A portrait marking method, device, and computer device based on streaming data

Abstract

This application relates to a portrait marking method, device, and computer device based on streaming data. The method includes: performing multi-level time series analysis on the historical active streaming data, historical interaction streaming data, and historical passive streaming data of a target object to generate an initial marked portrait; updating the features and labels of the initial marked portrait according to the current active streaming data, current interaction streaming data, and current passive streaming data of the target object to obtain the current marked portrait; mapping the streaming data of each classification scenario, the streaming data of each classification field, and the current marked portrait of the target object into the same embedding space to obtain a fused marked portrait; adjusting the portrait output style of the fused marked portrait according to the current active streaming data, current interaction streaming data, and current passive streaming data to obtain the target marked portrait. This method can improve the accuracy of image marking during the marking process and timely capture potential new patterns in the data.

Description

Technical Field

[0001] This application relates to the field of computer technology, and particularly to a method, device, and computer equipment for portrait labeling based on streaming data. Background Art

[0002] Portrait labeling in traditional technologies is mainly used for data tagging and classification. The system extracts key features from the original data, such as the basic information, behavior data, transaction records, etc. of users; further analyzes these features through machine learning models (such as classifiers, clustering algorithms, etc.) to generate different labels for describing the multi-dimensional attributes or categories of the data. These labels can help enterprises better understand users, market trends, and business needs, and are widely used in fields such as personalized recommendation, risk prediction, and customer segmentation.

[0003] However, traditional technologies rely too much on historical data and static features, and have poor adaptability to the dynamic changes of data and complex scenarios, resulting in a decrease in accuracy during the process of portrait labeling and an inability to timely capture potential new patterns in the data. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method, device, computer equipment, computer-readable storage medium, and computer program product for portrait labeling based on streaming data that can improve the accuracy during the process of portrait labeling and timely capture potential new patterns in the data.

[0005] In a first aspect, this application provides a method for portrait labeling based on streaming data, including:

[0006] Performing multi-level time series analysis on the historical active streaming data, historical interaction streaming data, and historical passive streaming data of the target object to generate an initial labeled portrait;

[0007] Updating the features and labels of the initial labeled portrait according to the current active streaming data, current interaction streaming data, and current passive streaming data of the target object to obtain the current labeled portrait;

[0008] Mapping the streaming data of each classification scenario, the streaming data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait;

[0009] Adjusting the portrait output style of the fused labeled portrait according to the current active streaming data, the current interaction streaming data, and the current passive streaming data to obtain the target labeled portrait.

[0010] In a second aspect, this application also provides a device for portrait labeling based on streaming data, including:

[0011] A marking image generation module, which is used to perform multi-level time series analysis on the historical active flow data, historical interaction flow data, and historical passive flow data of a target object, and generate an initial marking portrait;

[0012] A marking image update module, which is used to update the features and labels of the initial marking portrait according to the current active flow data, current interaction flow data, and current passive flow data of the target object, and obtain the current marking portrait;

[0013] An image data fusion module, which is used to map the flow data of each classification scenario, the flow data of each classification field, and the current marking portrait of the target object into the same embedding space to obtain a fused marking portrait;

[0014] An image style optimization module, which is used to adjust the portrait output style of the fused marking portrait according to the current active flow data, the current interaction flow data, and the current passive flow data, and obtain the target marking portrait.

[0015] In a third aspect, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0016] Perform multi-level time series analysis on the historical active flow data, historical interaction flow data, and historical passive flow data of a target object, and generate an initial marking portrait;

[0017] Update the features and labels of the initial marking portrait according to the current active flow data, current interaction flow data, and current passive flow data of the target object, and obtain the current marking portrait;

[0018] Map the flow data of each classification scenario, the flow data of each classification field, and the current marking portrait of the target object into the same embedding space to obtain a fused marking portrait;

[0019] Adjust the portrait output style of the fused marking portrait according to the current active flow data, the current interaction flow data, and the current passive flow data, and obtain the target marking portrait.

[0020] In a fourth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the following steps are implemented:

[0021] Perform multi-level time series analysis on the historical active flow data, historical interaction flow data, and historical passive flow data of a target object, and generate an initial marking portrait;

[0022] Update the features and labels of the initial labeled portrait based on the current active transaction data, current interaction transaction data, and current passive transaction data of the target object to obtain the current labeled portrait;

[0023] Map the transaction data of each classification scenario, the transaction data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait;

[0024] Adjust the portrait output style of the fused labeled portrait according to the current active transaction data, the current interaction transaction data, and the current passive transaction data to obtain the target labeled portrait.

[0025] In a fifth aspect, the present application also provides a computer program product, including a computer program, which when executed by a processor implements the following steps:

[0026] Perform multi-level time series analysis on the historical active transaction data, historical interaction transaction data, and historical passive transaction data of the target object to generate an initial labeled portrait;

[0027] Update the features and labels of the initial labeled portrait based on the current active transaction data, current interaction transaction data, and current passive transaction data of the target object to obtain the current labeled portrait;

[0028] Map the transaction data of each classification scenario, the transaction data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait;

[0029] Adjust the portrait output style of the fused labeled portrait according to the current active transaction data, the current interaction transaction data, and the current passive transaction data to obtain the target labeled portrait.

[0030] The above portrait labeling method, device, computer device, storage medium, and computer program product based on transaction data generate an initial labeled portrait by performing multi-level time series analysis on the historical active transaction data, historical interaction transaction data, and historical passive transaction data of the target object; update the features and labels of the initial labeled portrait according to the current active transaction data, current interaction transaction data, and current passive transaction data of the target object to obtain the current labeled portrait; map the transaction data of each classification scenario, the transaction data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait; adjust the portrait output style of the fused labeled portrait according to the current active transaction data, current interaction transaction data, and current passive transaction data to obtain the target labeled portrait.

[0031] Through multi-level time series analysis of the historical active, interactive, and passive transaction data of the target object, a preliminary labeled portrait can be constructed. This process provides a solid data foundation for the feature extraction of the target object, ensuring a comprehensive understanding of the behavior pattern of the target object. On this basis, the portrait is dynamically updated in combination with the current transaction data to ensure that the portrait can always reflect the latest behavior characteristics and change trends of the target object, avoiding the lag that may be brought by a static portrait. Further, mapping various types of transaction data (such as scenario and domain data) of the target object to the same embedding space as the current labeled portrait can achieve the comprehensive integration of multi-dimensional data, improving the accuracy and comprehensiveness of the portrait, so that the portrait not only stays at the single-dimensional description, but more comprehensively reflects the complex behavior characteristics of the target object. In addition, by further adjusting the fused portrait and optimizing the portrait output style according to the current transaction data, personalized portrait customization can be realized to meet the needs of different application scenarios. The combination of these technical means can improve the accuracy of the image in the labeling process, timely capture the potential new rules in the data, provide more powerful and flexible support for applications such as decision-making support and risk prediction, thus significantly enhancing the value and operability of data analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0033] Figure 1 It is an application environment diagram of the portrait labeling method based on transaction data in an embodiment;

[0034] Figure 2 It is a flow schematic diagram of the portrait labeling method based on transaction data in an embodiment;

[0035] Figure 3 It is a flow schematic diagram of the method for obtaining a fused labeled portrait in an embodiment;

[0036] Figure 4 It is a flow schematic diagram of the method for obtaining a cross-domain fusion vector in an embodiment;

