Precise marketing global video targeting promotion strategy method and system fusing AI technology

Abstract

The application relates to the technical field of advertisement delivery. A precise marketing global video targeting promotion strategy method and system fusing AI technology are provided, wherein the method comprises the following steps: performing cross-domain identity mapping processing on original user behavior streams collected from multiple platforms to generate identity unified data streams; performing behavior sequence standardization processing on the identity unified data streams to generate global unified user trajectories; performing dynamic response feature grouping processing on the global unified user trajectories to generate a dynamically updated marketing strategy unit set; acquiring a preset dynamic user portrait and performing tendency strategy coupling processing on the marketing strategy unit set to generate a tendency strategy set under brand constraints; and performing cross-domain distribution and negative feedback control processing on the tendency strategy set to output a multi-platform creative variant set, so that the technical effects of improving cross-platform identity integration capability, enhancing dynamic behavior response real-time performance and reducing creative distribution control lag are achieved.

Description

Technical Field

[0001] This invention relates to the field of advertising technology, and in particular to a precise marketing strategy, method and system for targeted video promotion that integrates AI technology. Background Technology

[0002] With the deep application of artificial intelligence technology in the marketing field, targeted video advertising has become a core means for brands to accurately reach their target users. This technology aims to generate personalized promotional strategies through multi-platform user behavior analysis, driving efficient conversion and brand growth.

[0003] Traditional technologies rely on single-platform behavioral data for cross-platform user identity management, lacking cross-domain identity mapping capabilities and failing to establish a unified identity identifier to integrate user behavior flows across multiple platforms. In implementing user segmentation strategies, static segmentation techniques with fixed-period updates struggle to respond in real-time to user behavior shifts, cannot promptly capture user interest migrations and sudden trending events, and exhibit delays in discovering new user groups. Regarding creative distribution and feedback processing, traditional cross-domain distribution technologies that separate creative generation and negative feedback handling into independent modules require manual intervention to analyze user negative feedback behavior data. Furthermore, the brand security rule base is independent of the strategy generation process, making it impossible to filter high-compliance-risk creatives in real-time. Summary of the Invention

[0004] Therefore, it is necessary to provide a precise marketing strategy and system that integrates AI technology to address the aforementioned technical issues, thereby improving cross-platform identity integration capabilities, enhancing the real-time responsiveness of dynamic behaviors, and reducing the lag in creative distribution control.

[0005] Firstly, this application provides a method for precise marketing and full-domain video targeted promotion strategy that integrates AI technology, the method including:

[0006] S1. Perform cross-domain identity mapping processing on the raw user behavior streams collected from multiple platforms to generate a unified identity data stream;

[0007] S2. Perform behavioral sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain;

[0008] S3. Perform dynamic response feature grouping processing on the unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units;

[0009] S4. Obtain the preset dynamic user profile and combine it with the marketing strategy unit set to perform biased strategy coupling processing to generate a biased strategy set under brand constraints;

[0010] S5. Perform cross-domain distribution and negative feedback control on the set of biased strategies to output a set of creative variations for multiple platforms.

[0011] In one embodiment, a preset dynamic user profile is obtained and coupled with a set of marketing strategy units to generate a set of brand-constrained biased strategies, including:

[0012] A matrix-based correlation process is performed on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index.

[0013] The response probability prediction process is performed on the two-dimensional strategy index to generate response probability data including creative form category, incentive mechanism type and reach time period interval;

[0014] The response probability data and the preset brand security rule base are filtered to generate a set of biased policies.

[0015] In one embodiment, a matrix-based correlation processing is performed on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index, including:

[0016] Discretize and encode the spending power tags in the preset dynamic user profile to generate a standardized spending power vector;

[0017] The features in the marketing strategy unit set are orthogonally projected to generate a dimension-reduced feature matrix.

[0018] Tensor product operations are performed on the standardized consumption capacity vector and the dimension-reduced feature matrix to generate the initial correlation matrix;

[0019] The initial correlation matrix is ​​sparsified and compressed to generate an optimized correlation matrix.

[0020] The optimized association matrix is ​​processed by constructing a two-dimensional index to generate a two-dimensional strategy index.

[0021] In one embodiment, the initial association matrix is ​​obtained using the following formula, including:

[0022]

[0023] Where M represents the initial correlation matrix, Let c represent the tensor product operation, c represent the standardized consumption capacity vector, and F represent the dimensionality-reduced feature matrix. This represents the matrix transpose operation.

[0024] In one embodiment, dynamic response feature clustering is performed on the unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units, including:

[0025] Perform time-series behavior encoding on the unified user trajectory across the entire domain to generate response feature vectors;

[0026] Drift-aware clustering is performed on the response feature vectors to generate initial marketing strategy units;

[0027] When the preset user behavior monitoring unit detects that the new data exceeds the preset cluster boundary, the data outside the boundary is processed to discover new groups and generate an extended marketing strategy unit.

[0028] The initial marketing strategy unit and the extended marketing strategy unit are aggregated and integrated to generate a marketing strategy unit set.

