Media information effect attribution and budget intelligent allocation method and system

Abstract

The application provides a media information effect attribution and budget intelligent allocation method and system, relates to the technical field of media information, and comprises the following steps: constructing a conversion path by analyzing a user touch sequence and calculating attribution contribution weights of each media channel, and then obtaining an attribution return on investment. Based on this, a budget allocation optimization model is constructed to maximize the overall return, and a target budget scheme is generated for execution. After execution, the conduction effect of budget adjustment is analyzed through a cause-effect tracking map, and feedback is fed back to optimize the attribution weight calculation, so that dynamic closed-loop optimization and accurate effect evaluation of budget allocation are realized.

Description

Technical Field

[0001] This invention relates to media information technology, and more particularly to a method and system for intelligent allocation of media information effect attribution and budget. Background Technology

[0002] In the field of digital marketing, media message effectiveness evaluation and budget allocation are core aspects of advertisers' marketing decisions. Current technologies typically rely on the collection and analysis of data from multiple media channels, including impressions, clicks, and corresponding costs. Simultaneously, to understand the entire user journey from ad exposure to final purchase or other conversion, the industry generally tracks and records the user's interaction sequence across multiple touchpoints—the conversion path. Based on this foundational data, the conventional approach is to use pre-defined attribution models, such as final click attribution, linear attribution, or location-based attribution rules, to assign contribution weights to each touchpoint or corresponding media channel along the path. By combining the calculated attribution conversions with the cost of goods sold, the return on investment (ROI) for each channel can be further calculated as a key indicator for evaluating its efficiency. Based on this, advertisers typically allocate their limited overall budget across various media channels manually or semi-automatically, using these efficiency metrics combined with historical experience or simple proportional rules, in order to improve overall marketing effectiveness.

[0003] However, the aforementioned conventional practices have significant limitations. A major drawback lies in the contradiction between static attribution rules and the dynamic marketing environment. Pre-set attribution models are often fixed and cannot sensitively capture the subsequent impact of budget allocation decisions on the user conversion path structure. When budgets for each channel are adjusted based on the results of a certain round of evaluation, the flow of users and the sequence of touchpoints across different media evolve, and the original attribution weights quickly become invalid. This results in attribution results based on historical paths failing to accurately reflect the true contribution of each channel after the budget adjustment. Another drawback is the isolation of budget allocation decisions. Traditional allocation methods typically aim at single, static optimization. When calculating ROI and allocating budgets accordingly, they fail to fully consider the chain reaction effect of budget changes on other downstream media channels through the complex user path network. This allocation method, which ignores the dynamic interactions between channels, easily leads to local optima rather than global optima in budget allocation, inhibiting the potential synergistic value of certain channels and ultimately limiting the potential for continuous improvement in the overall marketing ROI. Summary of the Invention

[0004] This invention provides a method and system for media information effect attribution and intelligent budget allocation, which can solve the problems in the prior art.

[0005] A first aspect of the present invention provides a method for media information effect attribution and intelligent budget allocation, comprising:

[0006] Acquire delivery and conversion data from multiple media channels within a preset time window. The delivery data includes the exposure, clicks, and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to conversion.

[0007] Based on the touchpoint sequence, a user conversion path diagram is constructed, and attribution rules are applied to each touchpoint in the user conversion path diagram to calculate the contribution weight of each media channel to the conversion, thereby obtaining the attribution conversion volume of each media channel.

[0008] The attribution conversion rate is correlated with the placement cost to obtain the attribution ROI of each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed.

[0009] The target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution.

[0010] Collect actual conversion data after campaign execution, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

[0011] Based on the touchpoint sequence, a user conversion path graph is constructed, and attribution rules are applied to each touchpoint in the user conversion path graph to calculate the contribution weight of each media channel to the conversion, resulting in the attribution conversion volume of each media channel, including:

[0012] The touchpoint sequence is arranged in chronological order to form a directed path chain. The directed path chain is clustered into path patterns. Multiple paths with similar touchpoint combination features and similar media channel sequence features are grouped into the same path pattern cluster. The typical path structure of each path pattern cluster is extracted as the pattern representative path. Based on the pattern representative path, a compressed user conversion path graph is constructed.

[0013] For each touchpoint in the user conversion path graph, retrieve all actual conversion paths containing the corresponding touchpoint within the path pattern cluster, statistically analyze the conversion success rate distribution when the corresponding touchpoint appears in different path positions, construct a position-sensitive contribution function based on the conversion success rate distribution, and use the position-sensitive contribution function to assign differentiated values ​​to the contribution weight of the corresponding touchpoint under different path structures.

[0014] The total contribution weight of the media channel to the conversion is obtained by weighting and aggregating the contribution weights of all touchpoints belonging to the same media channel under their respective position-sensitive contribution functions. The total contribution weight is then correlated with the total number of conversions associated with the media channel to obtain the attribution conversion volume.

[0015] The contact point sequences are arranged in chronological order to form directed path chains. Path pattern clustering is then performed on these directed path chains, grouping multiple paths with similar contact point combination features and similar media channel sequence features into the same path pattern cluster, including:

[0016] For each directed path chain, construct its touchpoint combination feature representation and media channel sequence feature representation, and fuse and map the touchpoint combination feature representation and the media channel sequence feature representation to obtain the path pattern feature vector of each directed path chain;

[0017] In the space of the path pattern feature vector, the path similarity between any two directed path chains is calculated. When the path similarity exceeds the dynamically adjusted similarity threshold, the corresponding two directed path chains are marked as path pairs with the same pattern. A path connectivity graph is constructed based on the path pairs with the same pattern. The path connectivity graph is decomposed into connected components, and all directed path chains contained in each connected component are merged into the same path pattern cluster.

[0018] The attribution conversion rate is correlated with the placement cost to obtain the attribution ROI for each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. According to the efficiency stratification status and a preset total budget constraint, a budget allocation optimization model is constructed with maximizing the attribution ROI as the objective function, including:

[0019] For each media channel, the attribution conversion volume is multiplied by the preset unit conversion value to obtain the attribution conversion value, and the ratio of the attribution conversion value to the placement cost is calculated to obtain the attribution return on investment.

[0020] The attribution ROI of each media channel is sorted and a quantile threshold is calculated. Media channels with an attribution ROI above the quantile threshold are marked as effective media channels and an uncovered path set is extracted from the user conversion path map. Media channels with an attribution ROI below the quantile threshold are marked as ineffective media channels and a detour path set is extracted from the user conversion path map.

