Network advertisement putting path optimization method based on causal inference

Abstract

The invention discloses a network advertisement putting path optimization method based on causal inference, and the method comprises the steps: S1, collecting historical advertisement putting data, and constructing an advertisement putting data set; s2, constructing a causal inference model based on the improved structure equation, and obtaining a causal relationship between the advertisement form and the advertisement effect; s3, designing an advertisement path selection strategy in combination with a causal relationship and a dobby machine algorithm, and obtaining an optimal advertisement form; s4, exploring a new advertisement putting path by using a dobby machine algorithm, and performing advertisement putting; s5, monitoring an advertisement putting process in real time, and dynamically adjusting an advertisement putting path according to a monitoring result; and S6, regularly updating the causal inference model according to the advertisement form, the user behavior and the change of the market environment. According to the method, an efficient and scientific optimization scheme can be provided in network advertisement putting path optimization, and remarkable technical values and economic benefits are brought to practical application.

Description

Technical Field

[0001] The present invention relates to the technical field of advertising delivery paths, and particularly to an optimization method for network advertising delivery paths based on causal inference. Background Art

[0002] With the rapid development of Internet advertising, the optimization of advertising delivery paths has become the key for advertisers to gain an advantage in the highly competitive market. However, traditional advertising delivery path optimization methods usually rely on historical data for decision-making and mostly use correlation-based analysis methods. These methods often ignore the causal relationships between advertising forms, resulting in causal biases in the optimization results, which in turn affect the accuracy and sustainability of advertising delivery effects.

[0003] Traditional advertising delivery optimization methods mostly rely on the multi-armed bandit algorithm based on statistics. This type of algorithm continuously probes the performance of different advertising forms in order to find the optimal delivery path. Although the multi-armed bandit algorithm can help advertisers select the optimal advertising form to a certain extent, its basic principle mainly relies on correlation-based analysis and does not consider the causal relationships between advertising effects. In this case, the algorithm tends to select those advertising forms that perform best in historical data. However, this selection does not necessarily reflect the true causal relationships between advertising forms. Since advertising effects are not only affected by the advertising forms themselves but may also be affected by various factors such as user behavior, time, and market environment, simply relying on correlation for decision-making easily leads to causal biases, which in turn affects the advertising delivery effect.

[0004] For example, traditional multi-armed bandit algorithms usually make selections by comparing indicators such as the click-through rate of each advertising form. However, the performance of an advertisement is not only related to its form but may also be affected by factors such as the time period when the advertisement is released and the characteristics of the user group. These factors may cause seemingly related advertising forms to actually have no causal relationship, and relying solely on correlation analysis may make the optimization decision limited. Therefore, advertisers often misjudge which advertising forms are more attractive to users, resulting in a decrease in advertising effects.

[0005] In addition, existing advertising optimization methods also lack a mechanism for dynamically adjusting advertising strategies. The changes in the advertising market, the diversity of user behavior, and the changes in the external environment all cause the advertising effects to fluctuate continuously. Traditional methods usually rely on static data analysis and cannot effectively adjust the advertising delivery path in a rapidly changing market environment. Advertisers may overly rely on the advertising form that has performed best historically and ignore other potential advertising forms or strategies, thus missing the opportunity to optimize advertising effects.

[0006] In response to the above problems, an optimization method for advertising placement paths based on causal inference has emerged. Causal inference can reveal the true causal relationship between advertising forms and advertising effects, helping advertisers avoid relying solely on correlation data when making placement decisions, thereby eliminating causal biases.

[0007] However, despite the significant advantages of causal inference in optimizing advertising placement paths, there is a lack of comprehensive methods in the existing technologies that combine causal inference with the multi-armed bandit algorithm. In the existing technologies, causal inference is mostly used alone and fails to be effectively integrated with the multi-armed bandit algorithm, making it impossible to consider both causal relationships and the dynamic adjustment of advertising placement paths during the optimization process. Therefore, the existing technologies cannot fully utilize the advantages of causal inference nor effectively improve the optimization effect of advertising placement strategies. There is an urgent need for an optimization method for advertising placement paths that combines causal inference with the multi-armed bandit algorithm to improve the accuracy and sustainability of advertising effects.

