A network advertisement delivery path optimization method based on causal inference

Abstract

The application discloses a network advertisement putting path optimization method based on causal inference, S1, collecting historical advertisement putting data, and constructing an advertisement putting dataset; S2, constructing a causal inference model based on an improved structural equation, and acquiring a causal relationship between advertisement forms and advertisement effects; S3, combining the causal relationship with a multi-arm tiger machine algorithm to design an advertisement path selection strategy, and acquiring an optimal advertisement form; S4, exploring a new advertisement putting path by using the multi-arm tiger machine algorithm, and putting the advertisement; S5, monitoring an advertisement putting process in real time, and dynamically adjusting the advertisement putting path according to a monitoring result; and S6, regularly updating the causal inference model according to changes in the advertisement forms, user behaviors and market environments. The application can provide an efficient and scientific optimization scheme in the network advertisement putting path optimization, and brings remarkable technical values and economic benefits for practical applications.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of advertisement delivery path, and particularly relates to a network advertisement delivery path optimization method based on causal inference. BACKGROUND

[0002] With the rapid development of Internet advertising, the optimization of advertisement delivery path has become the key for advertisers to gain an advantage in the competitive market. However, traditional advertisement 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 relationship between different forms of advertisements, resulting in causal bias in the optimization results, which in turn affects the accuracy and sustainability of the advertisement delivery effect.

[0003] Traditional advertisement delivery optimization methods mostly rely on statistical-based multi-armed bandit algorithms. Such algorithms try different advertisement forms to find the optimal delivery path, although multi-armed bandit algorithms can help advertisers choose the optimal advertisement form to some extent, their basic principle mainly relies on correlation-based analysis and does not consider the causal relationship between advertisement effects. In this case, the algorithm tends to choose the advertisement form that performs best in historical data, however, this choice may not necessarily reflect the true causal relationship between advertisement forms, as advertisement effects are not only affected by the advertisement form itself, but also by user behavior, time, market environment, and other factors. Simply relying on correlation to make decisions can easily lead to causal bias, which in turn affects the effect of advertisement delivery.

[0004] For example, traditional multi-armed bandit algorithms usually compare click-through rates and other indicators of different advertisement forms to make a choice, however, the performance of an advertisement is not only related to its form, but also affected by factors such as the time period of advertisement publication and user group characteristics. These factors may cause seemingly related advertisement forms to actually have no causal relationship, and relying solely on correlation analysis may limit the optimization decision, thus, advertisers often misjudge which advertisement forms are more attractive to users, which in turn leads to a decrease in advertisement effectiveness.

[0005] In addition, existing advertisement optimization methods also lack a mechanism for dynamically adjusting advertisement strategies. Changes in the advertising market, the diversity of user behavior, and changes in the external environment all make the advertisement effect fluctuate constantly. Traditional methods usually rely on static data analysis, and in a rapidly changing market environment, they cannot effectively adjust the advertisement delivery path in a timely manner. Advertisers may over-rely on the best-performing advertisement forms in history, while ignoring other potential advertisement forms or strategies, thus missing the opportunity to optimize the advertisement effect.

[0006] To solve the above problems, an advertisement delivery path optimization method based on causal inference emerges as the times require. Causal inference can reveal the real causal relationship between the advertisement form and the advertisement effect, helping the advertiser avoid relying only on correlation data when making delivery decisions, thereby eliminating causal bias.

[0007] However, although causal inference has significant advantages in advertisement delivery path optimization, there is a lack of comprehensive methods combining causal inference and multi-armed bandit algorithm in the prior art. In the prior art, causal inference is mostly used alone and cannot be effectively combined with multi-armed bandit algorithm, which cannot consider both causal relationship and dynamic adjustment of advertisement delivery path in the optimization process. Therefore, the prior art cannot fully utilize the advantages of causal inference and cannot effectively improve the optimization effect of advertisement delivery strategy. There is an urgent need for an advertisement delivery path optimization method combining causal inference and multi-armed bandit algorithm to improve the accuracy and sustainability of advertisement effect.

