Structured data corpus generation method and device, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a structured data corpus generation method and device, electronic equipment and a storage medium. The structured data corpus generation method comprises the steps that original structured data is collected; performing corpus field processing on the original structured data to obtain a target domain field corresponding to the original structured data; and determining a target structured data corpus corresponding to the original structured data based on the target domain field. According to the scheme of the embodiment, the target structured data corpus is high in generation and annotation efficiency, detailed and standard in content and higher in quality, a technical basis can be provided for effectively improving the efficiency and quality of large model training and reasoning, it is guaranteed that the large model achieves the same function with the high parameter quantity with the low parameter quantity, and the large model training and reasoning efficiency and quality are improved. And the method has important engineering value for actual deployment and application of a large model.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the fields of artificial intelligence and communication network applications, and in particular, to a method, apparatus, electronic device, and storage medium for generating structured data corpora. Background Art

[0002] In the field of artificial intelligence, the NLP (Natural Language Processing) large model is an autoregressive language model based on Transformer, with powerful language generation capabilities. Currently, its main application scenarios mainly focus on tasks such as dialogue chat, knowledge Q&A, text generation, text classification, sentiment analysis, and machine translation.

[0003] Language corpora are usually in text form, including standard specifications, regulations, basic knowledge materials, and various types of text data such as novels, news, blogs, and emails. They have a high degree of information richness and complexity, containing various semantic and context information in the text, such as the meaning of sentences and the structure of paragraphs. Moreover, the preprocessing and annotation of language corpora are relatively easy, mainly performing tasks such as word segmentation, part-of-speech tagging, and named entity recognition, which are relatively mature.

[0004] However, the corpora in communication networks are mostly structured data, such as performance statistics, alarms, logs, configuration commands, CHR (Call History Record), etc. Summary of the Invention

[0005] Embodiments of the present disclosure provide a method, apparatus, electronic device, and storage medium for generating structured data corpora.

[0006] In a first aspect, embodiments of the present disclosure provide a method for generating structured data corpora, the method may include:

[0007] Collecting original structured data;

[0008] Performing corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data;

[0009] Determining target structured data corpora corresponding to the original structured data based on the target domain fields.

[0010] In a second aspect, embodiments of the present disclosure provide a structured data corpus generation apparatus, which may include:

[0011] A collection module, configured to collect original structured data;

[0012] A processing module, configured to perform corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data;

[0013] A determination module, configured to determine a target structured data corpus corresponding to the original structured data based on the target domain fields.

[0014] In a third aspect, an embodiment of the present disclosure provides an electronic device, which includes:

[0015] One or more processors;

[0016] A memory, storing one or more programs thereon. When the one or more programs are executed by the one or more processors, the one or more processors implement the structured data corpus generation method;

[0017] One or more input / output (I / O) interfaces, connected between the processor and the memory, configured to implement information interaction between the processor and the memory.

[0018] In a fourth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the structured data corpus generation method is implemented.

[0019] In the embodiments of the present disclosure, by performing corpus domain processing on the original structured data to obtain target domain fields, and determining a target structured data corpus based on the target domain fields, the generation and annotation efficiency of the target structured data corpus is high, the content is detailed, standardized, and of higher quality, which can provide a technical basis for effectively improving the efficiency and quality of large model training and inference, ensure that the large model can achieve the same functions with a lower number of parameters as those with a higher number of parameters, and has important engineering value for the actual deployment and application of the large model. Description of the Drawings

[0020] In the drawings of the embodiments of the present disclosure:

[0021] Figure 1 It is a flowchart of the structured data corpus generation method provided by the embodiment of the present disclosure;

[0022] Figure 2 It is a flowchart of the method for performing corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data provided by the embodiment of the present disclosure;

[0023] Figure 3 It is a flowchart of the method for splitting or combining the original structured data based on a preset splitting rule library to obtain structured fields provided by the embodiment of the present disclosure;

[0024] Figure 4Flowchart of a method for selecting one or more domain fields from domain fields as alternative domain fields based on a preset selection rule library provided by an embodiment of the present disclosure;

[0025] Figure 5 Flowchart of a method for performing operations on alternative domain fields according to a preset operation rule library provided by an embodiment of the present disclosure;

[0026] Figure 6 Flowchart of a method for performing operations on alternative domain fields according to the operation content corresponding to the target third domain field name in the case where there are multiple operation contents provided by an embodiment of the present disclosure;

[0027] Figure 7 Block diagram of a structured data corpus generation device provided by an embodiment of the present disclosure;

[0028] Figure 8 Schematic diagram of the structure of a structured data corpus generation device provided by an embodiment of the present disclosure;

[0029] Figure 9 Flowchart of a method for generating a structured data corpus of performance statistical data provided by an embodiment of the present disclosure;

[0030] Figure 10 Block diagram of an electronic device provided by an embodiment of the present disclosure;

[0031] Figure 11 Block diagram of a computer-readable storage medium provided by an embodiment of the present disclosure. Detailed implementation manners

[0032] To enable those skilled in the art to better understand the technical solutions of the present disclosure, the communication perception data processing method and the computer-readable storage medium provided by the embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0033] The present disclosure will be described more fully hereinafter with reference to the accompanying drawings. However, the disclosed embodiments may be embodied in different forms and the present disclosure should not be construed as limited to the embodiments set forth below. On the contrary, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0034] The accompanying drawings of the embodiments of the present disclosure are used to provide a further understanding of the embodiments of the present disclosure, and constitute a part of the specification, and are used to explain the present disclosure together with the detailed embodiments, and do not constitute a limitation to the present disclosure. By describing the detailed embodiments with reference to the accompanying drawings, the above and other features and advantages will become more obvious to those skilled in the art.