[0037] Figure 5 It is a flow schematic diagram of the method for obtaining a spatial embedding vector in an embodiment;

[0038] Figure 6 It is a flow schematic diagram of the method for obtaining a non-linear relationship vector in an embodiment;

[0039] Figure 7 Schematic flowchart of the spatial embedding vector method in another embodiment;

[0040] Figure 8 Schematic flowchart of the method for obtaining the target labeled portrait in one embodiment;

[0041] Figure 9 Schematic flowchart of the method for obtaining the reconstructed labeled image in one embodiment;

[0042] Figure 10 Structural block diagram of the portrait labeling device based on streaming data in one embodiment;

[0043] Figure 11 Internal structure diagram of a computer device in one embodiment. Detailed implementation manners

[0044] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] A portrait labeling method based on streaming data provided by an embodiment of the present application can be applied to an application environment as shown in Figure 1 wherein, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or placed in the cloud or other network servers. The server 104 obtains the historical active streaming data, historical interaction streaming data and historical passive streaming data of the target object in the terminal 102; at the same time, obtains the current active streaming data, current interaction streaming data and current passive streaming data of the target object; at the same time obtains; further, the server 104 performs multi-level time series analysis on the historical active streaming data, historical interaction streaming data and historical passive streaming data of the target object to generate an initial labeled portrait; updates the features and labels of the initial labeled portrait according to the current active streaming data, current interaction streaming data and current passive streaming data of the target object to obtain the current labeled portrait; maps the streaming data of each classification scenario, streaming data of each classification field and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait; adjusts the portrait output style of the fused labeled portrait according to the current active streaming data, current interaction streaming data and current passive streaming data to obtain the target labeled portrait. Among them, the server 104 can be implemented by an independent server or a server cluster composed of multiple servers.

[0046] In an exemplary embodiment, as shown in Figure 2As shown, a portrait marking method based on streaming data is provided. Taking the server in Figure 1 as an example for illustration, the method includes the following steps 202 to 208. Among them:

[0047] Step 202, perform multi-level time series analysis on the historical active streaming data, historical interaction streaming data, and historical passive streaming data of the target object to generate an initial marked portrait.

[0048] Among them, the target object can be an individual or entity that needs to be characterized and studied during the data analysis process, such as a user, customer, device, account, etc., and usually has a series of quantifiable behavior data and characteristics.

[0049] Among them, the historical active streaming data can be the operation or behavior data actively initiated by the target object within a certain period in the past. For example, behavior data such as a user actively clicking on an advertisement, initiating a transaction request, or submitting a query.

[0050] Among them, the historical interaction streaming data can be the interaction behavior records between the target object and other entities. For example, social interactions between users, communication between users and service platforms, etc.

[0051] Among them, the historical passive streaming data can be the relevant records generated by the system or external environment automatically sending to the target object for behaviors that the target object did not actively initiate within a certain period in the past. For example, a user's browsing record, log data generated by the system background, etc.

[0052] Among them, the multi-level time series analysis can be a method for in-depth analysis of time series data in multiple dimensions and at multiple levels. By analyzing different time scales (such as hours, days, weeks, months) and different dimensions (such as active, interaction, passive), the dynamic changes and trends of the target object can be captured from multiple levels, providing an all-round perspective for portrait construction.

[0053] Among them, the initial marked portrait can be a preliminary portrait generated based on the historical data of the target object, including the basic behavior characteristics and labels of the target object.

[0054] Specifically, performing multi-level time series analysis on the historical active streaming data, historical interaction streaming data, and historical passive streaming data of the target object can divide different types of historical data in the time dimension in detail, capture the behavior patterns and potential rules of the target object. Through multi-dimensional data fusion, combined with data preprocessing and feature extraction techniques, an initial marked portrait of the target object is constructed, including its behavior characteristics, attribute labels, and historical trends, providing a basis for subsequent analysis.

[0055] Step 204: Update the features and labels of the initial labeled portrait based on the current active transaction data, current interaction transaction data, and current passive transaction data of the target object to obtain the current labeled portrait.

[0056] Among them, the current active transaction data can be the operation or behavior data actively initiated by the target object within the current time period. Different from the historical active transaction data, the current data reflects the active state and behavior trend of the target object at the latest time point.

[0057] Among them, the current interaction transaction data can be the interaction behavior records between the target object and other entities within the current time period, reflecting the social activity, interaction quality, and connection strength with the external environment of the target object within the current time period.

[0058] Among them, the current passive transaction data can be the behavior records automatically generated by the system or the external environment that the target object did not actively initiate within the current time period. It mainly reflects the potential interests or needs of the target object and is an indirect reflection of the current state of the target object.

[0059] Among them, the current labeled portrait can be the portrait updated based on the current data of the target object on the basis of the initial labeled portrait, which can more accurately reflect the behavior characteristics and state of the target object at the latest time point.

[0060] Specifically, based on the current active transaction data, interaction transaction data, and passive transaction data of the target object, use the update algorithm for analysis, identify the behavior changes or trends of the target object at the current moment, and combine machine learning models (such as incremental learning, online learning, etc.) to dynamically update the portrait, so that the current labeled portrait can accurately reflect the latest behavior characteristics and state changes of the target object.

[0061] Step 206: Map the transaction data of each classification scenario, transaction data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain the fused labeled portrait.

[0062] Among them, the transaction data of classification scenarios can be the transaction data related to the behavior of the target object collected for different business scenarios (such as marketing, customer service, transactions, etc.).

[0063] Among them, the transaction data of classification fields can be the transaction data collected according to different fields (such as finance, health, education, etc.) involved by the target object.

[0064] Among them, the fused labeled portrait can be the result of fusing the transaction data of the target object in various classification scenarios and classification fields with the current labeled portrait.

[0065] Specifically, the transaction data of the target object in different classification scenarios (such as marketing scenarios, service scenarios) and the transaction data in each classification field (such as finance, health, social, etc.) are combined with the current labeled portrait. Since data from different sources have different characteristics and structures, feature extraction and data preprocessing are required to unify the data format and dimensions. Then, high-dimensional embedding techniques (such as multidimensional scaling, autoencoders or transformers in deep learning models) are used to map these heterogeneous data into a unified embedding space. In this embedding space, various types of data can be comprehensively represented in the form of vectors, eliminating the original dimensional differences and structural inconsistencies, so that the fused labeled portrait can comprehensively and accurately represent all aspects of the target object's behavioral characteristics. This fusion process not only integrates historical and current data, but also combines the characteristics of different scenarios and fields, providing a multi-dimensional, cross-scenario panorama, thus greatly improving the accuracy and applicability of the target portrait.

[0066] Step 208: Adjust the portrait output style of the fused labeled portrait according to the current active transaction data, current interaction transaction data, and current passive transaction data to obtain the target labeled portrait.

[0067] Among them, the portrait output style can be the display form or structure of the target labeled portrait, which determines how the portrait is presented to the user or system. The adjustment of the output style can optimize the readability, accuracy, and operability of the portrait. Depending on the behavioral characteristics of the target object and the application scenario, the output style may vary.

[0068] Among them, the target labeled portrait can be the finally generated portrait that comprehensively reflects the behavioral characteristics and status of the target object.