[0008] Based on the above analysis, the present application proposes a network advertisement delivery path optimization method based on causal inference. The method effectively solves the problem of causal bias in traditional advertisement optimization methods by combining causal inference model and multi-armed bandit algorithm, and can more accurately analyze the causal relationship between advertisement form and effect, thereby optimizing the advertisement delivery path and improving the advertisement effect. SUMMARY

[0009] One object of the present application is to provide a network advertisement delivery path optimization method based on causal inference. The present application can provide an efficient and scientific optimization scheme in network advertisement delivery path optimization, bringing significant technical value and economic benefits to practical applications.

[0010] According to an embodiment of the present application, a network advertisement delivery path optimization method based on causal inference comprises the following steps:

[0011] S1, collect historical advertisement delivery data and construct an advertisement delivery dataset;

[0012] S2, construct a causal inference model based on an improved structural equation, and use the advertisement delivery data to obtain the causal relationship between the advertisement form and the advertisement effect by combining the causal inference model;

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

[0014] S4, explore new advertisement delivery paths using the multi-armed bandit algorithm, continuously optimize the advertisement effect, output the advertisement delivery path and perform advertisement delivery;

[0015] S5, real-time monitoring the advertising effect in the advertising process, and dynamically adjusting the advertising delivery path according to the monitoring result;

[0016] S6, periodically updating the causal inference model according to the changes of the advertising form, user behavior and market environment.

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

[0018] S11, collecting advertising delivery historical data, including the advertising form, user behavior data and advertising effect data of each advertising delivery;

[0019] S12, dividing the advertising delivery data into multiple time windows, each time window containing specific data of advertising form, user behavior and advertising effect;

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

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

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

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

[0024] S21, based on the collected historical advertising delivery data, and using an improved structural equation to construct a causal inference model between the advertising form and the advertising effect, and analyzing the causal influence of the advertising form on the advertising effect through causal inference technology;

[0025] S22, calculating the advertising effect combining the advertising form, potential control variables, external environmental 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] Wherein, 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 the regression coefficients, ∈ is the error term, θ1 and θ2 are the regression coefficients of the nonlinear term, A×Z is the interaction effect between the advertising form and the external environmental 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 Z is a lagging term for advertising formats. t-1 γ1 represents the lagged terms of external environmental variables, γ2 and γ3 are the regression coefficients of the lagged effects, 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] Where L(θ) is the likelihood function of the maximum likelihood estimate, and E i For the effect of the i-th advertisement, For the advertising effect calculated based on the causal inference model, σ 2 Let θ be the variance of the error term, θ be the parameter to be estimated, n be the total number of samples, and exp(.) be the natural exponential function.

[0034] S24. Based on the results of the causal inference model, calculate A for each type of advertising. i E on advertising effectiveness i causal effect C i :

[0035]

[0036] Among them, C i For advertising format A i E on advertising effectiveness i The causal effect, β1 is the advertising form A i The regression coefficients;

[0037] S25. Based on the calculated causal effect C i The causal effects of different advertising formats were ranked:

[0038] R = argsort(C1,C2,…,C…) n );

[0039] Where R represents advertising format A i The sorting results, C1, C2, ..., C n For the causal effects of different advertising formats, argsort represents the causal effects C. i Sorting operations;

[0040] S26. Obtain the priority of each ad format based on the sorting results.

[0041] Optionally, S3 includes the following steps:

[0042] S31. Combining causal effects and multi-armed slot machine algorithms, design an advertising path selection strategy:

[0043]

[0044] in, To select the optimal advertising format, For advertising format A i Historical expected return, λ is the weighting factor of causal effect in strategy selection, C i For advertising format A i causal effect, σ i For advertising format A i The uncertainty of the effect;

[0045] Advertising Format A i Historical expected return The expected value for each ad path is obtained by calculating a weighted average:

[0046]

[0047] Where n is the number of deliveries, R k (A i ) represents the return on the k-th ad placement;

[0048] S32. Based on the Bayesian update mechanism, for each selected ad format A... i The effect uncertainty σ i Update:

[0049]

[0050] in, For the updated ad format A i The effect uncertainty is denoted by τ, which is a prior uncertainty parameter used to adjust the weights of historical data.

[0051] S33. Based on the uncertainty of the updated advertising format effect Based on the exploration and utilization strategies of multi-armed slot machine algorithms, adjustments were made to the selection of advertising formats:

[0052]

[0053] in, For advertising Historical returns, C i For advertising causal effect For advertising The effect uncertainty, where n is the total number of trials, N i For advertising The number of selections, where β is the adjustment coefficient for exploration;

[0054] S34. After each ad campaign, update the causal effect C based on the ad performance. i And based on the updated causal effect C i Adjust the selection strategy for the multi-armed slot machine algorithm.