[0029] In one embodiment, when a preset user behavior monitoring unit detects that new data exceeds a preset clustering boundary, new group discovery processing is performed on the data outside the boundary to generate an extended marketing strategy unit, including:

[0030] The newly collected user behavior data from the user behavior monitoring unit undergoes boundary detection processing to generate a boundary exceedance flag.

[0031] Based on the boundary exceedance indicator, abnormal feature extraction processing is performed on data that exceeds the preset clustering boundary to generate a new group feature vector;

[0032] Density peak clustering is performed on the feature vectors of the new population to generate an initial new population cluster;

[0033] Perform response pattern verification on the initial new population cluster to generate a new population identifier that passes the verification.

[0034] The newly verified group identifiers are encapsulated into strategy units to generate extended marketing strategy units.

[0035] In one embodiment, the method further includes:

[0036] Based on the feedback data generated by the negative feedback control processing of S5, closed-loop strategy weight iteration processing is performed on the marketing strategy unit set and the bias strategy set. The updated marketing strategy unit set is fed back to S4, and the updated bias strategy set is fed back to S5.

[0037] Secondly, this application also provides a precision marketing full-domain video targeted promotion strategy system that integrates AI technology, the system including:

[0038] The identity mapping module is used to perform cross-domain identity mapping processing on the original user behavior streams collected from multiple platforms, and generate a unified identity data stream.

[0039] The behavior standardization module is used to perform behavior sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain.

[0040] The feature clustering module is used to dynamically respond to the feature clustering of the unified user trajectory across the entire domain, and generate a dynamically updated set of marketing strategy units.

[0041] The strategy coupling module is used to obtain a preset dynamic user profile and combine it with a set of marketing strategy units to perform biased strategy coupling processing, generating a set of biased strategies under brand constraints.

[0042] The cross-domain distribution module is used to perform cross-domain distribution and negative feedback control processing on the set of biased strategies, and output a set of creative variations for multiple platforms.

[0043] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0044] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0045] The precision marketing, omni-channel video-targeted promotion strategy and system provided in this application integrates AI technology to consolidate the original user behavior streams from multiple platforms through cross-domain identity mapping, generating a unified identity data stream. This directly breaks down platform data barriers and enhances cross-platform identity integration capabilities. The aforementioned data stream is standardized into a unified user trajectory across all domains, paving the way for subsequent analysis. Based on this trajectory, dynamic user segmentation is performed, behavioral boundaries are monitored in real time, and new groups are discovered, generating a dynamically updated set of marketing strategy units, effectively enhancing the real-time responsiveness to dynamic behaviors. Combining dynamic user profiles and strategy units, a set of brand-constrained biased strategies is generated, which are then distributed across domains with synchronous negative feedback control, allowing for real-time strategy adjustments and reducing the lag in creative distribution control. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 A flowchart of a precision marketing full-domain video targeted promotion strategy method integrating AI technology in one embodiment of the present invention;

[0048] Figure 2 This is a flowchart illustrating how to perform dynamic response feature clustering processing on a unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units, as described in one embodiment of the present invention.

[0049] Figure 3 This is a structural diagram of a precision marketing full-domain video targeted promotion strategy system that integrates AI technology, according to one embodiment of the present invention. Detailed Implementation

[0050] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0051] First, the application scenarios of the embodiments of this application are described. In the embodiments of this application, a method and system for precise marketing and targeted video promotion strategies integrating AI technology are provided, applicable to, but not limited to, brand-owned full-domain video advertising. For example, if a beauty brand needs to simultaneously promote new product short video ads on multiple platforms and improve conversion rates by accurately reaching target user groups on different platforms, this method and system can be used to achieve cross-platform user behavior integration and dynamic strategy adjustment.

[0052] The illustrative examples provided in this application demonstrate that the precise marketing full-domain video targeted promotion strategy and system integrating AI technology can also be applied to other application scenarios, such as full-domain video promotions conducted by 3C digital brands for new product launches on e-commerce platforms, short video platforms, and social forums, or video marketing activities conducted by FMCG brands on multiple channels to increase product repurchase rates. These are merely illustrative examples and do not limit the specific application scenarios.

[0053] like Figure 1 As shown, this application provides a method for precise marketing and full-domain video targeted promotion strategy that integrates AI technology. The method includes:

[0054] S1. Perform cross-domain identity mapping processing on the original user behavior streams collected from multiple platforms to generate a unified identity data stream.

[0055] For example, a multi-platform user raw behavior stream collection process is initiated. This process covers different types of user interaction scenarios, including short video platforms, social media platforms, and e-commerce platforms. The collection targets include users' raw behavior streams formed by video viewing behavior, content clicking behavior, page dwell behavior, interactive commenting behavior, and product browsing behavior on each platform. The collected multi-platform user raw behavior streams undergo data preprocessing. This preprocessing process removes invalid and abnormal behavior data. Invalid behavior data includes duplicate records generated due to system failures, and abnormal behavior data includes abnormal operation records exceeding the normal interaction frequency.

