Material file tag processing method and apparatus

Abstract

The embodiment of the application provides a kind of material file label processing method and device, method includes: receiving that user sends through front end and marks request, by preset multi-modal semantic recognition model, mark request is carried out comprehensive semantic recognition, and according to the result of comprehensive semantic recognition, determine corresponding mark template and return user place;Receive that user sends based on mark template and multiple field selection values and event query request, according to preset adaptive field filtering model, field selection value is filtered in label field range, and according to the result of label field range filtering, determine corresponding knowledge graph event rule model, obtain the label mutual constraint condition that knowledge graph event rule model output;According to label mutual constraint condition and target material file content feature, field selection value is carried out cross-modal linkage constraint, determines the corresponding material file label;The application can more efficiently, intelligent and automatically process material file label.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, specifically to a method and apparatus for processing material file tags. Background Technology

[0002] In today's era of rapid information and digital development, the management and processing of content files have become increasingly important. Various enterprises and organizations generate and use a large number of content files in their daily operations, including text, images, videos, and other formats. To improve the utilization rate and retrieval efficiency of content files, accurate and effective tagging methods are crucial. While existing technologies have made some progress in content file tagging, there are still many shortcomings in terms of tag accuracy, intelligence, and automation.

[0003] First, existing methods for processing source file tags typically rely on manual labeling. While intuitive, this method has significant drawbacks. Manual labeling is not only time-consuming and labor-intensive but also susceptible to human error, making it difficult to guarantee the accuracy and consistency of the tags. Especially when dealing with a large number of source files, manual labeling is extremely inefficient and cannot meet the demands for rapid processing and efficient management. Furthermore, different labelers may have different interpretations and annotations of the same source file, further impacting the consistency and reliability of the tags.

[0004] Secondly, existing labeling methods are insufficient in terms of intelligence and automation. While some methods incorporate machine learning and natural language processing techniques, most are limited to single-modal data processing. This means they can only handle single-form input data, such as plain text or images, and cannot effectively process complex source files containing multiple forms of information. In practical applications, many source files are multimodal, containing information in various forms such as text, images, audio, and video. Existing methods struggle to comprehensively process this multimodal data, thus limiting the accuracy and comprehensiveness of labeling.

[0005] Existing technologies also fall short in terms of flexibility and adaptability in tag processing. Different application scenarios and user needs may have different tag requirements, but existing methods typically use fixed tag templates and lack the ability to dynamically adjust according to specific needs. This makes it difficult for existing methods to provide suitable tag processing solutions when facing ever-changing application requirements. In addition, existing methods lack intelligence in the filtering and selection of tag fields, often requiring users to manually select and adjust, increasing the complexity and workload of use.

[0006] Existing technologies also have shortcomings in handling tag association and constraint processing. Tags in source files are often not independent; they are associated and constrained with each other. Existing methods typically rely on pre-defined rules and logic when processing these relationships, lacking intelligent processing capabilities based on data and knowledge graphs. This not only limits the flexibility and accuracy of tag processing but also makes it difficult to adapt to complex application scenarios and dynamically changing needs.

[0007] Furthermore, existing labeling methods have significant shortcomings in handling cross-modal data linkage constraints. Multimodal data often exhibit complex correlations and interactions, and effectively processing these correlations and constraints during labeling is a problem that existing methods struggle to solve. Traditional methods typically can only handle simple correlations between single-modal data, making it difficult to achieve linkage constraints and comprehensive processing across modal data. This not only affects the accuracy and comprehensiveness of labeling but also limits the application scope and effectiveness of labeling methods.

[0008] In summary, existing methods for processing material file tags still have many shortcomings and challenges in terms of manual tagging efficiency, intelligence and automation, multimodal data processing, tag flexibility and adaptability, tag association and constraint processing, and cross-modal linkage constraints. Summary of the Invention

[0009] To address the problems in the prior art, this application provides a method and apparatus for processing material file tags, which can process material file tags more efficiently, intelligently and automatically.

[0010] To solve at least one of the above problems, this application provides the following technical solution:

[0011] In a first aspect, this application provides a method for processing material file tags, comprising: receiving a tagging request sent by a user through a front end, performing comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determining a corresponding tagging template based on the result of the comprehensive semantic recognition, and returning it to the user, wherein the multimodal semantic recognition model is trained based on historical multimodal data;

[0012] The system receives multiple field selection values ​​and event query requests sent by the user based on the tagging template. It then filters the field selection values ​​for tag field ranges according to a preset adaptive field filtering model and determines the corresponding knowledge graph event rule model based on the result of the tag field range filtering. The system obtains the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0013] Based on the mutual constraints of the tags and the content characteristics of the target material file, cross-modal linkage constraints are applied to the field selection values ​​to determine the corresponding material file tags.

[0014] Further, before receiving the tagging request sent by the user through the front end and performing comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, the process includes:

[0015] Obtain multimodal data from a designated database that contains at least one of text, images, and audio / video, and perform noise cleaning and formatting normalization on the multimodal data;

[0016] The multimodal data, after noise cleaning and formatting normalization, is used as the model training set to train a multimodal network structure for a given convolutional neural network, thereby obtaining a multimodal semantic recognition model.

[0017] Furthermore, the step of receiving the tagging request sent by the user through the front end, and performing comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, includes:

[0018] Receive the tagging request sent by the user through the front end and perform syntax and format validation on the tagging request;

[0019] After the syntax and format verification is passed, the text information and / or associated image in the tagging request are input into a preset multimodal semantic recognition model for comprehensive semantic recognition to obtain the corresponding semantic recognition result.

[0020] Further, before receiving the multiple field selection values ​​and event query requests sent by the user based on the tagging template, and filtering the field selection values ​​for tag field ranges according to a preset adaptive field filtering model, the process includes:

[0021] Obtain historical tag field data from the designated database and perform data cleaning and formatting on the historical tag field data;

[0022] The historical label field data, after the data cleaning and formatting process, is used as the model training set and input into the random forest model to adaptively learn the field filtering rules, thus obtaining an adaptive field filtering model.

[0023] Further, the step of receiving multiple field selection values ​​and event query requests sent by the user based on the tagging template, and filtering the field selection values ​​for tag field ranges according to a preset adaptive field filtering model, includes:

[0024] Receive multiple structured field selection values ​​sent by the user based on the labeling template, and perform validity verification on the multiple structured field selection values;

[0025] After the legality verification is passed, the selected values ​​of the multiple structured fields are input into a preset adaptive field filtering model for range filtering to obtain a set of optional values ​​that conform to the business rules.

[0026] Further, before determining the corresponding knowledge graph event rule model based on the result of filtering according to the label field range, and obtaining the label mutual constraint conditions output by the knowledge graph event rule model, the process includes:

[0027] Semantic association knowledge between tag fields is extracted from historical tag data, and a knowledge graph is constructed in the form of a graph structure using the extracted semantic association knowledge between tag fields.

[0028] The knowledge graph is used as a model training set to train a preset graph attention network model for event rule reasoning, thereby obtaining a knowledge graph event rule model.

[0029] Further, the step of determining the corresponding knowledge graph event rule model based on the result of filtering the tag field range, and obtaining the tag mutual constraint conditions output by the knowledge graph event rule model, includes:

[0030] Based on the association rules between tag fields contained in the result of the tag field range filtering, determine the knowledge graph event rule model that best matches the current tag field range from the preset model library;

[0031] The user-selected tag field value is input into the knowledge graph event rule model for knowledge reasoning to obtain the corresponding tag mutual constraint conditions.

[0032] Secondly, this application provides a material file tag processing device, comprising:

[0033] The semantic analysis module is used to receive the tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0034] The tag constraint analysis module is used to receive multiple field selection values ​​and event query requests sent by the user based on the tagging template, filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model, and determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering, thereby obtaining the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0035] The material file tagging module is used to perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file, and to determine the corresponding material file tags.

[0036] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the material file tag processing method described above.

[0037] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the material file tag processing method described above.

[0038] Fifthly, this application provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the material file tag processing method described above.

