Digital delivery analysis method and platform based on multi-platform data integration

Abstract

The application provides a digital delivery analysis method and platform based on multi-platform data integration, which comprises the following steps: performing semantic mapping on user behavior data from two or more content platforms to generate unified behavior semantic labels; constructing a unified response vector of each user to the delivered content based on the labels; combining a user activity suppression mechanism and a time decay mechanism to generate a platform-neutral response vector of the delivered content; constructing a causal graph structure based on the response vector and a set of delivery variables to identify strategy variables that have a significant causal effect on the response to the content; and generating a structured delivery strategy suggestion by comprehensively considering the causal edge weight, historical response gain and implementation cost. The application realizes semantic alignment of cross-platform user behavior, unified quantification of content effect, causal identification of key influencing factors and automatic generation of strategies, and solves the problems of incomparable multi-platform data, strategy dependence on experience and lack of optimization basis in the prior art.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of data analysis, and particularly relates to a digital delivery analysis method and platform based on multi-platform data integration. BACKGROUND

[0002] In the current multi-platform digital marketing environment, brand parties face significant cross-platform data heterogeneity problems in content delivery, influencer selection and delivery effect evaluation. Natural differences such as short video and text platforms in content type, user interaction mode, recommendation link structure, etc. make it impossible to directly compare user behavior on different platforms. For example, short video platforms reflect user interest through completion ratio, likes and other instant behaviors, while text platforms use collection, long-time stay and message expansion for in-depth interaction as the main feedback. These differences make it difficult for marketing personnel to build a unified user reaction measurement system. At the same time, existing delivery analysis methods often stay at the static data presentation level of a single platform, and cannot achieve a structured, comparable and quantifiable user behavior abstraction method in a multi-platform environment. In addition, delivery effects are influenced by content structure, delivery time, influencer characteristics, platform attributes and other delivery factors. Existing systems cannot identify key variables that have a substantial impact from mixed delivery factors, and lack a causal modeling capability that can explain the influence path and strength between factors. Due to the lack of structured causal mechanisms, existing delivery optimization strategies still rely mainly on empirical judgment and manual screening, making it difficult to form targeted, interpretable, repeatable and logically based strategy adjustment paths. In actual delivery tasks, the Brief generation link is usually self-governed, lacks data support and structured methods, and different platforms have inconsistent strategies, which further exacerbates the waste of delivery resources and the risk of strategy deviation. Therefore, the lack of unified multi-platform behavior semantics, the lack of content response aggregation standards, the lack of structured causal identification mechanisms, and the inability of the strategy generation link to automatically link have become the core bottlenecks restricting digital delivery strategy optimization. SUMMARY

[0003] The purpose of the present application is to design a digital delivery analysis method and platform based on multi-platform data integration, which can realize unified expression of content effect, structured explanation of influencing factors and automatic generation of strategy configuration.

[0004] To achieve the above purpose, in the first aspect of the present application, a digital delivery analysis method based on multi-platform data integration is provided, which comprises:

[0005] Semantic mapping of user behavior data from two or more content platforms, mapping user behavior specific to each content platform into unified behavior semantic labels;

[0006] generating a unified response vector of each user to the launched content based on the behavioral semantic labels;

[0007] aggregating the unified response vectors of all users interacting with the same launched content, combining a user activity suppression mechanism and a time decay mechanism to generate a platform-neutral response vector of the launched content in a multi-platform environment;

[0008] constructing a causal graph structure based on the platform-neutral response vector and a set of launch variables corresponding to the launched content, and identifying a strategy variable having a significant causal impact on content response;

[0009] generating a structured launch strategy suggestion according to edge weight information in the causal graph structure, historical response gain, and strategy implementation cost.

[0010] Further, the behavioral semantic labels include shallow attention, emotional interaction, and deep identification, each corresponding to a three-dimensional vector representing user interest arousal intensity, emotional involvement, and conversion intention possibility.

[0011] Further, the unified response vector is obtained by assigning a preset weight to each behavioral semantic label and performing weighted summation, and the preset weight is set based on historical conversion performance.

