Method and system for returning and distributing operation data attribution ratio based on user behavior

Abstract

The present invention provides a method and system for returning and distributing attribution ratio of operation data based on user behavior, which relates to the field of information push technology, including real-time collection of user behavior data; extracting features from pre-processed behavior data, wherein the features include user active time period, click intensity and dwell time; constructing a feature matrix X and a target variable y; training the feature matrix X and the target variable y using a machine learning model to obtain the contribution weight of each operation data touch point; calculating the attribution ratio of each operation data touch point based on the contribution weight and in combination with a time-decaying Markov model; returning the calculated attribution ratio to an operation data platform and an operation data master in real time through an encrypted API interface; and dynamically adjusting the operation data delivery strategy and budget allocation according to the received attribution ratio by the operation data platform and the operation data master.

Description

Technical Field

[0001] The present invention relates to information push technology, and in particular to a method and system for returning and distributing attribution ratios of operational data based on user behavior. Background Art

[0002] Among the existing operational data attribution technologies, the most common methods include Last Click Attribution and Linear Attribution. Although these methods are simple and easy to use, they have obvious limitations:

[0003] Last click attribution: This method gives all the attribution power to the operational data of the user's last click, ignoring other operational data touchpoints that the user touched during the decision-making process. This method is too simplistic and cannot accurately reflect the overall effect of operational data.

[0004] Linear attribution: This method evenly distributes attribution to all touchpoints. Although it is fairer than last-click attribution, it still cannot accurately reflect the actual influence of each touchpoint.

[0005] The main disadvantage of these traditional methods is that they cannot accurately evaluate the actual contribution of each operational data touchpoint to user decision-making, resulting in inaccurate allocation of operational data budgets and affecting the efficiency and effectiveness of operational data delivery. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for returning and distributing attribution ratios of operational data based on user behavior, which can solve the problems in the prior art.

[0007] According to a first aspect of the embodiments of the present invention,

[0008] Provides a method for distributing the attribution ratio of operational data based on user behavior, including:

[0009] By deploying the tracking code on the client side, the user's behavior data on clicks and viewing time of the operational data is collected in real time; the behavior data is transmitted to the data processing center on the server side; the behavior data is pre-processed in the data processing center, including: deleting missing values ​​and duplicate data, and converting timestamps into a unified format;

[0010] Extract features from the preprocessed behavior data, the features including user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, where X contains the extracted features and y represents the user's conversion results; use a machine learning model to train the feature matrix X and the target variable y to obtain the contribution weight of each operational data touch point; based on the contribution weight, calculate the attribution ratio of each operational data touch point in combination with a time-decaying Markov model;

[0011] The calculated attribution ratio is transmitted back to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio.

[0012] In an optional embodiment,

[0013] The feature matrix X and the target variable y are trained using a machine learning model to obtain the contribution weights of each operational data touchpoint, including:

[0014] The user behavior data set is randomly divided into a training set and a test set according to a preset ratio; the feature matrix is ​​standardized to obtain a standardized feature matrix;

[0015] Initialize the random forest model and gradient boosting decision tree model, and set the initial hyperparameters; use the training set data to train the random forest model and gradient boosting decision tree model;

[0016] Use the test set to evaluate the performance of the random forest model and the gradient boosted decision tree model, and calculate the root mean square error and the coefficient of determination; optimize the hyperparameters of the random forest model and the gradient boosted decision tree model through grid search and cross-validation methods; extract feature importance from the optimized random forest model and the gradient boosted decision tree model; normalize the feature importance of the random forest model and the gradient boosted decision tree model; perform weighted average of the normalized feature importance of the random forest model and the gradient boosted decision tree model to obtain the integrated feature importance;

[0017] Establish a mapping relationship between features and operational data touchpoints; calculate the contribution weight of each operational data touchpoint based on the importance of integrated features and the mapping relationship between features and operational data touchpoints; normalize the obtained operational data touchpoint contribution weights to obtain the final operational data touchpoint contribution weights.

[0018] In an optional embodiment,

[0019] Based on the contribution weights, the attribution ratio of each operational data touchpoint is calculated in combination with the time decay Markov model, including:

[0020] An operation data contact point transfer probability matrix is ​​constructed based on the user behavior sequence, wherein the elements in the operation data contact point transfer probability matrix represent the transfer probability from one operation data contact point to another operation data contact point; a time decay function is introduced, wherein the time decay function is an exponential decay function, wherein the decay parameter is determined by cross-validation; the time decay function is applied to the operation data contact point transfer probability matrix to obtain an attenuated operation data contact point transfer probability matrix, wherein the attenuated transfer probability is the product of the original transfer probability and the time decay factor;

[0021] The power iteration method is used to solve the steady-state distribution of the attenuated operation data contact transfer probability matrix; based on the steady-state distribution and the predetermined operation data contact contribution weight, the attribution ratio of each operation data contact is calculated.

[0022] In an optional embodiment,

[0023] The method further comprises:

[0024] Smoothing the attenuated operation data contact transfer probability matrix to avoid zero probability problems;

[0025] Construct multi-order Markov chains to capture user behavior patterns in longer historical states; use grid search or Bayesian optimization methods to adaptively select the optimal time decay parameters;

[0026] Calculate the log-likelihood of the model for the test data to evaluate the model's fit; use the model to predict the next possible operational data touchpoint and calculate the prediction accuracy.

[0027] In an optional embodiment,

[0028] The calculated attribution ratio is sent back to the operation data platform and the operation data master in real time through an encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio, including:

[0029] Encrypt the operation data touchpoint attribution result by using an asymmetric encryption algorithm RSA to obtain an encrypted attribution result; receive the encrypted attribution result through a RESTful API interface, and forward the encrypted attribution result to a message queue;

[0030] Using Apache Kafka as the message queue to ensure high throughput and low latency processing of the encrypted attribution results; using Apache Flink to perform real-time data processing on the encrypted attribution results in the message queue, including decrypting the encrypted attribution results and performing preliminary analysis;

[0031] Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for the operational data delivery strategy, including budget allocation plans and bidding strategies; the optimization suggestions are sent to the operational data platform and the operational data master through a real-time feedback mechanism;

[0032] Use Redis to cache frequently accessed attribution result data to reduce database query pressure; implement an adaptive batch processing mechanism to dynamically adjust the batch size according to the system load, with the initial batch size of 100 and the maximum batch size of 1000;

[0033] Realize multi-tenant data isolation to ensure data security between different operational data masters; periodically update the optimization suggestions with an update cycle of 5 minutes;

[0034] When generating the optimization suggestions, the operational data touchpoints are divided into three bidding strategies: aggressive, moderate and conservative, according to the attribution ratio. The aggressive strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.2, the moderate strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.1 and less than or equal to 0.2, and the conservative strategy is adopted for operational data touchpoints with an attribution ratio less than or equal to 0.1;

[0035] Based on the optimization suggestions, the operational data delivery parameters, including budget allocation and bidding strategies, are dynamically adjusted to form a closed-loop optimization system.

