Data processing method and device based on big data and advertisement pushing

Abstract

The invention relates to a data processing method and device based on big data and advertisement pushing, and the method comprises the following steps: obtaining the historical behavior data of a user on a multi-channel platform, constructing a dynamic interest label map according to the historical behavior data, and depicting a user interest evolution process. Combining with a social relation network to analyze an interest propagation path, forming a user social interest diffusion trajectory, and introducing a time decay weighting mechanism to generate a dynamic interest decay curve. According to the method, user interests and advertisement materials are subjected to semantic similarity matching, a personalized advertisement recommendation list is generated, an optimal advertisement putting strategy is determined through multi-target optimization configuration and comprehensive consideration of display positions, opportunities and forms, accurate and efficient advertisement pushing is achieved, and the problems that a traditional user portrait method often depends on a static label system, and the user experience is poor are solved. The dynamic characteristic that the user interest changes along with time is difficult to reflect, so that the advertisement recommendation content lags behind the real intention of the user.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of big data, and particularly relates to a data processing method and device based on big data and advertisement pushing. BACKGROUND

[0002] With the rapid development of Internet technology and the popularity of intelligent terminals, the behavior data of users on various digital platforms presents an explosive growth. These multi-source heterogeneous behavior data contain rich user interest and demand information, which makes it possible to provide precise advertisement pushing. However, how to efficiently integrate and analyze these data to extract valuable user features is still one of the core challenges in the current big data application field. The traditional user portrait method often relies on a static label system and is difficult to reflect the dynamic characteristics of the change of user interest over time, resulting in that the recommended content of the advertisement lags behind the real intention of the user.

[0003] At the same time, the wide application of social networks makes the influence propagation between users an important factor affecting the consumption decision. Although the existing advertisement recommendation system starts to pay attention to the social attributes of users, it mostly stays at the shallow relationship mining level and lacks in-depth modeling of the social interest diffusion path and influence of users. This limitation makes it difficult for the advertisement pushing to accurately capture the potential interest propagation trend between users, thereby reducing the relevance and conversion rate of the advertisement content. In addition, the user interest has obvious timeliness characteristics, and the traditional recommendation algorithm ignores the interest decay mechanism, which easily causes the advertisement content to be out of touch with the current state of the user.

[0004] More complex is that, in the actual advertisement launching process, in addition to the content matching degree, the display position, timing and form and other multiple target optimization problems need to be considered comprehensively. At present, most systems still adopt a single target optimization strategy or simply fuse multiple targets through linear weighting when dealing with this problem, and lack of a unified framework for collaborative optimization mechanism. Therefore, how to build an advertisement recommendation and launching system that integrates user interest modeling, social influence analysis and timeliness adjustment has become the key to improving the efficiency of digital advertising and user experience. SUMMARY

[0005] The main purpose of the present application is to provide a data processing method based on big data and advertisement pushing, which solves the technical problem that the traditional user portrait method often relies on a static label system and is difficult to reflect the dynamic characteristics of the change of user interest over time, resulting in that the recommended content of the advertisement lags behind the real intention of the user.

[0006] To achieve the above purpose, the present application provides a data processing method based on big data and advertisement pushing, comprising the following steps: acquiring the historical behavior data of a target user group on multiple channel platforms; modeling interest preferences of a target user group based on the historical behavior data to obtain a dynamic interest label graph; performing social influence propagation analysis on the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory; performing time decay weighting on the user social interest diffusion trajectory to obtain a dynamic interest decay curve; performing semantic similarity sorting on a preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertisement recommendation list, and performing multi-target optimization configuration based on the personalized advertisement recommendation list to obtain an advertisement delivery strategy; wherein the advertisement delivery strategy includes a display position, a display time, and a display form.

[0007] Further, the historical behavior data of the target user group on the multi-channel platform includes: performing timestamp serialization processing on the interactive interface operation event stream of the target user group on the multi-channel platform to obtain a standardized behavior log set; performing semantic classification mapping on the behavior event types in the standardized behavior log set to obtain historical behavior data; wherein the historical behavior data includes browsing records, click behavior, purchase records, dwell time, page jump path, and social interaction information.

[0008] Further, the modeling of interest preferences of a target user group based on the historical behavior data to obtain a dynamic interest label graph includes the following steps: extracting multi-modal behavior features from the historical behavior data to obtain a user behavior semantic embedding vector sequence, and performing time series autoregressive analysis on the user behavior semantic embedding vector sequence to obtain a user behavior pattern primitive, wherein the user behavior pattern primitive includes a periodic behavior pattern, a burst behavior pattern, and a sustained attention pattern; based on the user behavior pattern primitive, performing dynamic topic modeling on the user behavior semantic embedding vector sequence to obtain a user interest topic evolution trajectory, and performing topological structuring processing on the user interest topic evolution trajectory to obtain a user interest topic network graph; performing node embedding and edge weight learning on the user interest topic network graph through a preset graph neural network mechanism to obtain a user interest label relationship matrix, and counting the label co-occurrence frequency in the user interest label relationship matrix to obtain a label collaborative filtering matrix; based on the label collaborative filtering matrix, performing label clustering and hierarchical organization on the user behavior pattern primitive to obtain a dynamic interest label graph.

[0009] Further, the dynamic interest label graph is used to analyze social influence propagation of the target user group, and a user social interest diffusion trajectory is obtained, including: The topology influence structure of the dynamic interest label graph is analyzed to obtain a label influence distribution graph, and node centrality calculation is performed based on the label influence distribution graph to obtain key label nodes, wherein the key label nodes include high-influence labels, bridge labels, and emerging labels. Based on the key label nodes, multi-level community discovery is performed on the social network structure of the target user group to obtain an interest community hierarchical structure, and boundary permeability analysis is performed on the interest community hierarchical structure to obtain a cross-group interest propagation channel graph; wherein the cross-group interest propagation channel graph includes strong connection propagation paths, weak connection propagation paths, and potential propagation barriers. The information flow of the cross-group interest propagation channel graph is evaluated by information entropy theory to generate an interest propagation dynamics model, and Monte Carlo simulation is performed based on the interest propagation dynamics model to obtain an interest diffusion probability field. Based on the interest diffusion probability field, the spatiotemporal mapping of individual interest evolution processes in the target user group is performed to obtain an interest trajectory vector sequence, and nonlinear time series analysis is performed on the interest trajectory vector sequence to obtain a user social interest diffusion trajectory, wherein the user social interest diffusion trajectory includes interest diffusion rate, diffusion direction change, and diffusion range boundary.