[0012] Further, in the aggregation process of the platform-neutral response vector, the activity of each user is subjected to nonlinear compression processing to reduce the dominant influence of high-activity users on the aggregation result.

[0013] Further, a time decay factor is introduced in the aggregation process of the platform-neutral response vector, so that the closer the user response is to the launch time of the launched content, the higher the weight.

[0014] Further, when generating the platform-neutral response vector, if the response dispersion of the launched content exceeds a preset threshold, only the response vectors of the top-ranked users within the time window are retained to recalculate the aggregation result.

[0015] Further, the construction process of the causal graph structure introduces a structure sparsity constraint term and a time stability bias term to suppress false causal paths and improve the interpretability of the causal model.

[0016] Further, the strategy variable includes influencer fan level, content publishing time interval, platform type, and content structure category.

[0017] Further, the structured launch strategy suggestion includes recommended influencer level, recommended publishing time period, recommended platform type, and recommended content structure type, and can be directly called and executed by the launch system.

[0018] In a second aspect of the present application, a digital delivery analysis platform based on multi-platform data integration is provided, comprising:

[0019] a behavior semantic alignment module for mapping user behavior data of two or more content platforms into unified behavior semantic labels and generating a unified response vector of each user to the delivered content;

[0020] a response aggregation module for aggregating the unified response vectors of all users interacting with the same delivered content, combining a user activity suppression mechanism and a time decay mechanism, and generating a platform-neutral response vector of the delivered content in a multi-platform environment;

[0021] a causal modeling module for constructing a causal graph structure based on the platform-neutral response vector and a set of delivery variables corresponding to the delivered content, and identifying strategy variables having a significant causal impact on content response;

[0022] a strategy generation module for generating a structured delivery strategy suggestion according to edge weight information in the causal graph structure, historical response gain, and strategy implementation cost.

[0023] The present application has at least the following beneficial technical effects:

[0024] To solve the above problems, the present application provides a digital delivery analysis method and platform based on multi-platform data integration, which abstracts multi-platform user behavior into unified behavior semantic labels and further constructs content-level response representation based on the semantic structure, so that user reactions on different platforms are comparable and consistent across platforms. On this basis, by introducing a causal modeling method with structural sparsity constraints and time dynamic correction mechanisms, the system can identify key delivery variables that truly affect content performance and clearly define the path relationships and impact strength of these variables on content response. Further, by combining causal edge weight, response change patterns in historical scenarios, and delivery cost, the present application generates a structured and callable strategy suggestion set, realizes the automatic generation of strategy from analysis to configuration, and makes the delivery strategy no longer dependent on manual adjustment experience. The present application forms a complete closed-loop method system from behavior standardization, content response abstraction, causal chain identification to delivery strategy output, solves the traditional pain points of multi-platform data being unable to be uniformly interpreted, delivery factors being difficult to identify effective contributions, and strategy output lacking basis in a structural hierarchical manner, and provides an innovative framework with high interpretability, high consistency, and direct deployment for digital delivery. BRIEF DESCRIPTION OF DRAWINGS

[0025] The application is further illustrated by the accompanying drawings, but the embodiments in the drawings do not constitute any limitation to the application, and other drawings can be obtained by those of ordinary skill in the art without creative labor on the basis of the following drawings.

[0026] Figure 1 A flow chart of the digital delivery analysis method based on multi-platform data integration of the application.

[0027] Figure 2 A platform framework diagram of the digital delivery analysis platform based on multi-platform data integration of the application. DETAILED DESCRIPTION

[0028] The embodiments of the application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference signs represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the application, and cannot be understood as a limitation to the application.