[0035] The present disclosure may be described with reference to plan views and / or cross-sectional views by means of ideal schematic diagrams of the present disclosure. Accordingly, the example illustrations may be modified according to manufacturing techniques and / or tolerances.

[0036] In the case of no conflict, the embodiments of the present disclosure and the features in the embodiments may be combined with each other.

[0037] The terms used in the present disclosure are only for describing specific embodiments and are not intended to limit the present disclosure. As used in the present disclosure, the term "and / or" includes any and all combinations of one or more of the related listed items. As used in the present disclosure, the singular forms "a" and "the" are also intended to include the plural forms unless the context clearly indicates otherwise. As used in the present disclosure, the terms "comprising", "made of", specify the presence of the stated features, wholes, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or their groups.

[0038] Unless otherwise defined, all terms (including technical and scientific terms) used in the present disclosure have the same meaning as commonly understood by one of ordinary skill in the art. It will also be understood that terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with their meaning in the context of the relevant art and the present disclosure, and will not be interpreted as having an idealized or overly formal meaning unless the present disclosure clearly so defines.

[0039] The present disclosure is not limited to the embodiments shown in the drawings, but includes modifications to the configurations formed based on the manufacturing process. Accordingly, the regions illustrated in the drawings have schematic properties, and the shapes of the regions shown in the figures illustrate the specific shapes of the regions of the elements, but are not intended to be restrictive.

[0040] In the field of artificial intelligence, the NLP (Natural Language Processing) large model is an autoregressive language model based on Transformer (transformer) and has powerful language generation capabilities. Currently, the main application scenarios are concentrated in tasks such as dialogue chat, knowledge Q&A, text generation, text classification, sentiment analysis, machine translation, etc.

[0041] Language-related corpora are usually in text form, including standard specifications, regulations, basic knowledge materials, and various types of text data such as novels, news, blogs, emails, etc. The information richness and complexity are relatively high, containing various semantic information and context information in the text, such as the meaning of sentences, the structure of paragraphs, etc. Moreover, the preprocessing and annotation of language-related corpora are relatively less difficult, mainly performing tasks such as word segmentation, part-of-speech tagging, and named entity recognition, which are relatively mature.

[0042] Most of the corpora in communication networks are structured data, such as performance statistics, alarms, logs, configuration commands, CHR (Call History Record), etc. Different from the above general corpora, the characteristics of structured data corpora are manifested in the following aspects:

[0043] 1. Structured corpora structure text data into forms such as tables, databases, log files, etc., containing more structured information.

[0044] 2. The information in structured corpora is organized according to certain standard specifications and formats, paying more attention to the standardization and normalization of information, emphasizing the accuracy and consistency of information, such as the description of information such as time, location, action, prompt, etc. with the same structure.

[0045] 3. The preprocessing and annotation of structured corpora are more difficult and require more technical and human input.

[0046] How NLP large models are trained and inferred based on the structured data corpora of communication networks is a current research hotspot and important direction in the industry. Its development and solution are also a systematic project and problem, including the generation of structured data corpora, the selection of training paradigms, model training, model ability evaluation, and deployment applications, etc. At present, there is no effective solution to improve the training and inference effects of NLP large models based on the structured data corpora of communication networks.

[0047] The application of NLP large models in the industry mainly includes processes such as corpus collection, corpus generation, the selection of training and inference paradigms (based on the large model hardware platform, select appropriate training and inference methods, such as data parallelism, tensor parallelism, or pipeline parallelism, etc.), model training / fine-tuning, model ability evaluation, model deployment, and inference applications. Among them, the corpus generation link is of great significance for large models in specific industries. High-quality corpora can enable the training process of NLP large models to converge quickly, and can effectively reduce the scale of NLP large model parameters, improving the accuracy, precision, and recall rate of NLP large models in industry applications.

[0048] The solution of this public implementation example analyzes how to provide high-quality corpora for the application of NLP large models in the industry, and specifically refers to the structured data corpora in communication networks. At present, what needs to be solved urgently is how to provide high-quality structured data corpora required by NLP large models.

[0049] In view of the above problems, in the embodiments of the present disclosure, the target domain fields are obtained by performing corpus domain processing on the original structured data, and the target structured data corpus is determined based on the target domain fields, so that the generation and annotation efficiency of the target structured data corpus is high, the content is detailed, standardized, and of higher quality, which can provide a technical basis for effectively improving the efficiency and quality of large model training and inference, ensure that the large model can achieve the same functions with a lower number of parameters as those with a higher number of parameters, and has important engineering value for the actual deployment and application of the large model.

[0050] The structured data corpus generation method of the embodiments of the present disclosure can be executed by any electronic device that needs to execute structured data corpus generation, such as a terminal device or a server. The terminal device may include, but is not limited to: in-vehicle devices, user equipment (UE), mobile devices, computing devices, wearable devices, etc. For example, it includes, but is not limited to, cellular phones, cordless phones, personal digital assistants (PDAs), portable computers, etc. The structured data corpus generation method can be implemented by a processor calling computer-readable program instructions stored in a memory, or can be implemented by a server.

