Data marking method and device, equipment and medium

Abstract

The invention provides a data marking method. The method can be applied to the technical fields of big data and artificial intelligence. The method comprises the steps of obtaining multiple pieces of multi-modal data, preprocessing the multiple pieces of multi-modal data, and generating multiple pieces of preprocessed text data; and performing vector conversion on the plurality of pieces of preprocessed text data to generate a plurality of pieces of feature vector data. Density clustering is carried out on the multiple pieces of feature vector data, a data sets are generated, and each data set comprises a first data label. Semantic clustering is conducted on the a first data labels through semantic analysis, and b second data labels are generated. And presetting a business knowledge graph, and performing knowledge fusion on the business knowledge graph and the b second data tags to generate b target data tags for data marking. The invention further provides a data marking device and equipment, a storage medium and a program product.

Description

Technical Field

[0001] The present disclosure relates to the field of big data technology, specifically to the field of artificial intelligence technology, and in particular to a data labeling method, device, equipment, medium and program product. Background Art

[0002] With the rapid development of computer technology, businesses, including banks, are driven by these new technologies to build content tagging systems and intelligent labeling models to improve operational efficiency through automated classification, thereby promoting user growth and asset expansion. Mainstream technical solutions fall into three categories: manual approaches, which typically rely on empirically configured rules and are supplemented by text matching tools; algorithmic approaches, which typically employ algorithmic models and require manual data labeling; and hybrid approaches, which integrate rules and statistical models but still suffer from issues such as heavy reliance on human intervention and weak dynamic adjustment.

[0003] Currently, manual labeling is mostly used in the financial field, resulting in coarse label granularity, which makes it difficult to support precision marketing. There are challenges in balancing the degree of automation and label quality. In particular, the accuracy rate drops significantly when targeting professional fields such as financial compliance, which restricts the actual effectiveness of data-driven business. Summary of the Invention

[0004] In view of the above problems, the present disclosure provides a data marking method, apparatus, device, medium and program product.

[0005] According to the first aspect of the present disclosure, a data labeling method is provided, which includes: acquiring multiple multimodal data, preprocessing the multiple multimodal data, and generating multiple preprocessed text data; performing vector conversion on the multiple preprocessed text data to generate multiple feature vector data; performing density clustering on the multiple feature vector data to generate a data sets, wherein each data set includes a first data label, wherein a is an integer, and a is greater than or equal to 2; performing semantic clustering on the a first data labels through semantic analysis to generate b second data labels, wherein b is an integer, and b is greater than 0 and b is less than a; and presetting a business knowledge graph, performing knowledge fusion on the business knowledge graph and the b second data labels to generate b target data labels for data labeling.

[0006] According to an embodiment of the present disclosure, the multiple multimodal data are preprocessed to generate multiple preprocessed text data, including: performing data cleaning on the multiple multimodal data to generate multiple data-cleaned multimodal data; standardizing the multiple cleaned multimodal data to generate multiple standardized multimodal data; performing cross-modal alignment on the multiple standardized multimodal data to generate multiple cross-modal aligned text data; and performing desensitization operation on the multiple cross-modal aligned text data to generate multiple preprocessed text data.

[0007] According to an embodiment of the present disclosure, vector conversion is performed on the multiple preprocessed text data to generate multiple feature vector data, including: performing a block encoding operation on the multiple preprocessed text data to generate multiple block text data; performing a vector pooling operation on the multiple block text data to generate multiple pooled vector data; performing a text metadata fusion operation on the multiple pooled vector data to generate multiple fused vector data; and performing a feature extraction operation on the multiple fused vector data to generate multiple feature vector data.

[0008] According to an embodiment of the present disclosure, density clustering is performed on the multiple feature vector data to generate a data set, including: density clustering is performed on the multiple feature vector data to generate a first-level structure tree; and data sets of each cluster in the first-level structure tree are obtained to generate a data set.

[0009] According to an embodiment of the present disclosure, semantic clustering is performed on a first data label through semantic analysis to generate b second data labels, including: extracting feature semantics from the a first data label to generate a first data label key semantics; and calculating the semantic similarity between each key semantic in the a first data label key semantics, merging the first data labels whose semantic similarity in the a first data label key semantics is higher than a preset threshold, and generating b second data labels.

[0010] According to an embodiment of the present disclosure, knowledge fusion is performed based on the business knowledge graph and the b second data tags to generate b target data tags, including: bottom-up semantic clustering of the first hierarchical structure tree based on the b second data tags to generate a second hierarchical structure tree; based on the business knowledge graph, a business knowledge graph hierarchical structure is constructed; predicate constraint pruning is performed on the business knowledge graph hierarchical structure from top to bottom to generate a target business hierarchical structure having the same level as the second hierarchical structure tree; and the b second data tags in the second hierarchical structure tree are fused with the business knowledge of the target business hierarchical structure to generate b target data tags having a business knowledge graph hierarchical structure.

[0011] According to an embodiment of the present disclosure, the method further includes: mining the b target data labels through an association rule algorithm to generate c mined data labels for data labeling of the multiple multimodal data, where c is an integer and c is greater than 1.

[0012] According to the second aspect of the present disclosure, a data labeling device is provided, which includes: a first generation module for acquiring multiple multimodal data, preprocessing the multiple multimodal data, and generating multiple preprocessed text data; a second generation module for performing vector conversion on the multiple preprocessed text data to generate multiple feature vector data; a third generation module for density clustering the multiple feature vector data to generate a data sets, wherein each data set includes a first data label, wherein a is an integer and a is greater than or equal to 2; a fourth generation module for performing semantic clustering on the a first data labels through semantic analysis to generate b second data labels, wherein b is an integer and b is greater than 0 and b is less than a; and a fifth generation module for presetting a business knowledge graph, performing knowledge fusion on the business knowledge graph and the b second data labels, and generating b target data labels for data labeling.

[0013] According to an embodiment of the present disclosure, the first generation module includes: a sixth generation module, which is used to perform data cleaning on the multiple multimodal data to generate multiple data-cleaned multimodal data; a seventh generation module, which is used to standardize the multiple cleaned multimodal data to generate multiple standardized multimodal data; an eighth generation module, which is used to perform cross-modal alignment on the multiple standardized multimodal data to generate multiple cross-modal aligned text data; and a ninth generation module, which is used to perform desensitization operation on the multiple cross-modal aligned text data to generate multiple preprocessed text data.