[0056] Cross-domain identity association rules are introduced. These rules combine multiple dimensions of association criteria, such as user device identification, account association information, and behavioral feature similarity. Based on these rules, the pre-processed original user behavior streams from multiple platforms undergo cross-domain identity mapping. During this mapping process, various behavioral data corresponding to the same user, scattered across different platforms, need to be associated and matched. This association and matching results in a dataset containing a unified user identifier and the user's complete behavioral records across multiple platforms; this dataset constitutes the unified identity data stream.

[0057] S2. Perform behavioral sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain.

[0058] For example, the behavioral data fields in the unified identity data stream are standardized, clarifying the core elements included in the behavioral data fields, such as behavioral type identifiers, behavioral occurrence timestamps, behavioral associated content identifiers, and behavioral interaction intensity parameters, to ensure consistent definitions of behavioral data fields from different platforms. The behavioral occurrence timestamps are then uniformly calibrated for time zones and converted to a preset standard time format, eliminating differences in the time dimension.

[0059] Based on the chronological order of the behavior timestamps, all behavior data corresponding to the same user identifier in the unified identity data stream are sorted to generate an ordered user behavior sequence. The interaction intensity parameters in the sorted user behavior sequence are standardized, and interaction intensity indicators from different platforms are converted into values ​​of a uniform magnitude using a preset intensity mapping rule. The user behavior data, after field normalization, time calibration, sequence sorting, and intensity standardization, are integrated to generate a dataset that fully reflects the continuous behavior process of users across multiple platforms; this dataset is the unified user trajectory across the entire domain.

[0060] S3. Perform dynamic response feature grouping processing on the unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units.

[0061] For example, key features such as behavioral frequency and preferences of the unified user trajectory across the entire domain are extracted, transformed into feature variables, and normalized to generate a standardized user behavior feature vector. This vector is initially clustered using a preset clustering algorithm to divide the user groups and match them with initial marketing strategy units. New data updates the unified user trajectory across the entire domain in real time, and boundary detection is used to determine whether the new data exceeds the boundaries of existing group features. If it does, new group features are extracted to generate a new vector, which is then matched with extended marketing strategy units after clustering verification. The initial and extended marketing strategy units are integrated and deduplicated to generate a dynamically updated set of marketing strategy units.

[0062] S4. Obtain the preset dynamic user profile and combine it with the marketing strategy unit set to perform biased strategy coupling processing to generate a biased strategy set under brand constraints.

[0063] For example, a pre-defined dynamic user profile database is invoked to extract core dimension information such as spending power and interest preferences, and then structured and organized. A dynamically updated set of marketing strategy units is retrieved, and key elements such as target group characteristics and promotion direction of each unit are analyzed. A correlation model between the two is established and a preference matching operation is performed to filter out highly suitable candidate strategy units. A pre-defined brand safety rule base and brand marketing objective parameters are introduced to perform compliance verification and target evaluation on the candidate strategy units, eliminating unqualified units. The candidate units that pass the verification are sorted by priority and integrated to form a structured strategy list, which is the set of preference strategies under brand constraints.

[0064] S5. Perform cross-domain distribution and negative feedback control on the set of biased strategies to output a set of creative variations for multiple platforms.

[0065] For example, the set of biased strategies under brand constraints is analyzed, and key distribution parameters such as target platform type and core creative appeal of each strategy are extracted. These parameters are then transformed into initial creative solutions based on multi-platform format specifications and traffic characteristics. A cross-domain distribution scheduling mechanism is established to push initial creative solutions to corresponding ad delivery interfaces according to the active user periods of each platform, while simultaneously collecting user interaction data and focusing on capturing negative feedback behavior data. Data is judged according to preset negative feedback threshold rules. Initial creative solutions with negative feedback rates exceeding the threshold are triggered to pause distribution and marked as needing optimization. Initial creative solutions that do not exceed the threshold are fine-tuned and optimized based on platform user preferences to generate adjusted creative solutions. The validated and optimized adjusted creative solutions are integrated, categorized by platform type, and a multi-platform creative variation set is generated.

[0066] This application provides an embodiment of a precision marketing full-domain video targeted promotion strategy method that integrates AI technology. By integrating the original user behavior streams from multiple platforms through cross-domain identity mapping, a unified identity data stream is generated, directly breaking down platform data barriers and improving cross-platform identity integration capabilities. This data stream is standardized into a unified user trajectory across all domains, paving the way for subsequent analysis. Based on this trajectory, dynamic grouping is performed, behavioral boundaries are monitored in real time, new groups are discovered, and a dynamically updated set of marketing strategy units is generated, effectively enhancing the real-time responsiveness to dynamic behaviors. Combining dynamic user profiles and strategy units, a brand-constrained set of biased strategies is generated, which is then distributed across domains with synchronous negative feedback control. Strategies are adjusted in real time, reducing the lag in creative distribution control.

[0067] In one embodiment, a preset dynamic user profile is obtained and coupled with a set of marketing strategy units to generate a set of brand-constrained biased strategies, including:

[0068] (1) Perform matrix-based association processing on the consumption capacity tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index.