[0069] Specifically, after generating the fused labeled portrait, it is necessary to adjust the output style of the portrait according to the current transaction data of the target object to better adapt to different business requirements and application scenarios. Based on the current active, interaction, and passive transaction data, identify the latest trends or changes in the behavior of the target object. These data can reflect real-time information in aspects such as its interests, preferences, and activity frequency. Combining these changes, make detailed adjustments to each feature value and display form of the fused portrait to optimize the readability and adaptability of the portrait. In different application scenarios, the display method of the portrait will be different. For example, in risk assessment, it may focus on highlighting potential risk points, while in a personalized recommendation system, it may focus on highlighting the preferences or needs of the target object. By adjusting the output style, the target labeled portrait better meets specific business requirements, improves the accuracy, operability, and practical value of the portrait, and thus provides stronger data support for decision-making and subsequent operations.

[0070] In the above-mentioned portrait labeling method based on streaming data, an initial labeled portrait is generated by performing multi-level time series analysis on the historical active streaming data, historical interaction streaming data, and historical passive streaming data of the target object; the features and labels of the initial labeled portrait are updated according to the current active streaming data, current interaction streaming data, and current passive streaming data of the target object to obtain the current labeled portrait; the streaming data of each classification scenario, the streaming data of each classification field, and the current labeled portrait of the target object are mapped into the same embedding space to obtain a fused labeled portrait; the portrait output style of the fused labeled portrait is adjusted according to the current active streaming data, current interaction streaming data, and current passive streaming data to obtain the target labeled portrait.

[0071] By performing multi-level time series analysis on the historical active, interaction, and passive streaming data of the target object, a preliminary labeled portrait can be constructed, which provides a solid data foundation for the feature extraction of the target object and ensures a comprehensive understanding of the behavior pattern of the target object. On this basis, the portrait is dynamically updated in combination with the current streaming data to ensure that the portrait can always reflect the latest behavior characteristics and change trends of the target object, avoiding the lag that may be brought by a static portrait. Further, mapping the various types of streaming data (such as scenario and field data) of the target object and the current labeled portrait into the same embedding space can achieve the comprehensive integration of multi-dimensional data, improving the accuracy and comprehensiveness of the portrait, so that the portrait not only stays at the single-dimensional description, but more comprehensively reflects the complex behavior characteristics of the target object. In addition, by further adjusting the fused portrait and optimizing the portrait output style according to the current streaming data, personalized portrait customization can be realized to meet the needs of different application scenarios. The combination of these series of technical means can improve the accuracy of the image during the labeling process, timely capture the potential new laws in the data, provide more powerful and flexible support for applications such as decision-making support and risk prediction, thus significantly enhancing the value and operability of data analysis.

[0072] In an exemplary embodiment, as Figure 3 shown, mapping the streaming data of each classification scenario, the streaming data of each classification field, and the current labeled portrait of the target object into the same embedding space to obtain a fused labeled portrait includes steps 302 to 306. Among them:

[0073] Step 302, perform cross-domain feature fusion on the streaming data of each classification scenario and the streaming data of each classification field to obtain each cross-domain fusion vector.

[0074] Among them, cross-domain feature fusion can be to integrate the data features from different scenarios and fields in order to comprehensively analyze the behavior characteristics of the target object from multiple perspectives.

[0075] Among them, the cross-domain fusion vector can be a high-dimensional vector representation obtained through cross-domain feature fusion, which contains the behavioral characteristics of the target object in different domains and scenarios.

[0076] Specifically, it is necessary to integrate the streaming data from different sources (including the streaming data of each classification scenario and the streaming data of the classification domain) into a unified representation. Each data type usually has different characteristics and structures, so feature extraction and standardization are performed on it to facilitate the fusion process. The feature alignment technology is used to convert different types of data into the same dimension and perform data normalization so that they can be compared in the same space. On this basis, through weighted averaging, concatenation, or using deep learning models (such as multi-layer perceptrons or autoencoders, etc.) for feature fusion, the key and auxiliary information in the data of each scenario and each domain is merged into each cross-domain fusion vector. Each vector not only contains the basic information of various types of data but also can effectively fuse the behavioral patterns of the target object in different scenarios and domains, forming a more comprehensive behavioral feature representation.

[0077] Step 304: Identify the non-linear relationship information from each cross-domain fusion vector to obtain each spatial embedding vector.

[0078] Among them, the non-linear relationship information can be the complex dependence relationships between the features in the data, and these relationships cannot be captured by traditional linear models (such as linear regression, simple weighted sums, etc.).

[0079] Among them, the spatial embedding vector can be a vector representation that maps high-dimensional and complex raw data to a low-dimensional continuous space through embedding learning technology.

[0080] Specifically, using technologies such as non-linear mapping and deep learning, such as multi-layer neural networks or convolutional neural networks, can capture the complex associations existing in the data in the high-dimensional feature space, especially the mutual influence of the target object between different scenarios and domains. At the same time, performing non-linear dimensionality reduction technologies (such as t-SNE or autoencoders) can also help identify the hidden structures and associations in the data, reveal the potential non-linear relationships between the behaviors of the target object, and extract the non-linear relationship information composed of these complex associations and non-linear relationships through methods such as activation functions and weight adjustment, and finally convert it into spatial embedding vectors.

[0081] Step 306: Map each spatial embedding vector and the current labeled portrait to the same embedding space to obtain the fused labeled portrait.

[0082] Specifically, the features from different sources (including the spatial embedding vectors extracted from the cross-domain fusion vectors and the existing current labeled portraits) are integrated into a comprehensive and unified vector representation through an embedding learning method. For example, techniques such as shared-weight networks or generative adversarial networks (GANs) in deep learning are used to map the features from different data sources into the same embedding space. In this embedding space, the current labeled portrait and the spatial embedding vectors share the same feature dimensions, enabling them to interact with each other and perform further analysis. The generated fused labeled portrait not only synthesizes the behavioral information of the target object in different scenarios and domains but also accurately reflects its latest state and dynamic changes, ultimately obtaining a portrait that can provide comprehensive support for subsequent analysis and decision-making.

[0083] In this embodiment, by performing cross-domain feature fusion on the classified scenario flow data and the classified domain flow data, the resulting cross-domain fusion vectors can effectively combine data features of different types, thereby enhancing the correlation and expressiveness between the data. Then, by identifying the non-linear relationship information in the cross-domain fusion vectors, the generated spatial embedding vectors further extract the latent patterns and relationships in the data, capable of capturing complex feature interactions more precisely. Finally, mapping these spatial embedding vectors and the current labeled portrait into the same embedding space, the generated fused labeled portrait not only has a high degree of personalization and customization visually but also can more accurately reflect the association between data features and user behavior. This process can significantly improve the adaptability, accuracy, and information transmission efficiency of the portrait, ultimately optimizing the user experience and enhancing the effect of data-driven decision-making.

[0084] In an exemplary embodiment, as Figure 4 shown, performing cross-domain feature fusion on each classified scenario flow data and each classified domain flow data to obtain each cross-domain fusion vector, including steps 402 to 406. Among them:

[0085] Step 402, performing feature extraction on each classified scenario flow data and each classified domain flow data to obtain each classified scenario flow feature vector and each classified domain flow feature vector.

[0086] Among them, the classified scenario flow feature vector can be a vector representation obtained by performing feature extraction on the flow data of the target object in a specific scenario.

[0087] Among them, the classified domain flow feature vector can be a vector representation obtained by performing feature extraction on the flow data of the target object in a specific domain.

