Label generation method and device and medium

Abstract

The invention provides a label generation method and device and a medium, and relates to the field of data processing. The method comprises the following steps: acquiring unstructured text data in an enterprise information system, wherein the unstructured text data is a natural language text which is not formatted; performing context analysis on the unstructured text data by utilizing a large language model to generate an initial label set; and processing the initial label set based on a preset rule to form a cross-hierarchy label structure, the cross-hierarchy label structure comprising at least two label layers with different semantic dimensions, and establishing an association relationship network between the label layers. According to the invention, the hysteresis of label updating is reduced.

Description

Technical Field

[0001] This application relates to the field of data processing, and in particular to a label generation method, apparatus and medium. Background Technology

[0002] In today's digital age, enterprises face profound changes in areas such as enterprise knowledge management, intelligent office systems, and organizational effectiveness analysis. A core challenge is how to effectively process and mine massive amounts of unstructured data, which exists in various forms such as text, images, audio, and video, and constitutes the vast majority of an enterprise's total data. Especially in the construction of internal knowledge bases, the fragmented, unstructured, and lacking of unified classification logic in daily reports, weekly reports, and project summaries generated by employees present significant difficulties for information retrieval, utilization, and value extraction.

[0003] Currently, processing unstructured data often relies on a manually pre-defined fixed labeling system. Specifically, this typically involves classifying and labeling the unstructured data using predefined rules or manually, which leads to delays in label updates. Summary of the Invention

[0004] This application provides a label generation method, device, and medium to reduce the lag in label updates.

[0005] Firstly, this application provides a label generation method, including:

[0006] Acquire unstructured text data from enterprise information systems, where unstructured text data is natural language text that has not undergone formatting.

[0007] Using a large language model, contextual analysis is performed on unstructured text data to generate an initial tag set;

[0008] The initial tag set is processed based on preset rules to form a cross-level tag structure. The cross-level tag structure contains at least two business tag layers with different semantic dimensions, and an association is established between the business tag layers.

[0009] In one possible implementation, a large language model is used to perform contextual analysis on unstructured text data to generate an initial tag set, including:

[0010] Unstructured text data is segmented to extract key event fragments;

[0011] Generate semantic vectors based on key event fragments;

[0012] The semantic vectors are matched with a preset tag library to generate an initial tag set.

[0013] In one possible implementation, a semantic vector is generated based on key event fragments, including:

[0014] Key event segments are vectorized to obtain event vectors;

[0015] Semantic vectors are obtained by semantically encoding the event vectors.

[0016] In one possible implementation, the initial tag set is processed based on preset rules to form a cross-level tag structure, including:

[0017] Perform semantic similarity calculations on the tags in the initial tag set to identify conflicting and normal tags;

[0018] Replace conflicting tags with semantically generalized unified tags;

[0019] Grouping uniform labels and normal labels yields grouped labels;

[0020] Based on business hierarchy rules, establish hierarchical relationships between group tags to obtain a cross-level tag structure.

[0021] In one possible implementation, after forming the cross-level tag structure, the method further includes:

[0022] Knowledge distillation is performed on the cross-level tag structure according to a preset cycle to generate hierarchical knowledge summaries;

[0023] Store the hierarchical knowledge summary in the database.

[0024] In one possible implementation, storing the hierarchical knowledge summary in a database includes:

[0025] Determine the access frequency of the hierarchical knowledge summary;

[0026] The storage compression rate of the hierarchical knowledge summary is dynamically adjusted based on the access frequency.

[0027] Based on storage compression ratio, storage tier knowledge summaries are written to the database.

[0028] In one possible implementation, storing the hierarchical knowledge summary in a database includes:

[0029] Determine the storage life of the hierarchical knowledge summary;

[0030] Based on storage cycles, storage-tiered knowledge summaries are written to the database.

[0031] In one possible implementation, the large language model is determined based on parameter pruning and quantization compression operations.