[0039] As can be seen from the above technical solution, this application provides a method and apparatus for processing material file tags. It receives tagging requests sent by users through a front-end, performs comprehensive semantic recognition on the tagging requests using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the comprehensive semantic recognition result, and returns it to the user. It also receives multiple field selection values ​​and event query requests sent by users based on the tagging template, filters the field selection values ​​for tag field ranges using a preset adaptive field filtering model, determines the corresponding knowledge graph event rule model based on the tag field range filtering result, and obtains the tag mutual constraint conditions output by the knowledge graph event rule model. Finally, it performs cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content characteristics of the target material file to determine the corresponding material file tags. This allows for more efficient, intelligent, and automated processing of material file tags. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is one of the flowcharts illustrating the material file tag processing method in the embodiments of this application;

[0042] Figure 2 This is a second flowchart illustrating the material file tag processing method in the embodiments of this application;

[0043] Figure 3 This is the third flowchart illustrating the material file tag processing method in the embodiments of this application;

[0044] Figure 4 This is the fourth flowchart illustrating the material file tag processing method in the embodiments of this application;

[0045] Figure 5 This is the fifth flowchart illustrating the material file tag processing method in the embodiments of this application;

[0046] Figure 6 This is the sixth flowchart illustrating the material file tag processing method in the embodiments of this application;

[0047] Figure 7 This is the seventh flowchart illustrating the material file tag processing method in the embodiments of this application;

[0048] Figure 8 This is a structural diagram of the material file tag processing device in the embodiments of this application;

[0049] Figure 9 This is a schematic diagram of the structure of the electronic device in the embodiments of this application.

[0050] Figure label:

[0051] Electronic device 9600, central processing unit 9100, memory 9140, communication module 9110, input unit 9120, audio processor 9130, display 9160, power supply 9170, buffer memory 9141, application / function storage unit 9142, data storage unit 9143, driver storage unit 9144, antenna 9111, speaker 9131, microphone 9132. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0053] The acquisition, storage, use, and processing of data in this application all comply with the relevant provisions of national laws and regulations.

[0054] In view of the problems existing in the prior art, this application provides a method and apparatus for processing material file tags. It receives tagging requests sent by users through a front-end, performs comprehensive semantic recognition on the tagging requests using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the comprehensive semantic recognition result, and returns it to the user; it receives multiple field selection values ​​and event query requests sent by users based on the tagging template, filters the field selection values ​​for tag field ranges using a preset adaptive field filtering model, determines the corresponding knowledge graph event rule model based on the tag field range filtering result, and obtains the tag mutual constraint conditions output by the knowledge graph event rule model; it performs cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content characteristics of the target material file, and determines the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0055] To enable more efficient, intelligent, and automated processing of material file tags, this application provides an embodiment of a material file tag processing method, see [link to embodiment]. Figure 1 The method for processing material file tags specifically includes the following:

[0056] Step S101: Receive a tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0057] Optionally, in this embodiment, in step S101, the system first receives a marking request sent by the user through the front end. The user may input the marking request through various devices (such as a computer, mobile phone, or tablet), and these requests typically contain text descriptions, images, or other multimodal data. The system needs to process this multimodal data in order to accurately understand the user's intent and provide a corresponding marking template.

[0058] To achieve this goal, the system uses a pre-defined multimodal semantic recognition model to perform comprehensive semantic recognition on tagging requests. This model is built using deep learning algorithms and trained on a large amount of historical multimodal data. This multimodal data includes text, images, videos, and audio, derived from various user interactions and historical tagging records. By jointly training on this data, the model can capture the correlations and semantic features between different modalities, thus enabling it to comprehensively analyze the multimodal input provided by the user when processing new tagging requests.

[0059] In its implementation, the system first preprocesses the received tagging requests. For example, if the request contains an image, the system uses image processing techniques to perform preprocessing operations such as denoising and normalization; if the request contains text, the system performs natural language processing operations such as word segmentation and part-of-speech tagging. The preprocessed multimodal data is then input into the multimodal semantic recognition model.

[0060] After receiving input data, multimodal semantic recognition models perform feature extraction and semantic understanding through their multi-layered neural network structure. The underlying networks typically include convolutional neural networks (CNNs) for image data and recurrent neural networks (RNNs) or transformers for text data. Through layer-by-layer processing of these networks, the model can extract high-level semantic features from images and text, and perform fusion and comprehensive analysis at a higher level.

[0061] By comprehensively recognizing semantics, the model can determine the user's main intent and needs in their tagging request. For example, if a user uploads a product image along with a descriptive text, the model can identify information such as the product category and brand in the image, and combine this with keywords and semantic relationships in the text description to determine the type of tagging template the user might need, such as a product tag template or a price tag template.

[0062] After determining the results of the comprehensive semantic recognition, the system selects the most suitable marking template from a pre-set marking template library. These templates are pre-designed and stored according to different marking needs, covering various possible marking scenarios and formats. The system returns the selected marking template to the user, who can view and confirm whether the template meets their requirements.

[0063] This tagging request processing method based on multimodal semantic recognition can effectively solve the problems of inaccurate understanding and unreasonable template selection in traditional single-modal tagging methods. By introducing multimodal data, the system can obtain more comprehensive and accurate semantic information, improving the accuracy and efficiency of tagging request processing.

[0064] In terms of technical effectiveness, this solution significantly improves the intelligence level of labeling request processing through the application of a multimodal semantic recognition model. First, the introduction of multimodal data allows the system to comprehensively consider various information from user input, resulting in more accurate and comprehensive semantic understanding. Second, the multimodal model trained on historical data can continuously optimize and improve its recognition capabilities; as the amount of data increases, the system's recognition performance also continuously improves. Finally, the automated template selection and return mechanism reduces user operation steps, improving user experience and labeling efficiency.

[0065] For example, suppose a user is an operations staff member at an e-commerce platform and needs to tag newly listed products. The user uploads several product images and enters a descriptive text. After receiving the tagging request, the system analyzes information such as product type and brand logo in the images using a multimodal semantic recognition model, and combines this with keywords in the text description such as "new product" and "promotion" to accurately identify the user's needs. The system then selects a suitable tag template for promoting the new product from its tagging template library and returns it to the user. The user can directly use this template for tagging, saving the time and effort of manually selecting a template.

[0066] In summary, step S101, by receiving user marking requests, using a multimodal semantic recognition model for comprehensive semantic recognition, and determining and returning marking templates based on the recognition results, solves the problems of low efficiency and poor accuracy in traditional marking methods, and significantly improves the intelligence of the marking process and user experience.

[0067] Step S102: Receive multiple field selection values ​​and event query requests sent by the user based on the tagging template; filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model; determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering; and obtain the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0068] Optionally, in this embodiment, in step S102, the system first receives multiple field selection values ​​and event query requests sent by the user based on the marking template. These field selection values ​​may include specific data items selected by the user in the marking template, such as product name, price, category, etc., while the event query requests may involve specific operations or retrievals that the user wants the system to perform, such as finding relevant product information or generating a specific type of report.

[0069] Upon receiving user input, the system first filters the selected field value using a pre-defined adaptive field filtering model to determine the range of tag fields. This model, trained on historical tag field data, dynamically adjusts and optimizes the range of tag fields based on the user's input. For example, if the user selects "electronic products" as the category tag, the model will automatically filter out other tag fields related to electronic products, such as brand, model, and technical parameters. This filtering mechanism effectively narrows the search space for tag fields, improving the efficiency and accuracy of subsequent processing.

[0070] In its implementation, the adaptive field filtering model first parses the user's input field selection values ​​and converts them into corresponding feature vectors. These feature vectors are then fed into a pre-trained neural network model, which extracts and classifies features through a multi-layered neural network structure, ultimately outputting an optimized range of labeled fields. This process considers not only the semantic information of the field selection values ​​themselves but also incorporates association patterns and co-occurrence relationships from historical data, thus achieving more intelligent and accurate label field filtering.

[0071] Next, the system determines the corresponding knowledge graph event rule model based on the results of the tag field range filtering. The knowledge graph event rule model is trained based on the semantic association knowledge between tag fields, enabling modeling and reasoning about the relationships between different tag fields. For example, for the tag field of the "electronic products" category, the knowledge graph event rule model would include semantic association rules between tag fields such as brand, model, and specifications, such as the correspondence between brand and model, and the dependency relationship of specifications.