[0014] According to an embodiment of the present disclosure, the second generation module includes: a tenth generation module, which is used to perform block encoding operations on the multiple preprocessed text data to generate multiple block text data; an eleventh generation module, which is used to perform vector pooling operations on the multiple block text data to generate multiple pooled vector data; a twelfth generation module, which is used to perform text metadata fusion operations on the multiple pooled vector data to generate multiple fused vector data; and a thirteenth generation module, which is used to perform feature extraction operations on the multiple fused vector data to generate multiple feature vector data.

[0015] According to an embodiment of the present disclosure, the third generation module includes: a fourteenth generation module, used to density cluster the multiple feature vector data to generate a first-level structure tree; and a fifteenth generation module, used to obtain the data set of each cluster in the first-level structure tree to generate a data set.

[0016] According to an embodiment of the present disclosure, the fourth generation module includes: a sixteenth generation module, which is used to extract feature semantics from the a first data labels to generate a first data label key semantics; and a seventeenth generation module, which is used to calculate the semantic similarity between each key semantic in the a first data label key semantics, merge the first data labels whose semantic similarity in the a first data label key semantics is higher than a preset threshold, and generate b second data labels.

[0017] According to an embodiment of the present disclosure, the fifth generation module includes: an eighteenth generation module, which is used to perform bottom-up semantic clustering on the first hierarchical structure tree based on the b second data labels to generate a second hierarchical structure tree; a nineteenth generation module, which is used to construct a business knowledge graph hierarchical structure based on the business knowledge graph; a twentieth generation module, which is used to perform predicate constraint pruning on the business knowledge graph hierarchical structure from top to bottom to generate a target business hierarchical structure with the same level as the second hierarchical structure tree; and a twenty-first generation module, which is used to fuse the b second data labels in the second hierarchical structure tree with the business knowledge of the target business hierarchical structure to generate b target data labels with a business knowledge graph hierarchical structure.

[0018] According to a third aspect of the present disclosure, an electronic device is provided, comprising: one or more processors; a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the above-mentioned data marking method.

[0019] According to a fourth aspect of the present disclosure, a computer-readable storage medium is provided, on which executable instructions or computer programs are stored. When the instructions or computer programs are executed by a processor, the processor executes the above-mentioned data marking method.

[0020] According to a fifth aspect of the present disclosure, a computer program product is further provided, comprising a computer program, which implements the above-mentioned data marking method when executed by a processor.

[0021] The technical solution disclosed in the present invention achieves the technical effect of breaking through the efficiency bottleneck of traditional manual labeling while ensuring the quality of content labels through technical means such as multimodal data preprocessing, vectorization, density clustering, semantic clustering and business knowledge fusion, and greatly improving the efficiency and accuracy of data labeling. In addition, due to the integration of business knowledge graphs, the technical effect of improving the business relevance, semantic accuracy and practicality of data labels can be achieved. It solves the problem that manual labeling results in coarse label granularity, making it difficult to support precision marketing, and faces challenges in balancing the degree of automation and label quality. In particular, the accuracy rate drops significantly when targeting professional content such as financial compliance, which restricts the actual effectiveness of data-driven business. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0023] Figure 1 The following schematically illustrates an application scenario of the data marking method and device according to an embodiment of the present disclosure;

[0024] Figure 2 The flowchart of the data marking method according to the embodiment of the present disclosure is schematically shown;

[0025] Figure 3 Schematically shows a flow chart of generating a plurality of pre-processed text data in a data tagging method according to an embodiment of the present disclosure;

[0026] Figure 4 Schematically shows a flow chart of generating a feature vector in a data marking method according to an embodiment of the present disclosure;

[0027] Figure 5 Schematically shows a flow chart of generating a data set in a data labeling method according to an embodiment of the present disclosure;

[0028] Figure 6 Schematically shows a flow chart for generating b second data labels in a data labeling method according to an embodiment of the present disclosure;

[0029] Figure 7 Schematically shows a flow chart for generating b target data labels in a data labeling method according to an embodiment of the present disclosure;

[0030] Figure 8 The following schematically shows a flow chart of a specific embodiment of the data marking method according to an embodiment of the present disclosure;

[0031] Figure 9 Schematically shows a structural block diagram of a data marking device according to an embodiment of the present disclosure; and

[0032] Figure 10 The block diagram schematically shows an electronic device suitable for implementing the data marking method according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0033] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the detailed description below, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0034] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0035] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0036] When expressions such as "at least one of A, B, and C, etc." are used, they should generally be interpreted in accordance with the meaning commonly understood by those skilled in the art (for example, "a system having at least one of A, B, and C" should include but is not limited to a system having A alone, B alone, C alone, A and B, A and C, B and C, and / or A, B, C, etc.).

[0037] The accompanying drawings illustrate some block diagrams and / or flow charts. It should be understood that some blocks in the block diagrams and / or flow charts, or combinations thereof, may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable control device, so that when executed by the processor, these instructions may create a device for implementing the functions / operations described in the block diagrams and / or flow charts.

[0038] First, let’s explain the technical terms that appear in this article as follows:

[0039] Content tagging system: This system systematically categorizes and tags content resources to create a multi-level, multi-dimensional tag set for precise content management, distribution, and analysis, improving operational efficiency. Its core focus is on converting unstructured content into structured data, supporting intelligent analysis and business decision-making.

[0040] Intelligent tagging: This process uses artificial intelligence algorithms (such as large models and natural language processing) to automatically generate tags for content, supporting manual intervention and dynamic optimization. The core technologies of intelligent tagging include text feature extraction, semantic understanding, and contextual analysis.

[0041] Universal tags: Standardized classification tags across business scenarios (such as finance, technology, and health) with wide applicability.

[0042] Featured tags: Special tags customized for specific business scenarios (such as bank product recommendations and user lifecycle management), such as "high net worth customer financial management" and "new user benefits".

[0043] Association rule algorithm is a data mining technology that mines implicit relationships between items from large-scale data sets.

[0044] An embodiment of the present disclosure provides a data labeling method, which includes: acquiring a plurality of multimodal data, preprocessing the plurality of multimodal data, and generating a plurality of preprocessed text data. Performing vector conversion on the plurality of preprocessed text data to generate a plurality of feature vector data. Density clustering is performed on the plurality of feature vector data to generate a data sets, wherein each data set includes a first data label, wherein a is an integer, and a is greater than or equal to 2. Semantic clustering is performed on the a first data labels through semantic analysis to generate b second data labels, wherein b is an integer, and b is greater than 0 and b is less than a. And a preset business knowledge graph is performed, and knowledge fusion is performed on the business knowledge graph and the b second data labels to generate b target data labels for data labeling.