[0008] Based on the above analysis, the present invention proposes an optimization method for online advertising placement paths based on causal inference. By combining a causal inference model with the multi-armed bandit algorithm, this method effectively solves the causal bias problem existing in traditional advertising optimization methods, can more accurately analyze the causal relationship between advertising forms and effects, thereby optimizing the advertising placement path and improving advertising effects. Summary of the Invention

[0009] An object of the present invention is to propose an optimization method for online advertising placement paths based on causal inference. The present invention can provide an efficient and scientific optimization scheme in optimizing online advertising placement paths, bringing significant technical value and economic benefits to practical applications.

[0010] An optimization method for online advertising placement paths based on causal inference according to an embodiment of the present invention includes the following steps:

[0011] S1. Collect historical advertising placement data and construct an advertising placement data set;

[0012] S2. Construct a causal inference model based on an improved structural equation, and use the advertising placement data in combination with the causal inference model to obtain the causal relationship between advertising forms and advertising effects;

[0013] S3. Design an advertising path selection strategy by combining the causal relationship with the multi-armed bandit algorithm to obtain the optimal advertising form, and adjust the optimal advertising form based on the Bayesian update mechanism;

[0014] S4. Use the multi-armed bandit algorithm to explore new advertising placement paths, continuously optimize advertising effects, output the advertising placement path, and conduct advertising placement;

[0015] S5. Monitor the advertising effect in real time during the advertising delivery process, and dynamically adjust the advertising delivery path according to the monitoring results;

[0016] S6. Regularly update the causal inference model according to changes in advertising forms, user behaviors, and market environments.

[0017] Optionally, the S1 includes the following steps:

[0018] S11. Collect historical advertising delivery data, including advertising forms, user behavior data, and advertising effect data for each advertising delivery;

[0019] S12. Divide the advertising delivery data into multiple time windows, and each time window contains specific data on advertising forms, user behaviors, and advertising effects;

[0020] S13. Preprocess the advertising form, user behavior data, and advertising effect data, including data cleaning, missing value filling, and outlier detection;

[0021] S14. Construct an advertising effect evaluation index system, and combine the advertising effect data to assign weights to each advertising delivery;

[0022] S15. Construct the data input format of the causal inference model according to the preprocessed data.

[0023] Optionally, the S2 includes the following steps:

[0024] S21. Based on the collected historical advertising delivery data, construct a causal inference model between the advertising form and the advertising effect using an improved structural equation, and analyze the causal impact of the advertising form on the advertising effect through causal inference technology;

[0025] S22. Calculate the advertising effect by combining the advertising form, potential control variables, external environment variables, and feedback variables of the advertising effect:

[0026] E = α + β1A + β2X + β3Z + γ1Y + δ1(A × Z) + θ1(A 2 ) + θ2(Z 2 ) + ∈;

[0027] where E is the advertising effect, A is the advertising form, X is the potential control variable, Z is the external environment variable, Y is the feedback variable of the advertising effect, α is the constant term, β1, β2, β3, γ1, δ1 are regression coefficients, ∈ is the error term, θ1 and θ2 are regression coefficients of non-linear terms, and A × Z is the interaction effect between the advertising form and the external environment variable;

[0028] The calculation method of the feedback variable Y of the advertising effect is:

[0029] Y = γ2At-1 +γ3Z t-1 +η;

[0030] Among them, A t-1 is the lag term of the advertising form, Z t-1 is the lag term of the external environmental variable, γ2 and γ3 are the regression coefficients of the lag effect, and η is the error term;

[0031] S23. Calculate and update the parameters in the causal inference model using the maximum likelihood estimation method:

[0032]

[0033] Among them, L(θ) is the likelihood function of the maximum likelihood estimation, E i is the advertising effect of the i-th one, is the advertising effect calculated based on the causal inference model, σ 2 is the variance of the error term, θ is the parameter to be estimated, n represents the total number of samples, and exp(.) is the natural exponential function;