[0072] The system inputs the filtered tag field range into the knowledge graph event rule model. Through the model's reasoning mechanism, the mutual constraints between the tag fields are determined. Specifically, the knowledge graph event rule model uses its built-in rule base and reasoning algorithm to perform association analysis on the input tag fields, identifying the logical relationships and constraints between them. For example, the system can determine that a certain brand of electronic products can only have specific models and specifications; these constraints will be returned to the user as output.

[0073] This processing method, based on adaptive field filtering and knowledge graph event rule reasoning, effectively solves the problems of inaccurate label field selection and incomplete identification of relationships in traditional methods. By introducing an adaptive field filtering model, the system can dynamically adjust the range of label fields, reduce interference from irrelevant fields, and improve the accuracy of label selection. Through the knowledge graph event rule model, the system can comprehensively identify and reason about the relationships and constraints between label fields, thereby providing more intelligent and efficient event processing capabilities.

[0074] In terms of technical effectiveness, this solution significantly improves the system's intelligence and processing efficiency through the combined application of an adaptive field filtering model and a knowledge graph event rule model. First, the adaptive field filtering model dynamically adjusts the range of tag fields based on user input, reducing interference from irrelevant fields and improving the accuracy and efficiency of subsequent processing. Second, the knowledge graph event rule model comprehensively identifies and infers the relationships and constraints between tag fields, thus providing more intelligent and efficient event processing capabilities. Finally, through training based on historical data and semantic association knowledge, these two models can be continuously optimized and improved; as the amount of data increases, the system's processing performance will continue to improve.

[0075] For example, suppose a user is an operations staff member at an electronics retailer who needs to select tags and query events for newly listed products. The user selects the "electronics" category in the tagging template and enters values ​​for fields such as brand and model, while simultaneously sending a query request asking the system to find promotional activities related to the product. Upon receiving the user's input, the system first uses an adaptive field filtering model to filter out tag fields related to electronics, such as brand, model, and specifications. Then, based on these field selection values, it determines the corresponding knowledge graph event rule model. This model, through internal semantic association rules and inference algorithms, identifies the mutual constraints between brand, model, and specifications, and returns a comprehensive report containing promotional activity information to the user. Based on the constraints and report content provided by the system, the user can perform more accurate and efficient tag selection and event processing, improving work efficiency and decision-making accuracy.

[0076] In summary, step S102, by receiving the user's input field selection values ​​and event query requests, uses an adaptive field filtering model to filter the range of tag fields and determines the mutual constraints of tags through a knowledge graph event rule model. This solves the problems of inaccurate tag selection and incomplete identification of associations in traditional methods, and significantly improves the intelligence level and processing efficiency of the system.

[0077] Step S103: Perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file to determine the corresponding material file tags.

[0078] Optionally, in this embodiment, in step S103, the system performs cross-modal linkage constraints on the user-provided field selection values ​​based on the tag mutual constraints obtained in the previous step and the content characteristics of the target material file, and finally determines the corresponding material file tags. The key to this step is to use cross-modal analysis of semantic association and content features to ensure that the generated tags accurately reflect the essential characteristics of the material file and user needs.

[0079] First, the system obtains the content features of the target material files. These material files may include various formats such as text, images, videos, or audio. For text files, the system applies natural language processing (NLP) techniques, such as word segmentation, part-of-speech tagging, entity recognition, etc., to extract the key content and themes in the file. For image and video files, the system uses computer vision techniques, such as convolutional neural networks (CNNs), to extract the objects, scenes, and other visual features in the images. Audio files are processed through speech recognition and audio feature extraction techniques to obtain the speech content and audio features therein.

[0080] While extracting the content features, the system also uses the mutual constraint conditions obtained in the previous step. These conditions are derived from the inference of the knowledge graph event rule model and contain the logical relationships and constraint conditions between the various tag fields. For example, for a set of tag fields, the system may identify that certain brands can only correspond to specific product models, or that there are dependencies between certain technical parameters.

[0081] Next, the system performs cross-modal linkage analysis on the field selection values and the content features of the target material files. In this process, the system converts the field selection values and the content features into unified feature representations, which can be represented using vector space models (such as word vectors, image feature vectors, etc.). By calculating the similarity and matching degree of these feature vectors, the system can identify the association between the field selection values and the content features.

[0082] In the cross-modal linkage analysis, the system comprehensively considers the semantic information of the field selection values, the actual performance of the content features, and the constraints of the mutual constraint conditions of the tags. For example, if the user selects "smartphone" as the product category and provides field selection values such as brand and model, and the target material file is a product introduction video of a smartphone, the system will determine whether the brand and model of the smartphone in the video match the user's selection values through object recognition and text parsing in the video content. At the same time, the system also checks the mutual constraint conditions of the tags to ensure that the relationship between the brand and the model conforms to the preset rules.

[0083] Through this cross-modal linkage constraint analysis, the system can determine the tags that best match the content features of the material files and the user's needs. Finally, the system generates a list of tags and annotates them on the target material files. These tags not only reflect the actual content of the material files but also conform to the user's field selection values and mutual constraint conditions, thus ensuring the accuracy and practicality of the tags.

[0084] This cross-modal linkage constraint processing method effectively solves the problems of insufficient accuracy and poor correlation in traditional single-modal tag generation methods. By introducing multimodal data analysis, the system can comprehensively consider multiple content features such as text, images, and audio, providing more comprehensive and accurate tag generation. In addition, through the application of mutual constraint conditions for tags, the system can ensure the logical relationship and semantic consistency between generated tags, improving the usability of tags and user satisfaction.

[0085] In terms of technical effectiveness, this solution significantly improves the accuracy and intelligence of media file tag generation through cross-modal linkage constraint analysis. First, multimodal data analysis enables the system to comprehensively capture the content features of media files, thus providing more accurate and comprehensive tags. Second, the application of mutual constraints on tags ensures the logical relationships and semantic consistency between tags, improving tag usability and user satisfaction. Finally, through training based on historical data and knowledge graphs, the system can continuously optimize and improve the tag generation algorithm; as the amount of data increases, the tag generation effect will continue to improve.

[0086] For example, suppose a user is a creative staff member at an advertising agency who needs to generate tags for a newly produced smartphone advertisement video. The user selects the "smartphone" category in the tagging template and enters values ​​for fields such as brand and model. After receiving the user's input, the system uses cross-modal constraint analysis to first extract visual and textual features from the advertisement video, identifying the smartphone brand and model in the video. Next, the system matches these features with the user's field selections, checking whether the logical relationship between the brand and model conforms to preset mutual constraints. Finally, the system generates a list of tags containing brand, model, technical parameters, etc., and labels it on the advertisement video. The user can directly use these tags for advertising promotion, improving work efficiency and tag accuracy.

[0087] In summary, step S103 determines the corresponding material file tags by applying cross-modal linkage constraints to the field selection values ​​based on the mutual constraints of tags and the content characteristics of the target material file. This solves the problems of insufficient accuracy and poor correlation in traditional tag generation methods, and significantly improves the accuracy and intelligence level of material file tag generation.

[0088] As described above, the material file tag processing method provided in this application can receive tagging requests sent by users through the front end, perform comprehensive semantic recognition on the tagging requests using a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user; receive multiple field selection values ​​and event query requests sent by users based on the tagging template, perform tag field range filtering on the field selection values ​​using a preset adaptive field filtering model, determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtain the tag mutual constraint conditions output by the knowledge graph event rule model; perform cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content features of the target material file, and determine the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0089] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 2 It can also specifically include the following:

[0090] Step S201: Obtain multimodal data containing at least one of text, images, and audio / video from a designated database, and perform noise cleaning and format normalization processing on the multimodal data;

[0091] Step S202: Use the multimodal data after noise cleaning and formatting normalization as the model training set to train the multimodal network structure of the set convolutional neural network to obtain a multimodal semantic recognition model.

[0092] Optionally, in this embodiment, in step S201, the system first obtains multimodal data containing at least one of text, images, and audio / video from a designated database. This data can originate from various scenarios, such as social media, internal enterprise document libraries, and user-uploaded multimedia files. Since this data may have issues such as inconsistent formats and noise interference, the system needs to perform noise cleaning and format normalization processing to ensure the accuracy and effectiveness of subsequent processing.