[0088] Specifically, in the feature extraction stage, first, preprocess the flowing water data of each classification scenario and the flowing water data of each classification field to remove noise and standardize it. Then, extract important features from the original data through statistical analysis, time series modeling, or deep learning methods (such as CNN, LSTM, etc.). For the flowing water data of classification scenarios, basic statistical features (such as mean, variance, etc.) can be extracted, while for data with time series properties, autoregressive models or LSTM are used to capture the temporal patterns in the data. In addition, using deep learning models (such as CNN) can automatically extract high-order features from the original data, and finally generate feature vectors for each dataset, including the flowing water feature vectors of classification scenarios and the flowing water feature vectors of classification fields.

[0089] Step 404, traverse the mutual relationships between the flowing water feature vectors of each classification scenario and the flowing water feature vectors of each classification field to obtain a classification feature vector relationship network.

[0090] Among them, the classification feature vector relationship network can be a network structure constructed based on the mutual relationships between the flowing water feature vectors of each classification scenario and the flowing water feature vectors of each classification field, which includes the closeness information between each flowing water feature vector of each classification scenario and each flowing water feature vector of each classification field directly.

[0091] Specifically, when constructing the classification feature vector relationship network, calculate the similarity between the flowing water feature vectors of classification scenarios and the flowing water feature vectors of classification fields, where metrics such as cosine similarity, Pearson correlation coefficient, or Euclidean distance are used. Construct a graph network through the similarity, where each feature vector represents a node, and the similarity between features determines the weight of the edge. This graph network can help identify the dependence relationships and influences between different features. In addition, using graph neural networks (GNN) can further identify the complex relationships between the feature vectors of the graph network, thereby obtaining a more accurate classification feature vector relationship network.

[0092] Step 406, mutually transfer the feature information of the flowing water feature vectors of each classification scenario and the flowing water feature vectors of each classification field in the classification feature vector relationship network whose vector relationship closeness value is greater than the preset closeness threshold to obtain each cross-domain fusion vector.

[0093] Among them, the vector relationship closeness value can be a numerical value that measures the similarity or correlation degree between two vectors.

[0094] Among them, the preset closeness threshold can be used to judge whether two vectors have sufficient similarity or relationship.

[0095] Specifically, based on the classification feature vector relationship network, determine the vector relationship closeness values between each classification scenario flow feature vector and each classification domain flow feature vector. Adopt transfer learning or self-attention mechanism to share the knowledge between the classification scenario flow feature vectors and the classification domain flow feature vectors with vector relationship closeness values greater than the preset closeness threshold, so that the data in the two domains can influence and complement each other. Through the self-attention mechanism, automatically adjust the weights of different features and dynamically select and fuse the most important feature information. Finally, use feature mapping (such as a fully connected layer or a convolutional layer) to fuse the two feature vectors with relationship closeness greater than the preset intimacy threshold into a cross-domain fusion vector, which contains rich information from the classification scenario and the classification domain and provides a comprehensive feature representation for subsequent analysis tasks.

[0096] In this embodiment, by extracting features from the classification scenario flow data and the classification domain flow data, the obtained classification scenario flow feature vectors and classification domain flow feature vectors provide multi-dimensional representations of the data and can comprehensively capture the key features of different categories and domains. Further, by traversing the mutual relationships between the feature vectors, the constructed classification feature vector relationship network reveals the correlation between different data sources and enhances the semantic understanding between the features. On this basis, information transfer is performed on the features with vector relationship closeness exceeding the preset threshold, effectively integrating the key information of the classification scenario flow feature vectors and the classification domain flow feature vectors, thereby generating a cross-domain fusion vector. This process can eliminate the differences between domains, improve the overall consistency and expression ability of the data, and thus provide more accurate and comprehensive feature information for subsequent analysis, prediction, and decision-making.

[0097] In an exemplary embodiment, as Figure 5 shown, identify the non-linear relationship information from each cross-domain fusion vector to obtain each spatial embedding vector, including steps 502 to 506. Among them:

[0098] Step 502, identify the non-linear relationship information from each cross-domain fusion vector to obtain each non-linear relationship vector.

[0099] Among them, the non-linear relationship information can be the complex interaction existing in the data, which cannot be represented by a simple linear model.

[0100] Among them, the non-linear relationship vector can be the representation obtained after non-linearly transforming the original data features.

[0101] Specifically, since the cross-domain fusion vectors may contain complex non-linear interactions, activation functions (such as ReLU, Sigmoid, tanh, etc.) are used to perform non-linear transformations on each cross-domain fusion vector. These activation functions can compress and expand the data, helping to capture complex patterns and features. To extract non-linear relationships from complex data, deep neural networks (DNNs), especially multi-layer perceptrons (MLPs), can also be used. Through their multi-layer structure, the non-linear dependencies between input features can be automatically learned. Among them, the MLP can extract complex non-linear features from the original input data through successive weight matrix transformations. Each cross-domain fusion vector processed in this way will be mapped to a new space, and the generated non-linear relationship vectors can effectively express the non-linear features in the data.

[0102] Step 504: Perform high-order feature interactions among the non-linear relationship vectors to obtain respective interaction embedding vectors.

[0103] Among them, high-order feature interaction can be a complex combination of multiple features. By combining and interacting with the original simple low-order features, deeper dependencies between them can be captured.

[0104] Among them, the interaction embedding vector can be a new feature vector obtained by performing high-order feature interactions among multiple features.

[0105] Specifically, a feature interaction layer (such as a multi-layer perceptron, a deep neural network, etc.) is introduced to capture the complex non-linear interactions between different features. In high-order feature interactions, each pair of non-linear relationship vectors will interact with other features through combination (such as weighted summation or concatenation) to form new and more expressive feature representations. The results of these interactions will generate interaction embedding vectors, which can better represent the complex relationships between the original features and provide richer information for subsequent processing.

[0106] Step 506: Adjust the adaptation range of each interaction embedding vector to obtain respective space embedding vectors.

[0107] Specifically, after high-order feature interactions, adjustment methods of regularization and constraint are then needed to adjust the adaptation range of the interaction embedding vectors so that they can have better generalization ability in different tasks or data scenarios. Through these adjustments, the interaction embedding vectors will be optimized into space embedding vectors, which can adapt to a wider range of scenarios and exhibit richer and more accurate data representations.

[0108] In this embodiment, by identifying non-linear relationship information from the cross-domain fusion vectors, more complex and deep patterns in the data can be captured, breaking through the limitations of linear relationships and enhancing the feature representation ability. Subsequently, through high-order feature interaction on the non-linear relationship vectors, the generated interaction embedding vectors effectively fuse the high-order dependency relationships among multiple features, further enhancing the understanding ability of complex data structures. On this basis, by adjusting the adaptation range of the interaction embedding vectors, the generated spatial embedding vectors can adaptively optimize the representation of the feature space, making the final vectors more accurately adapt to different data scenarios and requirements, thereby improving the generalization ability and expressiveness of the data in practical applications. Through this series of processes, the accuracy of the features can be greatly improved, providing more powerful support for data analysis and decision-making.

[0109] In an exemplary embodiment, as Figure 6 shown, identifying non-linear relationship information from each cross-domain fusion vector to obtain each non-linear relationship vector includes steps 602 to 604. Among them:

[0110] Step 602, perform high-order transformation on each cross-domain fusion vector to obtain each cross-domain high-order vector.