[0021] The budget expansion coefficient for the efficiency-enhancing media channel is calculated based on the number of paths and path length of the uncovered path set. The budget compression coefficient for the inefficient media channel is calculated based on the path connectivity and path conversion rate of the detourable path set. The budget expansion coefficient and the budget compression coefficient are used as budget adjustment boundary parameters to construct budget reallocation constraints.

[0022] Using the budget allocation amount of each media channel as the decision variable, the objective function is to maximize the attribution return on investment, and the budget redistribution constraint and the preset total budget constraint are used as constraints to construct a budget allocation optimization model.

[0023] The target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution, including:

[0024] The budget allocation optimization model is subjected to constraint relaxation and objective function convexization. An iterative solution mechanism is used to calculate the optimal budget allocation for each media channel. The difference between the optimal budget allocation and the current budget allocation is used to calculate the budget adjustment. Based on the positive or negative attribute of the budget adjustment, the media channels are classified into budget increment media channels and budget decrement media channels.

[0025] For media channels with incremental budgets, trace the set of uncovered paths in the user conversion path graph and extract the temporal characteristics of each touchpoint in the path and the combination characteristics of media channels to generate incremental budget execution strategies. For media channels with decremental budgets, trace the set of bypassable paths in the user conversion path graph and extract the alternative nodes of the path and the conversion retention conditions to generate decremental budget execution strategies.

[0026] The optimal budget allocation for the incremental budget media channel and the incremental budget execution strategy are encapsulated into an incremental budget delivery instruction, and the optimal budget allocation for the decremental budget media channel and the decremental budget execution strategy are encapsulated into a decremental budget delivery instruction. The incremental budget delivery instruction and the decremental budget delivery instruction are then sent to the delivery system of the corresponding media channel through the delivery interface.

[0027] The budget allocation optimization model is subjected to constraint relaxation and objective function convexity processing. An iterative solution mechanism is used to calculate the optimal budget allocation for each media channel, including:

[0028] Identify the nonlinear constraints in the budget allocation optimization model, introduce slack variables for the nonlinear constraints, and set an adaptive penalty coefficient according to the degree of constraint violation. The product of the adaptive penalty coefficient and the slack variables is superimposed as a penalty term on the objective function of the budget allocation optimization model to form a relaxed objective function. Perform convexity analysis on the relaxed objective function, and use piecewise linear approximation to reconstruct the relaxed objective function into a convex objective function for the non-convex regions identified by the convexity analysis.

[0029] A dual problem is established based on the convex objective function and the budget redistribution constraint. The dual variables of the dual problem are initialized. The original problem is solved under the current dual variable conditions to obtain the candidate budget allocation for each media channel. The dual variables are adjusted according to the degree to which the candidate budget allocation satisfies the constraint, and the original problem is solved again. When the dual gap is reduced to a preset accuracy threshold, the iteration stops and the optimal budget allocation for each media channel is determined.

[0030] Construct a causal tracking graph of budget changes and conversion path structure evolution for each media channel, and identify the transmission effect of budget adjustments on the conversion contribution of downstream media channels based on the causal tracking graph, including:

[0031] Collect budget change data and user conversion path map data of various media channels in multiple time periods, calculate the change in path topology centrality and the change in path traffic distribution entropy before and after budget changes of various media channels to construct a two-dimensional evolutionary descriptor.

[0032] The budget change data is used as an intervention variable, and the two-dimensional evolutionary descriptor is used as a response variable for time-series pairing. The time-series pairing with significant causal effects is identified by counterfactual inference. The time-series pairing is transformed into a directed causal edge. The propagation delay and propagation strength of the directed causal edge are calculated as edge weights to construct a causal tracking graph.

[0033] In the causal tracing graph, the intervention variable node corresponding to the target media channel of budget adjustment is located as the transmission source point. The downstream media channels are identified by traversing along the directed causal edge and the multi-hop transmission path of each downstream media channel is extracted. The path transmission attenuation coefficient is obtained by calculating the product of the transmission intensity and the cumulative sum of the transmission delay on the multi-hop transmission path.

[0034] A comprehensive transmission attenuation coefficient is obtained by nonlinearly weighting and aggregating all path transmission attenuation coefficients of the same downstream media channel. The comprehensive transmission attenuation coefficient is then coupled in three dimensions with the path flow distribution entropy change and attribution conversion increment in the two-dimensional evolution descriptor to obtain the transmission effect value.

[0035] A second aspect of the present invention provides a media information effect attribution and budget intelligent allocation system, comprising:

[0036] The data acquisition unit is used to acquire the delivery data and conversion data of multiple media channels within a preset time window. The delivery data includes the exposure, clicks and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to completion of conversion.

[0037] The attribution calculation unit is used to construct a user conversion path map based on the touchpoint sequence, and apply attribution rules to each touchpoint in the user conversion path map to calculate the contribution weight of each media channel to the conversion, so as to obtain the attribution conversion volume of each media channel.

[0038] The model building unit is used to correlate the attribution conversion volume with the placement cost to obtain the attribution ROI of each media channel, and to identify the efficiency stratification status of each media channel based on the attribution ROI. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed.

[0039] The budget allocation unit is used to obtain the target budget allocation scheme for each media channel by solving the budget allocation optimization model, and to distribute the target budget allocation scheme to the corresponding media channel for execution.

[0040] The causal tracking unit is used to collect actual conversion data after the campaign is launched, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

[0041] A third aspect of the present invention provides an electronic device, comprising:

[0042] processor;

[0043] Memory used to store processor-executable instructions;

[0044] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0045] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0046] The beneficial effects of this application are as follows:

[0047] This method enables precise quantitative evaluation of media channel campaign performance. By integrating campaign data with conversion data containing complete touchpoint sequences, it systematically acquires information about the entire user journey from initial contact to final conversion, laying a data foundation for subsequent analysis. Based on the touchpoint sequences, a user conversion path graph is constructed, and attribution rules are applied to calculate the contribution weight of each channel, thereby scientifically distributing the total conversion volume across all touchpoints and obtaining the attributed conversion volume for each media channel. This process overcomes the biases of traditional simple attribution models such as last click, and more realistically reflects the actual role of each channel in the user's decision-making path.