[0045] According to the disclosed embodiments, through the use of technical means such as multimodal data preprocessing, vectorization, density clustering, semantic clustering and business knowledge fusion, it is achieved that the efficiency bottleneck of traditional manual labeling is broken through while ensuring the quality of content labels, and the efficiency and accuracy of data labeling are greatly improved. In addition, due to the fusion of business knowledge graphs, the technical effect of improving the business relevance, semantic accuracy and practicality of data labels can be achieved. It solves the problem that manual labeling results in coarse label granularity, making it difficult to support precision marketing, and faces challenges in balancing the degree of automation and label quality. In particular, the accuracy rate drops significantly when targeting professional content such as financial compliance, which restricts the actual effectiveness of data-driven business.

[0046] Figure 1The following diagram schematically illustrates an application scenario of the data marking method and device according to an embodiment of the present disclosure. Figure 1 The examples shown are merely examples of scenarios in which the embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but do not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0047] like Figure 1 As shown, the application scenario 100 according to this embodiment may include a data tagging application scenario. A network 104 is used as a medium for providing a communication link between a first terminal device 101, a second terminal device 102, a third terminal device 103, and a server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables.

[0048] A user may use a first terminal device 101, a second terminal device 102, or a third terminal device 103 to interact with a server 105 via a network 104 to receive or send messages, etc. Various communication client applications may be installed on the first terminal device 101, the second terminal device 102, or the third terminal device 103, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (for example only).

[0049] The first terminal device 101 , the second terminal device 102 , and the third terminal device 103 may be various electronic devices having display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, and the like.

[0050] The server 105 may be a server that provides various services, such as a background management server (for example only) that supports websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103. The background management server may analyze and process received data such as user requests, and feed back processing results (e.g., web pages, information, or data obtained or generated based on user requests) to the terminal devices.

[0051] It should be noted that the data marking method provided in the embodiment of the present disclosure can generally be executed by the server 105. Accordingly, the data marking device provided in the embodiment of the present disclosure can generally be set in the server 105. The data marking method provided in the embodiment of the present disclosure can also be executed by a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105. Accordingly, the data marking device provided in the embodiment of the present disclosure can also be set in a server or server cluster that is different from the server 105 and can communicate with the first terminal device 101, the second terminal device 102, the third terminal device 103 and / or the server 105.

[0052] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as required.

[0053] The following will be based on Figure 1 The scene described by Figures 2 to 8 The data marking method of the disclosed embodiment is described in detail. It should be noted that the above application scenarios are only shown to facilitate understanding of the spirit and principles of the present disclosure, and the embodiments of the present disclosure are not limited in this respect. On the contrary, the embodiments of the present disclosure can be applied to any applicable scenario.

[0054] Figure 2 The flowchart of the data marking method according to the embodiment of the present disclosure is schematically shown.

[0055] like Figure 2 As shown, the method 200 includes steps S201 to S205.

[0056] Step S201: Acquire a plurality of multimodal data, preprocess the plurality of multimodal data, and generate a plurality of preprocessed text data.

[0057] For example, multiple multimodal data sets can include: matrix-based data extracted using document parsing tools, such as structured reports and tabular data; unstructured data such as social media and news, acquired in real time through distributed collection systems; image data from user-generated content locally pre-processed using edge computing devices, connected to professional data platforms through standardized interfaces; live streaming captured using low-latency transmission protocols; and audio data from recorded videos imported in batches through cloud storage services.

[0058] Figure 3 The flowchart of generating a plurality of pre-processed text data in the data labeling method according to an embodiment of the present disclosure is schematically shown.

[0059] like Figure 3 As shown, the method 300 includes steps S301 to S304.

[0060] Step S301 : performing data cleaning on the plurality of multimodal data to generate a plurality of cleaned multimodal data.

[0061] For example, domain-specific stopword libraries can be built to filter templated content and perform noise filtering on multiple multimodal data. After noise filtering, fuzzy time representations in the data documents can be converted to a standardized time format for spatiotemporal alignment for data cleaning.

[0062] Step S302 : normalizing the plurality of cleaned multimodal data to generate a plurality of standardized multimodal data.

[0063] For example, multiple cleaned multimodal data can be reconstructed based on computer vision technology to identify complex table structures and reconstruct logical relationships. Then, an automated quantitative unit conversion module can be deployed on the reconstructed table data to unify the units, complete standardization, and generate multiple standardized multimodal data.

[0064] Step S303: performing cross-modal alignment on the plurality of standardized multimodal data to generate a plurality of cross-modal aligned text data.

[0065] For example, multiple standardized multimodal data sets can be linked by establishing index relationships between text and images using a multimodal semantic matching model. Alternatively, a temporal alignment algorithm can be used to align the timelines of video and audio text to achieve audio-visual synchronization. Finally, text conversion is performed uniformly to generate multiple cross-modally aligned text data sets.

[0066] Step S304: performing a desensitization operation on the plurality of cross-modal aligned text data to generate a plurality of pre-processed text data.

[0067] For example, the privacy data of the multiple cross-modally aligned text data can be desensitized, specifically including: detecting the format of sensitive information through a pattern recognition engine and implementing dynamic masking; identifying entity information (such as name / institution) through a machine learning model and performing hierarchical desensitization.

[0068] The text data after the multiple cross-modal alignments can also be screened for compliance content, specifically including: forming a dynamically updated negative vocabulary library through a keyword enhancement model combined with regulatory rules; and detecting illegal expressions and their variant expressions through a semantic analysis model.

[0069] A hierarchical management and control mechanism can also be established for the multiple cross-modal aligned text data, specifically including: dynamic control of data access rights through a multi-level sensitivity classification system; and adding traceability identification to confidential content through digital watermarking technology.

[0070] By performing data cleaning, data standardization, data cross-domain alignment and desensitization on multimodal data, and generating pre-processed text data, the technical effects of improving data quality, enhancing data semantic relevance and improving data security compliance are achieved.

[0071] Return to reference Figure 2 In step S202, the plurality of pre-processed text data are vector-converted to generate a plurality of feature vector data.

[0072] Figure 4 The flowchart of generating a feature vector in the data marking method according to an embodiment of the present disclosure is schematically shown.

[0073] like Figure 4 As shown, the method 400 includes steps S401 to S404.