[0032] Secondly, this application provides a label generation apparatus, comprising:

[0033] The acquisition module is used to acquire unstructured text data from the enterprise information system, where the unstructured text data is natural language text that has not undergone formatting.

[0034] The generation module is used to perform contextual analysis on unstructured text data using a large language model to generate an initial set of tags;

[0035] The processing module is used to process the initial tag set based on preset rules to form a cross-level tag structure. The cross-level tag structure contains at least two business tag layers with different semantic dimensions, and establishes a relationship between the business tag layers.

[0036] In one possible implementation, the generation module is specifically used for:

[0037] Unstructured text data is segmented to extract key event fragments;

[0038] Generate semantic vectors based on key event fragments;

[0039] The semantic vectors are matched with a preset tag library to generate an initial tag set.

[0040] In one possible implementation, the generation module is specifically used for:

[0041] Key event segments are vectorized to obtain event vectors;

[0042] Semantic vectors are obtained by semantically encoding the event vectors.

[0043] In one possible implementation, the processing module is specifically used for:

[0044] Perform semantic similarity calculations on the tags in the initial tag set to identify conflicting and normal tags;

[0045] Replace conflicting tags with semantically generalized unified tags;

[0046] Grouping uniform labels and normal labels yields grouped labels;

[0047] Based on business hierarchy rules, establish hierarchical relationships between group tags to obtain a cross-level tag structure.

[0048] In one possible implementation, the processing module is further configured to:

[0049] Knowledge distillation is performed on the cross-level tag structure according to a preset cycle to generate hierarchical knowledge summaries;

[0050] Store the hierarchical knowledge summary in the database.

[0051] In one possible implementation, the processing module is specifically used for:

[0052] Determine the access frequency of the hierarchical knowledge summary;

[0053] The storage compression rate of the hierarchical knowledge summary is dynamically adjusted based on the access frequency.

[0054] Based on storage compression ratio, storage tier knowledge summaries are written to the database.

[0055] In one possible implementation, the processing module is specifically used for:

[0056] Determine the storage life of the hierarchical knowledge summary;

[0057] Based on storage cycles, storage-tiered knowledge summaries are written to the database.

[0058] In one possible implementation, the large language model is determined based on parameter pruning and quantization compression operations.

[0059] Thirdly, this application provides an electronic device, including: a memory and a processor;

[0060] The memory stores instructions that the computer executes;

[0061] The processor executes computer execution instructions stored in memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0062] Fourthly, this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the first aspect and / or various possible embodiments of the first aspect.

[0063] Fifthly, this application provides a computer program product, including a computer program that, when executed, implements the first aspect and / or various possible implementations of the first aspect.

[0064] This application provides a tag generation method, device, and medium, relating to the field of data processing. The method includes: acquiring unstructured text data from an enterprise information system, wherein the unstructured text data is unformatted natural language text; using a large language model to perform contextual analysis on the unstructured text data to generate an initial tag set; and processing the initial tag set based on preset rules to form a cross-level tag structure, which includes at least two business tag layers with different semantic dimensions, and establishes associations between the business tag layers. This application performs contextual analysis on the acquired unstructured text data to generate an initial tag set; secondly, it processes the initial tag set based on preset rules; and finally, it forms a cross-level tag structure based on the tags in the processed initial tag set. This cross-level tag structure can be directly used for subsequent structured data storage and multi-dimensional analysis. The entire process, through the combination of models and rules, achieves dynamic classification and tag management of unstructured data, thereby reducing the lag in tag updates. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] Figure 1 A flowchart illustrating the label generation method provided in this application embodiment. Figure 1 ;

[0067] Figure 2 A flowchart illustrating the label generation method provided in this application embodiment. Figure 2 ;

[0068] Figure 3 This is a schematic diagram of the label generation apparatus provided in the embodiments of this application;