[0093] Noise cleaning refers to removing irrelevant or harmful information from data. For text data, this might include removing stop words, special characters, and duplicate content; for image data, it might involve removing blur, noise, and low-resolution images; and for audio and video data, it might involve removing background noise, silent segments, and irrelevant audio content. The purpose of noise cleaning is to ensure the purity and quality of the data to facilitate subsequent feature extraction and model training.

[0094] Formatting and normalization refers to converting data of different formats into a unified representation. For text data, this might involve converting all text to lowercase, standardizing encoding formats, and standardizing punctuation; for image data, it might involve adjusting image size and color channel formats; and for audio and video data, it might involve standardizing audio sampling rates and video frame rates. Formatting and normalization ensures that data of different modalities have a consistent structure and representation, thus facilitating subsequent multimodal processing and analysis.

[0095] After noise cleaning and formatting normalization, the system uses this processed multimodal data as a model training set and proceeds to step S202. In this step, the system uses this training set data to train the set convolutional neural network (CNN) multimodal network structure, thereby obtaining a multimodal semantic recognition model.

[0096] In practical implementation, the first step is to construct a convolutional neural network (CNN) structure suitable for multimodal data processing. Traditional CNNs are typically used to process single-modal data, such as two-dimensional convolutional networks in image recognition. To handle multimodal data, the system needs to design a network structure capable of processing multiple modalities simultaneously, such as text, images, and audio. This can be achieved by introducing data processing branches for different modalities into the network. For example, text data can be processed through embedding layers and one-dimensional convolutional layers, image data through two-dimensional convolutional layers, and audio data through either one-dimensional or two-dimensional convolutional layers. After feature extraction layers for each modality, feature fusion layers can be used to fuse features from different modalities, thereby achieving comprehensive processing of multimodal data.

[0097] During model training, the system inputs multimodal data, after noise cleaning and formatting normalization, into a convolutional neural network. It calculates the feature representations of each modality through forward propagation and then fuses them in a feature fusion layer. Next, using backpropagation and gradient descent algorithms, the system continuously adjusts the network's weight parameters to minimize the loss function and improve the model's semantic recognition ability for multimodal data. During training, the system uses a certain proportion of validation set data for cross-validation to evaluate model performance and prevent overfitting.

[0098] The technical principle behind this multimodal network structure training lies in using convolutional neural networks to extract and fuse features from data of different modalities, thereby achieving comprehensive semantic recognition of multimodal data. Compared with traditional single-modal data processing methods, the multimodal network structure can leverage the complementarity of data from different modalities to improve the model's recognition accuracy and robustness.

[0099] This solution effectively addresses the problems of inconsistent multimodal data processing and severe noise interference inherent in traditional methods. Through noise cleaning and formatting normalization, the system obtains high-quality and consistent data input, thereby improving the effectiveness of subsequent model training. By designing and training a multimodal convolutional neural network structure, the system can comprehensively utilize data features from different modalities to achieve efficient semantic recognition of multimodal data.

[0100] In terms of technical effectiveness, this solution improves data quality and consistency through noise cleaning and formatting normalization of multimodal data, ensuring the effectiveness of model training. Through training with a multimodal convolutional neural network structure, the system achieves efficient semantic recognition of multimodal data, significantly improving the model's recognition accuracy and robustness. Furthermore, due to the introduction of feature fusion from multimodal data, the system exhibits stronger adaptability and recognition capabilities when handling complex scenarios and diverse data.

[0101] For example, suppose a user is a data analyst at a media company, responsible for processing and analyzing multimodal data from social media. The user retrieves a large amount of social media post data containing text, images, and videos from a pre-defined database. The system first cleans this data to remove irrelevant text, blurry images, and invalid audio clips. Next, the system formats and normalizes the data, converting text to a uniform encoding format, resizing images to a fixed size, and standardizing the sampling rate and frame rate of audio and video. This processed data serves as the model training set, inputting into a designed multimodal convolutional neural network for training, ultimately resulting in a multimodal semantic recognition model capable of comprehensively recognizing the semantics of text, images, and videos. Through this process, the user can efficiently analyze and understand various information within social media data, thereby making more accurate data-driven decisions.

[0102] In summary, steps S201 and S202, by acquiring multimodal data and performing noise cleaning and formatting normalization, and then training a multimodal convolutional neural network, obtain a multimodal semantic recognition model. This solves the problems of inconsistent multimodal data processing and noise interference, and improves the recognition accuracy and robustness of the model.

[0103] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 3 It can also specifically include the following:

[0104] Step S301: Receive the tagging request sent by the user through the front end and perform syntax and format verification on the tagging request;

[0105] Step S302: After the syntax and format verification is passed, the text information and / or associated image in the tagging request are input into the preset multimodal semantic recognition model for comprehensive semantic recognition to obtain the corresponding semantic recognition result.

[0106] Optionally, in this embodiment, in step S301, the system receives a marking request sent by the user through the front end and performs syntax and format validation on the request. The marking request submitted by the user through the front end interface typically contains text information and possible associated images, which need to undergo strict validation to ensure the integrity and standardization of the data.

[0107] First, after receiving a tagging request, the system parses the request content, extracting the text information and associated images. Syntax validation primarily targets the text information, including checking whether the text conforms to predetermined grammatical rules, such as sentence structure completeness, correct punctuation usage, and the presence of illegal characters. Format validation focuses on the correctness of the data format, such as whether the text encoding is correct and whether the image file is complete and conforms to supported formats (e.g., JPEG, PNG).

[0108] Syntax and format validation can be implemented using techniques such as predefined rules and regular expressions. For text syntax validation, the system can utilize Natural Language Processing (NLP) techniques, such as syntax parsers, to check the correctness of sentence grammatical structure. Format validation can be achieved by checking file header information and data integrity to ensure that image files are not corrupted and meet the specified format requirements.

[0109] After completing the syntax and format verification, if the request passes the verification, the system proceeds to step S302, whereby the text information and / or associated image in the tagging request are input into a preset multimodal semantic recognition model for comprehensive semantic recognition. The multimodal semantic recognition model is a deep learning model capable of processing and understanding multiple modalities of data (such as text and images), typically based on convolutional neural networks (CNNs) and other advanced neural network structures.

[0110] In step S302, the system first preprocesses the text information and associated images. For the text information, the system may need to perform word segmentation, part-of-speech tagging, and word vectorization to facilitate model input. For the images, the system may need to perform normalization, size adjustment, and other preprocessing steps. The preprocessed data is input into the multimodal semantic recognition model, which extracts features from the text and images and fuses the data from different modalities through a feature fusion layer to generate a comprehensive semantic representation.

[0111] The core of multimodal semantic recognition models lies in their ability to simultaneously understand and process text and image information. Through feature extraction and fusion, they achieve comprehensive semantic recognition of multimodal data. The training process typically involves supervised learning with a large amount of labeled data, continuously optimizing model parameters to enable accurate recognition and understanding of the semantics of multimodal data.

[0112] Through multimodal semantic recognition model processing, the system can generate semantic recognition results for tagging requests. These results can be semantic tags for text, category tags for images, or a combined tag combining both. The semantic recognition results will be fed back to the user, who can then further process and analyze the data based on these results.

[0113] The technical principle behind this series of steps lies in ensuring the integrity and standardization of the tagging request data through syntax and format verification, thereby improving the accuracy of subsequent semantic recognition. The multimodal semantic recognition model achieves a comprehensive understanding of text and image data through feature extraction and fusion, generating accurate semantic recognition results.

[0114] This solution effectively addresses potential issues such as non-standard data formats and syntax errors in user tagging requests, ensuring the quality of input data through rigorous validation steps. The multimodal semantic recognition model, by fusing multimodal data features, improves the accuracy and robustness of semantic recognition.

[0115] In terms of technical effectiveness, the system filters out unqualified labeling requests through syntax and format validation steps, ensuring the quality and standardization of input data. The application of a multimodal semantic recognition model enables the system to comprehensively process and understand text and image data, generating accurate semantic recognition results and improving the efficiency and accuracy of labeling.