[0074] Step S401 : performing block encoding operation on the plurality of pre-processed text data to generate a plurality of block text data.

[0075] Step S402: performing a vector pooling operation on the plurality of block text data to generate a plurality of pooled vector data.

[0076] For example, a triplet network structure model optimized by contrastive learning can be used as the core architecture to support block encoding and vector pooling operations for long texts, effectively capturing long-range semantic dependencies. To improve domain adaptability, the model undergoes secondary pre-training using domain-specific corpora to enhance the representation of specialized terminology, and an optional multimodal extension module can be added to handle mixed text and image content.

[0077] Step S403 : performing a text metadata fusion operation on the plurality of pooled vector data to generate a plurality of fused vector data.

[0078] For example, text metadata may include data such as publication time and author attributes. The metadata is used as an auxiliary vector to perform vector fusion on multiple pooled vector data to generate multiple fused vector data.

[0079] Step S404: performing a feature extraction operation on the plurality of fused vector data to generate a plurality of feature vector data.

[0080] For example, information retrieval and text mining techniques can be used to filter common vocabulary by weight, retain core entity features, and reduce noise interference.

[0081] By using technical means such as block encoding, vector pooling, metadata fusion and feature extraction, we can solve the bottleneck of long texts, capture local semantics, integrate block semantics, generate a global representation, inject business attributes of domain metadata, enhance interpretability, reduce dimensionality and remove redundancy, and optimize the adaptability of subsequent clustering.

[0082] Return to reference Figure 2 In step S203, density clustering is performed on the multiple feature vector data to generate a data sets, where each data set includes a first data label, where a is an integer and a is greater than or equal to 2.

[0083] Figure 5 The flowchart of generating a data set in the data labeling method according to an embodiment of the present disclosure is schematically shown.

[0084] like Figure 5 As shown, the method 500 includes steps S501-S502.

[0085] Step S501: performing density clustering on the plurality of feature vector data to generate a first-level structure tree.

[0086] For example, the automatic identification and optimization of label categories can be achieved based on the density clustering algorithm. A parameter self-adaptation strategy is adopted, and the optimal neighborhood radius ε of the density clustering algorithm is used. The minimum number of samples is adjusted according to the logarithmic function based on the data scale to improve clustering stability. For complex semantic distribution, after the first round of clustering, the density clustering algorithm is used to perform hierarchical sub-density clustering on large clusters, and a nested clustering tree is constructed to identify fine-grained patterns. In terms of outlier management, probabilistic attribution is assigned to outliers, and a manual correction interface is provided. Low-confidence samples are fed back to the iterative optimization process of the annotation system to form a closed-loop processing mechanism. By generating a cluster hierarchy structure tree and cluster labels through the density clustering algorithm, the cluster hierarchy structure tree is used as the first-level structure tree, and the cluster label is used as the first data label.

[0087] Step S502: Obtain the data set of each cluster in the first-level structure tree to generate a data set.

[0088] The technical means of generating cluster data sets by constructing a hierarchical structure tree through density clustering makes the data hierarchy clear and the label semantics clear, achieving the technical effect of improving the reliability and interpretability of the labels.

[0089] Return to reference Figure 2 In step S204, semantic clustering is performed on a first data label through semantic analysis to generate b second data label, where b is an integer, and b is greater than 0 and b is less than a.

[0090] Figure 6The flowchart of generating b second data labels in the data labeling method according to an embodiment of the present disclosure is schematically shown.

[0091] like Figure 6 As shown, the method 600 includes steps S601-S602.

[0092] Step S601: extract feature semantics from the a first data tags to generate a first data tag key semantics.

[0093] For example, natural language technology can be used to extract feature semantics from a first data tag to generate a first data tag key semantics.

[0094] Step S602, calculate the semantic similarity between each key semantic in the a first data tag key semantics, merge the first data tags whose semantic similarity in the a first data tag key semantics is higher than a preset threshold, and generate b second data tags.

[0095] In addition, a cluster label extraction model may be pre-trained based on historical data, a first data labels may be input into the pre-trained cluster label extraction model, and b second data cluster labels may be output.

[0096] By using the technical means of feature semantic extraction and semantic similarity merging of the first data tag, the reliability and quality of the tag are improved, and the efficiency of tag management is improved.

[0097] Return to reference Figure 2 In step S205, a business knowledge graph is preset, and knowledge fusion is performed on the business knowledge graph and the b second data tags to generate b target data tags for data labeling.

[0098] For example, a pre-defined business knowledge graph can include general tags and featured tags. General tags generate basic classification tags based on a hierarchical topic model, using a multi-level semantic architecture to divide the tag hierarchy: first-level tags focus on industry sectors (e.g., [Banking], [Securities], [Insurance]), while second-level tags extend to scenario-specific topics (e.g., [Fund Subscription Process], [Credit Delinquency Risk]).

[0099] Common tags can be preprocessed, including implementing dynamic optimization strategies to address text length differences: short texts (<200 words) use a broad topic distribution to avoid information sparsity; long texts (>1000 words) use sliding window segmentation (150-word windows) to capture local semantics, and use inverse document frequency (IDF) to filter out high-frequency invalid tags (such as "finance" and "business") to retain strong discriminative features. The system supports progressive optimization, automatically calibrating topic distribution weights based on clustering results, and improving tag semantic consistency.

[0100] Featured tags generate scenario-based segmentation tags through multi-level business rule library matching. The rule engine adopts a three-tier architecture: basic product rules define general business logic, scenario rules adapt to vertical field needs, and risk control rules embed compliance constraints.

[0101] Featured tags can be pre-processed, including integrating semantic enhancement mechanisms and using dependency analysis to verify tag rationality (for example, requiring that the "interest rate" in "interest rate adjustment" must be the object of a verb). Business personnel can use a visual interface to update the rule base in real time. New rules will take full effect after being verified through a phased release (e.g., trigger frequency > 50 times / day and accuracy > 95%).

[0102] Figure 7 The flowchart of generating b target data labels in the data labeling method according to an embodiment of the present disclosure is schematically shown.

[0103] like Figure 7 As shown, the method 700 includes steps S701 to S704.

[0104] Step S701 : performing bottom-up semantic clustering on the first hierarchical structure tree based on the b second data labels to generate a second hierarchical structure tree.

[0105] Step S702: construct a business knowledge graph hierarchical structure based on the business knowledge graph.