[0034] S24. According to the results of the causal inference model, calculate the causal effect C i of each advertising form A i on the advertising effect E i :

[0035]

[0036] Among them, C i is the causal effect of the advertising form A i on the advertising effect E i , and β1 is the regression coefficient of the advertising form A i ;

[0037] S25. According to the calculated causal effect C i , sort the causal effects of different advertising forms:

[0038] R = argsort(C1, C2,..., C n );

[0039] Among them, R represents the sorting result of the advertising form A i , C1, C2,..., C n are the causal effects of different advertising forms, and argsort is the operation of sorting according to the causal effect C i ;

[0040] S26. Obtain the priorities of each advertising form according to the sorting result.

[0041] Optionally, the S3 includes the following steps:

[0042] S31. Combining causal effects with the multi-armed bandit algorithm, design an advertising path selection strategy:

[0043]

[0044] in, To select the best advertising format, For advertising format A i The historical expected return, λ is the weight factor of causal effect in the selection strategy, C i For advertising format A i The causal effect of i For advertising format A i uncertainty of the effect;

[0045] Advertisement Format A i Historical expected return The expectation of each advertising path is calculated by weighted average:

[0046]

[0047] Among them, n is the number of times of delivery, R k (A i ) is the return of the kth advertising;

[0048] S32, based on the Bayesian update mechanism, for each selected advertising format A i The effect uncertainty σ i To update:

[0049]

[0050] in, For the updated advertising format A i The effect uncertainty of , τ is the prior uncertainty parameter, which is used to adjust the weight of historical data;

[0051] S33. Uncertainty of the effect of the updated advertising format Combined with the exploration and utilization strategy of the multi-armed bandit algorithm, the choice of advertising format is adjusted:

[0052]

[0053] in, For advertising The historical returns of C i For advertising The causal effect of For advertising The effect uncertainty is n, n is the total number of experiments, N i For advertising The number of selection times is α, and β is the adjustment coefficient for exploration;

[0054] S34. After each advertisement placement, update the causal effect C according to the advertisement effect i , and based on the updated causal effect C i Adjust the selection strategy of the multi-armed bandit algorithm.

[0055] Optionally, the S4 includes the following steps:

[0056] S41. Adopt the multi-armed bandit algorithm, and based on the causal effect and the historical advertisement placement performance, recalculate the expected return of each advertisement path newly through weighted average;

[0057] S42. When exploring a new advertisement path, adopt the causal effect as an additional evaluation index and incorporate the causal effect into the advertisement path selection strategy;

[0058] S43. According to the newly explored advertisement path and the result of the causal effect, update the advertisement path selection strategy in the multi-armed bandit algorithm, optimize the advertisement path exploration process, output the advertisement path and conduct advertisement placement.

[0059] Optionally, the S5 includes the following steps:

[0060] S51. Monitor the advertisement effect during the advertisement placement process in real time and collect real-time advertisement effect data including the immediate return of advertisement form A i and user behavior feedback;

[0061] S52. Update the expected return of each advertisement form A according to the real-time monitoring data i where,

[0062]

[0063] where, is the expected return of the updated advertisement form A i , is the historical expected return of advertisement form A i , is the real-time collected advertisement effect data, and α is the learning rate, which controls the influence degree of real-time data on the update;

[0064] S53. Re-evaluate the placement strategy of the advertisement form according to the real-time updated advertisement return information and the causal effect C i .

[0065] The beneficial effects of the present invention are:

[0066] (1) The network advertising placement path optimization method based on causal inference provided by the present invention has significant beneficial effects compared with traditional advertising placement optimization methods. First, the combination of causal inference and the multi-armed bandit algorithm makes advertising placement decisions more accurate. The traditional multi-armed bandit algorithm only relies on the correlation of historical data for decision-making, which may lead to causal bias, resulting in unsatisfactory advertising placement effects. However, by introducing a causal inference model, the present invention reveals the true causal relationship between advertising forms and advertising effects, avoiding the limitations of relying solely on correlation analysis and fundamentally improving the accuracy of advertising placement strategies.