[0029] In one or more embodiments, as shown in Figure 1 a digital delivery analysis method based on multi-platform data integration is disclosed, which comprises the following steps:

[0030] S1: performing semantic mapping on user behavior data from two or more content platforms, mapping user behaviors specific to each content platform into unified behavior semantic labels respectively; generating a unified response vector of each user to the delivered content based on the behavior semantic labels;

[0031] Specifically, this step is used to solve the problem of inconsistent user behavior data structure and semantic definition of different platforms. Current mainstream content platforms (such as short video platforms and text social platforms) have obvious differences in user interaction mode and indicators, for example, short video platforms often use "video completion rate" as a user interest intensity indicator, while text platforms pay more attention to "collection" or "long stay" behaviors. Without semantic normalization, it will not be possible to form a unified explanation of user behavior between multiple platforms, which will seriously restrict the effectiveness of subsequent strategy analysis and variable modeling. Therefore, this step performs unified semantic mapping on platform-specific user behavior data and further constructs a platform-neutral response vector, so that the reaction intensity of users to content has a unified measurement structure, which becomes the standard input for subsequent analysis and modeling.

[0032] The input received by this step is a set of user behavior data generated on a platform delivering content to the user, denoted as Behavior data is collected through data interfaces with each platform, such as the short video platform, which collects indicators including the completion ratio, the number of likes, and the number of comments, and the graphic platform, which collects indicators including the collection behavior, the page dwell time, and the forwarding behavior. All data is collected through platform-authorized interfaces, specifically, the user's interaction behavior on the launched content is pulled from the data interface channel at a fixed time every day, and after the data is stored in the database, it is processed by subsequent modules. The input data is not directly numerized, but first enters the semantic mapping process.

[0033] Platform-specific behaviors are first mapped to a set of unified behavior semantic labels, forming a semantic label set , where each label represents a semantic type of user behavior, such as "shallow attention", "emotional interaction", and "deep recognition". The semantic label is pre-set by a strategy expert and defined according to platform operation rules and user behavior intentions. For example, "completion ratio" is mapped to "deep browsing behavior", and "like behavior" is mapped to "emotional interaction behavior".

[0034] Subsequently, for each semantic label , a three-dimensional vector representation is obtained through a lookup table, denoted as , which represents the intensity of user interest, the degree of emotional involvement, and the possibility of content recognition or conversion intention. To integrate the overall contribution of different behaviors to user response, a weight coefficient is introduced, which represents the influence of behavior semantic label on the final response result. The weight coefficient is set based on historical conversion performance experience.

[0035] The calculation formula of the unified response vector is as follows:

[0036] ;

[0037] where represents the unified response vector of user to content , is the importance weight of label , and is the three-dimensional vector representation of label . The dimensions of each term remain consistent, so the structure of the left and right sides of the formula remains consistent.

[0038] For example, in a short video platform, a user watches a wedding decoration content completely and likes it, and the system identifies two labels, "deep browsing behavior" and "emotional interaction behavior". Assuming their vector representations are , , and the corresponding weights are , , then its response vector is:

[0039] ;

[0040] The response vector has three dimensions, respectively representing the intensity of interest arousal, the degree of emotional involvement, and the possibility of the user regarding the content as a conversion target. The vector structure is fixed, the semantics are stable, and it has the ability of cross-platform horizontal analysis.

[0041] S2: Aggregate the unified response vectors of all users who interact with the same content, combine the user activity suppression mechanism and the time decay mechanism, and generate a platform-neutral response vector of the content in the multi-platform environment;

[0042] Specifically, this step is responsible for connecting user-level behavior data and content-level response analysis. The output of the previous step is the response vector of the user on the specific content on each platform These vectors have completed the alignment between platforms in terms of behavior semantics. However, in actual delivery, the decision object is not the response of a single user, but the overall performance of the entire content on different platforms and different populations. Therefore, in order to achieve unified effect evaluation of the content, this step designs a content-level response representation with a differentiated aggregation mechanism, and integrates user activity suppression, response stability adjustment, and time weight modeling, etc. multiple dimensions of innovative structure for the reasonable integration of multi-platform response data.

[0043] The input of this step is the set of user-level response vectors , where each is a three-dimensional vector representing interest arousal, emotional involvement, and conversion intention. The vector set comes from the previous processing step, and all original user behavior data has been generated through semantic mapping and embedding, and is not processed again in this step. Another input is the behavior activity of each user in the statistical period , and the timestamp of the content being delivered , both of which come from the platform data log system. The activity is the number of all content behaviors participated by the user in the period, and the timestamp is the content delivery time recorded by the platform.