[0051] The solution of the embodiments of the present disclosure can be applied to any application of a large model in the communication network industry. For example, the solution of the embodiments of the present disclosure can be applied to 5G (the fifth generation of mobile communication technology), 6G (the sixth generation of mobile communication technology), and even application scenarios where the next-generation communication network is combined with a large model. For example, tasks such as using a large model to discriminate and analyze anomalies in a communication network, or using a large model to predict the development trend of communication network traffic, or generating an operation status report for a specified period through a large model.

[0052] The solution of the embodiments of the present disclosure will be introduced in detail below.

[0053] The embodiments of the present disclosure provide a structured data corpus generation method. Figure 1 The following is a flowchart of the structured data corpus generation method provided by the embodiments of the present disclosure. As Figure 1 shown, the method may include steps S101 - S103:

[0054] S101. Collect the original structured data.

[0055] In the embodiments of the present disclosure, raw data from the communication network itself can be collected. The raw data can include, but is not limited to: performance statistics data, alarm data, log data, CHR (Call History Record) data, configuration operations, and other data. And processes such as data cleaning and data formatting of the raw data are completed to obtain raw structured data.

[0056] In the embodiments of the present disclosure, the raw structured data can be further processed. The processing can include, but is not limited to: splitting, selecting, and calculating, etc. After these processes, the raw structured data is transformed into automatically annotated target corpus data (i.e., target domain fields), completing the processing and annotation of the raw structured data of the communication network. Based on the target domain fields, structured data corpus can be combined. The following introduces the detailed processing process of the raw structured data.

[0057] S102. Perform corpus domain processing on the raw structured data to obtain the target domain fields corresponding to the raw structured data.

[0058] In the embodiments of the present disclosure, the corpus domain processing can refer to processing the corpus in one or more fields, one or more ranges, or one or more dimensions. For example, transforming the corpus from one dimension to another dimension (such as, from one format to another format).

[0059] In the embodiments of the present disclosure, the domain fields refer to one or more fields in one or more fields, one or more ranges, or one or more dimensions obtained after performing corpus domain processing on the raw structured data.

[0060] In the embodiments of the present disclosure, Figure 2 is the method flow chart for performing corpus domain processing on the raw structured data provided by the embodiments of the present disclosure to obtain the target domain fields corresponding to the raw structured data. As Figure 2 shown, performing corpus domain processing on the raw structured data to obtain the target domain fields corresponding to the raw structured data can include steps S201 - S203:

[0061] S201. Split or combine the raw structured data to obtain domain fields.

[0062] In the embodiments of the present disclosure, splitting or combining the raw structured data to obtain domain fields can include:

[0063] Splitting or combining the raw structured data based on a preset splitting rule library to obtain structured fields;

[0064] Obtaining domain fields according to the structured fields.

[0065] In an embodiment of the present disclosure, the splitting rule library is a set of splitting rules for multiple telecommunication service scenarios that need to be trained and inferred in advance. Its main function is to define the splitting methods and principles for structured fields in the original structured data input by the domain splitter. The domain splitter is the execution object for splitting or combining the original structured data. Based on the splitting rule library, it can split or combine the data fields in the original structured data input into specific domain field execution entities; among them, the structured fields after splitting or combining are called domain fields.

[0066] In an embodiment of the present disclosure, Figure 3 is a flowchart of a method for splitting or combining original structured data based on a preset splitting rule library to obtain structured fields provided by an embodiment of the present disclosure. As Figure 3 shown, splitting or combining the original structured data based on the preset splitting rule library to obtain structured fields includes steps 301-303:

[0067] S301. Determine the splitting rules in the preset splitting rule library.

[0068] The splitting rules may include, but are not limited to: the first service scenario, the structured data type corresponding to the first service scenario, the structured fields corresponding to the structured data type, and the conversion format between the structured fields and the domain fields;

[0069] S302. Determine the structured data type to which the original structured data belongs according to the specified first service scenario;

[0070] S303. Based on the structured data type, split or combine structured fields from the original structured data so that the structured fields can be converted into domain fields based on the conversion format.

[0071] In an embodiment of the present disclosure, multiple service scenarios (i.e., the first service scenario) for training and inferring telecommunication services are preset in the splitting rule library. For example, it may include, but is not limited to, the computing power evaluation and prediction scenario.

[0072] In an embodiment of the present disclosure, the splitting rules may include multiple first service scenarios, one or more structured data types respectively corresponding to each first service scenario, one or more structured fields respectively corresponding to each structured data type, and the conversion format between each structured field and the domain field. The splitting rules can be preset in the terminal or server system in a way such as JSON (JavaScript Object Notation, a lightweight data interchange format) or XML (Extensible Markup Language).

[0073] In an embodiment of the present disclosure, as shown in Table 1, it is an embodiment of a splitting rule. Among them, the first service scenario includes a computing power evaluation and prediction scenario, the structured data type is performance statistics (5-minute granularity), and the structured fields may include: the number of PFCP (Packet Forwarding Control Protocol) session establishments, the number of PFCP session modifications, the number of PFCP session updates, the average traffic of service components, the maximum traffic of service components, the number of CPU cores of service components, the CPU load of service components, the average CPU frequency of service components, etc. The first domain field names of the converted domain fields may include: the number of session establishments, the number of session modifications, the number of session updates, the average traffic, the maximum traffic, the number of CPU cores, the CPU load, the average frequency, the computing power, the server model, the number of server CPU cores, the server memory size, and the server hard disk size.