[0048] This method enables dynamic optimization and intelligent decision-making in budget allocation. It correlates attribution conversions with campaign costs to calculate the attribution ROI, reflecting true efficiency. Based on this metric, channels are stratified by efficiency, and an optimization model is constructed with the goal of maximizing the overall attribution ROI within the total budget constraint. Solving the model yields the optimal budget allocation for each channel, which is then automatically deployed and executed, tilting the budget towards high-ROI channels and improving the overall efficiency of marketing fund utilization. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating the media information effect attribution and budget intelligent allocation method according to an embodiment of the present invention;

[0050] Figure 2 This is a flowchart illustrating the budget transmission effect identification based on causal tracking graphs, as described in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0053] Figure 1 This is a flowchart illustrating the media information effect attribution and budget intelligent allocation method according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes:

[0054] Acquire delivery and conversion data from multiple media channels within a preset time window. The delivery data includes the exposure, clicks, and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to conversion.

[0055] Based on the touchpoint sequence, a user conversion path diagram is constructed, and attribution rules are applied to each touchpoint in the user conversion path diagram to calculate the contribution weight of each media channel to the conversion, thereby obtaining the attribution conversion volume of each media channel.

[0056] The attribution conversion rate is correlated with the placement cost to obtain the attribution ROI of each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed.

[0057] The target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution.

[0058] Collect actual conversion data after campaign execution, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

[0059] In one optional implementation, a user conversion path graph is constructed based on the touchpoint sequence, and attribution rules are applied to each touchpoint in the user conversion path graph to calculate the contribution weight of each media channel to the conversion, resulting in the attribution conversion volume for each media channel, including:

[0060] The touchpoint sequence is arranged in chronological order to form a directed path chain. The directed path chain is clustered into path patterns. Multiple paths with similar touchpoint combination features and similar media channel sequence features are grouped into the same path pattern cluster. The typical path structure of each path pattern cluster is extracted as the pattern representative path. Based on the pattern representative path, a compressed user conversion path graph is constructed.

[0061] For each touchpoint in the user conversion path graph, retrieve all actual conversion paths containing the corresponding touchpoint within the path pattern cluster, statistically analyze the conversion success rate distribution when the corresponding touchpoint appears in different path positions, construct a position-sensitive contribution function based on the conversion success rate distribution, and use the position-sensitive contribution function to assign differentiated values ​​to the contribution weight of the corresponding touchpoint under different path structures.

[0062] The total contribution weight of the media channel to the conversion is obtained by weighting and aggregating the contribution weights of all touchpoints belonging to the same media channel under their respective position-sensitive contribution functions. The total contribution weight is then correlated with the total number of conversions associated with the media channel to obtain the attribution conversion volume.

[0063] After collecting the original touchpoint sequence containing user ID, touchpoint media channel identifier, touchpoint timestamp, and final conversion identifier, the touchpoint events are arranged in ascending order according to the timestamp corresponding to each user ID, forming a directed path chain from the first exposure to the final conversion. For users with multiple conversion records, the touchpoint sequence before each conversion is split into independent path chain instances.

[0064] For the entire set of directed path chains, path length, media channel type sequence, and touchpoint time interval distribution are extracted as path feature vectors. An edit distance-based sequence similarity metric is used to calculate the structural similarity score between any two path chains. When the edit distance of the media channel sequence between two paths is below a preset threshold and the KL divergence of the touchpoint time interval distribution is less than the tolerance value, the two paths are considered similar. A hierarchical clustering algorithm is used to merge similar paths, and the layer segmentation that minimizes the variance of the path structure within a cluster and maximizes the significance of the differences between clusters is selected from the clustering tree to complete the path pattern cluster division.

[0065] Within each path pattern cluster, the frequency distribution of media channel sequences for all paths within the cluster is calculated, and the most frequent channel permutations are selected as the typical touchpoint sequences for that cluster. Each touchpoint node in this typical sequence is labeled with its corresponding media channel type and average touchpoint position. Directed edges between nodes represent the temporal dependencies between preceding and subsequent touchpoints, forming a compressed user conversion path graph. This path graph, based on path pattern clusters, compresses the original massive number of discrete paths into a finite number of typical patterns, supporting subsequent efficient attribution calculations.

[0066] For the i-th touchpoint T_i in the path graph, traverse all actual conversion path instances within its corresponding path pattern cluster. Count the number of successful and failed conversions when this touchpoint appears at the beginning, middle, and end of the path, respectively, and calculate the conversion success rate P_pos for each position. Based on the differences in conversion success rates, construct a position-sensitive contribution function. , where d_i represents the path distance from the contact point to the conversion endpoint, and α and β are balance coefficients. This function reflects the non-uniform influence of the contact point position on the conversion driving force; the closer the contact point is to the conversion endpoint and the higher the conversion success rate, the higher its contribution weight.

[0067] For all touchpoints T_1, T_2, ..., T_n associated with the same media channel M_j, extract the position-sensitive contribution weights W_1, W_2, ..., W_n of each touchpoint in its actual occurrence path, and then sum them up using weighted averages. Calculate the total contribution weight of the media channel, where f_k is the frequency of touchpoint T_k in the cluster path. Multiply the total contribution weight C_j by the total number of conversions N_j actually associated with media channel M_j, and normalize by the sum of the total contribution weights of all media channels to obtain the attribution conversion volume of the media channel. This attribution conversion rate reflects the true contribution of media channels to the final conversion after considering differences in path structure and location effects, providing an accurate basis for performance evaluation for subsequent budget optimization and allocation.

[0068] In one optional implementation, the touchpoint sequence is arranged in chronological order to form a directed path chain. The directed path chain is then subjected to path pattern clustering, grouping multiple paths with similar touchpoint combination characteristics and similar media channel sequence characteristics into the same path pattern cluster, including:

[0069] For each directed path chain, construct its touchpoint combination feature representation and media channel sequence feature representation, and fuse and map the touchpoint combination feature representation and the media channel sequence feature representation to obtain the path pattern feature vector of each directed path chain;

[0070] In the space of the path pattern feature vector, the path similarity between any two directed path chains is calculated. When the path similarity exceeds the dynamically adjusted similarity threshold, the corresponding two directed path chains are marked as path pairs with the same pattern. A path connectivity graph is constructed based on the path pairs with the same pattern. The path connectivity graph is decomposed into connected components, and all directed path chains contained in each connected component are merged into the same path pattern cluster.

[0071] When performing path pattern clustering on user conversion paths, it is necessary to extract structured path features from the original touchpoint sequence. For each directed path chain arranged in chronological order, a touchpoint combination feature representation is first constructed. This feature representation is generated by statistically analyzing the types of touchpoints appearing in the path and their frequency of occurrence. For example, the "search ad - feed ad - direct access" contained in a certain path is transformed into a vector form [1, 1, 1, 0, 0], where each dimension corresponds to different touchpoint types in a preset touchpoint type library.