[0069] For example, spending power tags are extracted from dynamic user profiles, and these tags are discretized and encoded to transform spending power information at different levels into standardized vector forms, resulting in standardized spending power vectors. Simultaneously, a set of marketing strategy units is retrieved, and features corresponding to each marketing strategy unit are extracted. These features include promotional category attributes, discount levels, target audience scope, etc. Orthogonal projection processing is then applied to these features to reduce feature dimensionality and eliminate redundant information, resulting in a dimensionality-reduced feature matrix.

[0070] Tensor product operations are performed on the standardized consumption capacity vector and the dimensionality-reduced feature matrix to construct the correlation matrix between them, resulting in an initial correlation matrix. This initial correlation matrix is ​​then subjected to sparsity compression to remove invalid correlation data and optimize its structure for improved efficiency in subsequent processing, yielding an optimized correlation matrix. A two-dimensional index is constructed based on this optimized correlation matrix. One dimension of the index corresponds to the level of the standardized consumption capacity vector, and the other dimension corresponds to the feature category of the dimensionality-reduced feature matrix, generating a two-dimensional strategy index.

[0071] The features in the marketing strategy unit set include the attributes of the promoted product category, the level of discount, and the scope of the target group. Discretization encoding is used to transform the consumption capacity information at different levels into standardized vectors. Orthogonal projection is used to reduce the feature dimension and eliminate redundant information. Sparse compression is used to remove invalid association data in the initial association matrix.

[0072] (2) Perform response probability prediction processing on the dual-dimensional strategy index to generate response probability data including creative form category, incentive mechanism type and reach time interval.

[0073] For example, a two-dimensional strategy index is obtained, and the relationships within the index are analyzed to clarify the matching relationships between different consumer capacity levels and different marketing strategy unit characteristics. A pre-defined response probability prediction model is introduced, which is trained based on historical user behavior data and strategy response results. The correlation data from the two-dimensional strategy index is input into the response probability prediction model. The model calculates the probability of user response to the strategy under different correlation combinations. The prediction results need to cover multiple key strategy dimensions, including creative form category, incentive mechanism type, and reach time period. The predicted response probability data is structured and sorted according to response probability, generating response probability data that includes creative form category, incentive mechanism type, and reach time period.

[0074] Among them, the response probability prediction model is trained and generated based on historical user behavior data and policy response results. It is used to calculate the probability of users responding to policies under different association combinations. The structured organization is used to sort the prediction results according to the response probability.

[0075] (3) Filter the response probability data and the preset brand security rule base to generate a set of biased strategies.

[0076] For example, a pre-defined brand safety rule base is retrieved. This rule base includes constraint rules such as compliant content restrictions, competitor information avoidance, and value-oriented standards, clearly defining the judgment conditions and prohibited scope for each constraint rule. Response probability data is obtained, and each strategy information in the response probability data is broken down. The breakdown includes the presentation method corresponding to the creative form category, the implementation rules corresponding to the incentive mechanism type, and the time range corresponding to the reach period. Based on the constraint rules in the brand safety rule base, each broken strategy information is checked for compliance to determine whether it meets the constraint rule requirements. Strategy information that does not meet the constraint rules is removed, and strategy information that passes the compliance check is retained. The retained strategy information is then sorted a second time according to the response probability, and integrated to form a structured strategy list, which is the set of preferred strategies.

[0077] The brand safety rule library includes constraint rules such as compliant content restrictions, competitor information avoidance, and value orientation standards. Compliance verification is used to determine whether strategy information meets the requirements of constraint rules, and secondary sorting is used to ensure that the retained strategy information is arranged according to the response probability.

[0078] In one embodiment, a matrix-based correlation processing is performed on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index, including:

[0079] (1) Discretize the consumption capacity tags in the preset dynamic user profile to generate a standardized consumption capacity vector.

[0080] For example, a preset dynamic user profile is retrieved, and spending power tags are extracted from the dynamic user profile. According to preset encoding rules, the extracted spending power tags are discretized and encoded, converting spending power information of different intervals, levels, and ranges into vector elements of a unified format. The converted vector elements are then normalized to ensure that each element has a consistent magnitude and meets the standards for subsequent calculations, generating a standardized spending power vector.

[0081] (2) Perform orthogonal projection on the features in the marketing strategy unit set to generate a dimension-reduced feature matrix.

[0082] For example, a set of marketing strategy units is obtained, and each marketing strategy unit in the set is analyzed one by one. Features are extracted from each marketing strategy unit, including the category attributes of the promoted product, the level of discount, the target user group profile tags, and the type of promotion channel suitability. A preset orthogonal projection algorithm is introduced, and the extracted features are input into the orthogonal projection algorithm. The algorithm calculates and eliminates redundant correlations and dimensional overlaps between features. The features after eliminating redundancy are then subjected to dimensionality compression, retaining core feature information and reducing data complexity, thereby generating a dimensionality-reduced feature matrix.