[0074] In an embodiment of the present disclosure, for example, a set of original structured data is a performance statistics data file. After the domain splitter splits the input domain of this set of original structured data, the domain splitter, according to the first service scenario specified by the current training and inference tasks (such as the computing power evaluation and prediction scenario mentioned above), based on the contents of the structured fields of the above splitting rule, splits or combines the structured field data in the performance statistics data file to obtain the contents corresponding to the above structured fields, and converts the obtained contents corresponding to the structured fields into the domain fields required by this service scenario according to a preset conversion format. Among them, the conversion format and the contents of the corresponding first domain field names can be modified, deleted, and added according to different first service scenarios.

[0075] Table 1

[0076]

[0077] In an embodiment of the present disclosure, after the original structured data is processed by the above domain splitter, the data is initially split according to the scenario requirements to generate valid domain fields for this scenario.

[0078] S202. Select spare domain fields from the domain fields;

[0079] In an embodiment of the present disclosure, selecting spare domain fields from the domain fields includes:

[0080] Selecting one or more domain fields from the domain fields based on a preset selection rule library as spare domain fields.

[0081] In an embodiment of the present disclosure, selection rules corresponding to different training and inference service scenarios (i.e., the second service scenario) can be preset in the selection rule library. The domain selector can select corresponding domain fields based on the selection rules corresponding to the second service scenario to which it belongs and send them for arithmetic processing. The domain selector is an execution entity for selecting domain fields.

[0082] In an embodiment of the present disclosure, Figure 4 is a flowchart of a method for selecting one or more domain fields from domain fields as spare domain fields based on a preset selection rule library provided by an embodiment of the present disclosure. As Figure 4 shown, selecting one or more domain fields from domain fields as spare domain fields based on a preset selection rule library includes steps S401 - S404:

[0083] S401. Determine the selection rules in the preset selection rule library.

[0084] The selection rules can include but are not limited to: the second service scenario, all the second domain field names corresponding to the second service scenario, and whether the domain fields corresponding to each second domain field name need to be output.

[0085] S402. Determine the second service scenario corresponding to each domain field and all the second domain field names corresponding to the second service scenario.

[0086] S403. Determine the target second domain field name that matches each domain field and the information on whether it needs to be output corresponding to the target second domain field name.

[0087] S404. When the target second domain field name corresponds to the need to be output, select the domain field as a spare domain field.

[0088] In an embodiment of the present disclosure, the selection rules can be preset in the terminal or server system in a way such as JSON or XML.

[0089] In an embodiment of the present disclosure, first, determine the selection rules in the preset selection rule library and the second service scenario corresponding to the domain fields, and then obtain all the second domain field names corresponding to the second service scenario of each domain field; compare each domain field with the corresponding all second domain field names respectively to determine the target second domain field name that matches the domain field; according to the information on whether it needs to be output corresponding to the target second domain field name, it can be determined whether the domain field needs to be output; among them, when the information corresponding to the target second domain field name needs to be output, the domain field needs to be output, and when the information corresponding to the target second domain field name does not need to be output, the domain field does not need to be output; when the domain field needs to be output, select the domain field as a spare domain field.

[0090] In the embodiments of the present disclosure, as shown in Table 2, an embodiment of a selection rule is provided. Among them, the second service scenario includes a computing power evaluation and prediction scenario. The second domain field names may include: the number of session establishments, the number of session modifications, the number of session updates, the average traffic, the maximum traffic, the number of CPU cores, the CPU load, the average frequency, the computing power, the server model, the number of server CPU cores, the server memory size, the server hard disk size, etc. For each item of the second domain field names, whether the domain field corresponding to the second domain field name needs to output this item is respectively: the number of session establishments is needed, the number of session modifications is needed, the number of session updates is needed, the average traffic is needed, the maximum traffic is needed, the number of CPU cores is not needed, the CPU load is not needed, the average frequency is not needed, the computing power is needed, the server model is needed, the number of server CPU cores is needed, the server memory size is needed, the server hard disk size is needed, etc.

[0091] In the embodiments of the present disclosure, for example, after a group of domain field input domain selectors, the domain selector selects the specified domain fields (including: domain name (i.e., the above-mentioned second domain field name), domain content (i.e., the specific content of the domain field), whether it is needed) based on the above selection rule according to the second service scenario (such as the computing power evaluation and prediction scenario mentioned above) specified by the current training and inference tasks, and outputs them as alternative domain fields.

[0092] Table 2

[0093] Second business scenario Second domain field name Whether the domain field needs to be output Computing power evaluation and prediction Number of session establishments Required Number of session modifications Required Number of session updates Required Average traffic Required Maximum traffic Required Number of CPU cores Not required CPU load Not required Average frequency Not required Computing power Required Server model Required Number of CPU cores of the server Required Server memory size Required Server hard disk size Required

[0094] In the embodiments of the present disclosure, the domain selector sends the selected domain fields to the subsequent processing flow for processing (such as arithmetic processing).

[0095] S203. Perform arithmetic operations on the alternative domain fields to obtain the target domain fields.

[0096] In the embodiments of the present disclosure, performing arithmetic operations on the alternative domain fields to obtain the target domain fields includes:

[0097] Performing arithmetic operations on the alternative domain fields according to the arithmetic rules in the preset arithmetic rule library;

[0098] Taking the operation result as the target domain field.