[0055] Optionally, S4 includes the following steps:

[0056] S41. Using a multi-armed slot machine algorithm, based on causal effects and historical advertising performance, the expected return for each new advertising path is recalculated by weighted average.

[0057] S42. When exploring new advertising paths, use causal effect as an additional evaluation indicator and incorporate causal effect into the advertising path selection strategy.

[0058] S43. Based on the newly explored advertising paths and the results of causal effects, update the advertising path selection strategy in the multi-armed slot machine algorithm, optimize the advertising path exploration process, output the advertising path, and deliver the advertisement.

[0059] Optionally, S5 includes the following steps:

[0060] S51. Monitor the advertising effectiveness in real time during the advertising campaign and collect real-time advertising effectiveness data. Including advertising format A i Instant rewards and user behavior feedback;

[0061] S52, Based on real-time monitoring data Update each ad format A i Expected return

[0062]

[0063] in, For the updated ad format A i Expected return For advertising format A i Historical expected returns The data represents the advertising performance collected in real time, and α is the learning rate, which controls the degree to which real-time data influences the updates.

[0064] S53. Based on the real-time updated advertising return information and causal effect C i We need to reassess our advertising strategy.

[0065] The beneficial effects of this 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 multi-armed slot machine algorithm makes advertising placement decision more accurate. Traditional multi-armed slot machine algorithm only relies on the correlation of historical data to make decisions, which may lead to causal bias, resulting in unsatisfactory advertising placement effect. However, the present invention introduces a causal inference model to reveal the real causal relationship between advertising form and advertising effect, avoiding the limitations of relying solely on correlation analysis, and fundamentally improving the accuracy of advertising placement strategy.

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

[0068] (3) This invention adapts to changes in advertising formats, user behavior, and market environment by periodically updating the causal inference model, ensuring continuous optimization of the advertising delivery path. This feature gives the advertising delivery strategy long-term stability and adaptability, avoiding the rapid decline in advertising effectiveness due to market changes. By continuously optimizing the delivery path, advertisers can maintain a long-term advantage in fierce market competition. Attached Figure Description

[0069] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0070] Fig. 1 This is a flowchart of a network advertising delivery path optimization method based on causal inference proposed in this invention;

[0071] Fig. 2 This is a flowchart illustrating the combination of the causal inference model and the multi-armed slot machine algorithm in the network advertising delivery path optimization method based on causal inference proposed in this invention. Detailed Implementation

[0072] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0073] refer to Figs. 1-2A method for optimizing online advertising delivery paths based on causal inference includes the following steps:

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

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

[0076] S3. Design an advertising path selection strategy by combining causal relationships and multi-armed slot machine algorithms to obtain the optimal advertising format, and adjust the optimal advertising format based on the Bayesian update mechanism;

[0077] S4. Use the multi-armed slot machine algorithm to explore new advertising delivery paths, continuously optimize advertising performance, output advertising delivery paths and deliver ads.

[0078] S5. Monitor the advertising effect in real time during the advertising process and dynamically adjust the advertising delivery path based on the monitoring results;

[0079] S6. Update the causal inference model regularly based on changes in advertising formats, user behavior, and market environment.

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

[0081] S11. Collect historical data on ad placements, including ad formats, user behavior data, and ad performance data for each ad placement.

[0082] S12. Divide the advertising data into multiple time windows, each containing specific data on advertising format, user behavior, and advertising effectiveness.

[0083] S13. Preprocess the advertising format, user behavior data, and advertising performance data, including data cleaning, missing value imputation, and outlier detection;

[0084] S14. Construct an advertising effectiveness evaluation index system and assign weights to each advertising campaign based on advertising effectiveness data;

[0085] S15. Data input format for constructing a causal inference model based on preprocessed data.

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

[0087] S21. Based on the collected historical advertising data, and using an improved structural equation model to construct a causal inference model between advertising format and advertising effect, the causal impact of advertising format on advertising effect is analyzed through causal inference technology.