[0069] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0070] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0071] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0072] In enterprise knowledge management and intelligent office systems, the efficient processing and analysis of unstructured text data is a core requirement for improving organizational effectiveness. For example, in a large enterprise's project management scenario, employees need to submit hundreds of freely formatted text reports daily, covering task progress, problem feedback, customer communication records, etc. Traditional knowledge base management methods rely on manually preset fixed tags to classify data, but in real-world scenarios, the same event may involve multiple dimensions, and the emergence of new projects or new problems frequently disrupts the preset tagging system. Furthermore, enterprises need to generate periodic analysis reports based on dynamic data, but existing methods lack the ability to perform real-time structured governance of unstructured data, resulting in lagging data analysis and difficulty in cross-level correlation.

[0073] To address the aforementioned issues, this application provides a tag generation method, which involves acquiring unstructured text data from an enterprise information system; using a large language model to perform contextual analysis on the unstructured text data to generate an initial tag set; and processing the initial tag set based on preset rules to form a cross-level tag structure.

[0074] This application applies to the field of enterprise knowledge management and aims to address the challenges of processing unstructured text data such as employee daily reports, weekly reports, and meeting minutes. By automatically generating structured tags and supporting multi-dimensional analysis, it improves the efficiency of knowledge retrieval, utilization, and value extraction.

[0075] This application also applies to intelligent office systems, which provide timely and valuable decision-making support to management by generating periodic reports such as team performance summaries and customer relationship insights in real time.

[0076] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0077] Figure 1 A flowchart illustrating the label generation method provided in this application embodiment. Figure 1 ,like Figure 1 As shown, the method includes:

[0078] S101. Obtain unstructured text data from the enterprise information system, wherein the unstructured text data is natural language text that has not been formatted.

[0079] Unstructured text data refers to natural language text that has not undergone formatting, such as employee daily reports and meeting minutes.

[0080] There are several ways to obtain unstructured text data. For example, you can obtain unstructured text data through a data source interface. This is just an example.

[0081] Furthermore, an enterprise information system is an integrated system designed to support and manage all aspects of an enterprise's operations. It helps enterprises make more informed decisions, improve efficiency, and optimize business processes by collecting, processing, storing, and distributing information. Enterprise information systems include, but are not limited to, the following systems: Enterprise Resource Planning (ERP) systems, intelligent office systems, and supply chain management systems.

[0082] S102. Using a large language model, perform contextual analysis on unstructured text data to generate an initial tag set.

[0083] Contextual analysis refers to parsing the contextual semantics of unstructured text data using a large language model to extract key information or tags. The initial tag set refers to the set of preliminary classification tags generated by contextual analysis, such as Customer C and Requirement Change.

[0084] This step utilizes a large language model to parse the semantic context of unstructured text data and dynamically generates an initial set of labels, avoiding the lag of manually preset labels.

[0085] Current methods typically rely on high-parameter, large-scale language models to process massive amounts of data, resulting in high computational costs. Furthermore, high-parameter, large-scale language models are not suitable for low-resource scenarios (such as edge devices or small businesses).

[0086] To address this issue, embodiments of this application introduce a lightweight large language model in some examples. Specifically, in some examples, the large language model used is determined based on parameter pruning and quantization compression operations. This means that the model used in this example is a lightweight large language model. The lightweight large language model is determined based on parameter pruning and quantization compression operations. Parameter pruning refers to removing redundant weights in the large language model to reduce computational load, such as deleting low-importance neurons; quantization compression converts the large language model parameters from a high-precision format to a low-precision format, such as converting 32-bit floating-point numbers to 8-bit integers. This example significantly reduces the computational resource requirements of the large language model through parameter pruning and quantization compression. For example, in small business knowledge management scenarios, the lightweight large language model can complete tag generation locally, avoiding reliance on cloud computing power, thereby reducing data transmission latency and improving privacy and security. Ultimately, the lightweight large language model achieves low-cost, high-real-time knowledge processing, adapting to the needs of edge devices or low-resource scenarios.