[0010] Further, the node centrality calculation based on the label influence distribution graph obtains key label nodes, including: The node neighborhood density of the label influence distribution graph is calculated to obtain a label node connectivity matrix, and the local clustering coefficient of the label node connectivity matrix is analyzed to obtain a node clustering feature vector, wherein the node clustering feature vector includes direct connection strength, indirect connection density, and node group contact degree. The node clustering feature vector is divided into subgraphs by a spectral clustering method to obtain a label community structure graph, and cross-group connection analysis is performed based on the label community structure graph to obtain a bridge node vector, wherein the bridge node vector includes cross-group connection number, information flow rate, and bridge strength coefficient. The bridge node vector is fused with feature weights to obtain a node importance score table, and hierarchical sorting is performed based on the node importance score table to obtain a node influence level sequence, wherein the node influence level sequence includes core node marking, key path weight, and influence range coefficient. The node influence level sequence is threshold filtered to obtain key label nodes.

[0011] Further, the node neighborhood density calculation is performed on the label influence distribution diagram to obtain a label node connection degree matrix, comprising: The k-order neighborhood range is defined on the label influence distribution diagram to obtain a node adjacency relationship table, and the path distance of each node in the node adjacency relationship table is counted to obtain a node reachability matrix; The edge weight of the node reachability matrix is calculated through a multi-layer perception mechanism to obtain an edge connection strength vector, and connectivity analysis is performed based on the edge connection strength vector to obtain a node connectivity distribution diagram, wherein the node connectivity distribution diagram comprises the number of connected components, the maximum connected subgraph and the position identification of the cut point. The local density evaluation is performed on the node connectivity distribution diagram to obtain a node density distribution feature, and the regional aggregation degree analysis is performed based on the node density distribution feature to obtain a regional density gradient diagram. The neighborhood connection strength fusion is performed based on the regional density gradient diagram to obtain a label node connection degree matrix.

[0012] Further, the semantic similarity sorting is performed on the preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertisement recommendation list, comprising: The multi-modal content deconstruction and feature tensorization processing are performed on the advertisement material library to obtain an advertisement content feature tensor, and the sparse coding and manifold embedding are performed on the advertisement content feature tensor to obtain an advertisement semantic representation space. The trajectory segment decomposition and feature point extraction are performed on the user social interest diffusion trajectory through a spatio-temporal fusion mechanism to obtain an interest diffusion feature sequence, and the interest diffusion feature sequence is time-effectiveness weighted and decay compensated based on the dynamic interest decay curve to obtain a time-effectiveness modulated interest vector field. The cross-space mapping and similarity measurement are performed on the time-effectiveness modulated interest vector field based on the advertisement semantic representation space to obtain an advertisement-interest matching degree tensor, and the comprehensive score calculation is performed on the advertisement-interest matching degree tensor through multi-objective Pareto optimization to obtain an advertisement priority ranking vector, and the personalized advertisement recommendation list is constructed based on the advertisement priority ranking vector.

[0013] The application also provides a data processing device based on big data and advertisement pushing, comprising: An acquisition module is configured to acquire historical behavior data of a target user group on a multi-channel platform; A modeling module is configured to model interest preferences of the target user group based on the historical behavior data to obtain a dynamic interest label graph; An analysis module is configured to perform social influence propagation analysis on the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory. a weighting module configured to perform time decay weighting on the user social interest diffusion trajectory to obtain a dynamic interest decay curve; a sorting module configured to perform semantic similarity sorting on a preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertisement recommendation list, and perform multi-objective optimization configuration based on the personalized advertisement recommendation list to obtain an advertisement delivery strategy; wherein the advertisement delivery strategy comprises a display position, a display timing and a display form.

[0014] The application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of the preceding embodiments when executing the computer program.

[0015] The application further provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steps of the method according to any one of the preceding embodiments.

[0016] The application provides a data processing method based on big data and advertisement pushing, comprising the following steps: obtaining historical behavior data of a target user group on a multi-channel platform; performing interest preference modeling on the target user group based on the historical behavior data to obtain a dynamic interest label graph; performing social influence propagation analysis on the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory; performing time decay weighting on the user social interest diffusion trajectory to obtain a dynamic interest decay curve; performing semantic similarity sorting on a preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertisement recommendation list, and performing multi-objective optimization configuration based on the personalized advertisement recommendation list to obtain an advertisement delivery strategy. The technical problem that a traditional user portrait method often relies on a static label system and is difficult to reflect the dynamic characteristics of user interest changes over time, resulting in that an advertisement recommendation content lags behind a user's real intention is solved, the technical effect that a time decay weighting mechanism is introduced into a user social interest diffusion trajectory to construct a dynamic interest decay curve, the intensity of user interest changes over time can be scientifically measured, interference on a recommendation result caused by long-term inactive behavior is avoided, and the relevance and response rate of advertisement recommendation are improved is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a step schematic diagram of the data processing method based on big data and advertisement pushing in an embodiment of the application; Figure 2 is a structural block diagram of a data processing device based on big data and advertisement pushing in an embodiment of the application; Figure 3Fig. 1 is a structural schematic block diagram of a computer device according to an embodiment of the present application.

[0018] The purposes, functional features and advantages of the present application will be further illustrated in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0019] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0020] As shown in Fig. 1, the present application is a computer device according to an embodiment of the present application. Figure 1 Figure 1 is a data processing method based on big data and advertisement pushing in an embodiment of the present application, comprising the following steps: Step S1, obtaining historical behavior data of a target user group on a multi-channel platform.

[0021] Specifically, in the step of "obtaining historical behavior data of a target user group on a multi-channel platform", the core is to collect the past behavior records of the target user group from multiple sources and various types of digital platforms through a systematic data collection mechanism. These behavior data can include but are not limited to user browsing tracks, click behaviors, dwell time, interactive operations, search keywords and purchase records, etc., covering the digital activity tracks of users in different scenarios. In order to ensure that the obtained data can accurately reflect the interest preferences of users, a unified data format and label system must be used to standardize the behavior data from different platforms, and data cleaning techniques are used to remove noise and outliers, so as to improve the accuracy of subsequent modeling. For example, in an application scenario of e-commerce advertisement pushing, the system can collect the access frequency, collection behavior and order record of users on the product page from multiple channels such as mobile applications, web pages and small programs, so as to build a complete and time-dimensioned user behavior sequence, providing basic data support for subsequent interest modeling.

[0022] Step S2, interest preference modeling of the target user group based on the historical behavior data, obtaining a dynamic interest label graph.

[0023] ​Specifically, the interest preference modeling of the target user group based on the historical behavior data to obtain a dynamic interest label graph refers to, on the basis of the historical behavior data of users on the existing multi-channel platform, identifying and extracting the interest features of the users at different times and in different scenarios through data analysis and machine learning technology, and constructing a structured expression in the form of labels, so as to form a dynamic interest label system that can be updated with time and behavior changes. The label graph not only contains the long-term stable interest tendency of the user, but also reflects the short-term fluctuating interest hotspots, so that the user portrait is more detailed and real-time. For example, in the application scenario of e-commerce advertisement pushing, the system can automatically label the user with interest labels such as “sports shoes enthusiast” and “maternity and baby product follower” through time series analysis of the user's behaviors such as browsing goods, adding to shopping cart, collecting, commenting, etc. in mobile applications, web pages and other channels, combined with the understanding of search keywords through natural language processing technology, and dynamically adjusting the label weight according to the latest behavior, and then constructing a dynamic interest label graph with time dimension and hierarchical structure. This process ensures that the subsequent social influence propagation analysis and advertisement recommendation sorting have higher accuracy and timeliness.