[0036] In an optional embodiment,

[0037] Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for operational data delivery strategies. The optimization suggestions include budget allocation plans and bidding strategies, including:

[0038] Receive decrypted attribution results, historical performance data, and market competition data;

[0039] Using a data preprocessing module to merge, process missing values, and detect outliers on the attribution results, the historical performance data, and the market competition data, to obtain preprocessed data;

[0040] Based on the preprocessed data, calculating an efficiency score and a market share, wherein the efficiency score is a ratio of a conversion rate to a cost per click, and the market share is a ratio of the number of operational data impressions to the number of total market impressions;

[0041] The budget allocation is optimized using a convex optimization algorithm, where the objective function is to maximize the sum of the products of the efficiency score and the budget allocation, and the constraint is that the sum of the budget allocation is equal to the total budget;

[0042] Generate a bidding strategy based on the attribution ratio, market share, and budget share. An aggressive strategy is used when the attribution ratio is greater than 0.2 and the market share is less than 0.3. A moderate strategy is used when the attribution ratio is greater than 0.1 or the budget share is greater than 0.15. A conservative strategy is used in other cases.

[0043] Calculate the bid adjustment coefficient based on the bidding strategy, where the bid adjustment coefficient for the aggressive strategy is 1.2, the bid adjustment coefficient for the moderate strategy is 1.0, and the bid adjustment coefficient for the conservative strategy is 0.8;

[0044] The optimized performance is predicted using a random forest regression model, where the input features include budget, bid adjustment factor, and market share, and the output is the predicted return on investment;

[0045] Integrate the optimized budget allocation, bidding strategy and predicted performance to generate the final optimization recommendations and calculate the expected improvement;

[0046] Sort the optimization suggestions in descending order according to the expected improvement degree to obtain a sorted optimization suggestion list;

[0047] The sorted optimization suggestion list is output to the operation data master for guiding the adjustment of the operation data delivery strategy.

[0048] According to a second aspect of the embodiments of the present invention,

[0049] Provides a return distribution system for operational data attribution ratio based on user behavior, including:

[0050] The first unit is used to collect the user's behavior data on clicks and viewing time of the operation data in real time through the tracking code deployed on the client; transmit the behavior data to the data processing center on the server; pre-process the behavior data in the data processing center, including: deleting missing values ​​and duplicate data, and converting timestamps into a unified format;

[0051] The second unit is used to extract features from the preprocessed behavior data, wherein the features include user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, wherein X includes the extracted features and y represents the user's conversion result; train the feature matrix X and the target variable y using a machine learning model to obtain the contribution weight of each operation data touch point; and calculate the attribution ratio of each operation data touch point based on the contribution weight and in combination with a time decay Markov model;

[0052] The third unit is used to transmit the calculated attribution ratio to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio.

[0053] According to a third aspect of the embodiments of the present invention,

[0054] An electronic device is provided, comprising:

[0055] processor;

[0056] a memory for storing processor-executable instructions;

[0057] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0058] A fourth aspect of the embodiments of the present invention is:

[0059] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0060] The technical solution of this application can improve the efficiency of using the operating data budget: accurate attribution information helps the operating data owner to allocate the operating data budget more reasonably. Enhance the accuracy of operating data delivery: understanding the actual effect of each operating data touchpoint can more accurately target the target audience and optimize the operating data content. Improve the overall operating data ROI: improve the return on operating data by optimizing the operating data delivery strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 It is a flow chart of a method for returning and distributing attribution ratio of operation data based on user behavior according to an embodiment of the present invention;

[0062] Figure 2 The present invention is a schematic diagram of the structure of the system for returning and distributing the attribution ratio of operational data based on user behavior according to an embodiment of the present invention. DETAILED DESCRIPTION

[0063] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0064] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0065] Figure 1 FIG. 1 is a flow chart of a method for returning and distributing attribution ratio of operational data based on user behavior according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0066] S101. Collect the user's click and viewing time behavior data on the operation data in real time through the tracking code deployed on the client; transmit the behavior data to the data processing center on the server; pre-process the behavior data in the data processing center, including: deleting missing values ​​and duplicate data, and converting timestamps into a unified format;

[0067] S102. Extract features from the preprocessed behavior data, the features including user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, where X includes the extracted features and y represents the user's conversion result; train the feature matrix X and the target variable y using a machine learning model to obtain the contribution weight of each operational data touch point; calculate the attribution ratio of each operational data touch point based on the contribution weight and in combination with a time decay Markov model;

[0068] S103. The calculated attribution ratio is transmitted back to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio.

[0069] In an optional embodiment,

[0070] The feature matrix X and the target variable y are trained using a machine learning model to obtain the contribution weights of each operational data touchpoint, including:

[0071] The user behavior data set is randomly divided into a training set and a test set according to a preset ratio; the feature matrix is ​​standardized to obtain a standardized feature matrix;

[0072] Initialize the random forest model and gradient boosting decision tree model, and set the initial hyperparameters; use the training set data to train the random forest model and gradient boosting decision tree model;

[0073] Use the test set to evaluate the performance of the random forest model and the gradient boosted decision tree model, and calculate the root mean square error and the coefficient of determination; optimize the hyperparameters of the random forest model and the gradient boosted decision tree model through grid search and cross-validation methods; extract feature importance from the optimized random forest model and the gradient boosted decision tree model; normalize the feature importance of the random forest model and the gradient boosted decision tree model; perform weighted average of the normalized feature importance of the random forest model and the gradient boosted decision tree model to obtain the integrated feature importance;

[0074] Establish a mapping relationship between features and operational data touchpoints; calculate the contribution weight of each operational data touchpoint based on the importance of integrated features and the mapping relationship between features and operational data touchpoints; normalize the obtained operational data touchpoint contribution weights to obtain the final operational data touchpoint contribution weights.