[0111] Among them, high-order transformation can be to expand and transform the original feature vectors through mathematical operations and feature combination methods to generate more complex feature representations.

[0112] Among them, cross-domain high-order vectors can be vector representations obtained by performing high-order transformation on data features. These vectors not only contain the basic features in the original data but also fuse various high-order combination relationships of cross-domain data.

[0113] Specifically, use polynomial feature expansion (Polynomial Feature Expansion) or neural network hierarchies (such as fully connected layers) to perform high-order mapping on each cross-domain fusion vector. The new cross-domain high-order vectors obtained will contain more feature combinations and can better represent the complex patterns in the data. For example, for two original feature vectors A and B, high-order transformation can be achieved by calculating the product of A and B (A×B) or other high-order combinations to obtain a richer feature expression.

[0114] Step 604, perform non-linear transformation on each cross-domain high-order vector to obtain each non-linear relationship vector.

[0115] Among them, non-linear transformation can be to apply non-linear functions (such as ReLU, Sigmoid, tanh, etc.) to transform the feature vectors, so that the non-linear relationships in the data can be revealed and captured.

[0116] Specifically, activation functions (such as ReLU, Sigmoid, tanh, etc.) are usually used to perform non-linear transformations on cross-domain high-order vectors. The activation function can change the distribution of the feature space, enabling the model to output more complex and non-linear feature interactions during the calculation process. For example, using the ReLU activation function will "suppress" the negative value part to zero, enabling the model to focus on capturing the positive relationships of features. After non-linear transformation, the original cross-domain high-order vectors will be transformed into non-linear relationship vectors, which can effectively express the complex interaction relationships in cross-domain data and provide rich information for subsequent learning and prediction tasks.

[0117] In this embodiment, by performing high-order transformation on the cross-domain fusion vector, more complex feature relationships can be mined, the feature space is expanded, and the depth and breadth of data expression are improved. This high-order transformation can capture the potential high-order dependencies between features, enhancing the understanding and learning ability of complex data patterns. Subsequently, by performing non-linear transformation on the cross-domain high-order vector, the non-linear relationships in the data can be further mined, thus breaking through the limitations of the linear model and obtaining more accurate feature representations. This non-linear transformation can identify and learn more complex patterns and associations, improving the data expression ability. This processing process helps to better understand and utilize the deep features in cross-domain data, enhancing the performance and generalization ability in practical applications.

[0118] In an exemplary embodiment, as Figure 7 shown, adjusting the adaptation range of each interaction embedding vector to obtain each spatial embedding vector includes steps 702 to 710. Among them:

[0119] Step 702, performing ridge regression processing on each interaction embedding vector to obtain each regression regularization vector.

[0120] Among them, the regression regularization vector can be a feature vector processed by applying a regression regularization method (such as ridge regression).

[0121] Specifically, each interaction embedding vector will be used as input for ridge regression processing. Ridge regression is a linear regression method that adds an L2 regularization term to the loss function to penalize large parameter values in the model, thereby reducing the risk of overfitting. Specifically, the interaction embedding vector is input into the ridge regression model, and the ridge regression model will adjust the weights of the vector by minimizing the loss function with L2 regularization, enabling the objective function to fit the data with less parameter variation, thereby obtaining the regression regularization vector, which is the result of regularizing the original interaction embedding vector.

[0122] Step 704, regularizing the domain of each regression regularization vector to obtain each domain regularization vector.

[0123] Among them, the domain regularization vector can be a vector obtained by further optimizing and adjusting the feature vector after regression regularization to adapt to the feature distributions of different domains.

[0124] Specifically, domain regularization processing is performed on each regression regularization vector after ridge regression regularization. The goal of domain regularization is to adjust each regression regularization vector according to the characteristics of different domains, so that the features are more adaptable among different domains. This usually includes introducing domain-specific regularization terms, which can be achieved by designing domain-specific loss functions or regularization strategies. For example, using domain adaptation techniques or adding domain-related weights to the loss function to constrain the feature vectors of each domain. These regularization strategies will adjust the feature representation according to the differences between domains, so that the final feature vector can reflect the special requirements of the domain, and the obtained vector is the domain regularization vector for each domain.

[0125] Step 706: Regularize the scenarios of each domain regularization vector to obtain each scenario regularization vector.

[0126] Among them, the scenario regularization vector can be a vector obtained by optimizing the feature vector after domain regularization processing to adapt to a specific application scenario. The purpose of scenario regularization is to adjust the feature vector so that it has better expressive ability and discrimination in a specific scenario.

[0127] Specifically, further scenario regularization is performed on each domain regularization vector. The purpose of scenario regularization is to ensure that the feature vector has good adaptability and consistency in different application scenarios, which is achieved by introducing scenario-specific regularization strategies. For example, constraints are imposed on scenario-specific features or data distributions, which can be achieved by adding penalty terms for different scenarios to the loss function, or by applying scenario adaptation methods for optimization. The goal of scenario regularization is to make the feature vector have better expressive ability in different scenarios, and be able to more accurately reflect the feature differences and similarities between scenarios. Finally, each scenario regularization vector is obtained, which can better represent the feature information in a specific scenario.

[0128] Step 708: Use the triplet loss to constrain the semantic consistency of each scenario regularization vector to obtain each semantic constraint regularization vector.

[0129] Among them, the semantic constraint regularization vector can be a feature vector obtained after applying the triplet loss, aiming to ensure that samples that are semantically similar are closer in the feature space, while samples that are semantically different are farther apart.

[0130] Specifically, a scene regularization vector is selected from the scene regularization vectors as an anchor point, which represents the features of the current scene. The selection of the anchor point is usually based on the requirements of the current task or randomly selected from the scene regularization vectors. Then, a positive sample with similar semantics to the anchor point and a negative sample with significantly different semantics from the anchor point are selected from the scene regularization vectors. Through the optimization of the triplet loss, the goal is to minimize the distance between the anchor point scene regularization vector and the positive sample scene vector, while ensuring that the distance between the anchor point scene regularization vector and the negative sample scene vector is as large as possible, that is, to make the scene regularization vectors with similar semantics closer, and the scene regularization vectors with different semantics farther away, so as to obtain the semantic constraint regularization vector after semantic consistency constraint.

[0131] Step 710, according to the actual application target information, perform reinforcement learning constraint on each semantic constraint regularization vector to obtain each spatial embedding vector.

[0132] Among them, the actual application target information can be the optimization target set according to specific tasks and requirements during the calculation process.

[0133] Specifically, according to the actual application target information, perform reinforcement learning constraint on the semantic constraint regularization vector. The core of reinforcement learning is to optimize the policy by interacting with the environment to maximize the predefined reward function. In the application of this step, reinforcement learning guides the adjustment of the feature vector by designing the reward function, and the reward function is usually closely related to the actual application target (such as classification accuracy, recommendation effect, etc.). In specific implementation, reinforcement learning algorithms (such as Q-learning, policy gradient, etc.) can be used to adjust the semantic constraint regularization vector to improve its effect in actual applications. The vector optimized by reinforcement learning can better adapt to the requirements of the task and finally obtain the spatial embedding vector.