[0024] Step S3, performing social influence propagation analysis on the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory.

[0025] Specifically, the social influence propagation analysis of the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory refers to, on the basis of the constructed dynamic interest label graph, combining the social relationship network structure between users, modeling the propagation path and influence strength of the interest label in the social chain, identifying the role and influence range of different users in the interest propagation, and thus depicting the social diffusion process of the interest preference. This process usually relies on graph computing algorithm and social network analysis technology, uses the interaction behaviors (such as likes, forwards, comments, etc.) between users to quantify the propagation direction and attenuation law of the interest information, and introduces the time dimension into the propagation path to capture the dynamic evolution of the interest diffusion from the initial node to the surrounding users. For example, in the application scenario of e-commerce advertisement pushing, when a user is labeled as “sports shoes enthusiast” because of browsing and purchasing a pair of sports shoes, the system can track whether the interest label is propagated to the user's social circle through the user's likes, shares or purchase behaviors through the social friend relationship network, and record how the interest is propagated to the second-level and third-level users at different time nodes, and finally form a visual user social interest diffusion trajectory, which provides a basis for subsequent time decay weighting processing.

[0026] Step S4, performing time decay weighting on the user social interest diffusion trajectory to obtain a dynamic interest decay curve.

[0027] Specifically, the user social interest diffusion trajectory is time-decay weighted to obtain a dynamic interest decay curve, which means that on the basis of the existing user social interest diffusion trajectory, the influence strength in the interest propagation process is weighted and adjusted by introducing the time factor to reflect the natural law that the interest gradually weakens over time. In specific implementation, the system will use the timestamp information of each node (i.e. user) in the interest propagation path to receive the interest label, use mathematical models such as exponential decay function or Gaussian decay function to calculate the interest influence value of different time nodes, and arrange these decayed influence values in time sequence to form a dynamic interest decay curve that can reflect the trend of interest change. This process not only retains the path characteristics of interest propagation, but also enhances the prediction ability of the model for the current interest state of the user. For example, in the application scenario of e-commerce advertisement pushing, if a user temporarily showed interest in "outdoor camping equipment" because of a friend's sharing a week ago, but has not generated any related behavior since then, the system will automatically reduce the weight of the interest label according to the time decay function, thereby avoiding misjudgment of the outdated interest as the current preference, and ensuring that the subsequent advertisement recommendation is more in line with the actual needs of the user.

[0028] Step S5, based on the user social interest diffusion trajectory and the dynamic interest decay curve, the preset advertisement material library is sorted in terms of semantic similarity to obtain a personalized advertisement recommendation list, and based on the personalized advertisement recommendation list, a multi-objective optimization configuration is performed to obtain an advertisement delivery strategy; wherein the advertisement delivery strategy includes display position, display time and display form.

[0029] Specifically, based on the user social interest diffusion trajectory and the dynamic interest decay curve, the preset advertisement material library is subjected to semantic similarity sorting to obtain a personalized advertisement recommendation list, and based on the personalized advertisement recommendation list, multi-objective optimization configuration is performed to obtain an advertisement delivery strategy. It means that on the basis of completing user interest modeling and social communication analysis, the interest state of the user is matched and calculated with the content of the advertisement at the semantic level. Specifically, the system will use natural language processing technology or deep learning model to perform vector space mapping on the advertisement copy, label description in the advertisement material library and the current interest features of the user (including social communication path and time decay influence), and evaluate the correlation between the advertisement content and the user interest through cosine similarity algorithm and other algorithms, so as to generate a personalized advertisement recommendation list sorted by relevance. On this basis, a multi-objective optimization algorithm is further introduced, and multiple optimization objectives such as visibility of advertisement display position, user activity of display opportunity and interactive friendliness of display form are comprehensively considered, and under the premise of meeting the platform traffic distribution rules and the budget constraints of the advertiser, the optimal advertisement delivery strategy is generated. For example, in the application scene of e-commerce advertisement pushing, if a user shows high attention to "sports shoes" due to the influence of social friends in recent period, but the interest heat decreases with time, the system will preferentially recommend sports advertisement materials with high semantic matching degree, and according to the user browsing habit, the advertisement will be displayed in the form of combination of text and picture on the home page of the APP in the evening, so as to improve the click rate and conversion effect of the advertisement.

[0030] In specific embodiments, the historical behavior data of the target user group on the multi-channel platform includes: The interaction interface operation event stream of the target user group on the multi-channel platform is subjected to timestamp serialization processing to obtain a standardized behavior log set. The behavior event types in the standardized behavior log set are subjected to semantic classification mapping to obtain historical behavior data; wherein the historical behavior data includes browsing records, click behavior, purchase records, stay time, page jump path and social interaction information.

[0031] Specifically, the step of "obtaining historical behavior data of the target user group on the multi-channel platform" specifically includes two core processing procedures: one is to perform timestamp serialization processing on the interactive interface operation event stream of the target user group on the multi-channel platform to obtain a standardized behavior log set; and the other is to perform semantic classification mapping on the behavior event types in the standardized behavior log set, thereby forming structured historical behavior data. This step aims to unify the originally scattered and heterogeneous user behavior records into analyzable data assets to support subsequent interest modeling and advertisement recommendation processes. In specific implementation, the system first collects the event stream generated by the user in the interface interaction process from multiple channels (such as mobile applications, web pages, applets, etc.), such as clicking buttons, sliding pages, browsing product detail pages, etc., and marks each event with an accurate timestamp, sorts and serializes them according to the occurrence order, and forms a standardized behavior log set with time dimension. Subsequently, the system maps these original events to more business-meaningful behavior categories according to a pre-set behavior semantic classification system, such as "browsing records", "clicking behaviors", "purchase records", "dwell time", "page jump paths", and "social interaction information", etc., thereby generating historical behavior data that can be used for modeling. For example, in the application scenario of e-commerce advertisement pushing, when the user views the product detail page of a pair of sports shoes on the mobile terminal APP, clicks to add to the shopping cart, finally completes the order, and shares the product link on the social platform and gets friend likes, etc., these behaviors will be collected as operation event stream, which is converted into standard log after timestamp serialization, and identified as specific "browsing", "clicking", "purchasing" and "social interaction" behaviors through semantic mapping, and then a complete and structured user behavior portrait is constructed to provide high-quality data input for subsequent interest preference modeling.