[0083] The features in the marketing strategy unit set include the category attributes of the promoted products, the level of discount, the target user group profile tags, and the type of promotion channel. The orthogonal projection algorithm is used to eliminate redundant correlations and dimensional overlaps between features, while dimensional compression is used to retain core features and reduce data complexity.

[0084] (3) Perform tensor product operation on the standardized consumption capacity vector and the dimension-reduced feature matrix to generate the initial correlation matrix.

[0085] For example, the operational dimensions of the standardized consumption capacity vector and the dimension-reduced feature matrix are determined, clarifying whether the vector dimension of the standardized consumption capacity vector matches the row and column dimensions of the dimension-reduced feature matrix. If the dimensions do not match, the standardized consumption capacity vector is expanded in dimension, or the dimension of the dimension-reduced feature matrix is ​​adjusted to ensure that both meet the dimensionality requirements of tensor product operations. The tensor product operation tool is then invoked, and the adjusted standardized consumption capacity vector and dimension-reduced feature matrix are input into the tool. The operation is performed according to the tensor product operation rules to generate an initial correlation matrix.

[0086] (4) Perform sparsification and compression on the initial correlation matrix to generate an optimized correlation matrix.

[0087] For example, the data distribution of the initial association matrix is ​​analyzed to identify invalid association elements with values ​​of zero or close to zero. These elements represent combinations where there is no actual correlation between the standardized consumption capacity vector and the dimensionality-reduced feature matrix. A pre-defined sparsity algorithm is used to mark and remove invalid association elements from the initial association matrix, reducing the amount of matrix data. The matrix after removing invalid elements is then structurally reorganized to optimize the matrix storage format and data retrieval efficiency, generating an optimized association matrix.

[0088] (5) Perform two-dimensional index construction on the optimized correlation matrix to generate a two-dimensional strategy index.

[0089] For example, a two-dimensional index dimension is determined for the optimized association matrix. One dimension is set as the consumption capability level corresponding to the standardized consumption capability vector, and the other dimension is set as the strategy feature category corresponding to the dimensionality-reduced feature matrix. Based on the set dimensions, corresponding two-dimensional coordinates are assigned to each valid element in the optimized association matrix. The coordinates must accurately correspond to the consumption capability level and the strategy feature category. An index query mapping relationship is established, associating the two-dimensional coordinates with the element positions in the optimized association matrix to facilitate subsequent fast queries and invocations, generating a two-dimensional strategy index.

[0090] In one embodiment, the initial association matrix is ​​obtained using the following formula, including:

[0091]

[0092] Where M represents the initial correlation matrix, Let c represent the tensor product operation, c represent the standardized consumption capacity vector, and F represent the dimensionality-reduced feature matrix. This represents the matrix transpose operation.

[0093] Specifically, the study identifies two core elements: a standardized consumer spending power vector (obtained through discretization and encoding of dynamic user profile spending power tags, reflecting user spending power) and a dimensionality-reduced feature matrix (obtained through orthogonal projection of marketing strategy unit features, including key marketing strategy features). The dimensionality-reduced feature matrix is ​​transposed, swapping its rows and columns to obtain the transposed dimensionality-reduced feature matrix. A tensor product operation is then performed between the standardized consumer spending power vector and the transposed dimensionality-reduced feature matrix to establish their correlation. Subsequently, an initial correlation matrix is ​​obtained through the above operations, presenting the relationship between user spending power and marketing strategy features.

[0094] like Figure 2 As shown, dynamic response feature clustering is performed on the unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units, including:

[0095] S201: Perform time-series behavior encoding on the unified user trajectory across the entire domain to generate a response feature vector.

[0096] For example, a unified user trajectory across the entire domain is obtained. This trajectory includes user behavior records across various platforms and scenarios, including the temporal sequence and interactive content of actions such as browsing, clicking, and purchasing. The unified user trajectory is then processed using temporal behavior encoding. Each user action is converted into a specific encoded symbol or value according to chronological order, transforming a continuous sequence of user actions into a computable vector form, thus generating a response feature vector.

[0097] S202: Perform drift-aware clustering on the response feature vector to generate initial marketing strategy units.

[0098] For example, response feature vectors are obtained, and a drift-aware clustering algorithm is used, which can perceive changes in data distribution over time. The response feature vectors are input into the drift-aware clustering algorithm, which groups the vectors according to the similarity between vectors and the dynamic changes in data distribution. Vectors corresponding to user trajectories with similar behavioral characteristics are grouped into the same class, generating initial marketing strategy units.

[0099] S203: When the preset user behavior monitoring unit detects that the new data exceeds the preset cluster boundary, it performs new group discovery processing on the data outside the boundary and generates an extended marketing strategy unit.

[0100] For example, based on a preset user behavior monitoring unit, this unit continuously monitors the addition of new user behavior data. When the user behavior monitoring unit detects that the added data exceeds the preset cluster boundary, it performs new group discovery processing on the data outside the boundary. The new group discovery processing analyzes the behavioral patterns, feature distribution, etc. of the data outside the boundary to determine whether there is a new user group different from the existing initial marketing strategy unit. If so, it generates an extended marketing strategy unit based on this data.