[0067] (2) The present invention has strong dynamic adaptability. By real-time monitoring the effects of various advertising forms during the advertising placement process and dynamically adjusting the advertising placement path based on the monitoring data, the present invention can ensure that the advertising placement strategy is adjusted in a timely manner according to market changes and the diversity of user behaviors, thus avoiding the lag problem caused by over-reliance on historical data in traditional optimization methods. Advertisers can flexibly respond in a rapidly changing market environment and always maintain the optimization of advertising effects.

[0068] (3) By regularly updating the causal inference model to adapt to changes in advertising forms, user behaviors, and market environments, the present invention ensures the continuous optimization of the advertising placement path. This feature enables the advertising placement strategy to have long-term stability and adaptability, avoiding the rapid decay of advertising effects due to market changes. By continuously optimizing the placement path, advertisers can maintain long-term advantages in the fierce market competition. Description of the Drawings

[0069] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention. In the drawings:

[0070] Figure 1 is a flowchart of a network advertising placement path optimization method based on causal inference proposed by the present invention;

[0071] Figure 2 is a flowchart of the combination of the causal inference model and the multi-armed bandit algorithm in a network advertising placement path optimization method proposed by the present invention. Detailed Embodiments

[0072] Now, the present invention will be further described in detail with reference to the drawings. These drawings are all simplified schematic diagrams, only illustrating the basic structure of the present invention in a schematic manner, so they only show the components related to the present invention.

[0073] Refer to Figure 1 - Figure 2, A method for optimizing the network advertising placement path based on causal inference, comprising the following steps:

[0074] S1. Collect historical advertising placement data and construct an advertising placement dataset;

[0075] S2. Construct a causal inference model based on an improved structural equation, and use the advertising placement data in combination with the causal inference model to obtain the causal relationship between the advertising form and the advertising effect;

[0076] S3. Design an advertising path selection strategy by combining the causal relationship and the multi-armed bandit algorithm, obtain the optimal advertising form, and adjust the optimal advertising form based on the Bayesian update mechanism;

[0077] S4. Use the multi-armed bandit algorithm to explore new advertising placement paths, continuously optimize the advertising effect, output the advertising placement path and conduct advertising placement;

[0078] S5. Monitor the advertising effect during the advertising placement process in real time, and dynamically adjust the advertising placement path according to the monitoring results;

[0079] S6. Regularly update the causal inference model according to changes in the advertising form, user behavior, and market environment.

[0080] In this embodiment, S1 includes the following steps:

[0081] S11. Collect historical advertising placement data, including the advertising form, user behavior data, and advertising effect data for each advertising placement;

[0082] S12. Divide the advertising placement data into multiple time windows, and each time window contains specific data on the advertising form, user behavior, and advertising effect;

[0083] S13. Preprocess the advertising form, user behavior data, and advertising effect data, including data cleaning, missing value filling, and outlier detection;

[0084] S14. Construct an advertising effect evaluation index system, and combine the advertising effect data to assign weights to each advertising placement;

[0085] S15. Construct the data input format of the causal inference model according to the preprocessed data.

[0086] In this embodiment, S2 includes the following steps:

[0087] S21. Based on the collected historical advertising placement data, construct a causal inference model between the advertising form and the advertising effect using an improved structural equation, and analyze the causal impact of the advertising form on the advertising effect through causal inference technology;

[0088] S22. Calculate the advertising effect by combining the advertising form, potential control variables, external environmental variables, and feedback variables of the advertising effect:

[0089] E = α + β1A + β2X + β3Z + γ1Y + δ1(A × Z) + θ1(A 2 ) + θ2(Z 2 ) + ∈;

[0090] Among them, E is the advertising effect, A is the advertising form, X is the potential control variable, Z is the external environmental variable, Y is the feedback variable of the advertising effect, α is the constant term, β1, β2, β3, γ1, δ1 are regression coefficients, ∈ is the error term, θ1 and θ2 are the regression coefficients of the non - linear term, and A × Z is the interaction effect between the advertising form and the external environmental variable;