[0106] Step S703: perform predicate constraint pruning on the business knowledge graph hierarchical structure from top to bottom to generate a target business hierarchical structure having the same level as the second hierarchical structure tree.

[0107] Step S704: fuse the b second data tags in the second hierarchical structure tree with the business knowledge of the target business hierarchical structure to generate b target data tags with a business knowledge graph hierarchical structure.

[0108] Through the dual paths of knowledge fusion, namely, bottom-up semantic clustering of the first-level structure tree and top-down predicate constraint pruning of the business knowledge graph hierarchy, the data clustering tree and the business knowledge graph hierarchy are aligned, realizing the transition from data clustering to business cognition and improving the business interpretability of the generated target data labels.

[0109] Furthermore, after generating b target data labels, the b target data labels can be mined using an association rule algorithm to generate c mined data labels for use in data labeling of the plurality of multimodal data, where c is an integer greater than 1. By performing data label mining using an association rule algorithm, the comprehensiveness and robustness of data labeling can be improved.

[0110] Figure 8 The flowchart of a specific embodiment of the data marking method according to the embodiment of the present disclosure is schematically shown.

[0111] like Figure 8 As shown, the architecture of the specific embodiment 800 is divided into a data layer, an algorithm layer, and an application layer. The embodiment introduces a multimodal data acquisition module, data cleaning and standardization, a sensitive information filtering module, a labeling engine, a clustering analysis module, a real-time feedback learning module, a label management platform, and a business scenario adaptation module.

[0112] In client 801, users browse, select, and purchase financial products. The structured and unstructured data generated by these browsing and operations are sent to the multimodal data acquisition module 802. Multimodal data acquisition module 802 uses diverse acquisition strategies to collect structured and unstructured data from client 801, centralizes the data, and provides high-quality raw data input for subsequent processing. The data cleaning and standardization module 803 is used to build a multi-level data quality enhancement system. It uses a multi-level cleaning pipeline to automatically clean the data collected by the multimodal data acquisition module 802, eliminating data noise, repairing structural defects, and achieving spatiotemporal alignment and semantic consistency of the multimodal data. The sensitive information filtering module 804 obtains the cleaned data from the data cleaning and standardization module 803 and intelligently filters the sensitive information contained therein using multi-level detection technology. It removes unnecessary sensitive information and desensitizes and manages sensitive information that cannot be removed, ensuring that the data processing process meets industry compliance requirements. The cluster analysis module 805 obtains the filtered data from the sensitive information filtering module 804, performs semantic clustering and knowledge fusion operations, and realizes dynamic optimization of the data to be labeled.

[0113] The specific algorithm of cluster analysis module 805 includes the vectorization phase, the clustering phase, and the hierarchical partitioning phase. During the vectorization phase, a semantic encoding model is used to convert text into a structured vector representation to support subsequent cluster analysis. A triplet network structure model optimized by contrastive learning is used as the core architecture, supporting block encoding and vector pooling for long texts, effectively capturing long-range semantic dependencies. To improve domain adaptability, the model undergoes secondary pre-training using domain-specific corpora to enhance the representation of specialized terminology. An optional multimodal extension module is available to handle mixed text and image content. Regarding feature enhancement, text metadata (such as publication time and author attributes) is integrated as auxiliary vector dimensions. Information retrieval and text mining techniques are used to weight and filter common vocabulary, preserving core entity features and reducing noise interference. During the clustering phase, a density clustering algorithm is used to automatically identify and optimize label categories. A parameter adaptive strategy is employed, using the density clustering algorithm's optimal neighborhood radius ε and adjusting the minimum sample size using a logarithmic function based on data size to improve clustering stability. For complex semantic distributions, after the initial clustering, a density clustering algorithm is used to perform hierarchical sub-density clustering on large clusters, constructing a nested cluster tree to identify fine-grained patterns. For outlier management, probabilistic attribution is assigned to outliers, and a manual correction interface is provided. Low-confidence samples are fed back to the annotation system for iterative optimization, forming a closed-loop processing mechanism. During the hierarchical division phase, a dynamic hierarchical labeling system is constructed by combining a pre-configured business domain knowledge graph with semantic similarity calculations. This system employs a dual knowledge fusion approach: top-down, a pre-set hierarchical structure is imported and conflicting branches are pruned based on predicate constraints to ensure alignment with business specifications. Bottom-up, a bottom-up hierarchical clustering algorithm aggregates semantically similar clusters, combining cosine distance and divergence to measure similarity and generate a scalable hierarchical tree. The system includes a built-in dynamic optimization module that monitors concept drift through sliding window statistical tests and employs an incremental clustering strategy for local updates to newly added data, reducing overall computational overhead. The final output is a hierarchical labeling system that includes semantic center, confidence level, and rule association information, supporting visual interactive adjustment and versioning. The tag generation process simultaneously associates business domain knowledge graph entities to ensure strong consistency with business definitions, while also providing rule conflict detection and prioritization capabilities. Tagging Engine 806, based on an intelligent tag generation system, achieves multi-dimensional automatic annotation of text content through multi-stage semantic understanding and rule fusion, generating and outputting multi-level tags and confidence scores. After completing these steps, the tag data is output, enabling tag mining based on association rule algorithms.

[0114] The tag management platform 807 provides business personnel with intelligent collaboration tools to achieve full lifecycle management and closed-loop optimization of system-generated tags. The platform builds a multi-dimensional evaluation dashboard through an interactive evaluation and review mechanism. Business personnel can view tag generation quality indicators (such as confidence score, rule matching, and historical accuracy) through a visual dashboard, and quickly locate low-quality tags based on the tag health heat map generated by the dynamic scoring matrix and semantic consistency indicators. At the same time, it integrates the knowledge graph entity verification function, automatically highlights the association path between tags and business rules during manual review, provides conflict detection prompts, and recommends historical correction solutions for long-tail scenario tags. In the manual intervention and error correction mechanism, the platform provides drag-and-drop tag hierarchical editing, a semantic correction interface driven by dependency syntax analysis, and a mandatory association function for the business rule library. It supports batch processing and traceable version snapshot management to ensure the consistency of the tag system with the knowledge graph predicate constraints and business policies. After manual intervention and error correction operations are completed, the adjusted labels are written into the label database 810, and the downstream business scenarios are triggered to be updated synchronously. At the same time, the correction content and operation logs are encapsulated as training samples and transmitted to the real-time feedback learning module 808 to drive the incremental training of the large model. In the process, the hierarchical desensitization strategy of the sensitive information filtering module 804 is inherited, and digital watermarks are added to confidential labels to meet financial audit requirements.