[0075] Exemplarily, the detailed technical contents of using the machine learning model to train the feature matrix X and the target variable y to obtain the contribution weights of each operational data touchpoint are as follows:

[0076] First, the collected user behavior data set is randomly divided into 80% training set and 20% test set. Assuming there are 10,000 user behavior records, the training set will contain 8,000 records and the test set will contain 2,000 records.

[0077] Standardize the feature matrix using the Z-score standardization method. For each feature, calculate its mean and standard deviation, then subtract the mean from each feature value and divide it by the standard deviation. For example, for the feature of click count, if the original value is 100, the mean of the feature is 50, and the standard deviation is 25, then the standardized value is (100-50) / 25=2.

[0078] Initialize the random forest model and set the initial hyperparameters: the number of trees is 100, the maximum depth is 10, and the minimum number of leaf node samples is 5. At the same time, initialize the gradient boosting decision tree model and set the initial hyperparameters: the number of trees is 100, the learning rate is 0.1, and the maximum depth is 5.

[0079] The two models are trained using the training set data. The random forest model trains 100 decision trees in parallel, with each tree built on a randomly selected subset of features. The gradient boosted decision tree model uses an iterative approach, building one tree at a time to fit the residuals of the previous model.

[0080] Use the test set to evaluate the performance of both models. Calculate the root mean square error (RMSE) and the coefficient of determination (R²). Assume that the RMSE of the random forest model is 0.15 and the R² is 0.85; the RMSE of the gradient boosted decision tree model is 0.14 and the R² is 0.87.

[0081] Model hyperparameters were optimized by grid search and 5-fold cross validation. For the random forest model, the parameter combinations tried included: number of trees [100, 200, 300], maximum depth [10, 15, 20], minimum number of leaf node samples [3, 5, 7]. For the gradient boosted decision tree model, the parameter combinations tried included: number of trees [100, 200, 300], learning rate [0.05, 0.1, 0.2], maximum depth [3, 5, 7].

[0082] Select the parameter combination that makes the validation set performance optimal. Assume that the parameters of the optimized random forest model are: the number of trees is 200, the maximum depth is 15, and the minimum number of leaf node samples is 3; the parameters of the optimized gradient boosting decision tree model are: the number of trees is 200, the learning rate is 0.1, and the maximum depth is 5.

[0083] Feature importance is extracted from the optimized random forest model and gradient boosted decision tree model. The random forest model measures feature importance by calculating the average impurity reduction of each feature across all decision trees. The gradient boosted decision tree model measures feature importance by counting the number of times each feature is used as a split node.

[0084] The extracted feature importances are normalized so that the sum of the feature importances of each model is 1. For example, if the original importances of click counts, browsing time, and page depth in the random forest model are 0.5, 0.3, and 0.2, respectively, the normalized importances are still 0.5, 0.3, and 0.2.

[0085] The normalized feature importances of the random forest model and the gradient boosting decision tree model are weighted averaged, with each weight being 0.5, to obtain the integrated feature importance. For example, if the normalized importances of click counts, browsing time, and page depth in the gradient boosting decision tree model are 0.4, 0.4, and 0.2, then the integrated feature importances are (0.5*0.5+0.4*0.5, 0.3*0.5+0.4*0.5, 0.2*0.5+0.2*0.5), i.e. (0.45, 0.35, 0.2).

[0086] Establish a mapping relationship between features and operation data touchpoints. For example, map features such as "search operation data clicks" and "search operation data impressions" to the "search operation data" touchpoint, and map features such as "display operation data clicks" and "display operation data impressions" to the "display operation data" touchpoint.

[0087] The contribution weight of each operational data touchpoint is calculated based on the importance of the integrated features and the mapping relationship between the features and the operational data touchpoints. For each operational data touchpoint, the importance of all features related to it is added up. For example, if the integrated feature importance of "Search operational data clicks" and "Search operational data impressions" are 0.3 and 0.2 respectively, the contribution weight of the "Search operational data" touchpoint is 0.5.

[0088] The obtained operation data touch point contribution weights are normalized so that the sum of the contribution weights of all touch points is 1, and the final operation data touch point contribution weights are obtained. For example, if the original search operation data, display operation data, and social media operation data contribution weights are 0.5, 0.3, and 0.2, respectively, the normalized contribution weights are still 0.5, 0.3, and 0.2.

[0089] Through the above steps, the machine learning model was successfully used to train the feature matrix X and the target variable y, and the contribution weights of each operational data touchpoint were obtained. This method combines the advantages of the random forest and gradient boosting decision tree models. Through feature importance analysis and operational data touchpoint mapping, interpretable and reliable operational data touchpoint contribution weights were finally obtained. These weights can be directly used to guide the formulation and optimization of operational data delivery strategies, helping operational data owners to allocate budgets and resources more effectively.

[0090] In an optional embodiment,

[0091] Based on the contribution weights, the attribution ratio of each operational data touchpoint is calculated in combination with the time decay Markov model, including:

[0092] An operation data contact point transfer probability matrix is ​​constructed based on the user behavior sequence, wherein the elements in the operation data contact point transfer probability matrix represent the transfer probability from one operation data contact point to another operation data contact point; a time decay function is introduced, wherein the time decay function is an exponential decay function, wherein the decay parameter is determined by cross-validation; the time decay function is applied to the operation data contact point transfer probability matrix to obtain an attenuated operation data contact point transfer probability matrix, wherein the attenuated transfer probability is the product of the original transfer probability and the time decay factor;

[0093] The power iteration method is used to solve the steady-state distribution of the attenuated operation data contact transfer probability matrix; based on the steady-state distribution and the predetermined operation data contact contribution weight, the attribution ratio of each operation data contact is calculated.

[0094] In an optional embodiment,

[0095] The method further comprises:

[0096] Smoothing the attenuated operation data contact transfer probability matrix to avoid zero probability problems;

[0097] Construct multi-order Markov chains to capture user behavior patterns in longer historical states; use grid search or Bayesian optimization methods to adaptively select the optimal time decay parameters;

[0098] Calculate the log-likelihood of the model for the test data to evaluate the model's fit; use the model to predict the next possible operational data touchpoint and calculate the prediction accuracy.