[0101] S204: Perform set integration processing on the initial marketing strategy unit and the extended marketing strategy unit to generate a set of marketing strategy units.

[0102] For example, the initial marketing strategy units and extended marketing strategy units collected are aggregated and integrated. Duplicate units are removed to ensure that each unit has a unique strategic orientation. They are then combined into a complete set to generate a marketing strategy unit set.

[0103] In one embodiment, when a preset user behavior monitoring unit detects that new data exceeds a preset clustering boundary, new group discovery processing is performed on the data outside the boundary to generate an extended marketing strategy unit, including:

[0104] (1) Perform boundary detection processing on the newly added user behavior data collected by the user behavior monitoring unit and generate a boundary exceedance identifier.

[0105] For example, newly added user behavior data collected by the user behavior monitoring unit is acquired. This data includes the latest browsing, clicking, and interaction records of users on various platforms. Preset boundary detection rules are invoked to compare and analyze the newly added user behavior data with preset clustering boundaries to determine whether the data exceeds the preset clustering boundary range. If it exceeds the range, a boundary exceedance flag is generated to mark this situation; otherwise, no flag is generated.

[0106] Among them, the boundary detection rule is used to compare the newly added data with the preset cluster boundaries, and the boundary exceedance flag is used to mark the situation where the data exceeds the boundary.

[0107] (2) Based on the boundary exceedance identifier, perform abnormal feature extraction processing on the data that exceeds the preset cluster boundary to generate a new group feature vector.

[0108] For example, based on the generated boundary exceedance marker, data exceeding the preset cluster boundaries is located. This data represents the portion of new user behavior data that does not conform to existing clustering characteristics. Anomaly feature extraction is performed on the data exceeding the preset cluster boundaries. This extraction process identifies behavioral patterns, interaction frequencies, content preferences, and other features that significantly differ from existing group characteristics. The extracted anomaly feature information is then transformed into a structured vector form to generate a new group feature vector.

[0109] Among them, data that exceeds the preset clustering boundary refers to the part of the newly added user behavior data that does not conform to the existing clustering characteristics. Abnormal feature extraction processing is used to identify behavioral patterns and other information that are significantly different from the existing group characteristics.

[0110] (3) Perform density peak clustering on the feature vectors of the new population to generate the initial new population cluster.

[0111] For example, the generated new group feature vectors are obtained, and a density peak clustering algorithm is introduced. This algorithm determines the cluster centers by identifying density peak points in the dataset. The new group feature vectors are input into the density peak clustering algorithm, which calculates the local density and relative distance of each vector, divides the vectors into groups based on the density peak points, and groups vectors with similar features into the same group to generate an initial new group cluster.

[0112] (4) Perform response pattern verification on the initial new group cluster to generate a new group identifier that has passed the verification.

[0113] For example, a preset response pattern verification standard is invoked. This standard includes a group behavior consistency threshold, feature stability index, and policy adaptation probability parameters. Based on the response pattern verification standard, the behavioral patterns, feature persistence, and matching degree with potential policies of the initial new group cluster are evaluated. If the initial new group cluster meets all the requirements of the verification standard, a new group identifier that has passed verification is generated; otherwise, the initial new group cluster is discarded.

[0114] The response pattern verification criteria include group behavior consistency thresholds and feature stability indicators. The verification process is used to evaluate whether the behavior patterns of the initial new group cluster meet the requirements.

[0115] (5) Encapsulate the verified new group identifiers into strategy units to generate extended marketing strategy units.

[0116] For example, a verified new group identifier is obtained. Based on the characteristics of the new group corresponding to the identifier, corresponding promotional content direction, user reach method, incentive mechanism type, and other strategic elements are matched for it. The above strategic elements are integrated and encapsulated according to a preset unit structure to ensure that the encapsulated unit contains the characteristic description of the new group and the corresponding strategic solution, thereby generating an extended marketing strategy unit.

[0117] The strategy elements include the direction of promotional content and the way users are reached. The encapsulation process integrates the strategy elements according to a preset structure to generate an extended marketing strategy unit that includes feature descriptions and strategy solutions.

[0118] In one embodiment, the method further includes:

[0119] Based on the feedback data generated by the negative feedback control processing of S5, closed-loop strategy weight iteration processing is performed on the marketing strategy unit set and the bias strategy set. The updated marketing strategy unit set is fed back to S4, and the updated bias strategy set is fed back to S5.

[0120] For example, feedback data generated by the negative feedback control process of S5 is collected. This feedback data includes records of user negative feedback behavior during multi-platform creative variation campaigns and corresponding strategy conversion performance data. The feedback data is cleaned to remove invalid and abnormal data, resulting in standardized feedback data. A closed-loop strategy weight iteration model is constructed based on this standardized feedback data. The model includes weight calculation rules, iteration update thresholds, and other elements. The standardized feedback data is input into the model to calculate the current weight value of each strategy unit in the marketing strategy unit set and the weight percentage of each strategy in the biased strategy set.