[0072] The media channel sequence feature representation captures the sequential dependencies of media channels within the path, employing a positional encoding method to jointly encode the media channel identifier and its relative position within the path. Specifically, for a media channel m_j appearing at the i-th position in the path, the sequence is calculated... The hash mapping value is used to generate location-aware channel features, and the channel features of all locations in the path are concatenated in order to form a sequence feature vector.

[0073] The feature fusion layer inputs the touchpoint combination feature representation and the media channel sequence feature representation. This fusion layer employs a weighted concatenation mechanism, adaptively allocating fusion weights based on the ratio of touchpoint feature dimensions to sequence feature dimensions, generating a fixed-dimensional path pattern feature vector. This feature vector simultaneously contains information about the path's touchpoint composition and the temporal dependency information of the media channels.

[0074] In the path pattern feature vector space, cosine similarity is used to measure the path similarity between any two directed path chains, i.e., the cosine of the angle between the two feature vectors. When the similarity value exceeds a dynamically adjusted similarity threshold θ_t, the corresponding path pair is marked as a path pair with the same pattern. This threshold θ_t is dynamically updated based on the overall distribution characteristics of the current path set. Specifically, it is calculated by taking the median and standard deviation of the similarity of all path pairs, and setting θ_t = median + 0.5 × std to ensure that the threshold adaptively adjusts with the dispersion of the path distribution.

[0075] A path connectivity graph is constructed based on all identified path pairs with the same pattern. This graph uses directed path chains as nodes, and when two paths form a path pair with the same pattern, an undirected edge is added between the corresponding nodes. A depth-first search algorithm is performed on the constructed path connectivity graph to decompose the connected components. Each time, the graph is traversed starting from an unvisited node, and all reachable nodes are marked as belonging to the same connected component. All directed path chains contained in each connected component are merged into the same path pattern cluster. Paths within a cluster have highly similar touchpoint combination structures and media channel contact sequences.

[0076] After completing the path pattern clustering, the average conversion rate and average path length of the paths within each path pattern cluster are calculated as representative indicators of that pattern cluster. These indicators are used in the subsequent attribution rules to apply differentiated contribution weight allocation strategies for different path patterns, ensuring that the attribution results can reflect the actual differences in the impact of different path patterns on conversion.

[0077] In one optional implementation, the attribution conversion rate is correlated with the placement cost to obtain the attribution ROI for each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. According to the efficiency stratification status and a preset total budget constraint, a budget allocation optimization model is constructed with maximizing the attribution ROI as the objective function, including:

[0078] For each media channel, the attribution conversion volume is multiplied by the preset unit conversion value to obtain the attribution conversion value, and the ratio of the attribution conversion value to the placement cost is calculated to obtain the attribution return on investment.

[0079] The attribution ROI of each media channel is sorted and a quantile threshold is calculated. Media channels with an attribution ROI above the quantile threshold are marked as effective media channels and an uncovered path set is extracted from the user conversion path map. Media channels with an attribution ROI below the quantile threshold are marked as ineffective media channels and a detour path set is extracted from the user conversion path map.

[0080] The budget expansion coefficient for the efficiency-enhancing media channel is calculated based on the number of paths and path length of the uncovered path set. The budget compression coefficient for the inefficient media channel is calculated based on the path connectivity and path conversion rate of the detourable path set. The budget expansion coefficient and the budget compression coefficient are used as budget adjustment boundary parameters to construct budget reallocation constraints.

[0081] Using the budget allocation amount of each media channel as the decision variable, the objective function is to maximize the attribution return on investment, and the budget redistribution constraint and the preset total budget constraint are used as constraints to construct a budget allocation optimization model.

[0082] After obtaining the attribution conversion volume for each media channel, for each media channel i, its attribution conversion volume is first multiplied by a preset unit conversion value. This unit conversion value is determined based on the business type; for example, in an e-commerce scenario, it can be set as the average order amount, and in an education scenario, it can be set as the average course price. The attribution conversion value V_i is obtained through multiplication. Then, the attribution conversion value V_i is divided by the actual campaign cost C_i for that media channel to obtain the attribution return on investment (ROAS_i), which reflects the actual revenue generated per unit cost.

[0083] After obtaining the attribution ROAS for all media channels, the ROAS values ​​of all channels are sorted in descending order, and a preset quantile threshold is calculated. The 75th quantile can be chosen as the efficiency dividing line. Media channels with an attribution ROAS higher than this quantile threshold are marked as maximizing media channels, indicating that these channels have high investment efficiency and room for further investment. For maximizing media channels, all conversion paths are traversed in the user conversion path graph, and path combinations that do not include the channel's touchpoints are extracted to form an uncovered path set, which reflects the expansion potential of maximizing channels. Media channels with an attribution ROAS lower than the quantile threshold are marked as degrading media channels, indicating that these channels have relatively low output from investment. For degrading media channels, paths containing the channel's touchpoints are retrieved in the user conversion path graph. It is determined whether the path still maintains connectivity from the starting point to the conversion endpoint after removing the touchpoint. Paths that meet the connectivity condition are included in the set of detourable paths.

[0084] For media channels that enhance efficiency, the total number of paths in the uncovered path set is counted, and the average path length of these paths is calculated. A larger number of paths and a shorter average length indicate that the channel can quickly reach more potential conversion scenarios. The budget expansion coefficient α_i is obtained by weighted summing the number of paths and the reciprocal of the average path length. This coefficient is typically set between 1.2 and 2.0 to guide the upper limit of budget growth for this channel. For media channels that reduce efficiency, the set of alternative paths is traversed, and the connectivity retention rate of the path after removing the channel touchpoint is calculated. The historical average conversion rate of the alternative paths is also counted. A higher connectivity retention rate and a conversion rate of the alternative path closer to the original path conversion rate indicate a stronger substitutability for this channel. The budget compression coefficient β_i is obtained by combining the connectivity retention rate and the conversion rate ratio of the alternative path. This coefficient is typically set between 0.5 and 0.8, representing the lower limit of the proportion by which the channel's budget can be reduced.