[0099] For example, the detailed technical content of calculating the attribution ratio of each operational data touchpoint based on the contribution weight and in combination with the time decay Markov model is as follows:

[0100] First, we construct the operation data touchpoint transfer probability matrix based on the user behavior sequence. Assume that there are three types of operation data touchpoints: search operation data (S), display operation data (D), and social media operation data (M). By analyzing the operation data contact sequences of a large number of users, we count the number of transfers from one operation data touchpoint to another. For example, after analyzing 1,000 user behavior sequences, we get the following number of transfers:

[0101] S to S: 200 times, S to D: 150 times, S to M: 100 times;

[0102] D to S: 120 times, D to D: 180 times, D to M: 150 times;

[0103] M to S: 80 times, M to D: 100 times, M to M: 220 times;

[0104] Normalize these transfer times to get the operation data contact point transfer probability matrix:

[0105] S to S: 0.44, S to D: 0.33, S to M: 0.22;

[0106] D to S: 0.27, D to D: 0.40, D to M: 0.33;

[0107] M to S: 0.20, M to D: 0.25, M to M: 0.55;

[0108] Next, we introduce the time decay function to consider the impact of time on the effect of operational data. We use the exponential decay function as the time decay function, where the decay parameter is determined by cross-validation. Assuming that we use 5-fold cross-validation to test the performance of the decay parameter in the range of [0.1, 0.2, 0.3, 0.4, 0.5], we finally determine that the optimal decay parameter is 0.3.

[0109] Apply the time decay function to the operational data touchpoint transition probability matrix. For each pair of operational data touchpoint transitions, calculate the average time interval between them (in days), and then calculate the decay factor using the time decay function. For example, if the average time interval from S to D is 2 days, then the decay factor is e^(-0.3*2) ≈ 0.55. Multiply the original transition probability by the decay factor to get the attenuated transition probability.

[0110] Assume that the attenuated operational data contact transfer probability matrix is:

[0111] S to S: 0.35, S to D: 0.18, S to M: 0.11;

[0112] D to S: 0.16, D to D: 0.32, D to M: 0.20;

[0113] M to S: 0.10, M to D: 0.15, M to M: 0.44;

[0114] The power iteration method is used to solve the steady-state distribution of the contact transfer probability matrix of the attenuated operational data. The steps of the power iteration method are as follows:

[0115] 1. Initialize a random vector, for example [1 / 3, 1 / 3, 1 / 3].

[0116] 2. Multiply the vector by the transition probability matrix.

[0117] 3. Normalize the new vector obtained.

[0118] 4. Repeat steps 2 and 3 until the vector converges or the maximum number of iterations is reached.

[0119] Assume that after 100 iterations, the steady-state distribution is [0.28, 0.32, 0.40], which means that in the long run, the probability of users being exposed to search operation data, display operation data, and social media operation data is 28%, 32%, and 40%, respectively.

[0120] Finally, based on the steady-state distribution and the predetermined contribution weights of the operational data touchpoints, the attribution ratio of each operational data touchpoint is calculated. Assume that the predetermined contribution weights of the operational data touchpoints are: 0.5 for search operational data, 0.3 for display operational data, and 0.2 for social media operational data.

[0121] Multiplying the steady-state distribution by the contribution weight yields:

[0122] Search operation data: 0.28 * 0.5 = 0.14;

[0123] Display operation data: 0.32 * 0.3 = 0.096;

[0124] Social media operation data: 0.40 * 0.2 = 0.08;

[0125] Normalize these results to get the final attribution ratio:

[0126] Search operations data: 0.14 / (0.14 + 0.096 + 0.08) ≈ 0.44 (44%);

[0127] Display operation data: 0.096 / (0.14 + 0.096 + 0.08) ≈ 0.30 (30%);

[0128] Social media operation data: 0.08 / (0.14 + 0.096 + 0.08) ≈ 0.26 (26%);

[0129] This attribution method based on the time-decay Markov model takes into account the mutual influence between operational data touchpoints, the time decay effect, and the inherent contribution weight of each touchpoint. By constructing a transition probability matrix, introducing a time decay function, and solving the steady-state distribution, a more comprehensive and accurate attribution ratio of operational data touchpoints is finally obtained. These attribution ratios can help operational data owners better understand the actual contribution of each operational data channel, thereby optimizing operational data delivery strategies and budget allocation.

[0130] For example, based on the above results, the operation data owner may consider allocating more budget to search operation data because it contributes 44% of conversions. At the same time, although social media operation data accounts for the highest proportion in the steady-state distribution, due to its low contribution weight, the final attribution ratio is the smallest, which suggests that the operation data owner may need to improve the content or delivery strategy of social media operation data to increase its actual contribution.

[0131] The advantage of this method is that it combines the dynamic characteristics of the Markov model with the realistic consideration of time decay, and at the same time avoids the deviation that may be caused by relying entirely on user behavior sequences by introducing contribution weights. In this way, operational data owners can obtain more balanced and accurate multi-channel attribution results, providing strong support for data-driven marketing decisions.

[0132] In an optional embodiment,

[0133] The calculated attribution ratio is sent back to the operation data platform and the operation data master in real time through an encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio, including:

[0134] Encrypt the operation data touchpoint attribution result by using an asymmetric encryption algorithm RSA to obtain an encrypted attribution result; receive the encrypted attribution result through a RESTful API interface, and forward the encrypted attribution result to a message queue;

[0135] Using Apache Kafka as the message queue to ensure high throughput and low latency processing of the encrypted attribution results; using Apache Flink to perform real-time data processing on the encrypted attribution results in the message queue, including decrypting the encrypted attribution results and performing preliminary analysis;

[0136] Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for the operational data delivery strategy, including budget allocation plans and bidding strategies; the optimization suggestions are sent to the operational data platform and the operational data master through a real-time feedback mechanism;

[0137] Use Redis to cache frequently accessed attribution result data to reduce database query pressure; implement an adaptive batch processing mechanism to dynamically adjust the batch size according to the system load, with the initial batch size of 100 and the maximum batch size of 1000;

[0138] Realize multi-tenant data isolation to ensure data security between different operational data masters; periodically update the optimization suggestions with an update cycle of 5 minutes;

[0139] When generating the optimization suggestions, the operational data touchpoints are divided into three bidding strategies: aggressive, moderate and conservative, according to the attribution ratio. The aggressive strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.2, the moderate strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.1 and less than or equal to 0.2, and the conservative strategy is adopted for operational data touchpoints with an attribution ratio less than or equal to 0.1;

[0140] Based on the optimization suggestions, the operational data delivery parameters, including budget allocation and bidding strategies, are dynamically adjusted to form a closed-loop optimization system.

