevant academic papers, industry standards, and public network documents on the official websites of enterprises, campuses, and operators are crawled, and the relevant documents are parsed (filtered, cleaned), and then the data is converted into business intent to obtain the original business intent data, such as Figure 2 As shown, the following steps are included: S101. Collect real data from the intent-based network operation process: According to the network requirements and performance of the system in different business scenarios, extract relevant data indicators of network operation, including business name, business type, business start and end time, network performance requirements, and network functions, and convert these indicator data into corresponding network intent expression paradigms.

[0021] S102, interacting and expressing network requirements according to predefined task scenarios to generate user intent data, where the task scenarios include network configuration modification, service requests, performance optimization and other requirements, including the following steps: S1021. Build user interfaces using HTML, CSS, and JavaScript, and develop voice assistant systems using voice recognition APIs, such as Figure 3 As shown, user business intention expression data is collected through the voice assistant system.

[0022] S1022. Construct a reference model for intent expression to standardize intent expression and reduce the impact of different intent expression paradigms on the accuracy of the intent recognition system and subsequent intent strategy generation, such as Figure 4 As shown, the model consists of the following entity label groups: Intent user (customer, network administrator, service provider) entity: is the definer of declarative intent. It can be a technical user, such as a network expert, network administrator, service provider, etc., or a non-technical end user, such as a teacher or student in a school, or an administrative department in an enterprise, etc.

[0023] Network action entity: It is manifested as the service that needs to be presented or the way the network needs to reach a certain state. For example, providing services, limiting bandwidth, self-repair, etc.

[0024] Intent target entity: It is the user's needs, services or goals that need to be met, such as web browsing, online education, video conferencing, etc.

[0025] Target device or functional entity: represents the network function (such as data communication, resource sharing) or physical device (such as router, switch, etc.) used to deploy the intention, which is an optional entity.

[0026] Network performance entity: It is the network performance indicator that the user hopes to achieve. It is an optional entity, such as bandwidth size, delay limit, jitter requirement, etc.

[0027] Lifecycle (short-lived, continuous) entity: This is the duration of the intent. It is an optional entity and can be minutes, hours, or longer.

[0028] S103, obtaining relevant academic papers, industry standards, and websites of operators, educational institutions, enterprises, etc., about network problems, intent examples, and their solutions proposed by researchers, and obtaining the business intent of the corresponding network by parsing the data, including the following steps: S1031. Use web crawlers, keyword searches and snowballing methods to collect data on relevant documents or academic papers.

[0029] S1032. Clean the data, remove noise and irrelevant data, and retain relevant information.

[0030] S1033. Combine automated tools and manual review to parse the cleaned data, extract information related to business intent and convert it into corresponding network intent.

[0031] S2. Perform targeted preprocessing operations on the original business intent data collected in the initial data set. The preprocessing operations include removing semantic duplications, non-ASCII encoded characters, checking spelling errors, and removing expressions irrelevant to the intent. Then, the preprocessed business intent data is annotated with entities, relationships, slice types, and BIO, including the following steps: S201. Take appropriate preprocessing measures for data from different sources to solve various distortion effects on the original data collected from various channels and ensure the accuracy and reliability of the data, including the following steps: S2011. Remove semantically duplicated business intent data, that is, sentences that express the same meaning but in different languages.

[0032] S2012. Remove sentences consisting of non-ASCII codes such as Greek letters and pure mathematical symbols, or continuous punctuation marks.

[0033] S2013. Use a Python-based spell checker to detect and correct spelling errors in data and ensure the linguistic accuracy of the data.

[0034] S2014. Delete the additional information provided by the user when expressing his / her intention and only keep the key intention.

[0035] S202, performing entity labeling, relationship labeling, slice type labeling and BIO labeling on the preprocessed data to make it better for natural language processing tasks including named entity recognition, relationship extraction, etc., including the following steps: S2021. Label the type of entity involved in each piece of data according to predefined data categories.