[0134] In this embodiment, by performing ridge regression processing on the interactive embedding vector, the regularization of features can be effectively carried out, the risk of overfitting can be reduced, and it is ensured that it can be better generalized to unseen data. Then, through the gradual regularization processing of the domain and scene of the regression regularization vector, the expressiveness and robustness of the features are further optimized, and the adaptability and consistency in different domains and scenes are ensured. In this process, by using the triplet loss to constrain the semantic consistency of the scene regularization vector, it is ensured that the feature semantics in different scenes can be kept consistent, avoiding information loss or incorrect association. Finally, by performing reinforcement learning constraint on the semantic constraint regularization vector and further adjusting and optimizing the feature vector according to the actual application target, it is ensured that the finally obtained spatial embedding vector can accurately reflect the data features and achieve higher prediction accuracy and decision-making efficiency in actual applications.

[0135] In an exemplary embodiment, as Figure 8 shown, according to the current active transaction data, current interaction transaction data, and current passive transaction data, adjust the portrait output style of the fused labeled portrait to obtain the target labeled portrait, including steps 802 to 806. Among them:

[0136] Step 802, set the initial dynamic customization style information according to the current active transaction data, current interaction transaction data, and current passive transaction data.

[0137] Among them, the initial dynamic customization style information can be the setting information of the customized labeling image style obtained by analyzing data from different sources.

[0138] Specifically, extract key features from the input active transaction data (such as user active behaviors, such as clicks, searches, purchases, etc.), interaction transaction data (such as interaction data between the user and the system, such as browsing duration, click frequency, messages, etc.), and passive transaction data (such as user geographical location, device type, network environment, etc.). By analyzing information such as the trends, frequencies, and behavior patterns of the key feature data, generate the initial dynamic customization style information related to user behaviors and data patterns.

[0139] Step 804, reconstruct the image style of the fused labeled portrait according to the initial dynamic customization style information to obtain the reconstructed labeled image.

[0140] Among them, the reconstructed labeled image can be a new labeled image obtained after adjustment and transformation after applying the initial dynamic customization style information.

[0141] Specifically, reconstruct the original fused labeled portrait according to the extracted initial dynamic customization style information. In this process, the style of the fused labeled portrait (such as layout, color, font, graphic marks, etc.) will be adjusted targeted according to the characteristics of the current active transaction data, current interaction transaction data, and current passive transaction data. For example, if a certain user behavior is frequent, its characteristics may be highlighted by deepening the color, enlarging the icon, etc., and conversely, the prominence of relevant identifiers may be reduced. In addition, according to the differences in data sources, the presentation method of the image may be adapted to ensure that the image can accurately convey information in different scenarios, and obtain the reconstructed labeled image.

[0142] Step 806, optimize the image perception features of the reconstructed labeled image to obtain the target labeled portrait.

[0143] Among them, the optimization of image perception features can be achieved through a series of visual optimization techniques to improve the effect of the image at the user perception level. This process includes adjusting the contrast, brightness, color balance, element spacing, detail clarity, etc. of the image to make the image more visually coordinated, clear, and easy to understand.

[0144] Specifically, the reconstructed labeled image is further optimized in the direction of image perception features. The optimization process uses an image perception algorithm (such as optimization based on a visual perception model) to enhance the visual effect of the image, including adjusting various visual elements of the image, such as contrast, color balance, graphic spacing, layering of elements, edge clarity, etc., to make it more in line with the perception habits of the human eye, thereby improving the clarity and information transmission efficiency of the image. In addition, the display effect of the user on different devices is also considered during the optimization process, such as responsive design, to ensure that the image can maintain a good visual experience on various screen sizes and resolutions, and finally obtain a target labeled portrait that meets the user's needs.

[0145] In this embodiment, by combining the current active flow data, current interaction flow data, and current passive flow data, the set initial dynamic customization style information can generate personalized image style information according to the characteristics of different data sources. The application of this information provides precise guidance for the reconstruction of the fused labeled portrait, enabling the image style to better adapt to actual needs and scenario changes. Further, by optimizing the image perception features of the reconstructed labeled image, the visual expressiveness and information transmission effect of the image can be enhanced, ensuring the clarity and consistency of the image in every detail. The final obtained target labeled portrait can meet the requirements of actual applications in terms of accuracy and adaptability, helping to improve the user experience, optimize decision support, and increase the efficiency and accuracy of data-driven analysis.

[0146] In an exemplary embodiment, as Figure 9 shown, according to the initial dynamic customization style information, the image style of the fused labeled portrait is reconstructed to obtain a reconstructed labeled image, including steps 902 to 910. Among them:

[0147] Step 902, according to the initial dynamic customization style information, preliminarily reconstruct the image style of the fused labeled portrait to obtain an initial reconstructed image.

[0148] Among them, the initial reconstructed image can be the first reconstruction of the image style of the fused labeled portrait according to the set initial dynamic customization style information, providing a basic visual performance framework that reflects the preliminary mapping of data features.

[0149] Specifically, style parameters obtained by analyzing changes in user information (such as active, interactive, and passive streaming data) are extracted from the initial dynamic customization style information. These parameters may include colors, fonts, icons, layout structures, information display hierarchies, etc. Based on these preliminary style parameters, the visual style of the original fused labeled portrait is adjusted through design rules or algorithms. For example, the size of interactive elements may be increased or the color may be changed according to the user's interaction frequency to improve the visualization effect, or the layout of the image may be adjusted according to the user's activity status, and an initial reconstructed image is output. This image visually reflects the results of preliminary data features and style adjustments, providing a basis for subsequent optimization and fine-tuning.

[0150] Step 904, based on the initial reconstructed image, perform reverse reasoning on the original style information to obtain reverse dynamic customization style information.

[0151] Among them, the reverse dynamic customization style information can be the style information obtained by performing reverse reasoning on the initial reconstructed image.

[0152] Specifically, analyze the actual style of the initial reconstructed image, and extract the relationship between the original image style information and the reconstructed style through reverse reasoning methods, including quantitative analysis of each element in the initial reconstructed image (such as color matching, layout structure, font size and position, icon arrangement, etc.), and infer the design intention and data-driven mode behind the style from it. Reverse reasoning can reveal the mutual relationship between elements in image design, and how these visual adjustments respond to the input data features, obtaining reverse dynamic customization style information, which indicates which visual parameters need to be further optimized or adjusted after the preliminary style adjustment.

[0153] Step 906, based on the differences between the reverse dynamic customization style information and the initial dynamic customization style information, determine the style information difference data.

[0154] Among them, the style information difference data can be the quantitative result of the difference between the reverse dynamic customization style information and the initial dynamic customization style information, which clarifies the specific direction and gap of the style adjustment.

[0155] Specifically, by comparing the specific differences between the reverse dynamic customization style information and the initial dynamic customization style information, the system can identify the changes and deviations that occurred during the image reconstruction process, such as which visual elements such as colors, icon sizes, layout positions, etc. are different. The sources of these differences may include changes in data source errors, changes in user behavior pattern data errors, or style design itself adjustment errors, etc. This difference data helps to correct the errors in style changes and provides specific target information for further optimization of the style. The finally obtained style information difference data provides clear guidance and adjustment basis for subsequent customized style optimization.

[0156] Step 908: Set the optimized dynamic customized style information based on the style information difference data, the current active transaction data, the current interactive transaction data, and the current passive transaction data.