[0141] Exemplarily, the calculated attribution ratio is transmitted back to the operation data platform and the operation data master in real time through an encrypted API interface, and the detailed technical contents of dynamically adjusting the operation data delivery strategy and budget allocation according to the received attribution ratio are as follows:

[0142] First, use the RSA asymmetric encryption algorithm to encrypt the attribution results of the operational data touchpoints. Assume that there are three attribution ratios of operational data touchpoints: search operational data 44%, display operational data 30%, and social media operational data 26%. Use a 2048-bit RSA public key to encrypt these data and obtain an encrypted string, such as "8f3a9c2b7e1d6f4a5c8b2d9e7f3a1c5b...".

[0143] Build a RESTful API interface to receive encrypted attribution results. The API endpoint is designed as " / api / v1 / attribution" and uses the POST method to send encrypted data. After the API server receives the encrypted data, it immediately forwards it to the Kafka message queue to ensure high throughput and low latency processing.

[0144] The Kafka message queue is configured with 3 brokers, 5 partitions, and a replication factor of 2 to balance performance and reliability. Create a topic named "attribution-results" to store encrypted attribution results.

[0145] Use Apache Flink for real-time data processing. Configure a Flink cluster, including 3 TaskManagers, each with 4GB of memory and 2 processing slots. The Flink job consumes data from the Kafka message queue, decrypts the RSA-encrypted attribution results using the private key, and performs preliminary analysis, such as calculating the average attribution ratio change per hour.

[0146] Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for operational data delivery strategies. The optimization engine considers the following factors: historical performance data, current budget usage, competitor bids, market trends, etc. For example, for search operational data (attribution ratio 44%), it is recommended to increase the budget by 15% and the maximum bid by 10%; for display operational data (attribution ratio 30%), it is recommended to maintain the current budget and increase the maximum bid by 5%; for social media operational data (attribution ratio 26%), it is recommended to reduce the budget by 10% and maintain the current bid.

[0147] The real-time feedback mechanism sends optimization suggestions to the operation data platform and the operation data master through the WebSocket protocol, ensuring millisecond-level real-time performance. The client of the operation data platform and the operation data master implements WebSocket connection to receive real-time updated optimization suggestions.

[0148] Use Redis to cache frequently accessed attribution result data. Configure a Redis cluster, including 3 master nodes and 3 slave nodes, and enable the persistence mechanism. Set the cache expiration time to 10 minutes and use the LRU (least recently used) algorithm for cache elimination.

[0149] An adaptive batching mechanism is implemented, with the initial batch size set to 100. The system monitors processing latency and CPU usage, and when the processing latency exceeds 100 milliseconds or the CPU usage exceeds 80%, the batch size is increased by 50 until the maximum batch size of 1000 is reached. When the processing latency is less than 50 milliseconds and the CPU usage is less than 60%, the batch size is reduced by 50, but not less than the initial value of 100.

[0150] Implement multi-tenant data isolation and assign a unique tenant ID to each operational data master. At the database level, use Schema to isolate data from different tenants. At the application level, implement middleware to ensure that each query contains a tenant ID filter condition.

[0151] The optimization recommendations are updated every 5 minutes. The distributed scheduling system Apache Airflow is used to create a DAG (directed acyclic graph) workflow to trigger the optimization recommendation update task at a fixed time. The update process includes: recalculating the attribution ratio, generating new optimization recommendations, updating the Redis cache, and pushing new recommendations through WebSocket.

[0152] Operational data touchpoints are divided into three bidding strategies based on the attribution ratio:

[0153] 1. Aggressive strategy (attribution ratio > 0.2): For example, this strategy is used for search operation data (44%) and display operation data (30%). Increase the maximum bid by 15%, increase the display frequency, and give priority to obtaining high-quality traffic.

[0154] 2. Moderate strategy (0.1 < attribution ratio ≤ 0.2): For example, social media operation data (26%) adopts this strategy. Maintain the current highest bid and fine-tune the targeting conditions to increase conversion rate.

[0155] 3. Conservative strategy (attribution ratio ≤ 0.1): Assuming there is another operational data touchpoint, such as alliance operational data, with an attribution ratio of 8%, this strategy is adopted. Reduce the maximum bid by 10%, reduce investment in this channel, and transfer the budget to channels with better performance.

[0156] Based on the optimization suggestions, the system automatically adjusts the operational data delivery parameters. For example, for search operational data, the daily budget is increased from 1,000 yuan to 1,150 yuan, and the maximum bid per click is increased from 5 yuan to 5.5 yuan. At the same time, the keyword bid is adjusted, and the bid weight is increased for high-conversion rate keywords.

[0157] A closed-loop optimization system is implemented to continuously monitor the effects of adjusted operational data. The return on investment (ROI) and click-through rate (CTR) are calculated every hour. If the ROI drops by more than 10% or the CTR drops by more than 15%, an emergency adjustment mechanism is triggered to roll back to the last stable configuration.

[0158] Through the above detailed technical implementation, the system can achieve secure transmission, real-time processing, intelligent optimization and dynamic adjustment of attribution results. This closed-loop optimization mechanism can help the operation data platform and operation data owners to more effectively use the attribution analysis results, achieve continuous optimization of operation data delivery strategies, and maximize operation data effects and investment returns. At the same time, the system's high performance, security and scalability design ensure stable operation in large-scale operation data processing scenarios.