[0036] S2022. Label the relationships between entities in each piece of data. Each piece of data may contain multiple relationships.

[0037] S2023. Mark the slice type corresponding to each piece of data. The predefined slice types include enhanced mobile broadband eMBB, ultra-high reliability and low latency communication URLLC, and massive machine type communication mMTC.

[0038] S2024. By assigning a BIO tag to each word in the data, the entity type of each entity in the corresponding text is marked, where B represents the beginning of the entity, I represents the middle part of the entity, and O represents a non-entity word.

[0039] S2025. Store the data, store the collected original business intent data in csv format, and store the labeled structured data in json format.

[0040] S3. Use third-party detection tools to detect the relevance, duplication, and completeness of the data, and use the natural language processing method based on BERT named entity recognition to verify that the data can be used in the intent recognition scenario, including the following steps: S301. Analyze the relevance, repeatability and integrity of the data by using a third-party automated analysis tool DataProfiler based on Python to evaluate the reliability of the data in network intent identification, including the following steps: S3011. Check the relevance of the data, that is, the consistency of the input data format.

[0041] S3012. Analyze the repeatability of the data, which is expressed as the ratio of the number of repeated data to the total number of data.

[0042] S3013. Evaluate the integrity of the data, that is, the proportion of rows consisting entirely of non-null values ​​(blank rows) to the total number of rows.

[0043] S302: Using a BERT-based named entity recognition method to verify whether the data can be applied in the intent recognition scenario, so as to evaluate the performance of the intent recognition model. By using BERT-based named entity recognition and extracting key entities and related parameters in unstructured text, the model can effectively identify the user's network intent, including the following steps: Use the BERT model to encode the original input data x into The vector form of n Represents the input sequence token Quantity, vector Represents the hidden state vector after BERT encoding.

[0044] Then perform sequence modeling to obtain a new sequence form.

[0045] Given a predefined set of entity categories , where m represents the number of entity categories contained in the set E.

[0046] The goal is to maximize the conditional probability of the true label sequence among the combined scores of all possible label sequences, that is: (1); in, represents the score of each input sequence corresponding to each entity category, represents the actual annotation sequence, represents all possible label sequences.

[0047] Decode the function S to get the corresponding business intent entity: (2); The harmonic mean F1 value of precision and recall is used as the evaluation indicator of the model.

[0048] In order to observe the convergence of the model during the training process, the performance of the two models is plotted, such as Figure 5 As shown. Figure 5 (a) in the figure shows the convergence of the model during training, where the number of training steps is the ratio of the number of training samples to batch_size multiplied by the number of training cycles. In the experiment, batch_size is set to 8 and epoch is 1. Figure 5 As shown in (b), when the training step number is 1000, the BERT-BiLSTM-CRF model reaches convergence, and the F1 value is about 93%; while the BERT-CRF model reaches convergence when the training step number is 1200, and the F1 value is 90%.

[0049] Therefore, the present invention adopts the above-mentioned method of constructing, processing and verifying business intent data for intent-based networks, constructs business intent and network slice association data from three different methods, and ensures the reliability and applicability of the data through preprocessing, labeling and data quality verification, thereby providing important data support for the intent-based network closed-loop autonomous system, and further promoting the intelligent and automated development of the network.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for constructing, processing and verifying business intent data for an intent-based network, characterized in that: The following steps are involved: S1. Collect original business intent data from multiple sources to form an initial data set; S2. Preprocess the data in the initial data set and annotate the preprocessed business intent data; S3. Use third-party detection tools to detect the relevance, duplication, and completeness of the data, and verify that the data can be used in intent recognition scenarios through natural language processing methods based on BERT-based named entity recognition; In S3, the original input data is converted into Encoded as The vector form of Represents the input sequence Quantity, vector Represents the hidden state vector after BERT encoding; Then perform sequence modeling to obtain a new sequence form; Given a predefined set of entity categories , where m represents the number of entity categories contained in the set E; The goal is to maximize the conditional probability of the true label sequence among the combined scores of all possible label sequences, that is: （1）； in, represents the score of each input sequence corresponding to each entity category, represents the actual annotation sequence, represents all possible annotation sequences; Decode the function S to get the corresponding business intent entity: （2）； The harmonic mean F1 value of precision and recall is used as the evaluation indicator of the model.

2. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 1, characterized in that: In S1, the sources of original business intent data include recording corresponding operation data from intent-based network operations; expressing and converting user business intent according to predefined reference templates and combining speech-to-text APIs; crawling relevant academic papers, industry standards, and public network documents from official websites of enterprises, campuses, and operators, and obtaining original business intent data by parsing relevant documents or academic papers.

3. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 2, characterized in that: S1 includes the following steps: S101. Collecting real data from the intent-based network operation process: extracting relevant data indicators of network operation, including service name, service type, service start and end time, network performance requirements, and network functions, based on the network requirements and performance of the system in different service scenarios, and converting these indicator data into corresponding network intent expression paradigms; S102, interacting and expressing network requirements according to predefined task scenarios to generate user intent data, where the task scenarios include network configuration modification, service request, and performance optimization; S103. Obtain network problems, intent examples and their solutions raised by relevant academic papers, industry standards and websites of operators, educational institutions and enterprises, and obtain the business intent of the corresponding network by parsing the data.

4. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 3, characterized in that: S102 includes the following steps: S1021. Build a user interface using HTML, CSS, and JavaScript, develop a voice assistant system in combination with the voice recognition API, and collect user business intent expression data through the voice assistant system; S1022. Construct a reference model for intent expression to standardize intent expression and reduce the impact of different intent expression paradigms on the accuracy of the intent recognition system and subsequent intent strategy generation.

5. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 4, characterized in that: S103 includes the following steps: S1031. Collect data from relevant documents or academic papers using web crawlers, keyword searches and snowballing methods; S1032. Clean the data, remove noise and irrelevant data, and retain relevant information; S1033. Combine automated tools and manual review to parse the cleaned data, extract information related to business intent and convert it into corresponding network intent.

6. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 5, characterized in that: In S2, the preprocessing operations include removing semantic duplications, non-ASCII encoded characters, checking spelling errors, and removing expressions irrelevant to the intent. The preprocessed business intent data is labeled with entities, relationships, slice types, and BIO.

7. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 6, characterized in that: In S2, the preprocessed data is labeled with entities, relationships, slice types, and BIO, including the following steps: S2021. Label the type of entity involved in each piece of data according to predefined data categories; S2022. Annotate the relationships between entities in each piece of data, where each piece of data contains multiple relationships. S2023. Mark the slice type corresponding to each piece of data. The predefined slice types include enhanced mobile broadband (eMBB), ultra-high reliability and low latency communication (URLLC), and massive machine type communication (mMTC). S2024, by assigning a BIO tag to each word in the data, marking the entity type of each entity in the corresponding text, B represents the beginning of the entity, I represents the middle part of the entity, and O represents a non-entity word; S2025. Store the data, store the collected original business intent data in csv format, and store the labeled structured data in json format.

8. The method for constructing, processing and verifying service intent data for an intent-based network according to claim 7, characterized in that: S3 includes the following steps: S301. Use DataProfiler, a third-party automated analysis tool based on Python, to analyze the relevance, repeatability, and integrity of data and evaluate the reliability of data in network intent identification. Data relevance, i.e., consistency of input data format; The repeatability of the data is expressed as the ratio of the number of repeated data to the total number of data; The completeness of the data, that is, the proportion of rows consisting entirely of non-null values ​​to the total number of rows; S302. Use a BERT-based named entity recognition method to verify whether the data can be applied in the intent recognition scenario, evaluate the performance of the intent recognition model, and identify the user's network intent by using BERT-based named entity recognition and extracting key entities and related parameters in unstructured text.

Images