[0116] For example, suppose a user is an administrator of an e-commerce platform and needs to generate tags for product images and descriptions. The user submits a tagging request containing product description text and product images through the front-end interface. After receiving the request, the system first performs grammatical validation on the text description to ensure there are no grammatical errors, and then performs format validation on the image files to ensure they are complete and meet the format requirements. After successful validation, the system inputs the text description and product images into a multimodal semantic recognition model for processing. The model extracts and fuses features from the text and images to generate comprehensive semantic tags for the products, such as "smartphone," "brand name," and "model." The system then provides these tags back to the user, who can directly use these tags for product management and promotion.

[0117] In summary, steps S301 and S302, by receiving and verifying the user's labeling request and using a multimodal semantic recognition model for comprehensive semantic recognition, overcome the limitations of non-standard data input and single-modal processing, thereby improving the accuracy and efficiency of labeling.

[0118] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 4 It can also specifically include the following:

[0119] Step S401: Obtain historical tag field data from the designated database and perform data cleaning and formatting on the historical tag field data;

[0120] Step S402: Input the historical label field data after the data cleaning and formatting process into the random forest model as the model training set to adaptively learn the field filtering rules and obtain the adaptive field filtering model.

[0121] Optionally, in this embodiment, in step S401, the system retrieves historical tag field data from a designated database and performs data cleaning and formatting on this data. Historical tag field data typically contains a large amount of tagging information about different objects, such as product category tags, user behavior tags, and content category tags. This data may have redundancy, inconsistencies, or missing information, therefore it needs to be cleaned and formatted to ensure data accuracy and consistency.

[0122] Data cleaning refers to removing or correcting errors and irrelevant information from data. Specific operations include deleting duplicate tag records, filling in missing data, and correcting erroneous tag fields. For example, if some tag fields contain spelling errors or inconsistent formatting, the system will automatically correct them. For missing data, the system may fill in or interpolate based on existing data to ensure data integrity.

[0123] Formatting refers to converting tag data from different sources and formats into a unified standard format. This includes standardizing data encoding (such as converting all text tags to UTF-8 encoding), standardizing date formats, and standardizing numerical ranges. Formatting ensures that tag fields from different data sources have a consistent representation, facilitating subsequent processing and analysis.

[0124] After data cleaning and formatting, in step S402, the system will use the processed historical label field data as the model training set and input it into the random forest model for adaptive learning of field filtering rules. Random forest is an ensemble learning method that improves the accuracy and stability of the model by constructing multiple decision trees and combining their results.

[0125] In this step, the system first needs to construct the basic structure of the random forest model. The random forest model consists of multiple decision trees, each constructed using different training samples and feature subsets. This method effectively reduces overfitting and improves the model's generalization ability.

[0126] During training, the system inputs cleaned and formatted historical label field data into the random forest model. The model learns from this data and gradually establishes field filtering rules. Specifically, the random forest model constructs multiple decision trees based on the characteristics of the historical label field data, with each decision tree classifying and filtering based on different field features. By combining the results of multiple decision trees, the random forest model can adaptively learn the optimal field filtering rules.

[0127] The underlying principle of this adaptive learning technique is that, by learning from a large amount of historical data, the random forest model can identify and extract the regularity and importance of feature fields in the data, thereby establishing accurate field filtering rules. Compared with traditional manual rule-making methods, adaptive learning methods can handle complex data features more efficiently, improving the accuracy and robustness of field filtering.

[0128] This solution effectively addresses the problems of traditional field filtering rules being inflexible and unable to adapt to complex data features. Through data cleaning and formatting, the system ensures the quality and consistency of input data, thereby improving the effectiveness of model training. Through adaptive learning of the random forest model, the system can automatically generate field filtering rules based on historical data, improving the accuracy and efficiency of filtering.

[0129] In terms of technical effectiveness, the system improves data quality and consistency through data cleaning and formatting, ensuring the effectiveness of model training. The adaptive learning method of the random forest model enables the system to automatically generate accurate field filtering rules based on historical data, improving the accuracy and robustness of field filtering. Furthermore, because the random forest model can handle high-dimensional and complex feature data, the system exhibits stronger adaptability and processing capabilities when dealing with large-scale and complex scenarios.

[0130] For example, suppose a user is a data scientist at an e-commerce platform, responsible for optimizing the accuracy of product category tags. The user retrieves a large amount of historical product tag data from a designated database, which may contain redundancy and inconsistencies. The system first cleans this data, removing duplicate tag records, correcting erroneous tag fields, and filling in missing data. Next, the system formats the data, converting all tag fields to a uniform encoding format and standardizing date and numerical ranges. The processed data serves as the model training set, inputting into a random forest model for adaptive learning of field filtering rules. By learning from historical tag data, the model automatically generates optimized field filtering rules. These rules can be used to generate and classify new product tags, improving the accuracy and consistency of the tags.

[0131] In summary, steps S401 and S402, by acquiring and processing historical label field data and using a random forest model for adaptive learning of field filtering rules, solve the problems of data inconsistency and inflexible field filtering rules, thereby improving the accuracy and efficiency of field filtering.

[0132] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 5 It can also specifically include the following:

[0133] Step S501: Receive multiple structured field selection values ​​sent by the user based on the marking template, and perform legality verification on the multiple structured field selection values;

[0134] Step S502: After the legality verification is passed, the selected values ​​of the multiple structured fields are input into a preset adaptive field filtering model for range filtering to obtain a set of optional values ​​that conform to the business rules.

[0135] Optionally, in this embodiment, in step S501, the system receives multiple structured field selection values ​​sent by the user based on the tagging template, and performs validity checks on these selection values. The tagging template typically contains predefined fields and corresponding selection ranges. The user selects from these fields according to actual needs and submits the selection values ​​to the system. The structured field selection values ​​may involve various data types, such as strings, numbers, dates, etc., therefore, validity checks need to be performed differently for different data types.

[0136] In its implementation, the system first parses the user-submitted selection values ​​and matches them against the field definitions in the tagging template. For string selection values, the system checks for illegal characters and conforms to predefined format specifications; for numeric selection values, it verifies whether they are within a reasonable numerical range; and for date selection values, it verifies the validity of the date format and time range. Furthermore, the system checks whether the selection values ​​conform to business logic, such as whether there are duplicate or conflicting selections in multi-select fields.

[0137] The technical principle of validity verification is mainly based on predefined rules and regular expressions. Through these rules and expressions, the system can efficiently verify the validity of selected values, ensuring the accuracy and standardization of data input. In this process, the system not only verifies the validity of individual fields but also performs cross-field correlation verification to guarantee the logical consistency between selected values.

[0138] After the validity check passes, the system proceeds to step S502, where the selected values ​​for multiple structured fields are input into a preset adaptive field filtering model for range filtering. The adaptive field filtering model is a machine learning-based model that can automatically adjust the selection range of fields based on business rules and historical data, thereby generating a set of optional values ​​that meet business requirements.

[0139] Specifically, the system first needs to initialize and train the adaptive field filtering model. The model's training data mainly comes from historical labeling data and business rules. By learning from this data, the model can identify the distribution patterns of field selection values ​​and business constraints. The training process typically involves steps such as feature extraction, model building, and parameter optimization to ensure that the model can accurately filter and recommend appropriate selection values.

[0140] In actual operation, the system inputs the structured field selection values ​​submitted by the user into a trained adaptive field filtering model. The model calculates based on the input values ​​and business rules, filtering out selection values ​​that do not conform to the rules and generating a set of optional values ​​that meet business requirements. This process includes not only simple range filtering but may also involve complex business logic and relational constraints. For example, in product classification and tagging, if a user selects a certain product category, the model will recommend attributes and tags related to that product category based on historical data and business rules, filtering out irrelevant or conflicting selection values.

[0141] The technical principle behind adaptive field filtering models lies in their ability to automatically adjust the range of field selections by learning from a large amount of historical data and business rules, generating a set of optional values ​​that meet business needs. Compared to traditional manual rule-making methods, adaptive models can more efficiently handle complex business logic and data features, improving the accuracy and flexibility of value filtering and recommendation.