[0087] S103. Process the initial tag set based on preset rules to form a cross-level tag structure. The cross-level tag structure contains at least two business tag layers with different semantic dimensions, and establishes a relationship between the business tag layers.

[0088] After determining the initial tag set, it is processed based on preset rules to form a cross-level tag structure. For example, the initial tag set is filtered based on preset rules, and the retained tags are reorganized into a cross-level tag structure, such as Individual → Organization → Project. The reorganized cross-level tag structure can be directly used for subsequent structured data storage and multidimensional analysis.

[0089] This application embodiment performs contextual analysis on the acquired unstructured text data to generate an initial tag set; secondly, it processes the initial tag set based on preset rules; finally, it forms a cross-level tag structure based on the tags in the processed initial tag set. This cross-level tag structure can be directly used for subsequent structured data storage and multi-dimensional analysis. The entire process, through the combination of models and rules, achieves dynamic classification and tag management of unstructured data, thereby reducing the lag in tag updates.

[0090] Based on the above embodiments, a large language model is used to perform contextual analysis on unstructured text data to generate an initial tag set, including: segmenting the unstructured text data to extract key event fragments; generating semantic vectors based on the key event fragments; and matching the semantic vectors with a preset tag library to generate an initial tag set.

[0091] In this embodiment, to generate an initial tag set, the unstructured text data needs to be segmented to extract key event fragments. Segmentation refers to dividing the unstructured text data into independent segments based on semantics or punctuation, for example, dividing customer C's requirement change into customer C and the requirement change. Key event fragments contain independent semantic units containing key business information, such as the delay of project D.

[0092] Semantic vectors are generated using key event fragments to represent their semantic features. These generated semantic vectors are then matched with a pre-defined tag library to create an initial tag set. This entire process, through the synergy of the algorithm model and the tag library, achieves preliminary classification of unstructured data, laying the foundation for subsequent tag reassembly.

[0093] This application's embodiments significantly improve the accuracy and coverage of initial tag generation through segmented processing and semantic vector matching. For example, in enterprise project management scenarios, implicit business information in text (such as "customer C's requirement change") can be accurately parsed into the tags "customer C" and "requirement optimization," avoiding omissions or redundancy from manually preset tags. Furthermore, the semantic vector matching mechanism can dynamically adapt to the emergence of new business terms, ensuring real-time updates to the tag library and ultimately achieving flexibility and scalability in knowledge management.

[0094] Based on the above embodiments, in some examples, semantic vectors are generated based on key event fragments, including: vectorizing the key event fragments to obtain event vectors; and semantically encoding the event vectors to obtain semantic vectors.

[0095] In the example above, key event fragments are first vectorized to obtain corresponding event vectors. Then, semantic vectors are generated using a word embedding model or a pre-trained language model. For example, a word embedding model can convert requirement changes into fixed vectors, while a pre-trained language model can dynamically adjust the vector representation based on context (e.g., different vectors are generated for customer C's requirement change and project D's requirement change). Both methods extract semantic features from unstructured text data through algorithmic models, providing an accurate vector foundation for subsequent label matching.

[0096] This application's embodiments utilize either a word embedding model or a pre-trained language model to adapt to the semantic encoding needs of different scenarios. For example, in low-resource scenarios, the word embedding model can quickly generate fixed vectors, while in high-precision scenarios, the pre-trained language model can dynamically adjust the vector representation based on the context, significantly improving the accuracy of semantic matching.

[0097] Ultimately, the embodiments of this application enhance the flexibility and adaptability of initial label generation, providing a more reliable data foundation for subsequent cross-level label reorganization.