[0099] In the embodiments of the present disclosure, different arithmetic rules corresponding to different training and inference service scenarios (i.e., the third service scenario) can be preset in the arithmetic rule library. The domain arithmetic operator can perform arithmetic operations on the input alternative domain fields based on the arithmetic rules corresponding to the third service scenario to which it belongs. The domain arithmetic operator is the execution entity for performing arithmetic processing on the alternative domain fields.

[0100] In the embodiments of the present disclosure, the operation rules may include, but are not limited to: the third service scenario, all the third domain field names corresponding to the third service scenario, and the operation content corresponding to each third domain field name.

[0101] In the embodiments of the present disclosure, the operation content may include, but is not limited to, any one or more of the following: operation functions, threshold gates, and operation actions.

[0102] In the embodiments of the present disclosure, the operation functions may be provided by a preset function library, the threshold gates may be provided by a preset threshold gate library, and the operation actions may be provided by a preset operation action library.

[0103] In the embodiments of the present disclosure, the function library may provide all the operation functions required for the operation. For example, it may include, but is not limited to: sum function, mean function, maximum value, minimum value, mean square error function, activation function, standard deviation normalization, etc., to provide algorithm support for the operation.

[0104] In the embodiments of the present disclosure, the threshold gate library may preset the upper threshold or lower threshold of the key indicators and / or parameters in various training and inference service scenarios. For example, the system maximum value of the device throughput parameter, the minimum and maximum values of the single-user bandwidth, the maximum threshold of the resource occupancy rate, etc.

[0105] In the embodiments of the present disclosure, the operation action library may preset a set of operation action templates (i.e., templates of operations and / or actions) corresponding to the alternative domain fields in various training and inference service scenarios. The operation action template is a set of operation action command specification description files corresponding to the operations and / or actions, presenting and constraining the use of the operation action commands. The operation action commands may include, but are not limited to, configuration commands, database operation instructions, underlying system resource operation and maintenance instructions, etc., and can generate detailed operation instructions for specific domain fields.

[0106] In the embodiments of the present disclosure, the selection rules may be preset in the terminal or server system in a manner such as JSON or XML.

[0107] In an embodiment of the present disclosure, as shown in Table 3, it is an embodiment of an operation rule. Among them, the third service scenario includes a computing power evaluation and prediction scenario. The third domain field names can correspond to the following data items: the number of session establishments, the number of session modifications, the number of session updates, the average traffic, the maximum traffic, the number of CPU cores, the CPU load, the average frequency, the computing power, the server model, the number of server CPU cores, the server memory size, the server hard disk size, etc.; the operation content includes: threshold limits, operation functions, and operation actions. Each of the third domain field names corresponds to specific operation content. Since multiple data items corresponding to the third domain field names in the computing power evaluation and prediction scenario do not require numerical limitations, that is, the maximum and minimum values do not need to be limited, therefore, these multiple data items do not need to be compared with the threshold limits. In other scenarios, if one or more data items corresponding to the third domain field names require limiting the maximum and / or minimum values, then the corresponding threshold limits can be set, and the corresponding data items can be compared with the threshold limits.

[0108] Table 3

[0109]

[0110]

[0111] In an embodiment of the present disclosure, according to different service scenarios, the data items corresponding to the third and field names can be different, and the corresponding operation content can also be different. Moreover, the detailed threshold limits, operation functions, operation actions, etc. can all be different.

[0112] In an embodiment of the present disclosure, Figure 5 is a flowchart of a method for operating on a spare domain field according to a preset operation rule library provided by an embodiment of the present disclosure. As Figure 5 shown, operating on the spare domain field according to the preset operation rule library includes steps S501 - S504:

[0113] S501. Determine the operation rules in the preset operation rule library.

[0114] The operation rules include but are not limited to: the third service scenario, all the third domain field names corresponding to the third service scenario, and the operation content corresponding to each of the third domain field names.

[0115] S502. Determine the third service scenario corresponding to each spare domain field and all the third domain field names corresponding to the third service scenario.

[0116] S503. Determine the target third domain field name that matches the spare domain field.

[0117] S504. Perform an operation on the spare domain field according to the operation content corresponding to the target third domain field name, and use the operation result as the target domain field.

[0118] In the embodiments of the present disclosure, the operation rules in the preset operation rule library can be first determined, and then the following operations are performed for each spare domain field: determine the third service scenario corresponding to the spare domain field, and obtain all the third domain field names corresponding to the third service scenario of the spare domain field; compare the spare domain field with the third domain field names to determine the target third domain field name that matches the spare domain field; perform an operation on the spare domain field according to the operation content corresponding to the target third domain field name, and use the operation result as the target domain field.

[0119] In the embodiments of the present disclosure, after the alternative domain fields (which can be one or more and can be determined according to the corresponding third service scenario) that need to be operated are accessed, the domain operator can first confirm the third service scenario corresponding to the alternative domain fields. For example, in the computing power evaluation and prediction scenario, and select the names of the third domain fields corresponding to each alternative domain field from all the third domain field names corresponding to the third service scenario, and perform an operation on the corresponding alternative domain fields based on the operation content corresponding to the corresponding third domain field names.

[0120] In the embodiments of the present disclosure, performing an operation on the corresponding alternative domain fields can include the selection and determination of the operation function corresponding to the specific domain field, whether to compare with the threshold, the comparison rule and result generation logic, whether the alternative domain field needs to correspond to an operation description, and the selection of the operation action template, etc., which are the main logic control functions of the domain operator. The embodiments of the operation solution will be introduced in detail below.