[0088] S22. Calculate advertising effectiveness by combining advertising format, potential control variables, external environmental variables, and feedback variables of advertising effectiveness:

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

[0090] Where E is the advertising effect, A is the advertising format, X is the potential control variable, Z is the external environmental variable, Y is the feedback variable of advertising effect, α is the constant term, β1, β2, β3, γ1, δ1 are the regression coefficients, ∈ is the error term, θ1 and θ2 are the regression coefficients of the nonlinear term, and A×Z is the interaction effect between the advertising format and the external environmental variable.

[0091] The feedback variable Y for the advertising effect is calculated as follows:

[0092] Y = γ2A t-1 +γ3Z t-1 +η;

[0093] Among them, A t-1 Z is a lagging term for advertising formats. t-1 γ1 represents the lagged terms of external environmental variables, γ2 and γ3 are the regression coefficients of the lagged effects, and η is the error term;

[0094] S23. Calculate and update the parameters in the causal inference model using the maximum likelihood estimation method:

[0095]

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

[0097] S24. Based on the results of the causal inference model, calculate A for each type of advertising. i E on advertising effectiveness i causal effect C i :

[0098]

[0099] Among them, C i For advertising format A i E on advertising effectiveness i The causal effect, β1 is the advertising form A i The regression coefficients;

[0100] S25. Based on the calculated causal effect C i The causal effects of different advertising formats were ranked:

[0101] R = argsort(C1,C2,…,C…) n );

[0102] Where R represents advertising format A i The sorting results, C1, C2, ..., C n For the causal effects of different advertising formats, argsort represents the causal effects C. i Sorting operations;

[0103] S26. Obtain the priority of each ad format based on the sorting results.

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

[0105] S31. Combining causal effects and multi-armed slot machine algorithms, design an advertising path selection strategy:

[0106]

[0107] in, To select the optimal advertising format, For advertising format A i Historical expected return, λ is the weighting factor of causal effect in strategy selection, C i For advertising format A i causal effect, σ i For advertising format A i The uncertainty of the effect;

[0108] Advertising Format A i Historical expected return The expected value for each ad path is obtained by calculating a weighted average:

[0109]

[0110] Where n is the number of deliveries, R k (A i ) represents the return on the k-th ad placement;

[0111] S32. Based on the Bayesian update mechanism, for each selected ad format A... i The effect uncertainty σ i Update:

[0112]

[0113] in, For the updated ad format A i The effect uncertainty is τ, which is the prior uncertainty parameter used to adjust the weights of historical data.

[0114] S33. Based on the uncertainty of the updated advertising format effect Based on the exploration and utilization strategies of multi-armed slot machine algorithms, adjustments were made to the selection of advertising formats:

[0115]

[0116] in, For advertising Historical returns, C i For advertising causal effect For advertising The effect uncertainty, where n is the total number of trials, N i For advertising The number of selections, where β is the adjustment coefficient for exploration;

[0117] S34. After each ad campaign, update the causal effect C based on the ad performance. i And based on the updated causal effect C i Adjust the selection strategy for the multi-armed slot machine algorithm.

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

[0119] S41. Using a multi-armed slot machine algorithm, based on causal effects and historical advertising performance, the expected return for each new advertising path is recalculated by weighted average.

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

[0121] S43. Based on the newly explored advertising paths and the results of causal effects, update the advertising path selection strategy in the multi-armed slot machine algorithm, optimize the advertising path exploration process, output the advertising path, and deliver the advertisement.

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

[0123] S51. Monitor the advertising effectiveness in real time during the advertising campaign and collect real-time advertising effectiveness data. Including advertising format A i Instant rewards and user behavior feedback;

[0124] S52, Based on real-time monitoring data Update each ad format A iExpected return

[0125]

[0126] in, For the updated ad format A i Expected return For advertising format A i Historical expected returns The data represents the advertising performance collected in real time, and α is the learning rate, which controls the degree to which real-time data influences the updates.

[0127] S53. Based on the real-time updated advertising return information and causal effect C i We need to reassess our advertising strategy.

[0128] Example:

[0129] In an example, on a certain online advertising platform with a large user base, advertisers can showcase their products to users through various ad formats, such as image ads, video ads, and interactive ads. Based on users' historical behavior data, the advertising platform typically provides advertisers with historical performance data for different ad formats and suggests that they choose the best-performing ad formats. However, existing ad optimization methods rely on correlation analysis of historical data and fail to consider the causal relationship between ad formats and ad performance. This results in some well-performing ad formats not actually producing a real causal effect, but rather being affected by other factors.