[0091] The calculation method of the feedback variable Y of the advertising effect is:

[0092] Y = γ2A t-1 + γ3Z t-1 + η;

[0093] Among them, A t-1 is the lag term of the advertising form, Z t-1 is the lag term of the external environmental variable, γ2 and γ3 are the regression coefficients of the lag effect, and η is the error term;

[0094] S23. Use the maximum likelihood estimation method to calculate and update the parameters in the causal inference model:

[0095]

[0096] Among them, L(θ) is the likelihood function of the maximum likelihood estimation, E i is the advertising effect of the i - th, is the advertising effect calculated based on the causal inference model, σ 2 is the variance of the error term, θ is the parameter to be estimated, n represents the total number of samples, and exp(.) is the natural exponential function;

[0097] S24. According to the results of the causal inference model, calculate the causal effect C i of each advertising form A i on the advertising effect E i :

[0098]

[0099] Among them, C i is the causal effect of the advertising form A i on the advertising effect E i , and β1 is the regression coefficient of the advertising form A i ;

[0100] S25. Sort the causal effects of different advertising forms according to the calculated causal effect C: i ,

[0101] R = argsort(C1, C2, …, C n );

[0102] where R represents the sorting result of advertising form A i , C1, C2, …, C n are the causal effects of different advertising forms, and argsort is the operation of sorting according to the causal effect C i ;

[0103] S26. Obtain the priority of each advertising form according to the sorting result.

[0104] In this embodiment, S3 includes the following steps:

[0105] S31. Combine the causal effect with the multi-armed bandit algorithm to design an advertising path selection strategy:

[0106]

[0107] where is the optimal advertising form selected, is the historical expected return of advertising form A i , λ is the weight factor of the causal effect in the selection strategy, C i is the causal effect of advertising form A i , and σ i is the effect uncertainty of advertising form A i ;

[0108] The historical expected return of advertising form A i is obtained by calculating the weighted average of the expected values of each advertising path:

[0109]

[0110] where n is the number of ad placements, and R k (A i ) is the return of the k-th ad placement;

[0111] S32. Based on the Bayesian update mechanism, update the effect uncertainty σ i of the advertising form A i selected each time:

[0112]

[0113] where For the updated advertising form A i For the effect uncertainty, τ is the prior uncertainty parameter, which is used to adjust the weight of historical data;

[0114] S33. According to the effect uncertainty of the updated advertising form Combined with the exploration and exploitation strategies of the multi-armed bandit algorithm, adjust the selection of advertising forms:

[0115]

[0116] Among them, is the historical return of the advertising form , C i is the causal effect of the advertising form , is the effect uncertainty of the advertising form , n is the total number of trials, N i is the number of selections of the advertising form , and β is the adjustment coefficient for exploration;

[0117] S34. After each advertisement is placed, update the causal effect C i according to the advertisement effect, and based on the updated causal effect C i adjust the selection strategy of the multi-armed bandit algorithm.

[0118] In this embodiment, S4 includes the following steps:

[0119] S41. Adopt the multi-armed bandit algorithm, and based on the causal effect and historical advertisement placement performance, recalculate the expected return of each new advertisement path through weighted average;

[0120] S42. When exploring new advertisement paths, use the causal effect as an additional evaluation index and incorporate the causal effect into the advertisement path selection strategy;

[0121] S43. According to the results of the newly explored advertisement paths and causal effects, update the advertisement path selection strategy in the multi-armed bandit algorithm, optimize the advertisement path exploration process, output the advertisement paths and place advertisements.

[0122] In this embodiment, S5 includes the following steps:

[0123] S51. Monitor the advertisement effect during the advertisement placement process in real time and collect real-time advertisement effect data including the immediate return of advertising form A i and user behavior feedback;

[0124] S52. Update each advertising form A according to the real-time monitoring data i ​Expected return

[0125]

[0126] wherein is the updated advertising form A i Expected return is the advertising form A i Historical expected return is the advertising effect data collected in real time, and α is the learning rate, which controls the influence degree of real-time data on the update;

[0127] S53. According to the advertising return information updated in real time and the causal effect C i , re-evaluate the placement strategy of the advertising form.