[0121] In the embodiments of the present disclosure, when the operation content includes an operation function, performing an operation on the spare domain field according to the operation content corresponding to the target third domain field name includes:

[0122] Perform an operation on the spare domain field based on the constraint information included in the operation function to generate new content of the spare domain field, and use the new content of the spare domain field as the target domain field.

[0123] In the embodiments of the present disclosure, when the operation content includes a threshold, performing an operation on the spare domain field according to the operation content corresponding to the target third domain field name includes:

[0124] Perform a logical judgment on the spare domain field based on the threshold to generate a logical judgment result, and use the logical judgment result as the target domain field.

[0125] In an embodiment of the present disclosure, when the operation content includes an operation action, the spare domain field is operated according to the operation content corresponding to the target third domain field name, including:

[0126] Select an operation action template corresponding to the spare domain field;

[0127] Generate an operation action instruction corresponding to the spare domain field according to the operation action template;

[0128] Execute the operation action instruction and output an execution result, and use the execution result as the target domain field.

[0129] In an embodiment of the present disclosure, when there are multiple pieces of operation content, the spare domain field can be operated according to multiple pieces of operation content based on a preset operation logic.

[0130] In an embodiment of the present disclosure, Figure 6 is a flowchart of a method for operating a spare domain field according to the operation content corresponding to the target third domain field name when there are multiple pieces of operation content provided in an embodiment of the present disclosure, where the operation content includes: an operation function, a threshold, and an operation action; as Figure 6 shown, when there are multiple pieces of operation content, operating the spare domain field according to the operation content corresponding to the target third domain field name may include steps S601 - S612:

[0131] S601. Determine whether the spare domain field needs to select a corresponding operation function for operation based on the operation content.

[0132] S602. When it is determined that the spare domain field needs to select a corresponding operation function for operation, operate the spare domain field based on the constraint information included in the operation function, generate new content of the spare domain field as the target domain field, and update the spare domain field with the new content, then enter step S608; when it is determined that the spare domain field does not need to select a corresponding operation function for operation, enter step S603.

[0133] S603. Determine whether the spare domain field needs to perform a threshold judgment based on the operation content.

[0134] S604. When it is determined that the spare domain field needs to perform a threshold judgment, perform a logical judgment on the spare domain field based on the threshold, generate a logical judgment result as the target domain field, and enter step S605; when it is determined that the spare domain field does not need to perform a threshold judgment, enter step S605.

[0135] S605. Determine whether the spare domain field needs a corresponding operation action description based on the operation content.

[0136] S606. When it is determined that the spare domain field requires a corresponding operation action description, select the operation action template corresponding to the spare domain field, generate an operation action instruction corresponding to the spare domain field according to the operation action template, execute the operation action instruction, and output the execution result as the target domain field.

[0137] S607. When it is determined that the spare domain field does not require a corresponding operation or action description, output the spare domain field as the target domain field.

[0138] S608. Determine whether the updated spare domain field needs to be judged by a threshold based on the operation content.

[0139] S609. When it is determined that the updated spare domain field needs to be judged by a threshold, perform a logical judgment on the updated spare domain field based on the threshold, generate a logical judgment result as the target domain field, and enter step S610; when it is determined that the updated spare domain field does not need to be judged by a threshold, enter step S610.

[0140] S610. Determine whether the updated spare domain field requires a corresponding operation action description based on the operation content.

[0141] S611. When it is determined that the updated spare domain field requires a corresponding operation action description, select the operation action template corresponding to the updated spare domain field, generate an operation action instruction corresponding to the updated spare domain field according to the operation action template, execute the operation action instruction, and output the execution result as the target domain field.

[0142] S612. When it is determined that the updated spare domain field does not require a corresponding operation or action description, output the updated spare domain field as the target domain field.

[0143] In the embodiments of the present disclosure, one or more target domain fields can be obtained based on the above operations.

[0144] In the embodiments of the present disclosure, Figure 6 The judgment order in [the above] is only one embodiment. In different application scenarios, the judgment order of operation contents such as operation functions, threshold gates, and operation actions can be adjusted according to requirements.

[0145] S103. Determine the target structured data corpus corresponding to the original structured data based on the target domain field.

[0146] In the embodiments of the present disclosure, all the obtained target domain fields can be combined together to form a target structured data corpus.

[0147] The solutions of the embodiments of the present disclosure at least include the following advantages:

[0148] 1. A method for generating structured data corpus of communication network applied to large models is proposed for the first time, which is an exploration and innovation of the generation method of structured corpus data of communication network.

[0149] 2. A framework and process for generating structured data corpus of communication network are proposed. Through this framework, existing structured data such as performance statistics, alarms, logs, configuration commands, CHR, etc. in the communication network can be automatically converted into structured data corpus that can be recognized and used by large models, greatly reducing the labor cost of corpus conversion and annotation.

[0150] 3. The target structured data corpus generated by the method for generating structured data corpus of communication network proposed in the embodiment of the present disclosure is rich in content, standardized, and of high quality, which can effectively improve the efficiency and quality of large model training, ensure that the large model can achieve the same functions with a lower number of parameters as those with a higher number of parameters, and has important engineering value for the actual deployment and application of large models.

[0151] The embodiment of the present disclosure also provides a structured data corpus generation device 100. Figure 7 It is a block diagram of the structured data corpus generation device provided by the embodiment of the present disclosure. Figure 8 It is a schematic structural diagram of the structured data corpus generation device provided by the embodiment of the present disclosure; as Figure 7 、 Figure 8 shown, the structured data corpus generation device 100 may include:

[0152] A collection module 101, configured to collect original structured data.