[0159] In an optional embodiment,

[0160] Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for operational data delivery strategies. The optimization suggestions include budget allocation plans and bidding strategies, including:

[0161] Receive decrypted attribution results, historical performance data, and market competition data;

[0162] Using a data preprocessing module to merge, process missing values, and detect outliers on the attribution results, the historical performance data, and the market competition data, to obtain preprocessed data;

[0163] Based on the preprocessed data, calculating an efficiency score and a market share, wherein the efficiency score is a ratio of a conversion rate to a cost per click, and the market share is a ratio of the number of operational data impressions to the number of total market impressions;

[0164] The budget allocation is optimized using a convex optimization algorithm, where the objective function is to maximize the sum of the products of the efficiency score and the budget allocation, and the constraint is that the sum of the budget allocation is equal to the total budget;

[0165] Generate a bidding strategy based on the attribution ratio, market share, and budget share. An aggressive strategy is used when the attribution ratio is greater than 0.2 and the market share is less than 0.3. A moderate strategy is used when the attribution ratio is greater than 0.1 or the budget share is greater than 0.15. A conservative strategy is used in other cases.

[0166] Calculate the bid adjustment coefficient based on the bidding strategy, where the bid adjustment coefficient for the aggressive strategy is 1.2, the bid adjustment coefficient for the moderate strategy is 1.0, and the bid adjustment coefficient for the conservative strategy is 0.8;

[0167] The optimized performance is predicted using a random forest regression model, where the input features include budget, bid adjustment factor, and market share, and the output is the predicted return on investment;

[0168] Integrate the optimized budget allocation, bidding strategy and predicted performance to generate the final optimization recommendations and calculate the expected improvement;

[0169] Sort the optimization suggestions in descending order according to the expected improvement degree to obtain a sorted optimization suggestion list;

[0170] The sorted optimization suggestion list is output to the operation data master for guiding the adjustment of the operation data delivery strategy.

[0171] For example, based on the decrypted attribution results, the detailed technical content of using the strategy optimization engine to generate optimization suggestions for operational data delivery strategies is as follows:

[0172] First, the system receives the decrypted attribution results, historical performance data, and market competition data. Taking an operational data master as an example, the received data includes:

[0173] Attribution results: search operation data 44%, display operation data 30%, social media operation data 26%;

[0174] Historical performance data: impressions, clicks, conversions, spending, etc. for each operating data channel in the past 30 days;

[0175] Market competition data: average bid of each operation data channel, total market display times, etc.

[0176] The data preprocessing module processes these data. First, data from different sources are merged according to the operational data channel and time dimension. Then, missing values ​​are processed, such as using moving averages to fill missing historical data. Finally, outlier detection is performed, and the Z-score method is used to identify and process outliers. Data points exceeding 3 standard deviations are considered outliers and replaced with the median of the operational data channel.

[0177] Based on the preprocessed data, calculate the efficiency score and market share of each operational data channel. The efficiency score is calculated by dividing the conversion rate by the cost per click. For example, if the conversion rate of search operational data is 5% and the cost per click is 2 yuan, its efficiency score is 0.025. The market share is calculated by dividing the number of operational data impressions by the total number of market impressions. Assuming that the number of search operational data impressions is 1 million and the total number of market impressions is 5 million, its market share is 20%.

[0178] The budget allocation is optimized using a convex optimization algorithm. The goal is to maximize the sum of the product of the efficiency score and the budget allocation, while ensuring that the sum of the budget allocation is equal to the total budget. The gradient descent method is used to solve this optimization problem, with the number of iterations set to 1000 and the learning rate set to 0.01. Assuming the total budget is 10,000 yuan, the optimized budget allocation results are: 5,000 yuan for search operation data, 3,000 yuan for display operation data, and 2,000 yuan for social media operation data.

[0179] Generate a bidding strategy based on attribution ratio, market share, and budget share. The specific rules are as follows:

[0180] When the attribution ratio is greater than 0.2 and the market share is less than 0.3, an aggressive strategy is adopted;

[0181] A moderate strategy is used when the attribution ratio is greater than 0.1 or the budget share is greater than 0.15;

[0182] In other cases, adopt a conservative strategy.

[0183] For example, the attribution ratio of search operation data is 44%, the market share is 20%, and an aggressive strategy is adopted; the attribution ratio of display operation data is 30%, the budget share is 30%, and a moderate strategy is adopted; the attribution ratio of social media operation data is 26%, the budget share is 20%, and a moderate strategy is adopted.

[0184] The bid adjustment factor is calculated based on the bid strategy. The bid adjustment factor for the aggressive strategy is 1.2, the moderate strategy is 1.0, and the conservative strategy is 0.8. Therefore, the bid adjustment factor for search operation data is 1.2, and the bid adjustment factors for display operation data and social media operation data are both 1.0.

[0185] Use the random forest regression model to predict the performance after optimization. The input features include budget, bid adjustment factor, and market share, and the output is the predicted return on investment (ROI). Use the scikit-learn library to implement the random forest model, set the number of trees to 100, and the maximum depth to 10. The model is trained using historical data and then predicts the optimized solution. For example, the prediction results show that the ROI of search operation data is 3.5, the ROI of display operation data is 2.8, and the ROI of social media operation data is 2.2.

[0186] Integrate the optimized budget allocation, bidding strategy and predicted performance to generate the final optimization recommendations. Calculate the expected improvement, which is the ratio of the predicted ROI after optimization to the current ROI. For example, the expected improvement of search operation data is 1.2, display operation data is 1.1, and social media operation data is 1.05.

[0187] Sort the optimization suggestions in descending order according to the expected improvement degree to obtain a sorted optimization suggestion list:

[0188] 1. Search operation data: budget 5,000 yuan, aggressive strategy, bid adjustment coefficient 1.2, expected ROI 3.5

[0189] 2. Display operation data: budget 3,000 yuan, moderate strategy, bid adjustment coefficient 1.0, expected ROI 2.8

[0190] 3. Social media operation data: budget 2,000 yuan, moderate strategy, bid adjustment coefficient 1.0, expected ROI 2.2

[0191] The sorted optimization suggestion list is output to the operation data master to guide the adjustment of the operation data delivery strategy. The output format is JSON, which contains detailed optimization suggestions for each operation data channel, such as budget allocation, bidding strategy, bid adjustment coefficient, expected ROI, etc. At the same time, a visual report is generated, using a bar chart to show the comparison of budget allocation before and after optimization, and a line chart to show the change trend of expected ROI.