[0044] Considering that there may be a few high-frequency users dominating the results in the response sample, and the response data distribution is severely skewed, in order to improve the usability of the response vector and the explainability of the content effect, this step does not use the traditional weighted average method, but designs a response aggregation formula that integrates activity suppression and time-sensitive terms:

[0045] ;

[0046] in, Content The platform neutral response vector, Indicates user activity level Indicates user Reality and Content The timestamp of the interaction For content The timing of the release. This is the time decay factor, which controls the degree of influence of time distance on the response weight. Its value is typically [value missing]. Range. This formula is obtained through Non-linear compression of user activity is applied to prevent the behavior of super-active users from dominating the statistics and to ensure the representativeness of the response; and the following is introduced The time decay factor emphasizes that user reactions within a short time window after content delivery have a higher weight, which is closer to the "real audience's first perception" and enhances the timeliness of response evaluation.

[0047] To further prevent drastic shifts in the content response vector due to a few anomalous responses under small sample conditions, the system designs a vector stability regularization term to dynamically adjust the center offset of the aggregated vector. Each piece of content is defined. The response dispersion is:

[0048] ;

[0049] like If the threshold value exceeds an empirical threshold (e.g., 0.05), the system will automatically trigger an action. The sliding window reconstruction retains only the top-ranked user response vectors within the time window and recalculates the aggregation results to improve the stability and robustness of content response vectors. This mechanism can effectively suppress policy misjudgments caused by abnormal behavior within a short period of time. The output variables of this step are... It contains all content. A unified response representation in the current multi-platform environment.

[0050] S3: Based on the platform neutral response vector and the set of delivery variables corresponding to the delivered content, construct a causal graph structure to identify strategy variables that have a significant causal impact on the content response;

[0051] Specifically, after completing the first two steps, the platform behavior is normalized to a response vector. The construction and aggregation at the content level Next, the task of this step is based on A clear structure and logical causal relationship model is built to identify the key strategy variables that affect the delivery effect, thereby providing traceable decision-making basis for the next delivery strategy suggestion.

[0052] The input of this step is the content-level response vector generated in the previous stage , which is a three-dimensional vector representing the average level of user interest, emotional involvement, and conversion tendency, and has platform-neutral characteristics. In addition, the delivery variable set of the content Each variable represents an actual delivery strategy parameter, such as influencer fan level, content publishing time interval, platform type, content structure category, etc. These variables are derived from the delivery task record table of the delivery system, are structured fields, and have been set before the response vector is formed, meeting the causal precedence condition.

[0053] To achieve causal modeling, the system constructs a directed graph structure , where the nodes are all delivery variables and the three components of the response vector , , . To control the causal drift problem caused by excessive structural paths between variables, this step designs a causal graph objective function composed of a structure sparsity constraint term and a time stability bias term:

[0054] ;

[0055] where represents the residual term after linear modeling of the th component of the response vector as the dependent variable with as the independent variable; is the regression coefficient of this modeling path; is the indicator function, which is only included in the time bias when the impact strength of the path exceeds the threshold (such as 0.1); is the delivery variable setting time, is the median value of the actual response time of the content, is the weight of the time bias, is the sparsity control coefficient of the graph structure. represents the absolute value of the elements of the adjacency matrix , which is used to punish the structural complexity and improve the interpretability.

[0056] ​The structure combines the causal edge strength with the time dynamics, ensuring that when the policy variable changes significantly later than the response, the path weight will be suppressed, thus avoiding the interference of the "pseudo causal" path on the final structure. A threshold-driven edge selection mechanism is also introduced to control the complexity of the structure by combining the index function with the time decay term. The model does not rely on large-scale training, only linear modeling residual analysis and time information analysis, and has good implementability.

[0057] After the construction of the causal structure graph , the system constructs a set of strategy candidate variables according to the variables in the graph structure that have an effective causal path from each policy variable to the response component , and the path edge weight exceeds the threshold :

[0058] ;

[0059] The set of strategy variables is the input parameter required for the next step of strategy suggestion generation, and its members have been judged by the system to be core variables with the ability to truly affect the change of the response vector.