[0153] A processing module 102, configured to perform corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data.

[0154] A determination module 103, configured to determine the target structured data corpus corresponding to the original structured data based on the target domain fields.

[0155] In the embodiment of the present disclosure, the processing module 102 may include a domain splitter, a domain selector, and a domain operator.

[0156] In the embodiment of the present disclosure, the domain splitter (Field Splitter) is a logical function module or component for splitting data fields in structured data into specific domain fields. The domain splitter may include a splitting rule library and a field-splitted actuator. The field-splitted actuator is an execution entity for splitting or combining the original structured data according to the splitting rule library.

[0157] In the embodiments of the present disclosure, a Field Selector is a logical functional module or component for selecting field fields. The Field Selector may include a selection rule library and a Field-Selected Actuator, and the Field-Selected Actuator is an execution entity that selects field fields according to the selection rule library.

[0158] In the embodiments of the present disclosure, a Field Operator is a logical functional module or component for performing logical or mathematical operations on field fields and executing operation commands. The Field Operator includes an operation rule library and a Field Operation Actuator, and the Field Operation Actuator is an execution entity that performs field operations according to the operation rule library.

[0159] In the embodiments of the present disclosure, a detailed embodiment of the present disclosure is given below.

[0160] Training and inference scenario: Based on the performance statistical data file of a certain service collected at a 5-minute granularity, the corpus is processed and generated, and converted into a standard structured data corpus required by the large model. The data in this standard structured data corpus has been normalized and annotated.

[0161] Preconditions: The method flow for generating the structured data corpus proposed in the embodiments of the present disclosure has been running normally, and the splitting rules, selection rules, and operation rules in JSON format have been preset in advance, as shown in Tables 1, 2, and 3. Figure 9 The flowchart of the method for generating the structured data corpus of the performance statistical data provided for the embodiments of the present disclosure is as Figure 9 shown. The method flow for generating the structured data corpus in this embodiment includes steps S701 - S705:

[0162] S701, Collecting raw structured data, here it is the performance statistical data of a certain service collected at a 5-minute granularity.

[0163] S702, Based on the selected training and inference scenarios, the Field Splitter splits or combines the performance statistical data into structured fields according to the preset splitting rules, and converts the structured fields into field fields.

[0164] The converted field fields can be shown in the fourth column of Table 1.

[0165] S703, Based on the selected training and inference scenarios, the Field Selector selects different field fields according to the preset selection rules, and determines whether the field field needs to be finally output as a spare field field.

[0166] Whether the field field needs to be finally output is shown in the third column of Table 2.

[0167] S704. Based on the selected training and inference scenarios, the domain arithmetic unit performs arithmetic processing on the selected spare domain fields according to the preset arithmetic rules.

[0168] The arithmetic processing includes one or more of the following three processing actions:

[0169] The spare domain field is compared with the preset threshold and the result is output;

[0170] Function selection and arithmetic operations are performed on the spare domain field; and

[0171] The spare domain field executes the operation command and obtains the execution result.

[0172] Based on the arithmetic rules of each spare domain field, the domain arithmetic unit processes and labels the domain fields, and traverses all the spare domain fields.

[0173] S705. A file of a new target structured data corpus is formed according to the target domain fields after arithmetic processing for subsequent large model training and inference.

[0174] The embodiments of the present disclosure also provide an electronic device 200, Figure 10 which is a block diagram of the electronic device provided by the embodiments of the present disclosure, as Figure 10 shown. The electronic device 200 includes:

[0175] One or more processors 201;

[0176] A memory 202, on which one or more programs are stored. When the one or more programs are executed by the one or more processors 201, the one or more processors 201 implement the structured data corpus generation method;

[0177] One or more input / output I / O interfaces 203, connected between the processor 201 and the memory 202, configured to implement information interaction between the processor 201 and the memory 202.

[0178] Among them, the processor 201 is a device with data processing capabilities, which includes but is not limited to a central processing unit (CPU), etc.; the memory 202 is a device with data storage capabilities, which includes but is not limited to a random access memory (RAM, more specifically such as SDRAM, DDR, etc.), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a flash memory (FLASH); the I / O interface (read / write interface) 203 is connected between the processor 201 and the memory 202 and can implement information interaction between the processor 201 and the memory 202, which includes but is not limited to a data bus (Bus), etc.

[0179] In some embodiments, the processor 201, the memory 202, and the I / O interface 203 are interconnected via a bus 204 and further connected to other components of the computing device.

[0180] Embodiments of the present disclosure also provide a computer-readable storage medium 300. Figure 11 It is a block diagram of the computer-readable storage medium provided by the embodiments of the present disclosure. As Figure 11 shown, a computer program is stored on the computer-readable storage medium 300, and when the computer program is executed by a processor, the structured data corpus generation method described above is implemented.

[0181] Those of ordinary skill in the art can understand that all or some of the functional modules / units disclosed above can be implemented as software, firmware, hardware, and their appropriate combinations.

[0182] In a hardware implementation, the division between the functional modules / units mentioned above does not necessarily correspond to the division of physical components; for example, one physical component can have multiple functions, or one function or step can be executed by several physical components in cooperation.