[0128] Example:

[0129] In an online advertising platform, the user group of this advertising platform is huge. Advertisers can display products to users through different advertising forms, such as picture ads, video ads, and interactive ads, etc. Based on the historical behavior data of users, the advertising platform usually provides advertisers with historical effect data of different advertising forms and recommends the advertising forms with the best performance to them. However, the existing advertising placement optimization methods rely on the correlation analysis of historical data and fail to consider the causal relationship between the advertising form and the advertising effect, resulting in the fact that some advertising forms with good performance do not actually produce real causal effects, but are caused by the influence of other factors.

[0130] In order to overcome this causal bias problem, the implementer decides to apply the network advertising placement path optimization method based on causal inference of the present invention. In this embodiment, the implementer combines a causal inference model with a multi-armed bandit algorithm to more accurately evaluate the potential effect of the advertising form and achieve higher advertising effect and better return on investment by optimizing the advertising placement path.

[0131] In the initial stage of the advertising placement activity, the implementer first collected the historical advertising placement data provided by the platform, including the click-through rate, conversion rate, and advertising return on investment of different advertising forms, etc. These data include information such as the advertising display volume, user click volume, user behavior data, and advertising placement time period.

[0132] Based on these historical data, the implementer used a causal inference model to model the causal relationship between the advertising form and the advertising effect. At this stage, the implementer adopted a method based on the structural equation model to quantify the causal path between the advertising form, user behavior, and advertising effect. Through structural equation model analysis, the causal impact relationship of the advertising form on the advertising effect was revealed. The training results of the model showed that video advertising had a strong causal impact on the improvement of advertising effect, while the impact of picture advertising and interactive advertising was relatively small.

[0133] Next, by combining the causal inference model with the multi-armed bandit algorithm, the implementer evaluated the potential effects of different advertising forms according to the model and gave priority to those advertising forms with significant causal impacts in the selection of the advertising placement path. Through the exploration process of the multi-armed bandit algorithm, the implementer avoided relying solely on the historically best-performing advertising forms and instead selected those advertising forms that might bring higher advertising effects through continuous exploration and updating.

[0134] This advertising placement campaign was carried out from June 1, 2024, to June 30, 2024, on a certain online advertising platform. Before the placement campaign, the implementer used traditional advertising optimization methods and relied on historical data to select advertising forms. However, the advertising placement effect was not ideal, and the expected placement effect was not achieved.

[0135] According to the traditional method, the implementer initially selected picture advertising and video advertising as the main advertising forms. However, the selection of these advertising forms did not consider the causal relationship between the advertising form and the advertising effect. In the first three days of the advertising placement, the CTR of the picture advertising was 0.4%, and the conversion rate was 0.02%. While the CTR of the video advertising was 0.6%, and the conversion rate was 0.04%. However, these data could not fully reflect the true causal relationship between the advertising form and the effect, resulting in the inability to accurately judge the true effect of the advertising form.

[0136] Subsequently, the implementer adopted the network advertising placement path optimization method based on causal inference in the present invention and began to readjust the placement strategy on the fourth day of the advertising placement campaign. In the new advertising placement plan, the advertiser conducted a quantitative analysis of the causal relationship between different advertising forms and advertising effects based on the causal inference model and dynamically optimized the advertising placement path by combining the multi-armed bandit algorithm. The specific placement data is as follows:

[0137] Table 1 Specific placement data of the method of the present invention in the optimization of the advertising placement path

[0138]

[0139] As can be seen from the table, after applying the method of the present invention, the click-through rate (CTR) and conversion rate (CVR) of video ads have been continuously increasing. Especially on the 6th day, the CTR of video ads reached 1.21%, and the conversion rate also increased to 0.07%. At the same time, the return on investment (ROI) of video ads also increased by 1.8. Compared with 1.2 of the traditional method, the ROI has been significantly improved.