[0032] In specific embodiments, the interest preference modeling of the target user group based on the historical behavior data to obtain a dynamic interest label graph includes the following steps: extracting multi-modal behavior features in the historical behavior data to obtain a user behavior semantic embedding vector sequence, and performing time series autoregressive analysis on the user behavior semantic embedding vector sequence to obtain a user behavior pattern primitive, wherein the user behavior pattern primitive includes a periodic behavior pattern, a burst behavior pattern, and a sustained attention pattern; based on the user behavior pattern primitive, dynamically modeling the user behavior semantic embedding vector sequence to obtain a user interest theme evolution trajectory, and topologically structuring the user interest theme evolution trajectory to obtain a user interest theme network graph; The user interest topic network graph is subjected to node embedding and edge weight learning through a preset graph neural network mechanism to obtain a user interest label relationship matrix, and label co-occurrence frequencies in the user interest label relationship matrix are counted to obtain a label collaborative filtering matrix; The user behavior pattern primitives are subjected to label clustering and hierarchical organization based on the label collaborative filtering matrix to obtain a dynamic interest label graph.

[0033] Specifically, the step of "modeling interest preferences of a target user group based on the historical behavior data to obtain a dynamic interest label graph" is based on the user historical behavior data after acquisition and structured processing, and further constructs a dynamic label system that can reflect the evolution law of user interest through deep learning and graph computing technology. This process specifically includes multiple sub-steps, including extracting multi-modal features from user behavior sequences, identifying behavior pattern primitives, establishing interest theme evolution trajectories, and finally forming a dynamic interest label graph with topological structure and semantic association. First, the system extracts multi-modal behavior features from historical behavior data, including not only explicit behaviors such as user clicks, views, and purchases, but also implicit behavior information such as page dwell time, jump path, and interaction frequency. By vectorizing these behaviors, a sequence of user behavior semantic embedding vectors is generated, making different behaviors comparable and clusterable in high-dimensional space. Subsequently, the system uses time series autoregressive analysis to model these embedding vector sequences and identify three behavior pattern primitives: periodicity, suddenness, and persistence. For example, in the context of e-commerce advertising, if a user regularly browses baby products every week, their behavior exhibits periodicity; if they suddenly click on a brand's products in large quantities due to a promotion, it is a sudden behavior; and if they consistently focus on a certain type of sports equipment, it indicates a persistent interest pattern. Next, the system models the user behavior semantic embedding vector sequence based on the identified behavior pattern primitives to capture the trend of user interest evolution over time. This process typically uses time series modeling techniques such as LSTM or Transformer, combined with topic models such as Latent Dirichlet Allocation (LDA), to obtain user interest theme evolution trajectories. For example, a user may initially focus on "running shoes" and gradually expand to "outdoor hiking equipment" and then to "camping tents," reflecting the process of interest migration. To enhance the model's expressive power, the system further topologically structures the interest theme evolution trajectories, organizing them into an interest theme network graph with node connection relationships, where each node represents an interest theme and the edges represent the evolution or correlation strength between themes. Based on this, the system introduces a graph neural network mechanism to learn the user interest theme network graph in depth, specifically embedding the nodes in the graph and optimizing the weights of the edges to more accurately depict the relevance and influence paths between interest themes. In this way, the system can generate a user interest label relationship matrix, where each row / column corresponds to an interest label, and the matrix values reflect the correlation strength between labels. At the same time, the system also counts the frequency of label co-occurrence to construct a label collaborative filtering matrix, which measures the probability of different interest labels appearing together in the user group, and further mines potential interest combination patterns.Finally, the system performs label clustering and hierarchical organization on the user behavior pattern primitives based on the label collaborative filtering matrix, aggregates similar interest labels into higher-level interest categories, and divides the levels according to the evolution path of user interest, thereby constructing a dynamic interest label graph with dynamic updating capability. For example, in the e-commerce scenario, labels such as “sports shoes”, “hiking poles”, and “jackets” can be classified into the “outdoor sports” category, while “baby strollers”, “milk powder”, and “diapers” form the “baby products” category. The system can continuously adjust the label attribution and weight according to the latest user behavior, realizing real-time updating and personalized customization of the interest graph. This dynamic interest label graph provides a fine and time-sensitive interest representation basis for subsequent social influence propagation analysis and advertisement recommendation.

[0034] In specific embodiments, the social influence propagation analysis of the target user group based on the dynamic interest label graph includes: analyzing the topological influence structure of the dynamic interest label graph to obtain a label influence distribution map, and performing node centrality calculation based on the label influence distribution map to obtain key label nodes, wherein the key label nodes include high-influence labels, bridge labels, and emerging labels; based on the key label nodes, performing multi-level community discovery on the social network structure of the target user group to obtain an interest community hierarchical structure, and performing boundary permeability analysis on the interest community hierarchical structure to obtain a cross-group interest propagation channel map; wherein the cross-group interest propagation channel map includes strong connection propagation paths, weak connection propagation paths, and potential propagation barriers; evaluating the information flow of the cross-group interest propagation channel map through information entropy theory to generate an interest propagation dynamics model, and performing Monte Carlo simulation based on the interest propagation dynamics model to obtain an interest diffusion probability field; based on the interest diffusion probability field, performing spatiotemporal mapping on the individual interest evolution process in the target user group to obtain an interest trajectory vector sequence, and performing nonlinear time series analysis on the interest trajectory vector sequence to obtain a user social interest diffusion trajectory, wherein the user social interest diffusion trajectory includes interest diffusion rate, diffusion direction change, and diffusion range boundary.