[0098] Based on the above embodiments, the initial tag set is filtered and reorganized according to preset rules to form a cross-level tag structure, including: calculating the semantic similarity of the tags in the initial tag set to determine conflicting tags and normal tags; replacing conflicting tags with semantically generalized unified tags; grouping the unified tags and normal tags to obtain grouped tags; and establishing hierarchical associations between grouped tags based on business level rules to obtain a cross-level tag structure.

[0099] This application first calculates the semantic similarity of the tags in the initial tag set. The key to this step is to identify the potential relationships between tags, including highly similar synonyms or near-synonyms and tags that may have conflicting or ambiguous meanings. By accurately calculating the semantic similarity between tags, it is possible to clearly define which tags are normal and independent, and which tags are conflicting or redundant, thus laying a solid foundation for subsequent tag standardization.

[0100] For example, various techniques can be used to calculate semantic similarity between tags in the initial tag set. For instance, tags can be converted into high-dimensional vector representations based on word vector models, and then the semantic similarity can be quantified by calculating cosine similarity. Furthermore, conceptual relationships from dictionary matching, thesaurus data, and user behavior data can be combined to assist in determining semantic similarity.

[0101] Based on this, the identified conflicting labels are replaced with semantically generalized unified labels. This replacement operation is the core step in resolving label redundancy and inconsistency. Conflicting labels often arise from different sources, users, or time points using inconsistent terms to describe the same concept. Their existence can easily lead to inefficient information retrieval, biased analysis results, and user confusion. By introducing semantically generalized unified labels, such as unifying "phone," "mobile phone," and "smart terminal" as "mobile device," these conflicts can be effectively eliminated, the total number of labels can be reduced, and the cohesion and standardization of the label set can be significantly improved. Therefore, data organization and querying will be more concise and efficient. When browsing or searching, users do not need to worry about multiple expressions under the same concept, thus significantly improving user experience and data quality.

[0102] Optionally, the specific methods for replacing conflicting labels with semantically generalized unified labels can vary. For example, a new, more generalized unified label can be generated for the conflicting labels through text summarization, keyword extraction, or manual definition. Alternatively, a lightweight large language model can be used to automatically generate a generalized label that covers the semantics of all conflicting labels. Or, the most representative or most frequently used label can be directly selected from the set of conflicting labels as the unified label. It should be noted that this is only one example.

[0103] The unified tags and normal tags are grouped to obtain grouped tags. This step, after tag standardization, further performs preliminary structuring of all processed tags. Tags that are semantically related or belong to the same business category are grouped into a logical group; for example, shirts, pants, and skirts are grouped into the clothing group. This setup not only makes the large tag set more readable and manageable but also provides an intermediate organizational structure before establishing strict hierarchical relationships. This grouping preprocessing effectively reduces the difficulty of establishing complex hierarchical relationships later, making the tag system clearer, easier to understand and maintain, and laying a solid foundation for the final cross-hierarchical structure.

[0104] In some examples, the methods for grouping uniform tags and normal tags can include various strategies. For example, unsupervised clustering can be performed based on the semantic similarity of tags or common topics to automatically form different tag groups. Alternatively, grouping can be based on the attributes, uses, or categories of the content associated with the tags to make the grouping more in line with business logic.

[0105] Furthermore, based on business hierarchy rules, hierarchical relationships are established between grouped tags to obtain a cross-level tag structure. By introducing business hierarchy rules, such as electronic products including mobile devices, and mobile devices including smartphones and tablets, previously scattered or merely grouped tags can be organized into a logically rigorous, hierarchical tree or network structure. This cross-level structure greatly enhances the navigation capabilities of tags and the accuracy of information retrieval. It allows users to gradually refine from high-level general concepts to low-level specific tags, or trace from specific tags to their corresponding general categories, thereby achieving multi-dimensional and multi-granular data access and analysis.

[0106] It's important to note that the source and form of business hierarchy rules can be very flexible. For example, these rules can originate from existing product classification systems, industry standards, domain expert knowledge bases, or ontology structures within the enterprise. Furthermore, hierarchical relationships can be automatically discovered and constructed using graph algorithms or machine learning methods by analyzing data such as the co-occurrence frequency of tags, user navigation paths, or click behavior.