[0060] S4: Generate a structured delivery strategy suggestion based on the edge weight information in the causal graph structure, historical response gain, and policy implementation cost;

[0061] Specifically, after completing the causal structure modeling and identifying the set of key strategy variables , the task of this step is to generate a set of executable and optimized delivery strategy suggestions based on the structured analysis results. The starting point of this step is not to output a fixed template based on experience rules, but to combine the edge weight information of the causal structure graph , the multi-dimensional representation of the content response vector , and the platform resource allocation and historical execution cost to form a set of logical and clear strategy generation mechanisms suitable for multiple platforms and content types.

[0062] The main inputs of this step include three parts: first, the response vector of the content , which has been constructed in the second step, represents the overall interest, emotional participation, and conversion identity of the user under the platform neutrality; second, the causal graph structure obtained in the previous step, which contains all the directed edges between the policy variables and the components of the response vector, edge weights, and structure topology information; and third, the set of strategy candidate variables , which contains all variables that are determined to have significant causal impact on the response variable, and the variables are sourced from the content delivery configuration system, such as the level of the fan, the content duration interval, the delivery time period (such as the morning rush hour, lunch break, evening rush hour), platform identification (short video platform or text and image platform), etc. These variables are all structured fields, which are set by the platform operation system when creating a delivery task, and are loaded through a database scheduling interface.

[0063] Each variable generally has several optional values, denoted as , for example, the variable "delivery time period" can take "before 10 am", "12 noon", "after 8 pm", etc., and the variable "level of the expert" can take "L1 (below 10,000 fans)", "L2 (1-50,000)", "L3 (above 50,000)", etc. For each candidate value, the system will construct a response gain function based on backtracking analysis based on historical real delivery data , which represents the degree of improvement of the response vector in the dimension (such as "emotional engagement") compared to the average level of the platform after selecting the value in the historical sample.

[0064] The system will comprehensively calculate the structural weighted score of each value of the strategy variable based on the edge weight in the causal graph structure and the response gain:

[0065]

[0066] wherein represents the comprehensive score of the candidate value of the strategy variable, is the edge weight in the causal graph from the variable to the response dimension , indicating the direct impact strength of the variable on the dimension; is the response improvement value obtained from historical analysis; is the delivery implementation cost of the value, such as whether the expert's bid is too high, whether the time period is a high congestion period, whether it is limited by platform delivery rules, etc.; is the cost balance coefficient, which is generally set in the range of , and the specific value is set by the platform operation strategy.

[0067] For example, if the variable represents the level of the expert, the value is "below 10,000", the value is "10,000 to 50,000", and the value is "above 50,000". Suppose historical analysis finds that ​​The average promotion value of the user "conversion intention" dimension is the highest, and the corresponding execution cost is moderate (not high in offer, fast in response), so The score is the highest, and the value will be the final recommended option.

[0068] The system will score all The score is the highest, and the value will be the final recommended option.

[0069] ;

[0070] The final strategy recommendation is a structured object, including fields such as "recommended talent level: L2", "recommended time period: after 8pm", "recommended platform: short video platform", "recommended content structure: emotion-oriented text and image", etc. All recommended fields can be directly recognized by the delivery system, or transmitted to the content production module, talent screening module and platform delivery configuration module through the interface connection, to realize the automatic execution of the delivery task level.

[0071] In one or more embodiments, as Figure 2 shown, a digital delivery analysis platform based on multi-platform data integration is disclosed, which includes:

[0072] A behavior semantic alignment module for mapping user behavior data of two or more content platforms into unified behavior semantic labels and generating a unified response vector of each user to the delivered content;

[0073] A response aggregation module for aggregating the unified response vectors of all users interacting with the same delivered content, combining a user activity suppression mechanism and a time decay mechanism, and generating a platform-neutral response vector of the delivered content in a multi-platform environment;

[0074] A causal modeling module for constructing a causal graph structure based on the platform-neutral response vector and a set of delivery variables corresponding to the delivered content, and identifying strategy variables that have a significant causal impact on content response;

[0075] A strategy generation module for generating a structured delivery strategy recommendation based on edge weight information in the causal graph structure, historical response gains, and strategy implementation costs.