[0035] Specifically, the step of "carrying out social influence propagation analysis on the target user group based on the dynamic interest label graph to obtain a user social interest diffusion trajectory" is to further mine the propagation path, influence range and evolution trend of interest in the social chain based on the constructed dynamic interest label graph and in combination with the social network structure of the user. Through multi-level graph analysis and modeling technology, this process converts static interest labels into social interest diffusion trajectories with dynamic propagation characteristics. First, the system analyzes the topological influence structure of the dynamic interest label graph to identify key label nodes with different roles in the entire interest network. Specifically, the system evaluates the influence distribution of each label node in the entire graph through the centrality calculation method in graph theory (such as degree centrality, closeness centrality, betweenness centrality, etc.), thereby generating a label influence distribution graph. For example, in the application scenario of e-commerce advertisement pushing, the label "sports shoes" may be identified as a high-influence label due to its high frequency of occurrence and extensive connectivity; "outdoor camping" may be determined as a bridge label due to its location at the intersection of multiple interest categories; and a new label such as "smart running shoes" that has rapidly rising recent attention may be identified as a new label. These key label nodes provide a basis for further analyzing interest propagation paths. Subsequently, the system performs multi-level community discovery on the social network structure of the target user group based on these key label nodes. Through graph clustering algorithms (such as Louvain algorithm or spectral clustering), users are divided into several interest communities, and a hierarchical structure is formed according to interest similarity and interaction frequency. For example, in an e-commerce platform, a parent-child user community centered on "baby products" and a fitness enthusiast community centered on "sports equipment" may be identified. On this basis, the system further performs boundary permeability analysis on the interest community hierarchical structure to identify the possibility of information flow between different interest communities, thereby constructing a cross-community interest propagation channel graph. This graph contains three types of propagation paths: strong connection propagation paths (such as frequent interactions between friends within the same community), weak connection propagation paths (such as occasional likes or shares between communities), and potential propagation barriers (such as information blocking areas caused by too large interest differences). For example, if a fitness user shares a link to a protein powder product with a friend who is not in the fitness circle, and the friend clicks and purchases it, it indicates that there is an effective weak connection propagation path; otherwise, it may constitute a propagation barrier. Next, the system uses information entropy theory to quantitatively evaluate the information flow of the cross-community interest propagation channel graph and constructs an interest propagation dynamics model. The higher the information entropy, the more active the interest propagation on the path; otherwise, the propagation is limited. The system simulates the dynamics model multiple times through the Monte Carlo simulation method to simulate the diffusion process of interest in the social network, and finally generates an interest diffusion probability field.For example, in a simulation process, assuming a user shares a purchase link of a "mountain climbing backpack", the system can predict that about 35% of his friends will view the link within 24 hours, 15% will add it to the shopping cart, and 8% will complete the purchase, thus forming a specific diffusion probability distribution. Finally, the system maps the individual interest evolution process of the target user group based on the interest diffusion probability field. By embedding the user's interest state over time into the spatial dimension, an interest trajectory vector sequence is formed, and nonlinear time series analysis methods such as Hurst index analysis or wavelet transform are used to extract the core features of interest diffusion, including interest diffusion rate (growth rate of the number of affected users per unit time), diffusion direction change (path deviation of interest from one field to related or unrelated fields), and diffusion range boundary (the maximum user set that interest can reach). For example, in a promotion activity, the "jacket" interest tag affected 100 core users in the first hour after the activity started, 500 secondary users in the second hour, and 1200 edge users in the third hour, showing a trend of gradually slowing down the diffusion rate; at the same time, some users shifted their focus from "jacket" to "tent", "sleeping bag" and other related products, reflecting the direction of interest diffusion. In summary, by analyzing the social influence propagation of dynamic interest tag graphs, the system not only identifies the propagation path and key nodes of interest in the network, but also accurately depicts how interest evolves in both time and space dimensions, ultimately forming a dynamic and interpretable user social interest diffusion trajectory, providing solid data support for subsequent time decay weighted processing and ad recommendation optimization.

[0036] In specific embodiments, the node centrality calculation based on the tag influence distribution map to obtain key label nodes includes: The node neighborhood density calculation is performed on the tag influence distribution map to obtain a tag node connectivity matrix, and the local clustering coefficient of the tag node connectivity matrix is analyzed to obtain a node clustering feature vector, wherein the node clustering feature vector includes direct connection strength, indirect connection density, and node group connection degree; The node clustering feature vector is divided into subgraphs by a spectral clustering method to obtain a tag community structure graph, and cross-group connection analysis is performed based on the tag community structure graph to obtain a bridge node vector, wherein the bridge node vector includes cross-group connection number, information flow rate, and bridge strength coefficient; The bridge node vector is fused with feature weights to obtain a node importance score table, and hierarchical sorting is performed based on the node importance score table to obtain a node influence level sequence, wherein the node influence level sequence includes core node label, key path weight, and influence range coefficient; perform threshold screening based on the node influence level sequence to obtain a key label node.

[0037] Specifically, the step of "calculating the node centrality based on the label influence distribution diagram to obtain the key label node" is to further identify the label nodes with special influence in the interest propagation process through graph structure analysis and machine learning methods based on the constructed dynamic interest label graph. This process models multiple indicators such as local clustering characteristics, cross-group bridging ability, and comprehensive importance score of the label node connection matrix, and finally filters out key label nodes with high influence, strong connectivity, and extensive propagation range. In specific implementation, the system first calculates the neighborhood density of each label node in the label influence distribution diagram, i.e. counts the number of directly connected neighbor nodes and the closeness of their connection, to generate a label node connection matrix. On this basis, the system further analyzes the local clustering coefficient in the matrix to measure the ability of a label node to form a closed loop connection within its neighborhood range, and then extracts the node clustering feature vector. This vector contains three core dimensions: one is the direct connection strength, i.e. the number of edges between the node and its neighbors; two is the indirect connection density, reflecting the density of nodes that the node can influence through two-hop or three-hop paths; three is the node group connection degree, describing whether the node is between multiple subgraphs and has the potential to connect across communities. For example, in the application scenario of e-commerce advertising pushing, the "sports shoes" label may have up to 200 direct connection nodes (such as related brands, user reviews, and shopping behavior), and be associated with other interest topics through more than 500 indirect paths, while having a high group connection degree, indicating that it is not only active within the interest circle, but also has strong cross-domain influence. Subsequently, the system uses spectral clustering method to divide the above node clustering feature vector into subgraphs, i.e. divides the entire label graph into several label community structure graphs with similar clustering characteristics. This process helps to identify relatively independent but internally tightly connected interest clusters in the interest network, such as "maternal and infant products", "outdoor sports", "digital electronics", etc. After completing the construction of the label community structure graph, the system further analyzes the cross-group connection and identifies the bridge nodes that connect different interest communities, and generates a bridge node vector. This vector consists of three key indicators: one is the number of cross-group connections, i.e. the number of different interest communities connected by the node; two is the information flow rate, indicating the speed of the node in transferring interest information between different communities; three is the bridge strength coefficient, used to measure the stability and efficiency of the node as an intermediary in interest propagation. For example, in a promotion activity, the "smart watch" label may connect multiple interest communities such as "health monitoring", "exercise tracking", "fashion accessories", etc., and its information flow rate reaches 30 messages per hour per node, with a bridge strength coefficient of 0.78, indicating that it plays a highly efficient information transfer role in interest diffusion. Next, the system weights and fuses the features in the bridge node vector to generate a node importance score table.The score combines multiple dimensions such as direct connection strength, indirect connection density, inter-group connectivity, cross-group connection number, information flow rate, and bridge strength coefficient. Through a pre-set weight distribution mechanism (such as principal component analysis or expert scoring method), each feature is mapped to the same scale space, and finally a sortable node influence level sequence is formed. This sequence not only reflects the importance of each label node in the entire interest network, but also reflects its role in the interest propagation path. For example, in the e-commerce scenario, "baby milk powder" may be marked as a core node due to its high frequency and extensive connectivity, with an influence level score of 92 (out of 100); while "camping tent" has fewer connected nodes, but due to its bridging of multiple outdoor interest subgroups, the influence range coefficient is higher, with a score of 84, which is a node with a large weight in the key path. Finally, the system sets a threshold based on the node influence level sequence to filter the label nodes and extract key label nodes that meet specific criteria. These nodes usually include high-influence labels (such as scores higher than 90), bridge labels (cross-group connection number greater than 3), and emerging labels (score growth rate more than 10%). For example, during a certain promotional activity, the "smart running shoes" label was identified as an emerging label due to the addition of a large number of connected nodes in a short period of time, and the score increased from 75 to 88; while "hiking poles" were determined to be a bridge label due to their connection to the mother and baby, fitness, and outdoor interest communities. These key label nodes will become the core objects of subsequent social influence propagation analysis, providing key support for constructing user social interest diffusion trajectories.