[0130] To overcome this causal bias problem, the implementer decided to apply the causal inference-based network advertising delivery path optimization method of the present invention. In this embodiment, the implementer adopts a combination of causal inference model and multi-armed slot machine algorithm to more accurately evaluate the potential effect of advertising formats and achieve higher advertising effect and better return on investment by optimizing the advertising delivery path.

[0131] In the initial stage of the advertising campaign, the implementers first collected historical advertising data provided by the platform, including click-through rates, conversion rates, and return on investment for different ad formats. This data included information such as ad impressions, user clicks, user behavior data, and ad placement times.

[0132] Based on this historical data, the implementers used a causal inference model to model the causal relationship between advertising format and advertising effect. In this stage, the implementers adopted a structural equation modeling approach to quantify the causal path between advertising format, user behavior, and advertising effect. Through structural equation modeling analysis, the causal influence of advertising format on advertising effect was revealed. The training results of the model showed that video ads have a strong causal influence on improving advertising effect, while the influence of image ads and interactive ads is relatively small.

[0133] Next, combining the causal inference model with the multi-armed slot machine algorithm, the implementers evaluated the potential effects of different advertising formats based on the model, and prioritized those advertising formats with significant causal impact in the selection of advertising placement paths. Through the exploration process of the multi-armed slot machine algorithm, they avoided relying solely on the advertising formats with the best historical performance, and instead selected those advertising formats that might bring higher advertising effects through continuous trial and update.

[0134] This advertising campaign was conducted from June 1, 2024 to June 30, 2024, on a certain online advertising platform. Before the campaign, the implementer used traditional advertising optimization methods and relied on historical data to select advertising formats. However, the advertising results were not ideal and failed to achieve the expected results.

[0135] According to traditional methods, implementers initially chose image ads and video ads as the main ad formats. However, the selection of these ad formats did not take into account the causal relationship between ad format and ad performance. In the first three days of ad campaigns, image ads had a CTR of 0.4% and a conversion rate of 0.02%, while video ads had a CTR of 0.6% and a conversion rate of 0.04%. However, these data cannot fully reflect the true causal relationship between ad format and performance, making it impossible to accurately judge the true effectiveness of ad formats.

[0136] Subsequently, the implementers adopted the causal inference-based network advertising placement path optimization method of this invention and began to readjust the placement strategy on the fourth day of the advertising campaign. In the new advertising placement plan, the advertiser quantitatively analyzed the causal relationship between different advertising formats and advertising effects based on the causal inference model, and dynamically optimized the advertising placement path by combining the multi-armed slot machine algorithm. The specific placement data is as follows:

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

[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 continued to improve. 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, which is significantly higher than the 1.2 of the traditional method.

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

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

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

[0143] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for optimizing online advertising delivery paths based on causal inference, characterized in that, Includes the following steps: S1. Collect historical advertising data and build an advertising dataset; S2. Construct a causal inference model based on the improved structural equation model, and use advertising data in combination with the causal inference model to obtain the causal relationship between advertising format and advertising effect; S3. Design an advertising path selection strategy by combining causal relationships and the multi-armed slot machine algorithm to obtain the optimal advertising format, and adjust the optimal advertising format based on the Bayesian update mechanism; S4. Use the multi-armed slot machine algorithm to explore new advertising delivery paths, continuously optimize advertising effects, output advertising delivery paths, and deliver advertisements. S5. Monitor the advertising effectiveness in real time during the advertising process and dynamically adjust the advertising delivery path based on the monitoring results; S6. Update the causal inference model regularly based on changes in advertising format, user behavior, and market environment; S2 includes the following steps: S21. Based on the collected historical advertising data, and using an improved structural equation model to construct a causal inference model between advertising format and advertising effect, the causal impact of advertising format on advertising effect is analyzed through causal inference technology. S22. Calculate advertising effectiveness by combining advertising format, potential control variables, external environmental variables, and feedback variables of advertising effectiveness: ； in, The advertising effectiveness is calculated based on a causal inference model. As an advertisement, As a potential control variable, For external environment variables, As a feedback variable for advertising effectiveness, For constant terms, , , , , For regression coefficients, For error terms, and The regression coefficients of the nonlinear term are... The interaction effect between advertising formats and external environmental variables; The feedback variables of the advertising effect The calculation method is as follows: ； in, As a lagging factor for advertising formats, This is a lagged term of external environmental variables. and The regression coefficient for the lag effect is . This is the error term; S23. Calculate and update the parameters in the causal inference model using the maximum likelihood estimation method: )； in, The likelihood function for maximum likelihood estimation. For the effect of the i-th advertisement, For the effect of the i-th advertisement calculated based on the causal inference model, The variance of the error term. Let n be the parameter to be estimated, and n represent the total number of samples. It is a natural exponential function; S24. Based on the results of the causal inference model, calculate the advertising format for each type. On advertising effectiveness causal effect : ； in, For advertising The effect of the i-th advertisement calculated based on the causal inference model causal effect For advertising The regression coefficients; S25. Based on the calculated causal effect The causal effects of different advertising formats were ranked: ； in, Indicates advertising format The sorting results The causal effects of different advertising formats According to causal effect Sorting operations; S26. Obtain the priority of each ad format based on the sorting results.