[0140] Throughout the embodiments, the implementer not only solves the causal bias problem in the optimization of traditional advertising placement paths through the method of the present invention, but also significantly improves the accuracy of advertising placement effects and the return on investment of advertisers. By introducing the combination of a causal inference model and a multi-armed bandit algorithm, the present invention accurately analyzes the true causal relationship between advertising forms and advertising effects, avoids the limitations of relying solely on historical correlation data, and ensures that advertising placement decisions are more scientific and reasonable.

[0141] The present invention analyzes the causal effect of advertising forms on advertising effects through a causal inference model and combines it with a multi-armed bandit algorithm to effectively optimize the selection strategy of advertising forms. Based on real-time monitoring and feedback, advertisers can dynamically adjust the advertising placement path to adapt to changes in user behavior and market environment. Compared with traditional methods, the present invention can flexibly respond to fluctuations in advertising effects, thereby improving the flexibility and adaptability of advertising placement and ensuring the continuous optimization of advertising effects.

[0142] The present invention also ensures real-time adaptation to changes in advertising effects in different advertising placement scenarios by dynamically updating the causal inference model. By introducing causal effects and real-time feedback into the multi-armed bandit algorithm, the present invention effectively avoids the problem of over-reliance on a single advertising strategy and achieves the optimal adjustment of the advertising placement path. This method not only improves the ROI of advertising placement but also ensures that advertisers can continuously obtain the best advertising effects in a complex market environment.

[0143] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered by the protection scope of the present invention.

Claims

1. A method for optimizing the network advertising placement path based on causal inference, characterized in that It includes the following steps: S1. Collect historical advertising data and construct an advertising dataset; S2. Based on an improved structural equation, construct a causal inference model, and use the advertising data in combination with the causal inference model to obtain the causal relationship between the advertising form and the advertising effect; S3. Combine the causal relationship with the multi-armed bandit algorithm to design an advertising path selection strategy, obtain the optimal advertising form, and adjust the optimal advertising form based on the Bayesian update mechanism; S4. Use the multi-armed bandit algorithm to explore new advertising paths, continuously optimize the advertising effect, output the advertising path and conduct advertising; S5. Monitor the advertising effect during the advertising process in real time, and dynamically adjust the advertising path according to the monitoring results; S6. Regularly update the causal inference model according to changes in the advertising form, user behavior, and market environment.

2. The method for optimizing the network advertisement placement path based on causal inference according to claim 1, wherein The S1 includes the following steps: S11. Collect historical advertising data, including the advertising form, user behavior data, and advertising effect data for each advertising; S12. Divide the advertising data into multiple time windows, and each time window contains specific data on the advertising form, user behavior, and advertising effect; S13. Preprocess the advertising form, user behavior data, and advertising effect data, including data cleaning, missing value filling, and outlier detection; S14. Construct an advertising effect evaluation index system, and combine the advertising effect data to assign weights to each advertising; S15. Construct the data input format of the causal inference model according to the preprocessed data.