[0076] It is worth noting that the specific workflow of the digital delivery analysis platform based on multi-platform data integration provided by the embodiments of the present application is the same as the workflow of the digital delivery analysis method based on multi-platform data integration described in the above embodiments, and will not be repeated here.

[0077] The embodiment of the present application further provides a digital delivery analysis device based on multi-platform data integration, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the steps in the method embodiment of the digital delivery analysis based on multi-platform data integration as described above, for example Figure 1 steps S1-S4 described above; or the processor implements the functions of the modules in each platform embodiment described above.

[0078] For example, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the digital delivery analysis device based on multi-platform data integration.

[0079] The digital delivery analysis device based on multi-platform data integration can be a desktop computer, a notebook computer, a palm computer, a cloud server and other computing devices. The digital delivery analysis device based on multi-platform data integration can include, but is not limited to, a processor and a memory. Those skilled in the art can understand that the digital delivery analysis device based on multi-platform data integration can also include input and output devices, network access devices, buses and the like.

[0080] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASAC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The processor is the control center of the digital delivery analysis device based on multi-platform data integration, and connects all parts of the digital delivery analysis device based on multi-platform data integration through various interfaces and lines.

[0081] The memory can be used to store the computer program and / or modules, and the processor realizes various functions of the digital delivery analysis device based on multi-platform data integration by running or executing the computer program and / or modules stored in the memory, and calling data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required by a function, etc.; and the data storage area can store data created according to the running of the air conditioner controller, etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0082] The modules integrated in the digital delivery analysis device based on multi-platform data integration can be stored in a computer readable storage medium if they are realized in the form of software function units and sold or used as independent products. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. The computer program can implement the steps of the above-mentioned various method embodiments when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0083] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium, and the program can include the processes of the above-mentioned various method embodiments when executed. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0084] The above is the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can also make a number of improvements and refinements, these improvements and refinements are also considered to be within the scope of the present application.

Claims

1. A method for digital placement analysis based on multi-platform data integration, characterized in that, The method comprises: semantically mapping user behavior data from two or more content platforms to respective unified behavior semantic labels; generating a unified response vector of each user to the content being pushed based on the behavior semantic labels; aggregating the unified response vectors of all users interacting with the same content being pushed, combining a user activity suppression mechanism and a time decay mechanism to generate a platform-neutral response vector of the content being pushed in a multi-platform environment; the platform-neutral response vector is calculated as follows: ; wherein, representing the content platform-neutral response vector, representing the user activity, representing the user actual interaction with the content timestamp, the delivery time of the content is a time decay factor, controlling the degree of influence of the time distance on the response weight;​ the user activity is nonlinearly compressed in the aggregation process of the platform-neutral response vector to reduce the dominant influence of high-activity users on the aggregation result; the time decay mechanism introduces a time decay factor in the aggregation process of the platform-neutral response vector, so that the closer the user response is to the pushing time of the content being pushed, the higher the weight of the user response; based on the platform-neutral response vector and a set of pushing variables corresponding to the content being pushed, a causal graph structure is constructed to identify strategy variables that have a significant causal effect on the response to the content; the set of pushing variables corresponding to the content being pushed includes actual pushing strategy parameters derived from a pushing task record table of a pushing system; According to the edge weight information in the causal graph structure, historical response gain and policy implementation cost, a structured delivery policy suggestion is generated, specifically including that for each candidate value, the system will construct a response gain function based on backtracking analysis based on historical real delivery data , indicating the degree of improvement of the response vector compared with the platform average level in the first dimension after selecting the value in the historical sample; The system will integrate the edge weights in the causal graph structure with the response gain to compute a structural weighted score for each value of the policy variable: ; in Represents policy variables No. A comprehensive score of each candidate value. It is a causal graph from variables Pointing to response dimension The edge weight represents the strength of the direct influence of the variable on that dimension; It is the response improvement value derived from historical analysis; This is the cost of implementing this value; This is the cost balance coefficient; For all The scoring computation is performed, and the combination of values with the highest score is selected as the strategy recommendation output; wherein is a set of strategy candidate variables. 2.The digital delivery analysis method based on multi-platform data integration of claim 1, wherein, the behavior semantic labels include shallow attention, emotional interaction and deep identification, each behavior semantic label corresponds to a three-dimensional vector, and the three-dimensional vector represents user interest stimulation intensity, emotional involvement and conversion intention possibility. 3.The digital delivery analysis method based on multi-platform data integration of claim 1, wherein, The unified response vector is obtained by assigning a preset weight to each behavior semantic label and performing weighted summation, and the preset weight is set based on historical conversion performance. 4.The digital delivery analysis method based on multi-platform data integration of claim 1, wherein, When generating the platform-neutral response vector, if the response dispersion of the content being pushed exceeds a preset threshold, only the response vectors of the top users within a time window are retained to recalculate the aggregation result.