[0142] This solution effectively addresses the issues of validating user-selected values ​​and filtering within the scope of business rules. Through validity checks, the system ensures the accuracy and standardization of user input data, preventing subsequent processing problems caused by data errors. The adaptive field filtering model automatically generates a set of selectable values ​​that meet the requirements based on business rules and historical data, improving the efficiency and accuracy of the tagging process.

[0143] In terms of technical effectiveness, the system ensures the quality and consistency of user input data through rigorous validity checks. The application of an adaptive field filtering model allows the system to dynamically adjust the field selection range based on business needs, providing more accurate and flexible value recommendations. This not only improves the efficiency of user tagging but also enhances the system's adaptability in handling complex business scenarios.

[0144] For example, suppose a user is an administrator of an online education platform and needs to tag course content. The user selects values ​​for multiple structured fields based on a tagging template, such as course category, difficulty level, and target audience. Upon receiving these selections, the system first performs a validity check, verifying that each selection conforms to a predefined format and business logic. If the check passes, the system inputs these selections into an adaptive field filtering model for range filtering. Based on historical course data and business rules, the model filters out illogical selections and generates a set of optional values ​​that meet business requirements, such as recommending tags and attributes related to the course category. Finally, the system feeds back these recommended selections to the user, who can directly use these values ​​to tag courses, improving the efficiency and accuracy of tagging.

[0145] In summary, steps S501 and S502, by receiving and verifying the user's structured field selection values ​​and using an adaptive field filtering model for range filtering, solve the problems of non-standard data input and inaccurate selection value recommendations, thereby improving the accuracy and efficiency of the labeling process.

[0146] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 6 It can also specifically include the following:

[0147] Step S601: Extract semantic association knowledge between tag fields from historical tag data, and construct a knowledge graph in the form of a graph structure using the extracted semantic association knowledge between tag fields.

[0148] Step S602: Use the knowledge graph as a model training set to train the preset graph attention network model for event rule reasoning, and obtain the knowledge graph event rule model.

[0149] Optionally, in this embodiment, in step S601, the system extracts semantic association knowledge between tag fields from historical tag data and constructs this knowledge into a knowledge graph in the form of a graph structure. Historical tag data typically contains a large amount of tagging information about different entities (such as products, users, content, etc.), and there may be complex semantic associations between these tag fields. For example, some product tags may be associated with specific user behavior tags, or some content tags may be associated with specific topic tags.

[0150] In its implementation, the system first preprocesses historical tag data, including data cleaning, formatting, and standardization, to ensure data quality and consistency. Next, the system uses Natural Language Processing (NLP) and data mining techniques to extract semantic relationships between tag fields. For example, the system can identify relationships between tag fields through co-occurrence analysis and semantic similarity calculation. Furthermore, the system can utilize pre-trained language models (such as BERT) for semantic matching to further improve the accuracy of these relationships.

[0151] The extracted semantic relationships between tag fields need to be organized into a graph structure, i.e., a knowledge graph. A knowledge graph is a knowledge management method that uses a graph structure to represent entities and their relationships. Nodes represent entities (such as tag fields), and edges represent relationships between entities (such as semantic relationships). The process of building a knowledge graph includes node creation, edge definition, and graph optimization. Node creation refers to adding tag fields as nodes to the graph; edge definition refers to defining the relationships between nodes based on the extracted semantic relationships; and graph optimization involves adjusting the graph structure using graph algorithms to make it more compact and easier to query.

[0152] The completed knowledge graph not only visually displays the semantic relationships between label fields, but also serves as the data foundation for subsequent model training. In step S602, the system uses the knowledge graph as a model training set to train a preset graph attention network model for event rule reasoning, thereby obtaining a knowledge graph event rule model.

[0153] Graph Attention Network (GAT) is a deep learning model based on graph-structured data. By introducing an attention mechanism, it can better capture the complex relationships between nodes and their neighbors. During training, the system first inputs knowledge graph data into the GAT model. The model then aggregates and learns information about nodes and their neighbors through multiple graph attention layers. The attention mechanism allows the model to assign different weights to different neighboring nodes during aggregation, thereby more accurately capturing the semantic relationships between nodes.

[0154] During training, the system generates training samples based on predefined event rules and trains the graph attention network model using supervised learning. Event rules typically refer to specific association patterns between label fields; for example, certain combinations of label fields may trigger specific business events. Through training, the model can learn and recognize these event rules, enabling it to perform reasoning and prediction on knowledge graphs in practical applications.

[0155] This technology not only automates the extraction and construction of knowledge graphs from massive amounts of historical data, but also enables complex event rule reasoning through graph attention network models. Compared to traditional manual rule-making methods, this approach can handle large-scale and complex data more efficiently, significantly improving the accuracy and efficiency of knowledge management and event reasoning.

[0156] This solution effectively addresses the challenges of semantic association knowledge extraction and event rule reasoning from historical tag data. By constructing a knowledge graph, the system can intuitively display the semantic relationships between tag fields, providing a solid foundation for subsequent modeling and reasoning. Through training with a graph attention network model, the system can automatically learn and recognize complex event rules, improving the accuracy and efficiency of event reasoning.

[0157] In terms of technical effectiveness, the system automates the process of extracting semantic associations from historical tag data and performing event rule reasoning by constructing a knowledge graph and training a graph attention network model. The constructed knowledge graph not only visually displays the relationships between tag fields but also significantly improves the efficiency of data querying and analysis. The graph attention network model, by introducing an attention mechanism, can more accurately capture complex relationships between nodes, improving the accuracy and robustness of event reasoning.

[0158] For example, suppose a user is a data scientist at an online education platform, responsible for optimizing course tag management and the recommendation system. The user extracts semantic associations between course tags and user behavior tags from the platform's historical tag data, such as significant correlations between certain course tags and specific user behaviors (e.g., clicks, purchases, reviews). The system uses natural language processing and data mining techniques to identify and extract these associations, constructing a knowledge graph. Next, the system feeds this knowledge graph into a pre-defined graph attention network model for training, learning the event rules between course tags and user behavior tags. Through training, the system can identify and infer user preferences for specific course tags, providing accurate references for course recommendations and tag optimization.

[0159] In summary, steps S601 and S602 solve the problems of semantic association knowledge extraction and event rule reasoning by extracting semantic association knowledge from historical label data and constructing a knowledge graph, and then using a graph attention network model for event rule reasoning training, thereby improving the accuracy and efficiency of knowledge management and reasoning.

[0160] In one embodiment of the material file tag processing method of this application, see [link to relevant documentation]. Figure 7 It can also specifically include the following:

[0161] Step S701: Based on the association rules between tag fields contained in the result of the tag field range filtering, determine the knowledge graph event rule model that best matches the current tag field range from the preset model library;

[0162] Step S702: Input the user-selected tag field value into the knowledge graph event rule model for knowledge reasoning to obtain the corresponding tag mutual constraint conditions.

[0163] Optionally, in this embodiment, in step S701, the system determines the knowledge graph event rule model that best matches the current tag field range from a preset model library based on the association rules between tag fields contained in the tag field range filtering results. The tag field range filtering results refer to a set of tag fields that conform to business rules, obtained after validating and filtering the tag fields in previous steps. These sets contain association rules between tag fields, such as certain combinations of tag fields that may trigger specific business events or constraints.

[0164] In the specific implementation process, the system first needs to manage and classify the knowledge graph event rule models in the preset model library. The preset model library contains multiple trained knowledge graph event rule models, each corresponding to a specific range of label fields and association rules. The system parses the label field range filtering results, extracts the association rules, and matches them with the models in the model library. The matching process is typically based on similarity calculation and rule matching algorithms. The system calculates the matching degree between the label field range and each model, selecting the best-matching model for subsequent inference.

[0165] The core technology relies on similarity calculation and rule matching algorithms. Similarity calculation can employ methods such as Jaccard similarity coefficient and cosine similarity to measure the similarity between the label field range and the model. Rule matching algorithms are used to identify whether the association rules between label fields are consistent with the rules in the model. Through these techniques, the system can efficiently and accurately determine the most matching knowledge graph event rule model from a pre-set model library.

[0166] After determining the most suitable knowledge graph event rule model, the system proceeds to step S702, where the user-selected tag field values ​​are input into the model for knowledge reasoning to obtain the corresponding tag mutual constraints. Knowledge reasoning refers to using the knowledge graph and event rule model to analyze and infer the input tag field values ​​to generate mutual constraints. These constraints help users understand the relationships between tag fields, ensuring the rationality and consistency of tag selection.