[0192] In addition, the system also provides an interactive interface that allows operational data owners to fine-tune optimization suggestions according to their own needs. For example, operational data owners can manually adjust the budget allocation of a certain channel, and the system will update the budget and expected results of other channels in real time.

[0193] Through the above detailed technical implementation, the strategy optimization engine can generate scientific and reasonable operational data delivery strategy optimization suggestions based on multi-dimensional data. This method comprehensively considers attribution results, historical performance and market competition, and optimizes budget allocation and bidding strategies through mathematical optimization and machine learning algorithms. At the same time, the introduction of the prediction model makes the optimization suggestions more forward-looking and can provide more valuable decision support for operational data owners. This data-driven approach can help operational data owners allocate resources more accurately, improve the efficiency of operational data delivery, and ultimately achieve better operational data results and returns on investment.

[0194] Dynamic operational data attribution method based on user behavior Patent point: The present invention proposes a dynamic operational data attribution method based on user behavior data, which collects and analyzes users' various interactive behaviors on operational data and uses machine learning algorithms to dynamically calculate the attribution ratio of each operational data touchpoint. Innovation: Traditional operational data attribution methods (such as last-click attribution and linear attribution) cannot accurately evaluate the actual contribution of each operational data touchpoint to user decision-making. By introducing a machine learning algorithm, the present invention can dynamically evaluate the actual influence of each operational data touchpoint and improve the accuracy of attribution.

[0195] Real-time data processing and feedback mechanism Patent point: The present invention designs a real-time data processing and feedback mechanism to ensure that the operation data master and platform can obtain attribution information in real time and optimize the operation data strategy. Innovation: Traditional operation data attribution methods are usually post-analysis and cannot provide real-time feedback on the operation data effect. The present invention uses a real-time data processing and feedback mechanism to ensure that the operation data master and platform can obtain attribution information in a timely manner and optimize the operation data strategy.

[0196] Data preprocessing and feature extraction patent point: The present invention proposes an efficient data preprocessing and feature extraction method to extract features that are helpful for model analysis from raw data, such as user active time period, click intensity, dwell time, etc. Innovation: Through efficient data preprocessing and feature extraction methods, the training effect and prediction accuracy of the model can be improved, ensuring the reliability of the attribution results.

[0197] Machine learning model training and attribution calculation patent point: The present invention uses machine learning models (such as random forest, gradient boosting machine, etc.) to train attribution models, input user behavior characteristics, output the contribution weight of each touch point, and calculate the attribution ratio based on the contribution weight. Innovation: By introducing a machine learning model, it is possible to dynamically evaluate the actual influence of each operational data touch point and improve the accuracy and reliability of attribution.

[0198] Data encryption and secure transmission patent point: The present invention designs a data encryption and secure transmission mechanism to ensure that the data transmitted through the API is encrypted and processed to protect user privacy and data security. Innovation: Through the data encryption and secure transmission mechanism, it can effectively protect user privacy, enhance user trust, and ensure the security of data transmission.

[0199] Figure 2 FIG. 1 is a schematic diagram of a system for transmitting and distributing the attribution ratio of operational data based on user behavior according to an embodiment of the present invention. Figure 2 As shown, the system comprises:

[0200] The first unit is used to collect the user's behavior data on clicks and viewing time of the operation data in real time through the tracking code deployed on the client; transmit the behavior data to the data processing center on the server; pre-process the behavior data in the data processing center, including: deleting missing values ​​and duplicate data, and converting timestamps into a unified format;

[0201] The second unit is used to extract features from the preprocessed behavior data, wherein the features include user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, wherein X includes the extracted features and y represents the user's conversion result; train the feature matrix X and the target variable y using a machine learning model to obtain the contribution weight of each operation data touch point; and calculate the attribution ratio of each operation data touch point based on the contribution weight and in combination with a time decay Markov model;

[0202] The third unit is used to transmit the calculated attribution ratio to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio.

[0203] According to a third aspect of the embodiments of the present invention,

[0204] An electronic device is provided, comprising:

[0205] processor;

[0206] a memory for storing processor-executable instructions;

[0207] The processor is configured to call the instructions stored in the memory to execute the aforementioned method.

[0208] A fourth aspect of the embodiments of the present invention is:

[0209] A computer-readable storage medium is provided, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the aforementioned method is implemented.

[0210] The present invention may be a method, an apparatus, a system and / or a computer program product. The computer program product may include a computer-readable storage medium carrying computer-readable program instructions for executing various aspects of the present invention.

[0211] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements 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 returning and distributing the attribution ratio of operational data based on user behavior, characterized in that: include: By deploying tracking codes on the client side, users’ behavioral data on clicks and viewing time of operational data can be collected in real time; Transmitting the behavior data to a data processing center on the server side; Preprocessing the behavior data at the data processing center includes: deleting missing values ​​and duplicate data, and converting timestamps into a unified format; Extract features from the preprocessed behavior data, the features including user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, where X contains the extracted features and y represents the user's conversion results; use a machine learning model to train the feature matrix X and the target variable y to obtain the contribution weight of each operational data touch point; based on the contribution weight, calculate the attribution ratio of each operational data touch point in combination with a time-decaying Markov model; The calculated attribution ratio is transmitted back to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio; The feature matrix X and the target variable y are trained using a machine learning model to obtain the contribution weights of each operational data touchpoint, including: The user behavior data set is randomly divided into a training set and a test set according to a preset ratio; the feature matrix is ​​standardized to obtain a standardized feature matrix; Initialize the random forest model and gradient boosting decision tree model, and set the initial hyperparameters; use the training set data to train the random forest model and gradient boosting decision tree model; Use the test set to evaluate the performance of the random forest model and the gradient boosted decision tree model, and calculate the root mean square error and the coefficient of determination; optimize the hyperparameters of the random forest model and the gradient boosted decision tree model through grid search and cross-validation methods; extract feature importance from the optimized random forest model and the gradient boosted decision tree model; normalize the feature importance of the random forest model and the gradient boosted decision tree model; perform weighted average of the normalized feature importance of the random forest model and the gradient boosted decision tree model to obtain the integrated feature importance; Establish a mapping relationship between features and operational data touchpoints; calculate the contribution weight of each operational data touchpoint based on the importance of integrated features and the mapping relationship between features and operational data touchpoints; normalize the obtained operational data touchpoint contribution weights to obtain the final operational data touchpoint contribution weights; Based on the contribution weights, the attribution ratio of each operational data touchpoint is calculated in combination with the time decay Markov model, including: An operation data contact point transfer probability matrix is ​​constructed based on the user behavior sequence, wherein the elements in the operation data contact point transfer probability matrix represent the transfer probability from one operation data contact point to another operation data contact point; a time decay function is introduced, wherein the time decay function is an exponential decay function, wherein the decay parameter is determined by cross-validation; the time decay function is applied to the operation data contact point transfer probability matrix to obtain an attenuated operation data contact point transfer probability matrix, wherein the attenuated transfer probability is the product of the original transfer probability and the time decay factor; The power iteration method is used to solve the steady-state distribution of the attenuated operation data contact transfer probability matrix; based on the steady-state distribution and the predetermined operation data contact contribution weight, the attribution ratio of each operation data contact is calculated.