[0038] In specific embodiments, the node neighborhood density calculation on the label influence distribution map to obtain a label node connection degree matrix includes: Defining the k-order neighborhood range of the label influence distribution map to obtain a node adjacency relationship table, and counting the path distance of each node in the node adjacency relationship table to obtain a node reachability matrix; Calculating the edge weight of the node reachability matrix through a multi-layer perception mechanism to obtain an edge connection strength vector, and performing connectivity analysis based on the edge connection strength vector to obtain a node connectivity distribution map, wherein the node connectivity distribution map includes the number of connected components, the maximum connected subgraph, and the identification of cut points; Performing local density evaluation on the node connectivity distribution map to obtain node density distribution characteristics, and performing regional aggregation degree analysis based on the node density distribution characteristics to obtain a regional density gradient map; Fusing neighborhood connection strength based on the regional density gradient map to obtain a label node connection degree matrix.

[0039] Specifically, the step of "performing node neighborhood density calculation on the label influence distribution graph to obtain a label node connectivity matrix" is one of the basic links in the process of identifying key nodes in the dynamic interest label graph. This process aims to quantify the connection tightness of each label node in its local neighborhood through graph structure analysis, thereby providing support for subsequent clustering coefficient calculation, bridge node identification, and node importance scoring. In specific implementation, the system first defines the k-order neighborhood range based on the label influence distribution graph, i.e., sets a reasonable propagation depth (such as k=2) to determine the set of other nodes that each label node can influence within this neighborhood range, and generates a node adjacency relationship table accordingly. For example, in the application scenario of e-commerce advertisement pushing, the "sports shoes" label node may be directly connected to "running", "fitness", "outdoor equipment" and other first-level neighbor nodes, and these neighbors are connected to several second-level nodes, forming a neighborhood network containing multiple levels. Subsequently, the system counts the path distances between nodes in the adjacency relationship table to construct a node reachability matrix, where each item represents whether there is a reachable path between two label nodes and the path length. For example, if the shortest path between "sports shoes" and "hiking pole" is 2 hops, it means that there is an indirect connection between them. On this basis, the system introduces a multi-layer perception mechanism to evaluate the weight of each edge in the node reachability matrix. This mechanism usually combines user behavior intensity (such as click frequency, forwarding times) and semantic relevance (such as co-occurrence frequency between labels) to comprehensively judge the connection strength of the edge, thereby generating an edge connection strength vector. For example, a certain edge from "sports shoes" to "running APP" has been triggered 1200 times of user behavior in the past 30 days, and the semantic similarity score is 0.85, so its connection strength value will be significantly higher than that of other edges. The system further performs connectivity analysis based on the edge connection strength vector to identify the number of connected components, the range of the largest connected subgraph, and the location of key cut points in the entire graph, and finally generates a node connectivity distribution graph. For example, in a certain e-commerce scenario, the system finds that there are 7 connected components in the entire label graph, and the largest connected subgraph contains multiple highly interconnected interest topics such as "baby products", "children's education", and "parent-child activities", while the "milk powder" label is identified as a cut point because its failure may cause multiple subgraphs to disconnect. Next, the system performs local density evaluation on the node connectivity distribution graph, i.e., measures the local connection density of each label node in its connected subgraph. This evaluation method usually uses K-nearest neighbor algorithm or local clustering coefficient calculation to obtain the number of neighbor nodes around each node and their connection tightness, thereby generating node density distribution features.For example, the "smart watch" tag may have up to 45 high-connection-strength neighbor nodes within its neighborhood range, indicating that it is in an area with a high interest density; while the "baby stroller" has fewer connected nodes, but the connections between its neighbors are more closely connected, so it also has certain local density advantages. Subsequently, the system analyzes the regional aggregation based on the node density distribution characteristics, identifies the local areas with high-density connections in the entire tag atlas, and generates a regional density gradient map. This atlas can reflect the density change trend between different interest themes, helping to identify core interest areas, transition interest areas, and edge interest areas. For example, in a promotion activity, the system found that the "outdoor camping" area has the highest density, surrounded by "tent", "sleeping bag", "hiking stick" and other strongly associated tags, forming an obvious interest hotspot area; while "pet food" is in a relatively isolated low-density area, indicating that it has weak connections with other interest themes. Finally, the system fuses the neighborhood connection strength of each tag node based on the above regional density gradient map, considering factors such as local connection density, adjacency edge weight, and regional aggregation, and finally generates a tag node connectivity matrix. This matrix not only reflects the connection breadth of each tag node in the graph, but also reflects its connection quality and structural stability. For example, the "sports shoes" tag has a score of 92 points (out of 100) in the connectivity matrix, indicating that it not only has extensive connections, but also has stable and high-density connection paths; while "children's picture books" have more connected nodes, but due to the low density of the area they are in, the connection strength is generally low, so the score is only 68. This connectivity matrix provides a solid data foundation for subsequent clustering coefficient calculation, bridge node identification, and node importance sorting, enabling the system to more accurately identify tag nodes that play a key role in interest propagation.

[0040] In specific embodiments, the personalized advertisement recommendation list is obtained by performing semantic similarity sorting on a preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve, including: The advertisement material library is subjected to multi-modal content deconstruction and feature tensorization processing to obtain an advertisement content feature tensor, and the advertisement content feature tensor is subjected to sparse coding and manifold embedding to obtain an advertisement semantic representation space; The user social interest diffusion trajectory is subjected to trajectory segment decomposition and feature point extraction through a spatio-temporal fusion mechanism to obtain an interest diffusion feature sequence, and the interest diffusion feature sequence is subjected to time-effectiveness weighting and decay compensation based on the dynamic interest decay curve to obtain a time-effectiveness modulated interest vector field; The time-limited interest modulation vector field is mapped across spaces and similarity is measured based on the advertisement semantic representation space to obtain an advertisement-interest matching degree tensor, and the advertisement-interest matching degree tensor is comprehensively scored and calculated through multi-objective Pareto optimization to obtain an advertisement priority ranking vector, and a personalized advertisement recommendation list is constructed based on the advertisement priority ranking vector.