[0157] Among them, the optimized dynamic customized style information can be an optimization scheme generated by combining the style information difference data with the current transaction data flow information (such as active, interactive, and passive transaction data), aiming to perform more refined customization and adjustment of the image style to ensure that the image can better adapt to different application scenarios and requirements.

[0158] Specifically, combine the style information difference data with the current active transaction data, the current interactive transaction data, and the current passive transaction data to perform more refined setting of the image style. Analyze different difference information in the style information difference data, and the style setting algorithm adjusts the logic and method of setting the style by combining the style information difference data, the current active transaction data, the current interactive transaction data, and the current passive transaction data, so that the re-output customized data can better express the characteristics of the current data, generate the optimized dynamic customized style information, ensure that the image better matches the user's needs and data characteristics in visual design, and enhance the user experience.

[0159] Step 910: Perform secondary reconstruction on the image style of the fused labeled portrait according to the optimized dynamic customized style information to obtain the reconstructed labeled image.

[0160] Specifically, use the already generated optimized dynamic customized style information to perform secondary reconstruction on the original fused labeled portrait. The goal of secondary reconstruction is to eliminate the style errors found in the previous stage and apply the optimization requirements to the visual design of the image to ensure that the image can accurately reflect the changes in the data and can effectively convey information visually. Specific operations include further optimizing the color combination, element layout, font size, hierarchy of identifiers, etc. of the image to improve the visibility and readability of the image. The image after secondary reconstruction will more accurately meet the target requirements, and finally obtain the reconstructed labeled image. This image not only has an optimized effect visually but also can accurately display the characteristics and laws of the data, meeting the requirements of practical applications.

[0161] In this embodiment, by initially reconstructing the image style of the fused labeled portrait according to the initial dynamic customization style information, a balance can be found between personalized requirements and data characteristics, and a basic image style framework can be generated. Then, by performing reverse inference on the initially reconstructed image, reverse dynamic customization style information is obtained, further deeply exploring the potential rules of the image style and the data characteristics reflected, so as to provide more reference basis for style optimization. Based on the differences between the reverse dynamic customization style information and the initial style information, the differential data in the style can be accurately identified, and thus the image style can be refined. On this basis, by combining the current active flow data, the current interactive flow data, and the current passive flow data, the optimized dynamic customization style information is set, and through further optimization of the style, it is ensured that the generated reconstructed labeled image more accurately reflects the requirements and scenario changes of the actual application. The whole process helps to generate a highly customized labeled image that meets the actual requirements, improve the accuracy, adaptability, and visual effect of the image expression, and optimize the performance of the decision support system.

[0162] It should be understood that although the steps in the flowcharts involved in the above embodiments are shown in sequence according to the arrows, these steps do not necessarily have to be executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above embodiments may include multiple steps or multiple stages, and these steps or stages do not necessarily have to be executed at the same time, but can be executed at different times, and the execution order of these steps or stages does not necessarily have to be sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0163] Based on the same inventive concept, an embodiment of the present application also provides a portrait labeling device based on flow data for implementing the portrait labeling method based on flow data involved above. The implementation solution provided by this device to solve the problem is similar to the implementation solution described in the above method. Therefore, the specific limitations in one or more embodiments of the portrait labeling device based on flow data provided below can refer to the limitations on the portrait labeling method based on flow data in the above text, and will not be repeated here.

[0164] In an exemplary embodiment, as Figure 10 shown, a portrait labeling device based on flow data is provided, including: a labeled image generation module 1002, a labeled image update module 1004, an image data fusion module 1006, and an image style optimization module 1008, where:

[0165] The marking image generation module 1002 is used to perform multi-level time series analysis on the historical active flow data, historical interaction flow data, and historical passive flow data of the target object to generate an initial marking portrait;

[0166] The marking image update module 1004 is used to update the features and labels of the initial marking portrait according to the current active flow data, current interaction flow data, and current passive flow data of the target object to obtain the current marking portrait;

[0167] The image data fusion module 1006 is used to map the flow data of each classification scenario, the flow data of each classification field, and the current marking portrait of the target object into the same embedding space to obtain a fused marking portrait;

[0168] The image style optimization module 1008 is used to adjust the portrait output style of the fused marking portrait according to the current active flow data, current interaction flow data, and current passive flow data to obtain the target marking portrait.

[0169] In one embodiment, the image data fusion module 1006 is further used to perform cross-domain feature fusion on the flow data of each classification scenario and the flow data of each classification field to obtain cross-domain fusion vectors; identify non-linear relationship information from the cross-domain fusion vectors to obtain spatial embedding vectors; map the spatial embedding vectors and the current marking portrait into the same embedding space to obtain a fused marking portrait.

[0170] In one embodiment, the image data fusion module 1006 is further used to extract features from the flow data of each classification scenario and the flow data of each classification field to obtain flow feature vectors of each classification scenario and flow feature vectors of each classification field; traverse the mutual relationship between the flow feature vectors of each classification scenario and the flow feature vectors of each classification field to obtain a classification feature vector relationship network; mutually transfer the feature information of the flow feature vectors of each classification scenario and the flow feature vectors of each classification field in the classification feature vector relationship network whose vector relationship closeness value is greater than a preset closeness threshold to obtain cross-domain fusion vectors.

[0171] In one embodiment, the image data fusion module 1006 is further used to identify non-linear relationship information from the cross-domain fusion vectors to obtain non-linear relationship vectors; perform high-order feature interaction between the non-linear relationship vectors to obtain interaction embedding vectors; adjust the adaptation range of the interaction embedding vectors to obtain spatial embedding vectors.

[0172] In one embodiment, the image data fusion module 1006 is further used to perform high-order conversion on the cross-domain fusion vectors to obtain cross-domain high-order vectors; perform non-linear conversion on the cross-domain high-order vectors to obtain non-linear relationship vectors.

[0173] In one embodiment, the image data fusion module 1006 is further configured to perform ridge regression processing on each interactive embedding vector to obtain each regression regularization vector; regularize the neighborhood of each regression regularization vector to obtain each neighborhood regularization vector; regularize the scenario of each neighborhood regularization vector to obtain each scenario regularization vector; use the triplet loss to constrain the semantic consistency of each scenario regularization vector to obtain each semantic constraint regularization vector; and perform reinforcement learning constraint on each semantic constraint regularization vector according to the actual application target information to obtain each spatial embedding vector.

[0174] In one embodiment, the image style optimization module 1008 is further configured to set initial dynamic customization style information according to the current active flow data, the current interactive flow data, and the current passive flow data; reconstruct the image style of the fused labeled portrait according to the initial dynamic customization style information to obtain a reconstructed labeled image; and optimize the image perception features of the reconstructed labeled image to obtain a target labeled portrait.

[0175] In one embodiment, the image style optimization module 1008 is further configured to perform preliminary reconstruction on the image style of the fused labeled portrait according to the initial dynamic customization style information to obtain an initial reconstructed image; perform reverse inference on the original style information according to the initial reconstructed image to obtain reverse dynamic customization style information; determine style information difference data according to the difference between the reverse dynamic customization style information and the initial dynamic customization style information; set optimized dynamic customization style information according to the style information difference data, the current active flow data, the current interactive flow data, and the current passive flow data; and perform secondary reconstruction on the image style of the fused labeled portrait according to the optimized dynamic customization style information to obtain a reconstructed labeled image.