[0107] The aforementioned hierarchical organization not only optimizes the data retrieval path and improves information search efficiency, but also provides a clearer and more structured view for data analysis and decision-making, ultimately forming a highly structured tag system that can effectively manage tag conflicts and redundancy while providing powerful navigation and retrieval capabilities.

[0108] In this embodiment, after forming the cross-level tag structure, the method further includes: performing knowledge distillation on the cross-level tag structure according to a preset period to generate a hierarchical knowledge summary; and storing the hierarchical knowledge summary in a database. Knowledge distillation refers to extracting core information from data using model compression technology, such as generating a weekly team performance summary; hierarchical knowledge summary refers to summary information generated according to business levels.

[0109] This embodiment triggers knowledge distillation of cross-level tag structures at a preset period to generate hierarchical knowledge summaries, which are then stored in the corresponding database. There are multiple ways to store the hierarchical knowledge summaries in the corresponding database.

[0110] In one implementation, storing the hierarchical knowledge summary in a database includes: determining the access frequency of the hierarchical knowledge summary; dynamically adjusting the storage compression ratio of the hierarchical knowledge summary based on the access frequency; and storing the hierarchical knowledge summary in the database based on the storage compression ratio.

[0111] In this implementation, the storage strategy is determined based on the frequency of queries to the hierarchical knowledge digest. Specifically, the storage compression ratio of the hierarchical knowledge digest is dynamically adjusted based on the frequency of queries. For example, high-frequency hierarchical knowledge digests use a low compression ratio to ensure access speed, while low-frequency hierarchical knowledge digests use a high compression ratio to save space.

[0112] In another implementation, storing the hierarchical knowledge summary in a database includes: determining the storage period of the hierarchical knowledge summary; and storing the hierarchical knowledge summary in the database based on the storage period.

[0113] In this embodiment, the storage strategy is determined based on the storage period corresponding to the hierarchical knowledge digest. Specifically, the storage period corresponding to the hierarchical knowledge digest is first determined; for example, the storage period for high-frequency hierarchical knowledge digests is one week, while the storage period for low-frequency hierarchical knowledge digests is one month. Then, the stored hierarchical knowledge digests are updated according to the storage period.

[0114] This application embodiment triggers a lightweight Large Language Model (LLM) at a preset period to perform knowledge distillation on cross-level tag structures, generating hierarchical knowledge summaries. Simultaneously, a dynamic storage strategy is determined based on the hierarchical knowledge summaries. This dynamic storage strategy balances storage space and access efficiency, avoiding data lag caused by one-time processing, ultimately achieving continuous and efficient knowledge management.

[0115] Next, examples will be given to illustrate how to use the label generation method provided in the embodiments of this application. Figure 2 A flowchart illustrating the label generation method provided in this application embodiment. Figure 2 .like Figure 2 As shown, the method includes the following steps:

[0116] 1. Data Acquisition: Unstructured text data is collected in real time through the data source interface.

[0117] 2. Structured governance: Utilize lightweight LLM to perform structured governance on unstructured text data. For example, use LLM to segment unstructured text data, generate semantic vectors, and match initial labels to output an initial label set.

[0118] 3. Tag Reorganization: Detect and unify conflicts in the initial tag set, and group the tags according to business level rules to obtain grouped tags.

[0119] 4. Knowledge distillation: Based on weekly, monthly, and quarterly cycles, hierarchical knowledge summaries are generated from grouped tags, and storage strategies are dynamically adjusted.

[0120] 5. Multidimensional analysis: Users can trace back related data through any tags, supporting cross-level analysis.

[0121] Furthermore, on a weekly basis, data from the week and individual top-performing data are input into a lightweight large language model for analysis and summarization, and the results are stored in a database. It should be noted that the monthly and quarterly plans follow the same principle as the weekly plan.