[0167] In its implementation, the system first preprocesses the user-selected label field values ​​to ensure their format and content meet the model's input requirements. Next, the system inputs the preprocessed label field values ​​into the knowledge graph event rule model. The model uses a multi-layer graph neural network and attention mechanism to perform inference calculations on the input values. The inference process includes steps such as node feature aggregation, edge weight calculation, and rule matching. By calculating the nodes and edges in the knowledge graph, the model identifies the relationships and constraints between the label field values.

[0168] The technology is primarily based on Graph Neural Networks (GNNs) and the attention mechanism. GNNs can efficiently process graph-structured data, capturing complex relationships between nodes through node feature aggregation and edge weight calculation. The attention mechanism allows the model to assign different weights to different neighboring nodes during aggregation, thereby enabling more accurate inference and matching.

[0169] This solution effectively addresses the challenges of tag field range matching and knowledge reasoning. By identifying the most suitable knowledge graph event rule model from a pre-defined model library, the system can quickly select an appropriate model for reasoning based on the tag field range filtering results and association rules. Through knowledge reasoning, the system can generate mutual constraints between tag field values, helping users understand and manage the relationships between tag fields.

[0170] In terms of technical effectiveness, the system automates the process from label field range filtering results to mutual constraints on label field values ​​through label field range matching and knowledge reasoning. The matching knowledge graph event rule model not only accurately reflects the relationships between label fields but also significantly improves the efficiency of label selection and management. The knowledge reasoning process, through graph neural networks and attention mechanisms, efficiently and accurately generates mutual constraints on label field values, ensuring the rationality and consistency of label selection.

[0171] For example, suppose a user is a category administrator on an online retail platform and needs to add tags to products. The user selects multiple tag field values ​​based on product characteristics, such as brand, category, and price range. The system first performs validity checks and range filtering on these tag field values ​​to obtain a range of tag fields that conforms to business rules. Based on the association rules in the tag field range filtering results, the system determines the most matching knowledge graph event rule model from a pre-set model library. Next, the system inputs the user-selected tag field values ​​into this model for knowledge reasoning. The model generates mutual constraints between tag field values ​​through calculation and matching of the knowledge graph, such as the association conditions between brand and category, and the reasonableness of the price range. Finally, the system feeds back these constraints to the user, who can optimize and adjust the tag selection based on these conditions to ensure the accuracy and consistency of product tags.

[0172] In summary, steps S701 and S702, through label field range matching and knowledge reasoning, solve the problem of converting label field range filtering results into mutual constraints of label field values, thereby improving the accuracy and efficiency of label selection and management. The system achieves an efficient and accurate knowledge reasoning process through similarity calculation, rule matching, graph neural networks, and attention mechanisms, providing users with a reliable label management tool.

[0173] To enable more efficient, intelligent, and automated processing of material file tags, this application provides an embodiment of a material file tag processing apparatus for implementing all or part of the aforementioned material file tag processing method. See [link to embodiment]. Figure 8 The material file tag processing device specifically includes the following components:

[0174] The semantic analysis module 10 is used to receive a tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0175] The tag constraint analysis module 20 is used to receive multiple field selection values ​​and event query requests sent by the user based on the tagging template, filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model, and determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering, thereby obtaining the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0176] The material file tagging module 30 is used to perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file, and to determine the corresponding material file tags.

[0177] As described above, the material file tag processing device provided in this application embodiment can receive tagging requests sent by users through the front end, perform comprehensive semantic recognition on the tagging requests using a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user; receive multiple field selection values ​​and event query requests sent by users based on the tagging template, perform tag field range filtering on the field selection values ​​using a preset adaptive field filtering model, determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtain the tag mutual constraint conditions output by the knowledge graph event rule model; perform cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content features of the target material file, and determine the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0178] From a hardware perspective, in order to process material file tags more efficiently, intelligently, and automatically, this application provides an embodiment of an electronic device for implementing all or part of the material file tag processing method, wherein the electronic device specifically includes the following:

[0179] The system comprises a processor, memory, a communications interface, and a bus; wherein the processor, memory, and communications interface communicate with each other via the bus; the communications interface is used to realize information transmission between the material file tag processing device and core business systems, user terminals, and related databases and other related devices; the logic controller can be a desktop computer, tablet computer, or mobile terminal, etc., and this embodiment is not limited to these. In this embodiment, the logic controller can be implemented with reference to the embodiments of the material file tag processing method and the material file tag processing device in the embodiments, the contents of which are incorporated herein, and repeated details will not be described again.

[0180] It is understood that the user terminal may include smartphones, tablet computers, network set-top boxes, portable computers, desktop computers, personal digital assistants (PDAs), in-vehicle devices, smart wearable devices, etc. Among these, the smart wearable devices may include smart glasses, smartwatches, smart bracelets, etc.

[0181] In practical applications, the material file tag processing method can be partially executed on the electronic device side as described above, or all operations can be completed on the client device. The choice can be made based on the processing capabilities of the client device and the limitations of the user's usage scenario. This application does not impose any limitations on this. If all operations are completed on the client device, the client device may further include a processor.

[0182] The aforementioned client device may have a communication module (i.e., a communication unit) that can communicate with a remote server to achieve data transmission. The server may include a server on the task scheduling center side; in other implementation scenarios, it may also include a server on an intermediate platform, such as a server on a third-party server platform that has a communication link with the task scheduling center server. The server may include a single computer device, a server cluster consisting of multiple servers, or a distributed server structure.

[0183] Figure 9 This is a schematic block diagram illustrating the system configuration of the electronic device 9600 according to an embodiment of this application. Figure 9 As shown, the electronic device 9600 may include a central processing unit 9100 and a memory 9140; the memory 9140 is coupled to the central processing unit 9100. It is worth noting that... Figure 9 This is an example; other types of structures can also be used to supplement or replace this structure to achieve telecommunications functions or other functions.

[0184] In one embodiment, the material file tag processing method function can be integrated into the central processing unit 9100.

[0185] The central processing unit 9100 can be configured to perform the following controls:

[0186] Step S101: Receive a tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0187] Step S102: Receive multiple field selection values ​​and event query requests sent by the user based on the tagging template; filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model; determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering; and obtain the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0188] Step S103: Perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file to determine the corresponding material file tags.

[0189] As described above, the electronic device provided in this application embodiment receives a tagging request sent by a user through a front end, performs comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the result of the comprehensive semantic recognition, and returns it to the user; it receives multiple field selection values ​​and event query requests sent by the user based on the tagging template, performs tag field range filtering on the field selection values ​​using a preset adaptive field filtering model, determines the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtains the tag mutual constraint conditions output by the knowledge graph event rule model; it performs cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content features of the target material file, and determines the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0190] In another embodiment, the material file tag processing device can be configured separately from the central processing unit 9100. For example, the material file tag processing device can be configured as a chip connected to the central processing unit 9100, and the material file tag processing method function can be implemented through the control of the central processing unit.

[0191] like Figure 9As shown, the electronic device 9600 may further include: a communication module 9110, an input unit 9120, an audio processor 9130, a display 9160, and a power supply 9170. It is worth noting that the electronic device 9600 does not necessarily need to include these components. Figure 9 All components shown; in addition, the electronic device 9600 may also include Figure 9 For components not shown, please refer to existing technologies.

[0192] like Figure 9 As shown, the central processing unit 9100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device, which receives inputs and controls the operation of various components of the electronic device 9600.

[0193] The memory 9140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 9100 may execute the program stored in the memory 9140 to perform information storage or processing, etc.

[0194] Input unit 9120 provides input to central processing unit 9100. Input unit 9120 may be, for example, a keypad or touch input device. Power supply 9170 provides power to electronic device 9600. Display 9160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0195] The memory 9140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 9140 can also be some other type of device. The memory 9140 includes a buffer memory 9141 (sometimes referred to as a buffer). The memory 9140 may include an application / function storage unit 9142 for storing application programs and function programs or processes for executing the operation of the electronic device 9600 via the central processing unit 9100.