[0115] The real-time feedback learning module 808, connected to the label management platform 807, obtains original labels, corrected labels, and manual intervention records in real time. It then applies machine learning algorithms to analyze annotation deviation patterns and rule defects, locating semantic understanding blind spots and business logic vulnerabilities. Based on error attribution analysis, this module dynamically optimizes the feature extraction capabilities of the clustering analysis module 805 and the decision path of the labeling engine 806. It continuously iterates on the text vectorization strategy and rule triggering mechanism, transforming the error pattern from manual correction to algorithmic self-healing.

[0116] Tag Database 809 uses a layered architecture to manage the entire lifecycle of multimodal tags, supporting dynamic version control and fine-grained permission management. It provides real-time data services to downstream systems through vectorized storage and efficient retrieval mechanisms. Relying on real-time synchronization technology, it ensures strong consistency between business scenarios and the tag system, building a data bridge connecting model outputs and business applications.

[0117] Business scenario adaptation module 810 analyzes tag data in the tag database to build a demand-tag mapping model, identifying users' potential service needs and product preferences. It generates personalized marketing strategies based on tag association rules, such as binding high-frequency tags to new product recommendations or filtering compliant product pools by risk level tags to achieve targeted marketing push.

[0118] The operating principle of this specific embodiment 800 is as follows: When a user browses, selects, or purchases financial products on client 801, the structured and unstructured data generated by these browsing and operations is sent to the multimodal data acquisition module 802. The multimodal data acquisition module 802 collects structured and unstructured data from client 801 using diverse acquisition strategies, aggregates the data, and sends the aggregated data to the data cleaning and standardization module 803. The data cleaning and standardization module 803 obtains the aggregated data from the multimodal data acquisition module 802, performs automated cleaning on the collected data, removes data noise, repairs structural defects, and performs spatiotemporal alignment and semantic consistency processing on the multimodal data. The cleansed data is then sent to the sensitive information filtering module 804. The sensitive information filtering module 804 obtains the cleansed data from the data cleaning and standardization module 803 and intelligently filters the sensitive information using multi-level detection technology, removing unnecessary sensitive information and desensitizing and managing any sensitive information that cannot be removed. After confirming data compliance, the filtered data is sent to the clustering analysis module 805. The cluster analysis module 805 obtains filtered data from the sensitive information filtering module 804 and performs semantic clustering and knowledge fusion operations on the data to dynamically optimize the data to be tagged. After the cluster analysis is completed, the data to be tagged is sent to the tagging engine 806. The tagging engine 806 automatically annotates the data content to be tagged in multiple dimensions through multi-stage semantic understanding and rule fusion. After integration, the generated tags are generated and output to the tag management platform 807. Business personnel access the tag management platform 807 to analyze and review the generated tags. If the analysis finds that the generated tags contain non-compliant content, do not conform to the actual business scenario, or contain erroneous data, they manually intervene and correct the tags. The tags after manual intervention and correction are sent to the tag database storage 809. The business scenario adaptation module 810 parses the tag data in the tag database 809, constructs a demand-tag mapping model, identifies users' potential service needs and product preferences, and generates personalized marketing strategies based on tag association rules, achieving precise marketing and push notifications for products and services that meet user needs. Real-time feedback learning module 808 connects to label management platform 807, acquiring original labels, revised labels, and manual intervention records in real time. It then applies machine learning algorithms to analyze annotation deviation patterns and rule flaws, locating semantic understanding blind spots and business logic vulnerabilities. Based on these findings, real-time feedback learning module 808 dynamically optimizes the feature extraction capabilities of cluster analysis module 805 and the decision paths of labeling engine 806.

[0119] This specific embodiment 800 deeply integrates multimodal semantic alignment technology through an automated labeling engine to automatically extract related features from cross-modal data such as text, images, audio and video. Through the dual mechanisms of domain knowledge enhancement and knowledge graph entity verification, it breaks through the efficiency bottleneck of traditional manual labeling while ensuring the quality of content labels, and greatly improves the degree of automation of the labeling process. The system combines dynamic clustering algorithms with real-time feedback learning modules to enable the labeling system to have the ability to adapt to the migration of market hotspots. Through sliding window statistical tests, it can timely capture changes in user preferences and effectively improve the accuracy and dynamic response speed of digital marketing. Relying on a visual interactive platform and cross-modal alignment technology, unified label mapping and precise adaptation to business scenarios for multi-source data are achieved, which not only eliminates data barriers for cross-departmental collaboration, but also enhances the traceability of labeling decisions through explainable AI technology. Through the linkage of a hierarchical desensitization mechanism and a compliance rule library, a balance is established between sensitive information protection and business innovation needs, providing full-process intelligent support for digital marketing and user operations.

[0120] Figure 9 The structural block diagram of the data marking device according to an embodiment of the present disclosure is schematically shown.

[0121] like Figure 9 As shown, the device 900 includes: a first generating module 901 , a second generating module 902 , a third generating module 903 , a fourth generating module 904 and a fifth generating module 905 .

[0122] The first generating module 901 is configured to obtain a plurality of multimodal data, pre-process the plurality of multimodal data, and generate a plurality of pre-processed text data. In one embodiment, the first generating module 901 may be configured to execute step S201 described above.

[0123] The first generation module 901 includes: a sixth generation module, a seventh generation module, an eighth generation module and a ninth generation module.

[0124] The sixth generating module is configured to perform data cleaning on the plurality of multimodal data to generate a plurality of multimodal data after data cleaning. In one embodiment, the sixth generating module may be configured to execute step S301 described above, which will not be described in detail here.

[0125] The seventh generating module is configured to normalize the plurality of cleaned multimodal data to generate a plurality of standardized multimodal data. In one embodiment, the seventh generating module may be configured to execute step S302 described above, which will not be described in detail here.

[0126] The eighth generating module is configured to perform cross-modal alignment on the plurality of standardized multimodal data to generate a plurality of cross-modal aligned text data. In one embodiment, the eighth generating module may be configured to execute step S303 described above, which will not be described in detail here.

[0127] The ninth generating module is configured to perform a desensitization operation on the plurality of cross-modal aligned text data to generate a plurality of pre-processed text data. In one embodiment, the ninth generating module may be configured to execute step S304 described above, which will not be described in detail here.

[0128] The second generating module 902 is configured to perform vector conversion on the plurality of pre-processed text data to generate a plurality of feature vector data. In one embodiment, the second generating module 902 may be configured to execute step S202 described above.