[0122] Furthermore, individual top information content can be extracted from unstructured text data according to the time dimension, and the top information of all people can be classified and integrated. Based on the top classification results, the integrated data is determined, the integrated data is input into a large language model for analysis and summary, and the analysis and summary results are stored in a database.

[0123] In summary, it can be seen that the input data of the lightweight large language model is real-time dynamic structured data, which does not require reliance on large parameter models and more tokens; the embodiments of this application analyze and summarize the data in the database periodically, refresh the periodic data and save the database, and the database stores hierarchical data.

[0124] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0125] Figure 3 This is a schematic diagram of the structure of the label generation device provided in the embodiments of this application, such as... Figure 3 As shown, the label generation device provided in this embodiment includes:

[0126] The acquisition module 301 is used to acquire unstructured text data from the enterprise information system, wherein the unstructured text data is natural language text that has not been formatted.

[0127] The generation module 302 is used to perform contextual analysis on unstructured text data using a large language model to generate an initial tag set;

[0128] The processing module 303 is used to process the initial tag set based on preset rules to form a cross-level tag structure. The cross-level tag structure contains at least two business tag layers with different semantic dimensions, and establishes a relationship between the business tag layers.

[0129] In one possible implementation, the generation module 302 is specifically used for:

[0130] Unstructured text data is segmented to extract key event fragments;

[0131] Generate semantic vectors based on key event fragments;

[0132] The semantic vectors are matched with a preset tag library to generate an initial tag set.

[0133] In one possible implementation, the generation module 302 is specifically used for:

[0134] Key event segments are vectorized to obtain event vectors;

[0135] Semantic vectors are obtained by semantically encoding the event vectors.

[0136] In one possible implementation, the processing module 303 is specifically used for:

[0137] Perform semantic similarity calculations on the tags in the initial tag set to identify conflicting and normal tags;

[0138] Replace conflicting tags with semantically generalized unified tags;

[0139] Grouping uniform labels and normal labels yields grouped labels;

[0140] Based on business hierarchy rules, establish hierarchical relationships between group tags to obtain a cross-level tag structure.

[0141] In one possible implementation, the processing module 303 is further configured to:

[0142] Knowledge distillation is performed on the cross-level tag structure according to a preset cycle to generate hierarchical knowledge summaries;

[0143] Store the hierarchical knowledge summary in the database.

[0144] In one possible implementation, the processing module 303 is specifically used for:

[0145] Determine the access frequency of the hierarchical knowledge summary;

[0146] The storage compression rate of the hierarchical knowledge summary is dynamically adjusted based on the access frequency.

[0147] Based on storage compression ratio, storage tier knowledge summaries are written to the database.

[0148] In one possible implementation, the processing module 303 is specifically used for:

[0149] Determine the storage life of the hierarchical knowledge summary;

[0150] Based on storage cycles, storage-tiered knowledge summaries are written to the database.

[0151] In one possible implementation, the large language model is determined based on parameter pruning and quantization compression operations.

[0152] The label generation device provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0153] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. For example, a processing module can be a separate processing element, or it can be integrated into an integrated circuit within the above device. Alternatively, it can be stored as program code in the device's memory, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. During implementation, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0154] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented by calling program code through a processing element, that processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to implement a System-On-a-Chip (SOC).

[0155] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 provided in this application embodiment may include: a processor 401, and a memory 402 communicatively connected to the processor, wherein:

[0156] The memory stores instructions that the computer executes;

[0157] The processor executes computer execution instructions stored in memory to implement the method described in the foregoing method embodiments.

[0158] It should be understood that processor 401 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor. Memory 402 may include high-speed random access memory (RAM), and may also include non-volatile memory (NVM), such as at least one disk storage device, or a USB flash drive, external hard drive, read-only memory, disk, or optical disc, etc.