[0176] Each module in the above portrait labeling device based on flow data can be implemented in whole or in part by software, hardware, and their combination. Each of the above modules can be embedded in the processor of the computer device in hardware form or be independent of it, or can be stored in the memory of the computer device in software form so that the processor can call and execute the operations corresponding to each of the above modules.

[0177] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 11As shown in the figure. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store server data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a portrait marking method based on streaming data.

[0178] Those skilled in the art can understand that Figure 11 the structure shown in the figure is only a block diagram of some structures related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine some components, or have different component arrangements.

[0179] In one embodiment, a computer device is further provided, including a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps in the above method embodiments are implemented.

[0180] In one embodiment, a computer-readable storage medium is provided, storing a computer program. When the computer program is executed by the processor, the steps in the above method embodiments are implemented.

[0181] In one embodiment, a computer program product or a computer program is provided. The computer program product or the computer program includes computer instructions, and the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the steps in the above method embodiments.

[0182] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0183] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random access memories (ReRAM), magnetoresistive random access memories (MRAM), ferroelectric random access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, etc., without limitation.

[0184] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0185] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the appended claims.

Claims

1. A portrait marking method based on flow data, characterized in that: The method comprises: Perform multi-level time series analysis on the historical active flow data, historical interactive flow data, and historical passive flow data of the target object to generate an initial labeling portrait; According to the current active pipeline data, the current interactive pipeline data and the current passive pipeline data of the target object, the features and labels of the initial marking portrait are updated to obtain the current marking portrait; Mapping the flow data of each classification scene, the flow data of each classification field and the current marking portrait of the target object into the same embedding space to obtain a fused marking portrait; According to the current active pipeline data, the current interactive pipeline data and the current passive pipeline data, the image output style of the fused marking image is adjusted to obtain a target marking image; The step of adjusting the image output style of the fused marking image according to the current active pipeline data, the current interactive pipeline data, and the current passive pipeline data to obtain the target marking image includes: Setting initial dynamic customized style information according to the current active flow data, the current interactive flow data and the current passive flow data; Reconstructing the image style of the fused marking image according to the initial dynamic customized style information to obtain a reconstructed marking image; Performing image perception feature optimization on the reconstructed marking image to obtain the target marking portrait; The step of reconstructing the image style of the fused marking image according to the initial dynamic customized style information to obtain the reconstructed marking image includes: Preliminarily reconstructing the image style of the fused marked image according to the initial dynamically customized style information to obtain an initial reconstructed image; According to the initial reconstructed image, reverse reasoning is performed on the original style information to obtain reverse dynamic customized style information; Determining style information difference data according to a difference between the reverse dynamically customized style information and the initial dynamically customized style information; According to the style information difference data, the current active flow data, the current interactive flow data and the current passive flow data, setting and optimizing the dynamic customized style information; According to the optimized dynamically customized style information, the image style of the fused marking image is reconstructed to obtain the reconstructed marking image.

2. The method according to claim 1, characterized in that The step of mapping the classification scene flow data, classification field flow data and the current marking image of the target object into the same embedding space to obtain a fused marking image includes: Perform cross-domain feature fusion on the pipeline data of each classification scene and the pipeline data of each classification field to obtain each cross-domain fusion vector; Identifying nonlinear relationship information from each of the cross-domain fusion vectors to obtain each spatial embedding vector; Each of the spatial embedding vectors and the current marking image are mapped into the same embedding space to obtain the fused marking image.

3. The method according to claim 2, characterized in that The cross-domain feature fusion of each of the classification scene pipeline data and each of the classification field pipeline data to obtain each cross-domain fusion vector includes: Perform feature extraction on the pipeline data of each classification scene and the pipeline data of each classification field to obtain a pipeline feature vector of each classification scene and a pipeline feature vector of each classification field; Traversing the mutual relationship between the flow feature vectors of each classification scene and the flow feature vectors of each classification field to obtain a classification feature vector relationship network; The feature information of each of the classification scene flow feature vectors and each of the classification field flow feature vectors whose vector relationship closeness value in the classification feature vector relationship network is greater than a preset closeness threshold is mutually migrated to obtain each of the cross-field fusion vectors.

4. The method according to claim 2, characterized in that: The step of identifying nonlinear relationship information from each of the cross-domain fusion vectors to obtain each spatial embedding vector includes: Identifying nonlinear relationship information from each of the cross-domain fusion vectors to obtain each nonlinear relationship vector; Performing high-order feature interactions on the nonlinear relationship vectors to obtain interaction embedding vectors; The adaptation range of each of the interactive embedding vectors is adjusted to obtain each of the spatial embedding vectors.

5. The method according to claim 4, characterized in that The step of identifying nonlinear relationship information from each of the cross-domain fusion vectors to obtain each nonlinear relationship vector includes: Performing high-order transformation on each of the cross-domain fusion vectors to obtain each cross-domain high-order vector; Nonlinear transformation is performed on each of the cross-domain high-order vectors to obtain each of the nonlinear relationship vectors.

6. The method according to claim 4, characterized in that The step of adjusting the adaptation range of each of the interactive embedding vectors to obtain each of the spatial embedding vectors includes: Performing ridge regression processing on each of the interactive embedding vectors to obtain each regression regularization vector; Regularizing the domain of each regression regularization vector to obtain a regularization vector for each domain; Regularizing the scenes of the regularized vectors of the fields to obtain regularized vectors of the scenes; Using a triplet loss to constrain the semantic consistency of each of the scene regularization vectors to obtain each semantically constrained regularization vector; According to the actual application target information, each of the semantic constraint regularization vectors is subjected to reinforcement learning constraints to obtain each of the spatial embedding vectors.

7. A portrait marking device based on flow data, characterized in that: The device comprises: The marking image generation module is used to perform multi-level time series analysis on the historical active flow data, historical interactive flow data, and historical passive flow data of the target object to generate an initial marking image; A marking image updating module, used for updating the features and labels of the initial marking image according to the current active pipeline data, the current interactive pipeline data and the current passive pipeline data of the target object to obtain a current marking image; An image data fusion module, used to map the pipeline data of each classification scene, the pipeline data of each classification field and the current marking image of the target object into the same embedding space to obtain a fused marking image; An image style optimization module, used to adjust the image output style of the fused marking image according to the current active pipeline data, the current interactive pipeline data and the current passive pipeline data to obtain a target marking image; The step of adjusting the image output style of the fused marking image according to the current active pipeline data, the current interactive pipeline data, and the current passive pipeline data to obtain the target marking image includes: Setting initial dynamic customized style information according to the current active flow data, the current interactive flow data and the current passive flow data; Reconstructing the image style of the fused marking image according to the initial dynamic customized style information to obtain a reconstructed marking image; Performing image perception feature optimization on the reconstructed marking image to obtain the target marking portrait; The step of reconstructing the image style of the fused marking image according to the initial dynamic customized style information to obtain the reconstructed marking image includes: Preliminarily reconstructing the image style of the fused marked image according to the initial dynamically customized style information to obtain an initial reconstructed image; According to the initial reconstructed image, reverse reasoning is performed on the original style information to obtain reverse dynamic customized style information; Determining style information difference data according to a difference between the reverse dynamically customized style information and the initial dynamically customized style information; According to the style information difference data, the current active flow data, the current interactive flow data and the current passive flow data, setting and optimizing the dynamic customized style information; According to the optimized dynamically customized style information, the image style of the fused marking image is reconstructed to obtain the reconstructed marking image.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

Images