[0085] The budget expansion coefficient for efficiency-enhancing media channels and the budget compression coefficient for inefficiency-reducing media channels are used as boundary parameters for budget reallocation constraints. For efficiency-enhancing channels, the new budget is set to not exceed the original budget multiplied by the budget expansion coefficient; for inefficiency-reducing channels, the new budget is set to not be lower than the original budget multiplied by the budget compression coefficient. Based on this, the budget allocation amount x_i for each media channel is used as the decision variable, and the objective function is constructed as the maximization expression of the weighted sum of the attributable ROI of all channels. The constraints include budget reallocation constraints, the total budget of all channels equaling the preset total budget constraint, and the non-negativity constraint of each channel's budget. This budget allocation optimization model can be solved using linear programming or convex optimization algorithms to obtain the target budget allocation scheme for each media channel that satisfies efficiency maximization, realizing the scientific flow and optimal allocation of budget resources among channels at different efficiency levels.

[0086] In one optional implementation, the target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution, including:

[0087] The budget allocation optimization model is subjected to constraint relaxation and objective function convexization. An iterative solution mechanism is used to calculate the optimal budget allocation for each media channel. The difference between the optimal budget allocation and the current budget allocation is used to calculate the budget adjustment. Based on the positive or negative attribute of the budget adjustment, the media channels are classified into budget increment media channels and budget decrement media channels.

[0088] For media channels with incremental budgets, trace the set of uncovered paths in the user conversion path graph and extract the temporal characteristics of each touchpoint in the path and the combination characteristics of media channels to generate incremental budget execution strategies. For media channels with decremental budgets, trace the set of bypassable paths in the user conversion path graph and extract the alternative nodes of the path and the conversion retention conditions to generate decremental budget execution strategies.

[0089] The optimal budget allocation for the incremental budget media channel and the incremental budget execution strategy are encapsulated into an incremental budget delivery instruction, and the optimal budget allocation for the decremental budget media channel and the decremental budget execution strategy are encapsulated into a decremental budget delivery instruction. The incremental budget delivery instruction and the decremental budget delivery instruction are then sent to the delivery system of the corresponding media channel through the delivery interface.

[0090] After obtaining the budget allocation optimization model, the non-convex constraints in the model are relaxed, converting the original integer constraints into continuous variable constraints. Simultaneously, the nonlinear terms in the objective function are transformed into a convex function through logarithmic transformation or Taylor expansion. An augmented Lagrange function is constructed using the Lagrange multiplier method, and iterative solutions are obtained through gradient descent or Newton's method. In each iteration, the budget allocation for each media channel is updated until the change between two adjacent iterations falls below a preset threshold or the maximum number of iterations is reached. Finally, convergence yields the optimal budget allocation for each media channel.

[0091] The optimal budget allocation obtained from the solution is compared with the current budget allocation being implemented by each media channel on a channel-by-channel basis, and the difference between the two is calculated as the budget adjustment amount. When the budget adjustment amount for a media channel is positive, the channel is marked as a budget-increase type media channel, indicating that the budget for this channel needs to be increased; when the budget adjustment amount is negative, the channel is marked as a budget-decrease type media channel, indicating that the budget for this channel needs to be reduced.

[0092] For media channels identified as having incremental budgets, the user touchpoints not yet reached by these channels are located in the user conversion path map, forming a set of uncovered paths. The temporal characteristics of each touchpoint in these paths are extracted, including the touchpoint's position in the conversion path, the time interval distribution between touchpoints, and the time period characteristics of the touchpoints. Simultaneously, media channel combination characteristics are extracted, including the collaborative occurrence patterns of this channel with other channels in the conversion path and the conversion transfer probability between channels. Based on these characteristics, an incremental budget execution strategy is generated. The strategy includes suggested ad placement time distribution, suggested audience targeting parameters, suggested bid adjustment ranges, and suggestions for joint ad placements with other media channels to create synergy.

[0093] For media channels identified as having reduced budgets, the user conversion path graph is searched for conversion paths containing that channel, and alternative paths are identified where users can achieve conversions through other touchpoint sequences even without going through the channel's touchpoints. Alternative nodes in these alternative paths are extracted; the media channels corresponding to these nodes can take over some of the conversion tasks from the reduced channels. The conversion retention conditions for these alternative paths are analyzed, including historical conversion rate levels, the number of touchpoints required for each alternative path, and the average conversion time for each alternative path. Based on this analysis, a budget reduction execution strategy is generated. This strategy includes a suggested timeline for gradually reducing ad frequency, suggestions for pausing ad campaigns during inefficient periods, recommendations for retaining core audience identifiers, and suggestions for transferring the reduced budget to alternative channels.

[0094] The optimal budget allocation for media channels with incremental budgets is packaged with the corresponding incremental budget execution strategy text to form a structured incremental budget delivery instruction. This instruction includes a target budget field, an execution period field, and strategy parameter fields. Similarly, the optimal budget allocation for media channels with decremental budgets is packaged with the decremental budget execution strategy to form a decremental budget delivery instruction. By calling the delivery interfaces provided by each media channel, using RESTful APIs or message queues, the incremental budget delivery instruction is sent to the delivery system of the media channel that needs to increase its budget, and the decremental budget delivery instruction is sent to the delivery system of the media channel that needs to reduce its budget. Each delivery system automatically adjusts its delivery parameters and executes the new budget allocation scheme after receiving the instruction.

[0095] In one optional implementation, the budget allocation optimization model undergoes constraint relaxation and objective function convexity processing, and an iterative solution mechanism is used to calculate the optimal budget allocation for each media channel, including:

[0096] Identify the nonlinear constraints in the budget allocation optimization model, introduce slack variables for the nonlinear constraints, and set an adaptive penalty coefficient according to the degree of constraint violation. The product of the adaptive penalty coefficient and the slack variables is superimposed as a penalty term on the objective function of the budget allocation optimization model to form a relaxed objective function. Perform convexity analysis on the relaxed objective function, and use piecewise linear approximation to reconstruct the relaxed objective function into a convex objective function for the non-convex regions identified by the convexity analysis.

[0097] A dual problem is established based on the convex objective function and the budget redistribution constraint. The dual variables of the dual problem are initialized. The original problem is solved under the current dual variable conditions to obtain the candidate budget allocation for each media channel. The dual variables are adjusted according to the degree to which the candidate budget allocation satisfies the constraint, and the original problem is solved again. When the dual gap is reduced to a preset accuracy threshold, the iteration stops and the optimal budget allocation for each media channel is determined.

[0098] After obtaining the attribution ROI and efficiency stratification of each media channel, it is necessary to solve the constructed budget allocation optimization model. This model often includes constraints such as upper and lower limits of media channel budgets, fixed total budgets, and nonlinear response of ROI. These constraints make the model non-convex, making it difficult to directly obtain the global optimal solution.