[0196] The memory 9140 may also include a data storage unit 9143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 9144 of the memory 9140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0197] The communication module 9110 is a transmitter / receiver that sends and receives signals via the antenna 9111. The communication module 9110 (transmitter / receiver) is coupled to the central processing unit 9100 to provide input signals and receive output signals, which is the same as in a conventional mobile communication terminal.

[0198] Based on different communication technologies, multiple communication modules 9110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module 9110 (transmitter / receiver) is also coupled to a speaker 9131 and a microphone 9132 via an audio processor 9130 to provide audio output via the speaker 9131 and receive audio input from the microphone 9132, thereby realizing typical telecommunications functions. The audio processor 9130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 9130 is coupled to a central processing unit 9100, enabling on-device recording via the microphone 9132 and on-device playback of stored audio via the speaker 9131.

[0199] Embodiments of this application also provide a computer-readable storage medium capable of implementing all steps of the material file tag processing method with a server or client as the execution subject in the above embodiments. The computer-readable storage medium stores a computer program that, when executed by a processor, implements all steps of the material file tag processing method with a server or client as the execution subject in the above embodiments. For example, when the processor executes the computer program, it implements the following steps:

[0200] Step S101: Receive a tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0201] Step S102: Receive multiple field selection values ​​and event query requests sent by the user based on the tagging template; filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model; determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering; and obtain the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0202] Step S103: Perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file to determine the corresponding material file tags.

[0203] As described above, the computer-readable storage medium provided in this application embodiment receives a tagging request sent by a user through a front end, performs comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the result of the comprehensive semantic recognition, and returns it to the user; it receives multiple field selection values ​​and event query requests sent by the user based on the tagging template, performs tag field range filtering on the field selection values ​​using a preset adaptive field filtering model, determines the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtains the tag mutual constraint conditions output by the knowledge graph event rule model; it performs cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content features of the target material file, and determines the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0204] Embodiments of this application also provide a computer program product capable of implementing all steps of the material file tag processing method with the execution subject being a server or client in the above embodiments. When this computer program / instruction is executed by a processor, it implements the steps of the material file tag processing method. For example, the computer program / instruction implements the following steps:

[0205] Step S101: Receive a tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data.

[0206] Step S102: Receive multiple field selection values ​​and event query requests sent by the user based on the tagging template; filter the field selection values ​​for tag field ranges according to a preset adaptive field filtering model; determine the corresponding knowledge graph event rule model based on the result of the tag field range filtering; and obtain the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields.

[0207] Step S103: Perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file to determine the corresponding material file tags.

[0208] As described above, the computer program product provided in this application receives a tagging request sent by a user through a front end, performs comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the result of the comprehensive semantic recognition, and returns it to the user; it receives multiple field selection values ​​and event query requests sent by the user based on the tagging template, performs tag field range filtering on the field selection values ​​using a preset adaptive field filtering model, determines the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtains the tag mutual constraint conditions output by the knowledge graph event rule model; it performs cross-modal linkage constraints on the field selection values ​​based on the tag mutual constraint conditions and the content characteristics of the target material file, and determines the corresponding material file tags, thereby enabling more efficient, intelligent, and automated tag processing.

[0209] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0210] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0211] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0212] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0213] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. A method for processing material file tags, characterized in that, The method includes: The system receives a tagging request sent by a user through the front end, performs comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, determines the corresponding tagging template based on the result of the comprehensive semantic recognition, and returns it to the user. The multimodal semantic recognition model is trained based on historical multimodal data. Receiving multiple field selection values ​​and event query requests sent by the user based on the tagging template, and filtering the field selection values ​​for tag field range according to a preset adaptive field filtering model, includes: receiving multiple structured field selection values ​​sent by the user based on the tagging template, and performing legality verification on the multiple structured field selection values; after the legality verification passes, inputting the multiple structured field selection values ​​into the preset adaptive field filtering model for range filtering to obtain a set of optional values ​​that conform to business rules; Based on the result of the label field range filtering, a corresponding knowledge graph event rule model is determined, and the label mutual constraint conditions output by the knowledge graph event rule model are obtained. The adaptive field filtering model is trained based on historical label field data, and the knowledge graph event rule model is trained based on semantic association knowledge between label fields. This includes: determining the knowledge graph event rule model that best matches the current label field range from a preset model library based on the label field association rules contained in the label field range filtering result; inputting the user-selected label field value into the knowledge graph event rule model for knowledge reasoning to obtain the corresponding label mutual constraint conditions. Based on the mutual constraints of the tags and the content characteristics of the target material file, cross-modal linkage constraints are applied to the field selection values ​​to determine the corresponding material file tags.

2. The material file tag processing method according to claim 1, characterized in that, Before receiving the tagging request sent by the user through the front end and performing comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, the process includes: Obtain multimodal data from a designated database that contains at least one of text, images, and audio / video, and perform noise cleaning and formatting normalization on the multimodal data; The multimodal data, after noise cleaning and formatting normalization, is used as the model training set to train a multimodal network structure for a given convolutional neural network, thereby obtaining a multimodal semantic recognition model.

3. The material file tag processing method according to claim 1, characterized in that, The process of receiving a tagging request sent by a user through a front-end, and performing comprehensive semantic recognition on the tagging request using a preset multimodal semantic recognition model, includes: Receive the tagging request sent by the user through the front end and perform syntax and format validation on the tagging request; After the syntax and format verification is passed, the text information and / or associated image in the tagging request are input into a preset multimodal semantic recognition model for comprehensive semantic recognition to obtain the corresponding semantic recognition result.

4. The material file tag processing method according to claim 1, characterized in that, Before receiving multiple field selection values ​​and event query requests sent by the user based on the tagging template, and filtering the field selection values ​​for tag field ranges according to a preset adaptive field filtering model, the process includes: Obtain historical tag field data from the designated database and perform data cleaning and formatting on the historical tag field data; The historical label field data, after the data cleaning and formatting process, is used as the model training set and input into the random forest model to adaptively learn the field filtering rules, thus obtaining an adaptive field filtering model.

5. The material file tag processing method according to claim 1, characterized in that, Before determining the corresponding knowledge graph event rule model based on the result of filtering according to the tag field range, and obtaining the tag mutual constraint conditions output by the knowledge graph event rule model, the process includes: Semantic association knowledge between tag fields is extracted from historical tag data, and a knowledge graph is constructed in the form of a graph structure using the extracted semantic association knowledge between tag fields. The knowledge graph is used as a model training set to train a preset graph attention network model for event rule reasoning, thereby obtaining a knowledge graph event rule model.

6. A material file tag processing device, characterized in that, The device includes: The semantic analysis module is used to receive the tagging request sent by the user through the front end, perform comprehensive semantic recognition on the tagging request through a preset multimodal semantic recognition model, determine the corresponding tagging template based on the result of the comprehensive semantic recognition, and return it to the user. The multimodal semantic recognition model is trained based on historical multimodal data. The tag constraint analysis module is used to receive multiple field selection values ​​and event query requests sent by the user based on the tagging template, and to filter the field selection values ​​for tag field range according to a preset adaptive field filtering model. This includes: receiving multiple structured field selection values ​​sent by the user based on the tagging template, and performing legality verification on the multiple structured field selection values; after the legality verification passes, inputting the multiple structured field selection values ​​into the preset adaptive field filtering model for range filtering to obtain a set of optional values ​​that conform to business rules; determining the corresponding knowledge graph event rule model based on the result of the tag field range filtering, and obtaining the tag mutual constraint conditions output by the knowledge graph event rule model. The adaptive field filtering model is trained based on historical tag field data, and the knowledge graph event rule model is trained based on semantic association knowledge between tag fields. This includes: determining the knowledge graph event rule model that best matches the current tag field range from a preset model library based on the tag field association rules contained in the result of the tag field range filtering; inputting the tag field values ​​selected by the user into the knowledge graph event rule model for knowledge reasoning to obtain the corresponding tag mutual constraint conditions. The material file tagging module is used to perform cross-modal linkage constraints on the field selection values ​​based on the mutual constraints of the tags and the content characteristics of the target material file, and to determine the corresponding material file tags.

7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the material file tag processing method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the material file tag processing method according to any one of claims 1 to 5.

Images