[0183] Some or all physical components can be implemented as software executed by a processor, such as a central processing unit (CPU), a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media include, but are not limited to, random access memory (RAM, more specifically SDRAM, DDR, etc.), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory (FLASH), or other disk memories; compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical disc memories; magnetic cassettes, tapes, magnetic disk storage, or other magnetic memories; any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0184] The present disclosure has disclosed exemplary embodiments, and although specific terms are employed, they are used only and should be interpreted only as general illustrative meanings and not for the purpose of limitation. In some instances, it will be apparent to those skilled in the art that, unless otherwise expressly specified, features, characteristics, and / or elements described in connection with a particular embodiment may be used singly or in combination with features, characteristics, and / or elements described in connection with other embodiments. Accordingly, those skilled in the art will understand that various forms and details of changes may be made without departing from the scope of the present disclosure as set forth by the appended claims.

Claims

1. A method for generating structured data corpus, characterized in that, The method includes: Collecting original structured data; Performing corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data; Determining a target structured data corpus corresponding to the original structured data based on the target domain fields.

2. The structured data corpus generation method according to claim 1, characterized in that The performing corpus domain processing on the original structured data to obtain target domain fields corresponding to the original structured data includes: Splitting or combining the original structured data to obtain domain fields; Selecting standby domain fields from the domain fields; Performing operations on the standby domain fields to obtain the target domain fields.

3. The structured data corpus generation method according to claim 2, wherein The splitting or combining the original structured data to obtain domain fields includes: Splitting or combining the original structured data based on a preset splitting rule library to obtain structured fields; Obtaining the domain fields according to the structured fields.

4. The structured data corpus generation method according to claim 3, wherein The splitting or combining the original structured data based on a preset splitting rule library to obtain structured fields includes: Determining splitting rules in the preset splitting rule library; the splitting rules include: a first business scenario, a structured data type corresponding to the first business scenario, structured fields corresponding to the structured data type, and a conversion format between the structured fields and the domain fields; Determining the structured data type to which the original structured data belongs according to a specified first business scenario; Based on the structured data type, splitting or combining the structured fields from the original structured data so that the structured fields are converted into the domain fields based on the conversion format.

5. The structured data corpus generation method according to claim 2, wherein The selecting standby domain fields from the domain fields includes: Selecting one or more domain fields from the domain fields based on a preset selection rule library as the standby domain fields.

6. The structured data corpus generation method according to claim 5, wherein The selecting one or more domain fields from the domain fields based on a preset selection rule library includes: Determining selection rules in the preset selection rule library; the selection rules include: a second business scenario, all second domain field names corresponding to the second business scenario, and whether each domain field name corresponding domain field needs to be output; Determining the second business scenario corresponding to each domain field and all the second domain field names corresponding to the second business scenario; Determining a target second domain field name matching each domain field and information on whether the target second domain field name needs to be output; In the case where the target second domain field name needs to be output, selecting the domain field as the standby domain field.

7. The structured data corpus generation method according to claim 2, characterized in that The performing operations on the standby domain fields to obtain target domain fields includes: Performing operations on the standby domain fields according to a preset operation rule library; Taking the operation result as the target domain field.

8. The structured data corpus generation method according to claim 7, wherein, The performing operations on the standby domain fields according to a preset operation rule library includes: Determining operation rules in the preset operation rule library; the operation rules include: a third business scenario, all third domain field names corresponding to the third business scenario, and operation content corresponding to each third domain field name; Determine the third service scenario corresponding to each of the said spare domain fields and all the names of the third domain fields corresponding to the third service scenario; Determine the target third domain field name that matches the spare domain field; Perform an operation on the spare domain field according to the operation content corresponding to the target third domain field name, and use the operation result as the target domain field.

9. The structured data corpus generation method according to claim 8, wherein The operation content includes any one or more of the following: operation functions, threshold limits, and operation actions.

10. The structured data corpus generation method according to claim 9, characterized in that, When the operation content includes the operation function, the performing an operation on the spare domain field according to the operation content corresponding to the target third domain field name includes: Performing an operation on the spare domain field based on the constraint information included in the operation function to generate new content of the spare domain field, and using the new content of the spare domain field as the target domain field.

11. The structured data corpus generation method according to claim 9, wherein, When the operation content includes the threshold limit, the performing an operation on the spare domain field according to the operation content corresponding to the target third domain field name includes: Performing a logical judgment on the spare domain field based on the threshold limit to generate a logical judgment result, and using the logical judgment result as the target domain field.

12. The structured data corpus generation method according to claim 9, wherein, When the operation content includes the operation action, the performing an operation on the spare domain field according to the operation content corresponding to the target third domain field name includes: Select an operation action template corresponding to the spare domain field; Generate an operation action instruction corresponding to the spare domain field according to the operation action template; Execute the operation action instruction and output an execution result, and use the execution result as the target domain field.

13. A structured data corpus generation device, characterized in that, Including: An acquisition module for acquiring original structured data; A processing module for performing corpus domain processing on the original structured data to obtain the target domain field corresponding to the original structured data; A determination module for determining the target structured data corpus corresponding to the original structured data based on the target domain field.

14. An electronic device, characterized in that, The electronic device includes: One or more processors; A memory on which one or more programs are stored. When the one or more programs are executed by the one or more processors, the one or more processors implement the structured data corpus generation method according to any one of claims 1-12; One or more input / output I / O interfaces connected between the processor and the memory and configured to implement information interaction between the processor and the memory.

15. A computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the structured data corpus generation method according to any one of claims 1-12 is implemented.

Images