[0041] Specifically, the step of "performing semantic similarity sorting on the preset advertisement material library based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertisement recommendation list" is the core link of precise advertisement recommendation in the entire data processing process. Through the fusion of user social behavior evolution characteristics and time decay mechanism, combined with the multi-modal semantic expression of advertisement content, a highly personalized and dynamically adjusted advertisement matching and sorting system is constructed. In the specific implementation process, the system first performs multi-modal content deconstruction and feature tensorization processing on the preset advertisement material library, that is, it extracts structured features for different modal contents such as text description, image elements, video segments, and audio information in the advertisement, and unifies them into a high-dimensional tensor form to form an advertisement content feature tensor. For example, in the application scenario of e-commerce advertisement pushing, an advertisement about "smart sports watch" may contain product introduction text, wearing effect picture, product function short video, and user evaluation voice information, etc. The system encodes these information into corresponding feature vectors respectively, and combines them into a unified tensor structure in a higher dimension for subsequent analysis and use. Subsequently, the system performs sparse coding and manifold embedding processing on the advertisement content feature tensor to compress redundant information and retain key semantic features, and finally generates an advertisement semantic representation space. Each advertisement in this space corresponds to a low-dimensional but semantically rich embedding vector, so that the semantic relationship between advertisements can be measured by the distance between vectors. For example, the system can map 100,000 advertisements into a 512-dimensional semantic space, in which the distance between "running shoes" and "sports water bottle" advertisements is relatively close, while the distance between them and "baby milk powder" advertisements is significantly increased, reflecting the difference in semantic relevance. At the same time, the system models and analyzes the user social interest diffusion trajectory through a spatio-temporal fusion mechanism to extract interest evolution features with time sequence characteristics. Specifically, the system divides the user interest diffusion path into multiple trajectory segments and extracts key feature points (such as interest outbreak time, propagation path change node, etc.) from each segment to form an interest diffusion feature sequence. On this basis, the system introduces a dynamic interest decay curve as a time weight function to weight and decay the interest diffusion feature sequence, generating a time-efficiency modulated interest vector field. For example, a user first paid attention to "mountain backpack" due to a friend's sharing three days ago, but browsed the same kind of commodity page again two days ago, and clicked on the related advertisement link today. The interest intensity shows a trend of first rising, then falling, and then rising again with time, and the system will weight the interest intensity of different time nodes according to the dynamic decay curve to ensure that the current interest state is accurately captured.Next, the system performs cross-space mapping on the time-modulated interest vector field based on the constructed advertisement semantic representation space, that is, compares the user interest feature vector with the advertisement semantic vector one by one, and calculates the matching degree between the advertisement and the user interest by using cosine similarity, Euclidean distance or a deep matching network, to finally generate an advertisement-interest matching degree tensor. The tensor not only reflects the relevance of the advertisement content and the current interest of the user, but also comprehensively considers the influence of the social communication path and the time decay factor. For example, the similarity score of the "mountain climbing backpack" advertisement with a user interest vector is 0.87, while the "children's scooter" advertisement is only 0.32, indicating that the former is more in line with the user's interest preference. Finally, the system calculates the comprehensive score of the advertisement-interest matching degree tensor by using a multi-objective Pareto optimization algorithm, that is, on the basis of ensuring that the advertisement content and the user interest are highly matched, multiple optimization objectives such as the advertisement principal budget allocation, platform display resource constraints, advertisement exposure frequency restrictions, etc. are considered, to find an optimal balance between multiple objectives. For example, within a certain advertisement delivery period, the system needs to select the advertisement combination that can best improve the click-through rate and conversion rate under the premise of meeting the daily maximum display of 1000 times for A brand and covering the core user group for B brand. After Pareto optimization, the system generates an advertisement priority ranking vector containing the top 50 advertisements, and constructs a personalized advertisement recommendation list accordingly. In summary, this step realizes the high personalization and intelligence of advertisement recommendation by deeply integrating the advertisement content semantic expression, the user social interest evolution path and the time decay regulation mechanism. In the actual application of e-commerce advertisement pushing, this recommendation mechanism based on dynamic interest modeling and social communication analysis can significantly improve the advertisement click-through rate, user satisfaction and advertiser ROI, and provides strong support for digital marketing.

[0042] The data processing method based on big data and advertisement pushing in the embodiment of the application is described above, and the data processing system based on big data and advertisement pushing in the embodiment of the application is described below. Please refer to Figure 2 The data processing system based on big data and advertisement pushing in the embodiment of the application includes one embodiment: The acquisition module 21 is configured to acquire historical behavior data of a target user group on a multi-channel platform. The modeling module 22 is configured to model the interest preferences of the target user group based on the historical behavior data, to obtain a dynamic interest label graph. The analysis module 23 is configured to perform social influence propagation analysis on the target user group based on the dynamic interest label graph, to obtain a user social interest diffusion trajectory. The weighting module 24 is configured to perform time decay weighting on the user social interest diffusion trajectory, to obtain a dynamic interest decay curve. The sorting module 25 is used to sort the preset advertising material library by semantic similarity based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertising recommendation list, and perform multi-objective optimization configuration based on the personalized advertising recommendation list to obtain an advertising delivery strategy; wherein the advertising delivery strategy includes display location, display timing and display form.

[0043] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0044] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design 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 computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0045] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0046] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0047] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods 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, and when executed, can include the processes of the above-mentioned embodiment methods. Any reference to memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM, etc.

[0048] It should be noted that in this document, the terms "comprising", "including", or any other variant thereof are intended to cover a non-exclusive inclusion, such that a process, device, article or method that comprises a list of elements does not only include those elements, but can also include other elements not expressly listed or inherent to such process, device, article or method. Without more limitations, the element defined by the statement "comprising a" does not exclude the presence of additional identical elements in the process, device, article or method that includes the element.

[0049] The above description is only the preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation, or direct or indirect application in other related technical fields, based on the content of the present application specification and drawings, are also included in the patent protection scope of the present application.

Claims

1. A data processing method based on big data and advertising push, characterized in that: The following steps are involved: Obtain historical behavior data of target user groups on multi-channel platforms; Based on the historical behavior data, interest preference modeling is performed on the target user group to obtain a dynamic interest tag map; Performing social influence propagation analysis on the target user group based on the dynamic interest tag graph to obtain a user social interest diffusion trajectory; Performing time decay weighting on the user's social interest diffusion trajectory to obtain a dynamic interest decay curve; Based on the user social interest diffusion trajectory and the dynamic interest decay curve, the preset advertising material library is sorted by semantic similarity to obtain a personalized advertising recommendation list, and multi-objective optimization configuration is performed based on the personalized advertising recommendation list to obtain an advertising delivery strategy; wherein, the advertising delivery strategy includes display location, display timing and display format.