2. The method according to claim 1, characterized in that The method further comprises: Smoothing the attenuated operation data contact transfer probability matrix to avoid zero probability problems; Construct multi-order Markov chains to capture user behavior patterns in longer historical states; use grid search or Bayesian optimization methods to adaptively select the optimal time decay parameters; Calculate the log-likelihood of the model for the test data to evaluate the model's fit; use the model to predict the next possible operational data touchpoint and calculate the prediction accuracy.

3. The method according to claim 1, characterized in that The calculated attribution ratio is sent back to the operation data platform and the operation data master in real time through an encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio, including: Encrypt the operation data touchpoint attribution result by using an asymmetric encryption algorithm RSA to obtain an encrypted attribution result; receive the encrypted attribution result through a RESTful API interface, and forward the encrypted attribution result to a message queue; Using Apache Kafka as the message queue to ensure high throughput and low latency processing of the encrypted attribution results; using Apache Flink to perform real-time data processing on the encrypted attribution results in the message queue, including decrypting the encrypted attribution results and performing preliminary analysis; Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for the operation data delivery strategy, and the optimization suggestions include budget allocation plans and bidding strategies; the optimization suggestions are sent to the operation data platform and the operation data master through a real-time feedback mechanism.

4. The method according to claim 3, characterized in that The method further comprises: Use Redis to cache frequently accessed attribution result data to reduce database query pressure; implement an adaptive batch processing mechanism to dynamically adjust the batch size according to the system load, with the initial batch size of 100 and the maximum batch size of 1000; Realize multi-tenant data isolation to ensure data security between different operational data masters; periodically update the optimization suggestions with an update cycle of 5 minutes; When generating the optimization suggestions, the operational data touchpoints are divided into three bidding strategies: aggressive, moderate and conservative, according to the attribution ratio. The aggressive strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.2, the moderate strategy is adopted for operational data touchpoints with an attribution ratio greater than 0.1 and less than or equal to 0.2, and the conservative strategy is adopted for operational data touchpoints with an attribution ratio less than or equal to 0.1; Based on the optimization suggestions, the operational data delivery parameters, including budget allocation and bidding strategies, are dynamically adjusted to form a closed-loop optimization system.

5. The method according to claim 1, characterized in that Based on the decrypted attribution results, the strategy optimization engine is used to generate optimization suggestions for operational data delivery strategies. The optimization suggestions include budget allocation plans and bidding strategies, including: Receive decrypted attribution results, historical performance data, and market competition data; Using a data preprocessing module to merge, process missing values, and detect outliers on the attribution results, the historical performance data, and the market competition data, to obtain preprocessed data; Based on the preprocessed data, calculating an efficiency score and a market share, wherein the efficiency score is a ratio of a conversion rate to a cost per click, and the market share is a ratio of the number of operational data impressions to the number of total market impressions; The budget allocation is optimized using a convex optimization algorithm, where the objective function is to maximize the sum of the products of the efficiency score and the budget allocation, and the constraint is that the sum of the budget allocation is equal to the total budget; Generate a bidding strategy based on the attribution ratio, market share, and budget share. An aggressive strategy is used when the attribution ratio is greater than 0.2 and the market share is less than 0.

3. A moderate strategy is used when the attribution ratio is greater than 0.1 or the budget share is greater than 0.

15. A conservative strategy is used in other cases. Calculate the bid adjustment coefficient based on the bidding strategy, where the bid adjustment coefficient for the aggressive strategy is 1.2, the bid adjustment coefficient for the moderate strategy is 1.0, and the bid adjustment coefficient for the conservative strategy is 0.8; The optimized performance is predicted using a random forest regression model, where the input features include budget, bid adjustment factor, and market share, and the output is the predicted return on investment; Integrate the optimized budget allocation, bidding strategy and predicted performance to generate the final optimization recommendations and calculate the expected improvement; Sort the optimization suggestions in descending order according to the expected improvement degree to obtain a sorted optimization suggestion list; The sorted optimization suggestion list is output to the operation data master for guiding the adjustment of the operation data delivery strategy.

6. A user behavior-based operational data attribution ratio return distribution system, used to implement the method described in any one of claims 1 to 5, characterized in that: include: The first unit is used to collect user behavior data on clicks and viewing time of operational data in real time through the tracking code deployed on the client; Transmitting the behavior data to a data processing center on the server side; Preprocessing the behavior data at the data processing center includes: deleting missing values ​​and duplicate data, and converting timestamps into a unified format; The second unit is used to extract features from the preprocessed behavior data, wherein the features include user active time period, click intensity and dwell time; construct a feature matrix X and a target variable y, wherein X includes the extracted features and y represents the user's conversion result; train the feature matrix X and the target variable y using a machine learning model to obtain the contribution weight of each operation data touch point; and calculate the attribution ratio of each operation data touch point based on the contribution weight and in combination with a time decay Markov model; The third unit is used to transmit the calculated attribution ratio to the operation data platform and the operation data master in real time through the encrypted API interface; the operation data platform and the operation data master dynamically adjust the operation data delivery strategy and budget allocation according to the received attribution ratio.

7. An electronic device, characterized in that: include: processor; a memory for storing processor-executable instructions; The processor is configured to call the instructions stored in the memory to execute the method described in any one of claims 1 to 5.

8. A computer-readable storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images