[0129] The second generating module 902 includes: a tenth generating module, an eleventh generating module, a twelfth generating module and a thirteenth generating module.

[0130] The tenth generating module is configured to perform block encoding on the plurality of pre-processed text data to generate a plurality of block text data. In one embodiment, the tenth generating module may be configured to execute step S401 described above, which will not be described in detail here.

[0131] The eleventh generating module is configured to perform a vector pooling operation on the plurality of block text data to generate a plurality of pooled vector data. In one embodiment, the eleventh generating module may be configured to execute step S402 described above, which will not be described in detail herein.

[0132] The twelfth generating module is configured to perform a text metadata fusion operation on the plurality of pooled vector data to generate a plurality of fused vector data. In one embodiment, the twelfth generating module may be configured to execute step S403 described above, which will not be described in detail herein.

[0133] The thirteenth generating module is configured to perform a feature extraction operation on the plurality of fused vector data to generate a plurality of feature vector data. In one embodiment, the thirteenth generating module may be configured to execute step S404 described above, which will not be described in detail here.

[0134] The third generating module 903 is configured to perform density clustering on the plurality of feature vector data to generate a data sets, where each data set includes a first data label, where a is an integer greater than or equal to 2. In one embodiment, the third generating module 903 may be configured to execute step S203 described above.

[0135] The third generation module 903 includes: a fifteenth generation module and a sixteenth generation module.

[0136] The fifteenth generating module is configured to perform density clustering on the plurality of feature vector data to generate a first-level structure tree. In one embodiment, the fifteenth generating module may be configured to execute step S501 described above, which will not be described in detail here.

[0137] The sixteenth generating module is configured to obtain a data set of each cluster in the first hierarchical structure tree and generate a data set. In one embodiment, the sixteenth generating module can be configured to execute step S502 described above, which will not be described in detail here.

[0138] The fourth generation module 904 is configured to perform semantic clustering on the a first data labels through semantic analysis to generate b second data labels, where b is an integer greater than 0 and less than a. In one embodiment, the fourth generation module 904 can be configured to execute step S204 described above.

[0139] The fourth generating module 904 includes: a seventeenth generating module and an eighteenth generating module.

[0140] The seventeenth generation module is configured to extract feature semantics from the a first data tags and generate a first data tag key semantics. In one embodiment, the seventeenth generation module may be configured to execute step S601 described above, which will not be described in detail here.

[0141] An eighteenth generation module is configured to calculate semantic similarity between each key semantic element in the a first data tag key semantic elements, merge first data tags whose semantic similarity among the a first data tag key semantic elements exceeds a preset threshold, and generate b second data tags. In one embodiment, the eighteenth generation module can be configured to execute step S602 described above and will not be further described here.

[0142] The fifth generation module 905 is used to preset a business knowledge graph, perform knowledge fusion on the business knowledge graph and the b second data tags, and generate b target data tags for data labeling. In one embodiment, the fifth generation module 905 can be used to execute step S205 described above.

[0143] The fifth generation module 905 includes: an eighteenth generation module, a nineteenth generation module, a twentieth generation module and a twenty-first generation module.

[0144] The eighteenth generation module is configured to perform bottom-up semantic clustering on the first hierarchical structure tree based on the b second data tags to generate a second hierarchical structure tree. In one embodiment, the eighteenth generation module may be configured to execute step S701 described above, which will not be described in detail here.

[0145] The nineteenth generation module is used to construct a business knowledge graph hierarchical structure based on the business knowledge graph. In one embodiment, the nineteenth generation module can be used to execute step S702 described above, which will not be repeated here.

[0146] The twentieth generation module is used to perform predicate constraint pruning on the business knowledge graph hierarchical structure from top to bottom to generate a target business hierarchical structure having the same level as the second hierarchical structure tree. In one embodiment, the twentieth generation module can be used to execute step S703 described above, which will not be repeated here.

[0147] The twenty-first generation module is configured to fuse the b second data tags in the second hierarchical structure tree with the business knowledge of the target business hierarchical structure to generate b target data tags having a business knowledge graph hierarchical structure. In one embodiment, the twenty-first generation module can be configured to execute step S704 described above and will not be further described here.

[0148] According to embodiments of the present disclosure, any multiple modules among the first generation module 901, the second generation module 902, the third generation module 903, the fourth generation module 904, and the fifth generation module 905 can be combined into a single module, or any one of these modules can be split into multiple modules. Alternatively, at least part of the functionality of one or more of these modules can be combined with at least part of the functionality of other modules and implemented in a single module. According to embodiments of the present disclosure, at least one of the first generation module 901, the second generation module 902, the third generation module 903, the fourth generation module 904, and the fifth generation module 905 can be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application-specific integrated circuit (ASIC), or can be implemented in hardware or firmware through any other reasonable means of circuit integration or packaging, or can be implemented in any one of the three implementation methods of software, hardware, and firmware, or any appropriate combination of any of these. Alternatively, at least one of the first generation module 901, the second generation module 902, the third generation module 903, the fourth generation module 904 and the fifth generation module 905 can be at least partially implemented as a computer program module, which can perform corresponding functions when executed.

[0149] Figure 10 The block diagram schematically shows an electronic device suitable for implementing the data marking method according to an embodiment of the present disclosure.

[0150] like Figure 10As shown, the electronic device 1000 according to an embodiment of the present disclosure includes a processor 1001, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 1002 or a program loaded from a storage portion 1008 into a random access memory (RAM) 1003. The processor 1001 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 1001 may also include onboard memory for caching purposes. The processor 1001 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0151] Various programs and data required for the operation of the electronic device 1000 are stored in the RAM 1003. The processor 1001, the ROM 1002, and the RAM 1003 are connected to each other via a bus 1004. The processor 1001 performs various operations of the method flow according to the embodiment of the present disclosure by executing the programs in the ROM 1002 and / or the RAM 1003. It should be noted that the programs may also be stored in one or more memories other than the ROM 1002 and the RAM 1003. The processor 1001 may also perform various operations of the method flow according to the embodiment of the present disclosure by executing the programs stored in the one or more memories.