2. The data processing method based on big data and advertising push according to claim 1, characterized in that: The acquisition of historical behavior data of the target user group on the multi-channel platform includes: Perform timestamp serialization on the target user group's interactive interface operation event streams on multi-channel platforms to obtain a standardized behavior log set; Perform semantic classification mapping on the behavioral event types in the standardized behavioral log set to obtain historical behavioral data; wherein, the historical behavioral data includes browsing history, click behavior, purchase history, stay time, page jump path and social interaction information.

3. The data processing method based on big data and advertising push according to claim 1, characterized in that: The method of performing interest preference modeling on the target user group based on the historical behavior data to obtain a dynamic interest tag map includes the following steps: Extracting multimodal behavior features from the historical behavior data to obtain a user behavior semantic embedding vector sequence, and performing time series autoregressive analysis on the user behavior semantic embedding vector sequence to obtain user behavior pattern primitives, wherein the user behavior pattern primitives include periodic behavior patterns, sudden behavior patterns, and sustained attention patterns; Performing dynamic topic modeling on the user behavior semantic embedding vector sequence based on the user behavior pattern primitives to obtain a user interest topic evolution trajectory, and performing topological structural processing on the user interest topic evolution trajectory to obtain a user interest topic network graph; Performing node embedding and edge weight learning on the user interest topic network graph through a preset graph neural network mechanism to obtain a user interest tag relationship matrix, and then counting the tag co-occurrence frequency in the user interest tag relationship matrix to obtain a tag collaborative filtering matrix; Based on the tag collaborative filtering matrix, the user behavior pattern primitives are clustered and hierarchically organized to obtain a dynamic interest tag map.

4. The data processing method based on big data and advertising push according to claim 1, characterized in that: The performing of social influence propagation analysis on the target user group based on the dynamic interest tag graph to obtain a user social interest diffusion trajectory includes: Analyze the topological influence structure of the dynamic interest tag graph to obtain a tag influence distribution map, and calculate the node centrality based on the tag influence distribution map to obtain key tag nodes, wherein the key tag nodes include high-influence tags, bridging tags, and emerging tags; Based on the key tag nodes, multi-level community discovery is performed on the social network structure of the target user group to obtain a hierarchical structure of interest communities, and boundary permeability analysis is performed on the hierarchical structure of interest communities to obtain a cross-group interest propagation channel graph; wherein the cross-group interest propagation channel graph includes strong connection propagation paths, weak connection propagation paths, and potential propagation barriers; Evaluate the information fluidity of the cross-group interest propagation channel graph using information entropy theory, generate an interest propagation dynamics model, and perform Monte Carlo simulation based on the interest propagation dynamics model to obtain an interest diffusion probability field; Based on the interest diffusion probability field, the individual interest evolution process in the target user group is mapped in time and space to obtain an interest trajectory vector sequence, and the interest trajectory vector sequence is subjected to nonlinear time series analysis to obtain the user social interest diffusion trajectory, wherein the user social interest diffusion trajectory includes the interest diffusion rate, the diffusion direction change and the diffusion range boundary.

5. The data processing method based on big data and advertising push according to claim 4 is characterized in that: The node centrality calculation based on the label influence distribution graph to obtain key label nodes includes: Calculating the node neighborhood density of the label influence distribution graph to obtain a label node connectivity matrix, and analyzing the local clustering coefficient of the label node connectivity matrix to obtain a node clustering feature vector, wherein the node clustering feature vector includes direct connection strength, indirect connection density, and node group connection degree; The node aggregation feature vector is sub-divided into subgraphs by a spectral clustering method to obtain a label community structure graph, and a cross-group connection analysis is performed based on the label community structure graph to obtain a bridging node vector, wherein the bridging node vector includes the number of cross-group connections, the information flow rate, and the bridging strength coefficient; Performing feature weight fusion on the bridge node vectors to obtain a node importance score table, and performing hierarchical sorting based on the node importance score table to obtain a node influence level sequence, wherein the node influence level sequence includes a core node label, a critical path weight, and an influence range coefficient; Threshold screening is performed based on the node influence level sequence to obtain key label nodes.

6. The data processing method based on big data and advertising push according to claim 5, characterized in that: The node neighborhood density calculation is performed on the label influence distribution graph to obtain a label node connectivity matrix, including: Defining the k-order neighborhood range of the label influence distribution map to obtain a node adjacency table, and counting the path distance of each node in the node adjacency table to obtain a node reachability matrix; Calculating the edge weights of the node reachability matrix through a multi-layer perception mechanism to obtain an edge connection strength vector, and performing connectivity analysis based on the edge connection strength vector to obtain a node connectivity distribution graph, wherein the node connectivity distribution graph includes the number of connected components, the largest connected subgraph, and cut point location identifiers; Performing local density evaluation on the node connectivity distribution map to obtain node density distribution characteristics, and performing regional aggregation analysis based on the node density distribution characteristics to obtain a regional density gradient map; Neighborhood connection strength fusion is performed based on the regional density gradient map to obtain a label node connectivity matrix.

7. The data processing method based on big data and advertising push according to claim 1, characterized in that: The step of sorting a preset advertising material library by semantic similarity based on the user's social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertising recommendation list includes: Performing multimodal content deconstruction and feature tensor quantization processing on the advertising material library to obtain an advertising content feature tensor, and performing sparse coding and manifold embedding on the advertising content feature tensor to obtain an advertising semantic representation space; The user's social interest diffusion trajectory is decomposed into segments and feature points are extracted through a spatiotemporal fusion mechanism to obtain an interest diffusion feature sequence, and the interest diffusion feature sequence is weighted and attenuated based on the dynamic interest attenuation curve to obtain a time-modulated interest vector field; Based on the advertising semantic representation space, the time-modulated interest vector field is cross-space mapped and similarity measured to obtain an advertising-interest matching tensor, and a comprehensive score calculation is performed on the advertising-interest matching tensor through multi-objective Pareto optimization to obtain an advertising priority ranking vector, and a personalized advertising recommendation list is constructed based on the advertising priority ranking vector.

8. A data processing device based on big data and advertising push, characterized in that: include: The acquisition module is used to obtain historical behavior data of the target user group on multi-channel platforms; A modeling module is used to perform interest preference modeling on the target user group based on the historical behavior data to obtain a dynamic interest tag map; An analysis module is used to perform social influence propagation analysis on the target user group based on the dynamic interest tag graph to obtain a user social interest diffusion trajectory; A weighting module, configured to perform time-attenuation weighting on the user's social interest diffusion trajectory to obtain a dynamic interest decay curve; A sorting module is used to sort the preset advertising material library by semantic similarity based on the user social interest diffusion trajectory and the dynamic interest decay curve to obtain a personalized advertising recommendation list, and perform multi-objective optimization configuration based on the personalized advertising recommendation list to obtain an advertising delivery strategy; wherein the advertising delivery strategy includes display location, display timing and display format.

9. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

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

Images