2. The method for optimizing online advertising delivery paths based on causal inference according to claim 1, characterized in that, S1 includes the following steps: S11. Collect historical data on ad placements, including ad formats, user behavior data, and ad performance data for each ad placement. S12. Divide the advertising data into multiple time windows, each containing specific data on advertising format, user behavior, and advertising effectiveness. S13. Preprocess the advertising format, user behavior data, and advertising performance data, including data cleaning, missing value imputation, and outlier detection; S14. Construct an advertising effectiveness evaluation index system and assign weights to each advertising campaign based on advertising effectiveness data; S15. Data input format for constructing a causal inference model based on preprocessed data.

3. The method for optimizing online advertising delivery paths based on causal inference according to claim 1, characterized in that, S3 includes the following steps: S31. Combining causal effects and multi-armed slot machine algorithms, design an advertising path selection strategy: )； in, To select the optimal advertising format, For advertising Historical expected returns The weighting factor for causal effects in strategy selection. For advertising causal effect For advertising The uncertainty of the effect; Advertising formats Historical expected return The expected value for each ad path is obtained by calculating a weighted average: ； Where n is the number of deliveries. The return on the k-th ad placement; S32. Based on the Bayesian update mechanism, for each selected ad format... Uncertainty of effects Update: ； in, For the updated advertising format The effect is uncertain. This is a priori uncertainty parameter used to adjust the weights of historical data; S33. Based on the uncertainty of the updated advertising format effect Based on the exploration and utilization strategies of multi-armed slot machine algorithms, adjustments were made to the selection of advertising formats: )； in, For the updated advertising format Historical returns For the updated advertising format The causal effect, where n is the total number of trials. For the updated advertising format Number of selections, Adjustment coefficients for exploration; S34. Update the causal effect based on the advertising performance after each ad campaign. And based on updated causal effects Adjust the selection strategy for the multi-armed slot machine algorithm.

4. The method for optimizing online advertising delivery paths based on causal inference according to claim 3, characterized in that, S4 includes the following steps: S41. Using a multi-armed slot machine algorithm, based on causal effects and historical advertising performance, the expected return for each new advertising path is recalculated by weighted average. S42. When exploring new advertising paths, use causal effect as an additional evaluation indicator and incorporate causal effect into the advertising path selection strategy. S43. Based on the newly explored advertising paths and the results of causal effects, update the advertising path selection strategy in the multi-armed slot machine algorithm, optimize the advertising path exploration process, output the advertising path, and deliver the advertisement.

5. The method for optimizing online advertising delivery paths based on causal inference according to claim 4, characterized in that, S5 includes the following steps: S51. Monitor the advertising effectiveness in real time during the advertising campaign and collect real-time advertising effectiveness data. Including advertising formats Instant rewards and user behavior feedback; S52, Based on real-time monitoring data Update each ad format Expected return : ； in, For the updated advertising format Expected return For advertising Historical expected returns For real-time collection of advertising performance data, The learning rate controls the degree to which real-time data influences the updates; S53. Based on the real-time updated advertising return information and causal effects We need to reassess our advertising strategy.

Images