[0099] To address the nonlinear constraints in the model, constraint relaxation techniques are employed. Specifically, when the budget allocation value x_i for a certain media channel needs to satisfy nonlinear constraints... When introducing slack variables Rewrite the constraints as Slack variables allow constraints to be violated to a certain extent, but penalties must be imposed on these violations. The penalty coefficient is set using an adaptive mechanism: a small penalty coefficient is set in the early stages of iteration to maintain the search range of solutions, and the penalty coefficient is gradually increased as iterations progress, forcing the solution to converge towards the feasible region. Adaptive penalty coefficient. Based on the constraint violation amount in the k-th iteration Dynamically adjust the penalty coefficient, multiplying it by a factor when the number of constraint violations exceeds a set threshold. Apply all penalty items. The functions are superimposed on the original objective function to form the relaxed objective function.

[0100] Even after relaxation, the objective function still exhibits non-convex regions. These regions are identified through convexity analysis, which determines non-convexity by calculating the second derivative matrix (Hessian matrix) of the objective function. If the Hessian matrix is ​​not positive definite in a certain region, then that region is non-convex. For these non-convex regions, a piecewise linear approximation method is used for convexification. The domain of the non-convex function is divided into several sub-intervals, and the original function is approximated by a linear function within each sub-interval. The approximation accuracy is controlled by increasing the number of segments; more segments result in smaller approximation errors but increase the number of variables. After completing the piecewise linear approximation, the convex objective function is obtained.

[0101] Based on the convex objective function and constraints such as the total budget not exceeding a preset upper limit and the budgets for each channel being non-negative, a dual problem is established. The dual problem transforms the inequality constraints in the original problem into dual variables, and the optimal solution to the original problem is indirectly obtained by solving the dual problem. When initializing the dual variables, they can be set as a zero vector or based on historical empirical values. Under the current dual variable conditions, the extreme points of the Lagrange function with respect to the original variables are solved to obtain the candidate budget allocations for each media channel.

[0102] After obtaining the candidate budget allocations, examine their satisfaction with the constraints of the original problem, calculate the violation of each constraint, and if a constraint is severely violated, increase the corresponding dual variable to strengthen its effect; if a constraint has significant slack, decrease the corresponding dual variable. The adjustment range of the dual variable is determined based on the degree of constraint violation and the preset step size parameter. After adjusting the dual variable, resolve the original problem to obtain new candidate budget allocations.

[0103] The dual gap is calculated to determine whether the iteration has converged. The dual gap is defined as the difference between the objective function value of the primal problem and the objective function value of the dual problem. When the dual gap shrinks to a preset accuracy threshold, such as 0.01%, it indicates that the solutions to the primal problem and the dual problem are sufficiently close. At this point, the iteration is stopped and the current candidate budget allocation is determined as the optimal budget allocation for each media channel, thus completing the budget optimization allocation calculation.

[0104] In one optional implementation, a causal tracking graph is constructed to trace the relationship between budget changes and conversion path structure evolution across various media channels. Based on this causal tracking graph, the transmission effect of budget adjustments on the conversion contribution of downstream media channels is identified, including:

[0105] Collect budget change data and user conversion path map data of various media channels in multiple time periods, calculate the change in path topology centrality and the change in path traffic distribution entropy before and after budget changes of various media channels to construct a two-dimensional evolutionary descriptor.

[0106] The budget change data is used as an intervention variable, and the two-dimensional evolutionary descriptor is used as a response variable for time-series pairing. The time-series pairing with significant causal effects is identified by counterfactual inference. The time-series pairing is transformed into a directed causal edge. The propagation delay and propagation strength of the directed causal edge are calculated as edge weights to construct a causal tracking graph.

[0107] In the causal tracing graph, the intervention variable node corresponding to the target media channel of budget adjustment is located as the transmission source point. The downstream media channels are identified by traversing along the directed causal edge and the multi-hop transmission path of each downstream media channel is extracted. The path transmission attenuation coefficient is obtained by calculating the product of the transmission intensity and the cumulative sum of the transmission delay on the multi-hop transmission path.

[0108] A comprehensive transmission attenuation coefficient is obtained by nonlinearly weighting and aggregating all path transmission attenuation coefficients of the same downstream media channel. The comprehensive transmission attenuation coefficient is then coupled in three dimensions with the path flow distribution entropy change and attribution conversion increment in the two-dimensional evolution descriptor to obtain the transmission effect value.

[0109] like Figure 2 As shown, the method includes:

[0110] Data is collected from various media channels over multiple consecutive time periods, with each period set to a 24-hour granularity. Budget change data records the amount and timing of budget adjustments for each media channel within each period, in yuan. User conversion path data collects the sequence of touchpoints for all users who have completed conversions within the period, including the media channel identifier corresponding to each touchpoint, the timestamp of the touchpoint, and the temporal connections between touchpoints.

[0111] For each media channel, the change in path topology centrality before and after the budget change is calculated. Path topology centrality is the degree centrality index of the media channel as a node in the conversion path graph, i.e., the ratio of the frequency of the channel appearing as a touchpoint in all conversion paths to the total number of paths. The degree centrality values ​​of the period before and after the budget change are extracted, and the difference between the two is the change in path topology centrality. The change in path traffic distribution entropy is used to quantify the change in the uniformity of conversion traffic distribution across different channels. The proportion of conversion paths undertaken by each media channel before and after the budget change is calculated, and the information entropy formula is used to calculate the proportion distribution. The higher the entropy value, the more uniform the traffic distribution. The change in path traffic distribution entropy is obtained by subtracting the entropy value before the budget change from the entropy value after the budget change. The two-dimensional vector composed of these two changes constitutes the two-dimensional evolutionary descriptor of the media channel.

[0112] Budget changes were used as intervention variables, specifically the amount of budget adjustment for a particular media channel within a given time period. Two-dimensional evolutionary descriptors were used as response variables, representing changes in path topological centrality and path flow distribution entropy observed over several periods following the budget adjustment. For time-series pairing, the budget adjustment date was recorded as time zero, and two-dimensional evolutionary descriptor values ​​for the subsequent 1 to 7 periods were extracted. A counterfactual inference framework was used to construct a virtual control group, which assumed the expected value of the evolutionary descriptor for the media channel without budget adjustment, and synthesized by matching it with historical data from other channels without budget adjustments. The observed two-dimensional evolutionary descriptors and the virtual control group values ​​were subjected to a significance test. If the difference passed the 95% confidence level test, the causal effect of the time-series pairing was considered significant; otherwise, it was discarded.