[0121] Based on the weights calculated by the model, the weights of the marketing strategy unit set are adjusted, increasing the weight ratio of high-conversion, low-negative-feedback strategy units, decreasing or eliminating the weights of poor-performing strategy units, and generating an updated marketing strategy unit set which is then fed back to S4. Based on the model weights, the weights of the biased strategy set are iterated, strengthening the weights of strategies with high matching degrees to user needs and weakening the weights of strategies with low matching degrees, generating an updated biased strategy set which is then fed back to S5.

[0122] The feedback data includes records of negative feedback behaviors such as users skipping ads and data on strategy conversion effectiveness.

[0123] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0124] In one embodiment, such as Figure 3 As shown, this application also provides a precision marketing full-domain video targeted promotion strategy system 300 that integrates AI technology. The system 300 includes:

[0125] The identity mapping module 301 is used to perform cross-domain identity mapping processing on the original user behavior streams collected from multiple platforms to generate a unified identity data stream.

[0126] The behavior standardization module 302 is used to perform behavior sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain.

[0127] The feature clustering module 303 is used to perform dynamic response feature clustering processing on the unified user trajectory across the entire domain, and generate a dynamically updated set of marketing strategy units.

[0128] The strategy coupling module 304 is used to obtain a preset dynamic user profile and combine it with the marketing strategy unit set to perform biased strategy coupling processing to generate a biased strategy set under brand constraints.

[0129] The cross-domain distribution module 305 is used to perform cross-domain distribution and negative feedback control processing on the set of biased strategies, and output a set of creative variations for multiple platforms.

[0130] Specifically, the identity mapping module 301 collects raw user behavior streams from multiple platforms, including user browsing, clicking, and other behavior records on various platforms. After preprocessing to remove invalid and abnormal data, it performs cross-domain identity mapping according to cross-domain identity association rules, associating and matching the behavior data of the same user across multiple platforms to generate a unified identity data stream. The raw user behavior stream includes user behavior records from multiple platforms, and the cross-domain identity association rules are used to associate data from different platforms.

[0131] The behavior standardization module 302 acquires a unified identity data stream, standardizes behavior data fields to ensure consistent definitions, calibrates timestamps to eliminate time discrepancies, sorts behaviors by time to form an ordered sequence, standardizes interaction intensity parameters, and integrates them to generate a unified user trajectory across the entire domain. The behavior data fields include behavior type, timestamp, etc., and standardization processing ensures data consistency and orderliness.

[0132] The feature clustering module 303 performs time-series behavior encoding on the unified user trajectory across the entire domain to generate response feature vectors, which are then used for drift-aware clustering to generate initial marketing strategy units. When new data is detected exceeding the cluster boundary, new group discovery processing is performed on the data outside the boundary to generate extended marketing strategy units, which are then integrated to generate a dynamically updated set of marketing strategy units. Specifically, the time-series behavior encoding transforms the behavior sequence into a feature vector, and the new group discovery processing includes the needs of new user groups.

[0133] The strategy coupling module 304 extracts core information such as spending power and interest preferences from dynamic user profiles, analyzes the key elements of the marketing strategy unit set, filters candidate strategy units through correlation matching, and combines brand safety rules with marketing objective verification and evaluation to generate a set of biased strategies under brand constraints after integration and ranking. In this process, the core information of the dynamic user profile is correlated and matched with the elements of the strategy units, and brand rules ensure strategy compliance.

[0134] The cross-domain distribution module parses the set of biased strategies, transforms them into initial creative solutions adapted to multiple platforms, pushes them according to platform patterns, and collects negative feedback data. Distribution is paused for solutions exceeding thresholds, while qualified solutions are optimized and integrated to output a multi-platform creative variation set.

[0135] The strategy coupling module 304 is also used for:

[0136] A matrix-based correlation process is performed on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index.

[0137] The response probability prediction process is performed on the two-dimensional strategy index to generate response probability data including creative form category, incentive mechanism type and reach time period interval;

[0138] The response probability data and the preset brand security rule base are filtered to generate a set of biased policies.

[0139] The strategy coupling module 304 is also used for:

[0140] Discretize and encode the spending power tags in the preset dynamic user profile to generate a standardized spending power vector;

[0141] The features in the marketing strategy unit set are orthogonally projected to generate a dimension-reduced feature matrix.

[0142] Tensor product operations are performed on the standardized consumption capacity vector and the dimension-reduced feature matrix to generate the initial correlation matrix;

[0143] The initial correlation matrix is ​​sparsified and compressed to generate an optimized correlation matrix.

[0144] The optimized association matrix is ​​processed by constructing a two-dimensional index to generate a two-dimensional strategy index.