5. The method of claim 1, wherein, The construction process of the causal graph structure introduces a structure sparsity constraint term and a time stability bias term to suppress false causal paths and improve the interpretability of the causal model, wherein the causal graph structure is obtained by optimizing the following function: ; where, represents the residual term after linear modeling the response vector's first component as the dependent variable with as the independent variable; is the regression coefficient of this modeling path; is the indicator function, only when the influence strength of this path exceeds the threshold is counted into the time offset; is the time set for the delivery variable, is the median value of the actual response time of the content, is the weight of the time offset term, is the sparse control coefficient of the graph structure represents the element absolute value sum of the adjacency matrix , used to punish the structure complexity and improve the interpretability. 6.The digital delivery analysis method based on multi-platform data integration of claim 1, wherein, The strategy variables include influencer fan level, content publishing time interval, platform type and content structure category.

7. The method of claim 1, wherein, The structured pushing strategy suggestion includes recommended influencer level, recommended publishing time period, recommended platform type and recommended content structure type, and can be directly called and executed by the pushing system.

8. A digital placement analytics platform based on multi-platform data integration, characterized in that, The platform comprises: a behavior semantic alignment module for mapping user behavior data from two or more content platforms to unified behavior semantic labels and generating a unified response vector of each user to the content being pushed; a response aggregation module for aggregating the unified response vectors of all users interacting with the same content being pushed, combining a user activity suppression mechanism and a time decay mechanism to generate a platform-neutral response vector of the content being pushed in a multi-platform environment; the platform-neutral response vector is calculated as follows: ; wherein, representing the content platform-neutral response vector, representing the user activity, representing the user actual interaction with the content timestamp, the time of delivery of the content is a time decay factor, controlling the degree of influence of the temporal distance on the response weight;​ The user activity is nonlinearly compressed for each user in the aggregation of the platform-neutral response vector, so as to reduce the dominant influence of high-activity users on the aggregation result; and the time decay mechanism is a time decay factor introduced in the aggregation of the platform-neutral response vector, so that the weight of the user response closer to the delivery time of the delivered content is higher. A causal modeling module is configured to construct a causal graph structure based on a set of delivery variables corresponding to the platform-neutral response vector and the delivered content, and identify a strategy variable having a significant causal effect on the response to the content; and the set of delivery variables corresponding to the delivered content includes actual delivery strategy parameters derived from a delivery task record table of a delivery system. The policy generation module is configured to generate a structured delivery policy suggestion according to the edge weight information in the causal graph structure, historical response gain, and policy implementation cost. Specifically, for each candidate value, the system constructs a response gain function based on backtracking analysis based on historical real delivery data , which represents the degree of improvement of the response vector in the first dimension compared to the platform average level after selecting the value in the historical sample. The system will integrate the edge weights in the causal graph structure with the response gain to compute a structural weighted score for each value of the policy variable: ; in Represents policy variables No. A comprehensive score of each candidate value. It is a causal graph from variables Pointing to response dimension The edge weight represents the strength of the direct influence of the variable on that dimension; It is the response improvement value derived from historical analysis; This is the cost of implementing this value; This is the cost balance coefficient; For all The scoring computation is performed, and the combination of values with the highest score is selected as the strategy recommendation output; wherein is a set of strategy candidate variables.

Images