[0113] The identified significant temporal pairs are transformed into directed causal edges, with the starting point of the media channel node for budget adjustment and the ending point for the media channel node corresponding to the response variable. Transmission delay is defined as the time difference between the budget adjustment time and the time when a significant response change is observed, measured in periods. Transmission strength is a standardized value of the change amplitude of the two-dimensional evolutionary descriptor, specifically the weighted Euclidean distance between the change in path topological centrality and the change in path flow distribution entropy; the weighting coefficients are determined through regression fitting of historical data. Transmission delay and transmission strength together serve as the edge weights of the directed causal edges, completing the construction of the causal tracking graph.

[0114] The target media channels for budget adjustments are identified as nodes in the causal tracing graph, serving as the source points. A breadth-first search is performed along the directed causal edges to identify all reachable downstream media channel nodes. For each downstream channel, all multi-hop transmission paths originating from the source point are extracted, with the path length representing the number of causal edges traversed. The transmission strength product of a single path is calculated by multiplying the transmission strength values ​​of each edge along the path sequentially. The cumulative transmission delay is the arithmetic sum of the transmission delays of each edge along the path. The path transmission attenuation coefficient is defined as the natural logarithm of the cumulative transmission delay plus one, obtained by dividing the cumulative transmission strength product by the cumulative transmission delay, to avoid numerical anomalies when the delay is zero.

[0115] When aggregating the transmission attenuation coefficients of multiple paths within the same downstream media channel, an exponentially weighted average method is used. The weights are the normalized result of an exponential function of the product of path transmission strengths, with higher-strength paths assigned greater weights. The aggregation yields the comprehensive transmission attenuation coefficient for that downstream channel. The change in path traffic allocation entropy and the attribution conversion increment for that downstream channel are extracted. The attribution conversion increment is the difference between the channel's attribution conversion volume after budget adjustment and before adjustment. The three-dimensional coupling calculation formula is: the transmission effect value equals the comprehensive transmission attenuation coefficient multiplied by the change in path traffic allocation entropy, multiplied by the attribution conversion increment, with the unit being the contribution value of attribution conversion volume. This transmission effect value quantifies the specific impact of budget adjustments on the downstream channel's conversion contribution through path structure evolution.

[0116] A second aspect of the present invention provides a media information effect attribution and budget intelligent allocation system, comprising:

[0117] The data acquisition unit is used to acquire the delivery data and conversion data of multiple media channels within a preset time window. The delivery data includes the exposure, clicks and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to completion of conversion.

[0118] The attribution calculation unit is used to construct a user conversion path map based on the touchpoint sequence, and apply attribution rules to each touchpoint in the user conversion path map to calculate the contribution weight of each media channel to the conversion, so as to obtain the attribution conversion volume of each media channel.

[0119] The model building unit is used to correlate the attribution conversion volume with the placement cost to obtain the attribution ROI of each media channel, and to identify the efficiency stratification status of each media channel based on the attribution ROI. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed.

[0120] The budget allocation unit is used to obtain the target budget allocation scheme for each media channel by solving the budget allocation optimization model, and to distribute the target budget allocation scheme to the corresponding media channel for execution.

[0121] The causal tracking unit is used to collect actual conversion data after the campaign is launched, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

[0122] A third aspect of the present invention provides an electronic device, comprising:

[0123] processor;

[0124] Memory used to store processor-executable instructions;

[0125] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0126] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0127] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for media information effect attribution and intelligent budget allocation, characterized in that, include: Acquire delivery and conversion data from multiple media channels within a preset time window. The delivery data includes the exposure, clicks, and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to conversion. Based on the touchpoint sequence, a user conversion path diagram is constructed, and attribution rules are applied to each touchpoint in the user conversion path diagram to calculate the contribution weight of each media channel to the conversion, thereby obtaining the attribution conversion volume of each media channel. The attribution conversion rate is correlated with the placement cost to obtain the attribution ROI of each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed. The target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution. Collect actual conversion data after campaign execution, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

2. The method according to claim 1, characterized in that, Based on the touchpoint sequence, a user conversion path graph is constructed, and attribution rules are applied to each touchpoint in the user conversion path graph to calculate the contribution weight of each media channel to the conversion, resulting in the attribution conversion volume of each media channel, including: The touchpoint sequence is arranged in chronological order to form a directed path chain. The directed path chain is clustered into path patterns. Multiple paths with similar touchpoint combination features and similar media channel sequence features are grouped into the same path pattern cluster. The typical path structure of each path pattern cluster is extracted as the pattern representative path. Based on the pattern representative path, a compressed user conversion path graph is constructed. For each touchpoint in the user conversion path graph, retrieve all actual conversion paths containing the corresponding touchpoint within the path pattern cluster, statistically analyze the conversion success rate distribution when the corresponding touchpoint appears in different path positions, construct a position-sensitive contribution function based on the conversion success rate distribution, and use the position-sensitive contribution function to assign differentiated values ​​to the contribution weight of the corresponding touchpoint under different path structures. The total contribution weight of the media channel to the conversion is obtained by weighting and aggregating the contribution weights of all touchpoints belonging to the same media channel under their respective position-sensitive contribution functions. The total contribution weight is then correlated with the total number of conversions associated with the media channel to obtain the attribution conversion volume.

3. The method according to claim 2, characterized in that, The contact point sequences are arranged in chronological order to form directed path chains. Path pattern clustering is then performed on these directed path chains, grouping multiple paths with similar contact point combination features and similar media channel sequence features into the same path pattern cluster, including: For each directed path chain, construct its touchpoint combination feature representation and media channel sequence feature representation, and fuse and map the touchpoint combination feature representation and the media channel sequence feature representation to obtain the path pattern feature vector of each directed path chain; In the space of the path pattern feature vector, the path similarity between any two directed path chains is calculated. When the path similarity exceeds the dynamically adjusted similarity threshold, the corresponding two directed path chains are marked as path pairs with the same pattern. A path connectivity graph is constructed based on the path pairs with the same pattern. The path connectivity graph is decomposed into connected components, and all directed path chains contained in each connected component are merged into the same path pattern cluster.