3. A method for optimizing the network advertising delivery path based on causal inference according to claim 1, characterized in that The S2 includes the following steps: S21. Based on the collected historical advertising data, and use an improved structural equation to construct a causal inference model between the advertising form and the advertising effect, and analyze the causal impact of the advertising form on the advertising effect through causal inference technology; S22. Calculate the advertising effect by combining the advertising form, potential control variables, external environment variables, and feedback variables of the advertising effect: E = α + β1A + β2X + β3Z + γ1Y + δ1(A×Z) + θ1(A 2 ) + θ2(Z 2 ) + ∈; Among them, E is the advertising effect, A is the advertising form, X is the potential control variable, Z is the external environment variable, Y is the feedback variable of the advertising effect, α is the constant term, β1, β2, β3, γ1, δ1 are regression coefficients, ∈ is the error term, θ1 and θ2 are the regression coefficients of the non-linear term, and A×Z is the interaction effect between the advertising form and the external environment variable; The calculation method of the feedback variable Y of the advertising effect is: Y = γ2A t-1 + γ3Z t-1 + η; Among them, A t-1 is the lag term of the advertising form, Z t-1 is the lag term of the external environmental variable, γ2 and γ3 are the regression coefficients of the lag effect, and η is the error term; S23. Use the maximum likelihood estimation method to calculate and update the parameters in the causal inference model: Among them, \(L(\theta)\) is the likelihood function of the maximum likelihood estimation, and \(E\) i is the advertising effect of the \(i\)-th one, is the advertising effect calculated based on the causal inference model, and \(\sigma\) 2 is the variance of the error term, \(\theta\) is the parameter to be estimated, \(n\) represents the total number of samples, and \(\exp(.)\) is the natural exponential function; S24. Calculate each advertising form A according to the results of the causal inference model i for the advertising effect E i of the causal effect C i : Among them, C i is the advertising form A i on the advertising effect E i of the causal effect, and β1 is the regression coefficient of the advertising form A i ; S25. Sort the causal effects of different advertising forms according to the calculated causal effect C i : R = argsort(C1, C2, …, C n ); Among them, R represents the sorting result of advertising form A i , and C1, C2, …, C n are the causal effects of different advertising forms, and argsort is the operation of sorting by the causal effect C i ; S26. Obtain the priority of each advertising form according to the sorting result.

4. A method for optimizing the network advertising placement path based on causal inference according to claim 1, characterized in that The S3 includes the following steps: S31. Combine the causal effect with the multi-armed bandit algorithm to design an advertising path selection strategy: Among them, is the selected optimal advertising form, is advertising form A i is the historical expected return of advertising form A, λ is the weight factor of the causal effect in the selection strategy, C i is advertising form A i is the causal effect of advertising form A, σ i is advertising form A i is the effect uncertainty of advertising form A; Advertising form A i Historical expected return The expected value of each advertising path is obtained by weighted average calculation: where n is the number of ad placements, and R k (A i ) is the return of the k-th ad placement; S32. Update the uncertainty of the effect σ of the advertising form A selected each time based on the Bayesian update mechanism i of i : Among them, is the updated advertising form A i is the uncertainty of the effect, τ is the prior uncertainty parameter, which is used to adjust the weight of historical data; S33. According to the updated uncertainty of the advertising form effect Combine the exploration and exploitation strategies of the multi-armed bandit algorithm to adjust the selection of advertising forms: Among them, is the historical return of the advertising form, C is the causal effect of the advertising form i is the advertising form is the causal effect is the advertising form is the effect uncertainty of the advertising form, n is the total number of trials, N i is the advertising form is the number of selections of the advertising form, and β is the adjustment coefficient for exploration; S34. After each advertisement placement, update the causal effect C according to the advertisement effect i , and based on the updated causal effect C i adjust the selection strategy of the multi-armed bandit algorithm.

5. A method for optimizing the network advertisement placement path based on causal inference according to claim 1, characterized in that The S4 includes the following steps: S41. Use the multi-armed bandit algorithm to recalculate the expected return of each new advertising path through weighted average based on the causal effect and historical advertising performance; S42. When exploring new advertising paths, use the causal effect as an additional evaluation index and incorporate the causal effect into the advertising path selection strategy; S43. Update the advertisement path selection strategy in the multi-armed bandit algorithm according to the results of the explored new advertisement paths and causal effects, optimize the advertisement path exploration process, output the advertisement paths, and conduct advertisement placement.

6. The method for optimizing the network advertisement delivery path based on causal inference according to claim 1, wherein The said S5 includes the following steps: S51. Monitor the advertising effect in real time during the advertising placement process and collect real-time advertising effect data including the immediate return of advertising form A i and user behavior feedback; S52. According to the real-time monitoring data Update the expected return of each advertising form A i ​ Among them, is the expected return of the updated advertising form A i the expected return of is the historical expected return of advertising form A i the historical expected return of is the advertising effect data collected in real time, and α is the learning rate, which controls the influence degree of real-time data on the update; S53. According to the advertising return information updated in real time and the causal effect C i , re-evaluate the placement strategy of the advertising format.

Images