[0152] According to an embodiment of the present disclosure, electronic device 1000 may further include an input / output (I / O) interface 1005, which is also connected to bus 1004. Electronic device 1000 may also include one or more of the following components connected to I / O interface 1005: an input section 1006 including a keyboard, mouse, etc.; an output section 1007 including devices such as a cathode ray tube (CRT), liquid crystal display (LCD), and speakers; a storage section 1008 including a hard disk; and a communication section 1009 including a network interface card such as a LAN card or modem. Communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to I / O interface 1005 as needed. Removable media 1011, such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed in drive 1010 as needed, so that computer programs read from the removable media can be installed into storage section 1008 as needed.

[0153] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments, or may exist independently and not be incorporated into the device / apparatus / system. The computer-readable storage medium carries one or more programs, and when executed, implements the method according to the embodiments of the present disclosure.

[0154] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, and may include, for example, but not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include ROM 1002 and / or RAM 1003 described above and / or one or more memories other than ROM 1002 and RAM 1003.

[0155] The embodiments of the present disclosure also include a computer program product, which includes a computer program containing program code for executing the method shown in the flowchart. When the computer program product is executed in a computer system, the program code is used to enable the computer system to implement the data marking method provided by the embodiments of the present disclosure.

[0156] The computer program executes the above functions defined in the system / device of the embodiment of the present disclosure when the processor 1001 executes the computer program. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0157] In one embodiment, the computer program may be stored on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may be transmitted and distributed in the form of a signal on a network medium, downloaded and installed via the communication portion 1009, and / or installed from the removable medium 1011. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0158] In such an embodiment, the computer program can be downloaded and installed from a network via the communication section 1009, and / or installed from the removable medium 1011. When the computer program is executed by the processor 1001, the above-described functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the systems, devices, means, modules, units, etc. described above can be implemented by computer program modules.

[0159] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computer programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, languages ​​such as Java, C++, Python, "C" or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect via the Internet).

[0160] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0161] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways, even if such combinations and / or couplings are not explicitly described in this disclosure. In particular, the features described in the various embodiments and / or claims of this disclosure may be combined and / or coupled in various ways without departing from the spirit and teachings of this disclosure. All such combinations and / or couplings are intended to fall within the scope of this disclosure.

[0162] The embodiments of the present disclosure are described above. However, these embodiments are for illustrative purposes only and are not intended to limit the scope of the present disclosure. Although each embodiment has been described separately above, this does not mean that the measures in each embodiment cannot be used in combination to advantage. The scope of the present disclosure is defined by the appended claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make various substitutions and modifications, which should all fall within the scope of the present disclosure.

Claims

1. A data marking method, characterized in that: The method includes: Acquiring a plurality of multimodal data, preprocessing the plurality of multimodal data, and generating a plurality of preprocessed text data; Performing vector conversion on the plurality of preprocessed text data to generate a plurality of feature vector data; Performing density clustering on the plurality of feature vector data to generate a data sets, wherein each data set includes a first data label, wherein a is an integer and a is greater than or equal to 2; Performing semantic clustering on the a first data labels through semantic analysis to generate b second data labels, where b is an integer, and b is greater than 0 and b is less than a; and A business knowledge graph is preset, and knowledge fusion is performed on the business knowledge graph and the b second data tags to generate b target data tags for data labeling.

2. The method according to claim 1, characterized in that Preprocessing the plurality of multimodal data to generate a plurality of preprocessed text data includes: performing data cleaning on the plurality of multimodal data to generate a plurality of cleaned multimodal data; Normalizing the plurality of cleaned multimodal data to generate a plurality of standardized multimodal data; Performing cross-modal alignment on the plurality of standardized multimodal data to generate a plurality of cross-modal aligned text data; and A desensitization operation is performed on the multiple cross-modal aligned text data to generate multiple preprocessed text data.

3. The method according to claim 1, characterized in that Performing vector conversion on the plurality of pre-processed text data to generate a plurality of feature vector data, including: Performing block encoding operations on the plurality of preprocessed text data to generate a plurality of block text data; Performing a vector pooling operation on the plurality of block text data to generate a plurality of pooled vector data; Performing a text metadata fusion operation on the plurality of pooled vector data to generate a plurality of fused vector data; and A feature extraction operation is performed on the multiple fused vector data to generate multiple feature vector data.

4. The method according to claim 1, wherein Density clustering is performed on the plurality of feature vector data to generate a data set, including: Performing density clustering on the plurality of feature vector data to generate a first-level structure tree; and The data set of each cluster in the first-level structure tree is obtained to generate a data set.

5. The method according to claim 4, characterized in that Perform semantic clustering on a first data label through semantic analysis to generate b second data labels, including: Performing feature semantic extraction on the a first data tags to generate a first data tag key semantics; and The semantic similarity between each key semantic in the a first data tag key semantics is calculated, and the first data tags whose semantic similarity in the a first data tag key semantics is higher than a preset threshold are merged to generate b second data tags.

6. The method according to claim 5, characterized in that Performing knowledge fusion based on the business knowledge graph and the b second data tags to generate b target data tags includes: Performing bottom-up semantic clustering on the first hierarchical structure tree based on the b second data labels to generate a second hierarchical structure tree; Based on the business knowledge graph, build a hierarchical structure of the business knowledge graph; Performing predicate constraint pruning on the business knowledge graph hierarchical structure from top to bottom to generate a target business hierarchical structure having the same level as the second hierarchical structure tree; and The b second data tags in the second hierarchical structure tree are integrated with the business knowledge of the target business hierarchical structure to generate b target data tags with a business knowledge graph hierarchical structure.

7. The method according to any one of claims 1 to 6, characterized in that The method further includes: The b target data labels are mined by an association rule algorithm to generate c mined data labels for data labeling of the multiple multimodal data, where c is an integer and c is greater than 1.

8. A data marking device, characterized in that: The device includes: A first generating module is configured to obtain a plurality of multimodal data, preprocess the plurality of multimodal data, and generate a plurality of preprocessed text data; A second generating module is used to perform vector conversion on the plurality of pre-processed text data to generate a plurality of feature vector data; a third generating module, configured to perform density clustering on the plurality of feature vector data to generate a data sets, wherein each data set includes a first data label, wherein a is an integer and a is greater than or equal to 2; a fourth generating module, configured to perform semantic clustering on the a first data labels through semantic analysis to generate b second data labels, wherein b is an integer, and b is greater than 0 and b is less than a; and The fifth generation module is used to preset a business knowledge graph, perform knowledge fusion on the business knowledge graph and the b second data tags, and generate b target data tags for data labeling.

9. An electronic device comprising: one or more processors; a memory for storing one or more computer programs, The method further comprises the step of executing the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program or instruction is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images