4. The method according to claim 1, characterized in that, The attribution conversion rate is correlated with the placement cost to obtain the attribution ROI for each media channel. Based on the attribution ROI, the efficiency stratification status of each media channel is identified. According to the efficiency stratification status and a preset total budget constraint, a budget allocation optimization model is constructed with maximizing the attribution ROI as the objective function, including: For each media channel, the attribution conversion volume is multiplied by the preset unit conversion value to obtain the attribution conversion value, and the ratio of the attribution conversion value to the placement cost is calculated to obtain the attribution return on investment. The attribution ROI of each media channel is sorted and a quantile threshold is calculated. Media channels with an attribution ROI above the quantile threshold are marked as effective media channels and an uncovered path set is extracted from the user conversion path map. Media channels with an attribution ROI below the quantile threshold are marked as ineffective media channels and a detour path set is extracted from the user conversion path map. The budget expansion coefficient for the efficiency-enhancing media channel is calculated based on the number of paths and path length of the uncovered path set. The budget compression coefficient for the inefficient media channel is calculated based on the path connectivity and path conversion rate of the detourable path set. The budget expansion coefficient and the budget compression coefficient are used as budget adjustment boundary parameters to construct budget reallocation constraints. Using the budget allocation amount of each media channel as the decision variable, the objective function is to maximize the attribution return on investment, and the budget redistribution constraint and the preset total budget constraint are used as constraints to construct a budget allocation optimization model.

5. The method according to claim 1, characterized in that, The target budget allocation scheme for each media channel is obtained by solving the budget allocation optimization model, and the target budget allocation scheme is then distributed to the corresponding media channels for execution, including: The budget allocation optimization model is subjected to constraint relaxation and objective function convexization. An iterative solution mechanism is used to calculate the optimal budget allocation for each media channel. The difference between the optimal budget allocation and the current budget allocation is used to calculate the budget adjustment. Based on the positive or negative attribute of the budget adjustment, the media channels are classified into budget increment media channels and budget decrement media channels. For media channels with incremental budgets, trace the set of uncovered paths in the user conversion path graph and extract the temporal characteristics of each touchpoint in the path and the combination characteristics of media channels to generate incremental budget execution strategies. For media channels with decremental budgets, trace the set of bypassable paths in the user conversion path graph and extract the alternative nodes of the path and the conversion retention conditions to generate decremental budget execution strategies. The optimal budget allocation for the incremental budget media channel and the incremental budget execution strategy are encapsulated into an incremental budget delivery instruction, and the optimal budget allocation for the decremental budget media channel and the decremental budget execution strategy are encapsulated into a decremental budget delivery instruction. The incremental budget delivery instruction and the decremental budget delivery instruction are then sent to the delivery system of the corresponding media channel through the delivery interface.

6. The method according to claim 5, characterized in that, The budget allocation optimization model is subjected to constraint relaxation and objective function convexity processing. An iterative solution mechanism is used to calculate the optimal budget allocation for each media channel, including: Identify the nonlinear constraints in the budget allocation optimization model, introduce slack variables for the nonlinear constraints, and set an adaptive penalty coefficient according to the degree of constraint violation. The product of the adaptive penalty coefficient and the slack variables is superimposed as a penalty term on the objective function of the budget allocation optimization model to form a relaxed objective function. Perform convexity analysis on the relaxed objective function, and use piecewise linear approximation to reconstruct the relaxed objective function into a convex objective function for the non-convex regions identified by the convexity analysis. A dual problem is established based on the convex objective function and the budget redistribution constraint. The dual variables of the dual problem are initialized. The original problem is solved under the current dual variable conditions to obtain the candidate budget allocation for each media channel. The dual variables are adjusted according to the degree to which the candidate budget allocation satisfies the constraint, and the original problem is solved again. When the dual gap is reduced to a preset accuracy threshold, the iteration stops and the optimal budget allocation for each media channel is determined.

7. The method according to claim 1, characterized in that, Construct a causal tracking graph of budget changes and conversion path structure evolution for each media channel, and identify the transmission effect of budget adjustments on the conversion contribution of downstream media channels based on the causal tracking graph, including: Collect budget change data and user conversion path map data of various media channels in multiple time periods, calculate the change in path topology centrality and the change in path traffic distribution entropy before and after budget changes of various media channels to construct a two-dimensional evolutionary descriptor. The budget change data is used as an intervention variable, and the two-dimensional evolutionary descriptor is used as a response variable for time-series pairing. The time-series pairing with significant causal effects is identified by counterfactual inference. The time-series pairing is transformed into a directed causal edge. The propagation delay and propagation strength of the directed causal edge are calculated as edge weights to construct a causal tracking graph. In the causal tracing graph, the intervention variable node corresponding to the target media channel of budget adjustment is located as the transmission source point. The downstream media channels are identified by traversing along the directed causal edge and the multi-hop transmission path of each downstream media channel is extracted. The path transmission attenuation coefficient is obtained by calculating the product of the transmission intensity and the cumulative sum of the transmission delay on the multi-hop transmission path. A comprehensive transmission attenuation coefficient is obtained by nonlinearly weighting and aggregating all path transmission attenuation coefficients of the same downstream media channel. The comprehensive transmission attenuation coefficient is then coupled in three dimensions with the path flow distribution entropy change and attribution conversion increment in the two-dimensional evolution descriptor to obtain the transmission effect value.

8. A media information effect attribution and budget intelligent allocation system, used to implement the method of any one of claims 1-7, characterized in that, include: The data acquisition unit is used to acquire the delivery data and conversion data of multiple media channels within a preset time window. The delivery data includes the exposure, clicks and delivery cost of each media channel, and the conversion data includes the sequence of touchpoints from user contact to completion of conversion. The attribution calculation unit is used to construct a user conversion path map based on the touchpoint sequence, and apply attribution rules to each touchpoint in the user conversion path map to calculate the contribution weight of each media channel to the conversion, so as to obtain the attribution conversion volume of each media channel. The model building unit is used to correlate the attribution conversion volume with the placement cost to obtain the attribution ROI of each media channel, and to identify the efficiency stratification status of each media channel based on the attribution ROI. Based on the efficiency stratification status and the preset total budget constraint, a budget allocation optimization model with maximizing the attribution ROI as the objective function is constructed. The budget allocation unit is used to obtain the target budget allocation scheme for each media channel by solving the budget allocation optimization model, and to distribute the target budget allocation scheme to the corresponding media channel for execution. The causal tracking unit is used to collect actual conversion data after the campaign is launched, construct a causal tracking graph between budget changes and conversion path structure evolution of each media channel, identify the transmission effect of budget adjustment on the conversion contribution of downstream media channels based on the causal tracking graph, and feed the transmission effect back to the contribution weight calculation of the attribution rule.

9. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 7.

Images