[0145] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0146] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0147] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0148] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A precise marketing strategy and method integrating AI technology for targeted video promotion across all domains, characterized in that: The method includes: S1. Perform cross-domain identity mapping processing on the raw user behavior streams collected from multiple platforms to generate a unified identity data stream; S2. Perform behavior sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain; S3. Perform dynamic response feature clustering processing on the unified user trajectory across the entire domain to generate a dynamically updated set of marketing strategy units; S4. Obtain the preset dynamic user profile and perform biased strategy coupling processing in combination with the marketing strategy unit set to generate a biased strategy set under brand constraints; S5. Perform cross-domain distribution and negative feedback control processing on the set of biased strategies to output a multi-platform creative variant set.

2. The precise marketing full-domain video targeted promotion strategy method integrating AI technology according to claim 1, characterized in that, The process of obtaining a preset dynamic user profile and combining it with the marketing strategy unit set to perform biased strategy coupling processing to generate a biased strategy set under brand constraints includes: A matrix-based correlation process is performed on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index. The two-dimensional strategy index is processed for response probability prediction to generate response probability data including creative form category, incentive mechanism type and reach time interval; The response probability data and the preset brand security rule base are filtered to generate the set of biased policies.

3. The precise marketing full-domain video targeted promotion strategy method integrating AI technology according to claim 2, characterized in that, The step of performing matrix-based correlation processing on the spending power tags in the dynamic user profile and the features in the marketing strategy unit set to generate a two-dimensional strategy index includes: The spending power tags in the preset dynamic user profile are discretized and encoded to generate a standardized spending power vector. The features in the marketing strategy unit set are subjected to orthogonal projection processing to generate a dimension-reduced feature matrix; Tensor product operation is performed on the standardized consumption capacity vector and the dimensionality-reduced feature matrix to generate an initial correlation matrix; The initial correlation matrix is ​​subjected to sparsification and compression to generate an optimized correlation matrix; The optimized correlation matrix is ​​subjected to a two-dimensional index construction process to generate the two-dimensional strategy index.

4. The precise marketing full-domain video targeted promotion strategy method integrating AI technology according to claim 3, characterized in that, The initial correlation matrix is ​​obtained using the following formula, including: Where M represents the initial correlation matrix, denoted by , c represents the standardized consumption capacity vector, F represents the dimension-reduced feature matrix, and T represents the matrix transpose operation.

5. The precise marketing full-domain video targeted promotion strategy method integrating AI technology according to claim 1, characterized in that, The step of dynamically responding to the unified user trajectory across the entire domain and performing feature clustering to generate a dynamically updated set of marketing strategy units includes: The unified user trajectory across the entire domain is subjected to time-series behavior encoding processing to generate a response feature vector; The response feature vector is subjected to drift-aware clustering to generate initial marketing strategy units; When the preset user behavior monitoring unit detects that the new data exceeds the preset cluster boundary, the data outside the boundary is processed to discover new groups and generate an extended marketing strategy unit. The initial marketing strategy unit and the extended marketing strategy unit are integrated to generate the marketing strategy unit set.

6. The precise marketing full-domain video targeted promotion strategy method integrating AI technology according to claim 5, characterized in that, When the preset user behavior monitoring unit detects that new data exceeds the preset clustering boundary, it performs new group discovery processing on the data outside the boundary to generate an extended marketing strategy unit, including: The newly collected user behavior data by the user behavior monitoring unit is subjected to boundary detection processing to generate a boundary exceedance flag; Based on the boundary exceedance identifier, abnormal feature extraction processing is performed on the data that exceeds the preset clustering boundary to generate a new group feature vector; Density peak clustering is performed on the feature vectors of the new population to generate an initial new population cluster; The initial new group cluster is subjected to response pattern verification processing to generate a new group identifier that passes the verification. The newly verified group identifier is encapsulated into a strategy unit to generate the extended marketing strategy unit.

7. The method for precise marketing and full-domain video targeted promotion strategy integrating AI technology according to claim 1, characterized in that, The method further includes: Based on the feedback data generated by the negative feedback control processing in S5, closed-loop strategy weight iteration processing is performed on the marketing strategy unit set and the preference strategy set. The updated marketing strategy unit set is fed back to S4, and the updated preference strategy set is fed back to S5.

8. A precision marketing full-domain video targeted promotion strategy system integrating AI technology, characterized in that: The system includes: The identity mapping module is used to perform cross-domain identity mapping processing on the original user behavior streams collected from multiple platforms, and generate a unified identity data stream. The behavior standardization module is used to perform behavior sequence standardization processing on the identity unified data stream to generate a unified user trajectory across the entire domain. The feature clustering module is used to perform dynamic response feature clustering processing on the unified user trajectory across the entire domain, and generate a dynamically updated set of marketing strategy units. The strategy coupling module is used to obtain a preset dynamic user profile and perform biased strategy coupling processing in combination with the marketing strategy unit set to generate a biased strategy set under brand constraints. The cross-domain distribution module is used to perform cross-domain distribution and negative feedback control processing on the set of biased strategies, and output a set of creative variations for multiple platforms.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the precise marketing full-domain video targeted promotion strategy method that integrates AI technology as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the precise marketing full-domain video targeted promotion strategy method that integrates AI technology as described in any one of claims 1 to 7.

Images