[0159] Optionally, the electronic device 400 may also include a communication interface 403. In specific implementations, if the communication interface 403, memory 402, and processor 401 are implemented independently, they can be interconnected via a bus to complete communication. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc., but this does not imply that there is only one bus or one type of bus.

[0160] Optionally, in a specific implementation, if the communication interface 403, memory 402 and processor 401 are integrated on a single integrated circuit, then the communication interface 403, memory 402 and processor 401 can communicate through an internal interface.

[0161] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed, are used to implement the methods described in any of the foregoing embodiments.

[0162] It is understood that the computer-readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0163] An exemplary computer-readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the computer-readable storage medium. Of course, the computer-readable storage medium can also be a component of the processor. The processor and the computer-readable storage medium can reside in an ASIC. Alternatively, the processor and the computer-readable storage medium can exist as discrete components in an electronic device.

[0164] The integrated modules implemented as software functional modules described above can be stored in a computer-readable storage medium. These software functional modules, stored in a computer-readable storage medium, include several instructions to cause an electronic device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application.

[0165] This application also provides a computer program product, including a computer program that, when executed, implements the method described in any of the foregoing embodiments.

[0166] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this application.

[0167] It should be further noted that although the steps in the flowchart are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowchart may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0168] In the above embodiments, the descriptions of each embodiment have their own emphasis. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments. The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as these combinations of technical features do not contradict each other, they should be considered within the scope of this specification.

[0169] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this application are indicated by the following claims.

[0170] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A label generation method, characterized in that, include: Acquire unstructured text data from an enterprise information system, wherein the unstructured text data is natural language text that has not undergone formatting. Using a large language model, contextual analysis is performed on the unstructured text data to generate an initial tag set; The initial tag set is processed based on preset rules to form a cross-level tag structure. The cross-level tag structure contains at least two business tag layers with different semantic dimensions, and an association is established between the business tag layers.

2. The method according to claim 1, characterized in that, The step of using a large language model to perform contextual analysis on the unstructured text data and generating an initial tag set includes: The unstructured text data is segmented to extract key event fragments; Based on the key event fragments, generate semantic vectors; The semantic vector is matched with a preset tag library to generate the initial tag set.

3. The method according to claim 2, characterized in that, The generation of semantic vectors based on the key event fragments includes: The key event segments are vectorized to obtain event vectors; The event vector is semantically encoded to obtain the semantic vector.

4. The method according to any one of claims 1 to 3, characterized in that, The process of processing the initial tag set based on preset rules to form a cross-level tag structure includes: Semantic similarity is calculated for the tags in the initial tag set to determine conflicting tags and normal tags; Replace the conflicting tags with semantically generalized unified tags; The unified label and the normal label are grouped to obtain grouped labels; Based on business hierarchy rules, a hierarchical relationship is established between the group tags to obtain the cross-level tag structure.

5. The method according to any one of claims 1 to 3, characterized in that, After forming the cross-level tag structure, the following is also included: The cross-level tag structure is subjected to knowledge distillation according to a preset period to generate a hierarchical knowledge summary. The hierarchical knowledge summary is stored in the database.

6. The method according to claim 5, characterized in that, The step of storing the hierarchical knowledge summary in the database includes: Determine the access frequency of the hierarchical knowledge summary; The storage compression rate of the hierarchical knowledge summary is dynamically adjusted based on the access frequency. Based on the storage compression ratio, the hierarchical knowledge summary is stored in the database.

7. The method according to claim 5, characterized in that, The step of storing the hierarchical knowledge summary in the database includes: Determine the storage period for the hierarchical knowledge summary; Based on the storage period, the hierarchical knowledge summary is stored in the database.

8. The method according to any one of claims 1 to 3, characterized in that, The large language model is determined based on parameter pruning and quantization compression operations.

9. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed, are used to implement the method as described in any one of claims 1-8.

Images