CN-122020028-A - Advertisement analysis method, device, equipment and medium based on Bayesian network model

Abstract

The application relates to the technical field of big data analysis and artificial intelligence, and discloses an advertisement analysis method, device, equipment and medium based on a Bayesian network model, wherein the method comprises the steps of respectively extracting characteristics of a current advertisement putting strategy, current advertisement user behavior data and current advertisement conversion results by adopting a characteristic extraction model to respectively obtain the characteristics of the current advertisement putting strategy, the current advertisement user behavior data and the current advertisement conversion results; and splicing the features of the putting strategy of the current advertisement, the features of the user behavior data of the current advertisement and the features of the conversion result of the current advertisement to obtain the multi-modal features of the current advertisement, inputting the multi-modal features of the current advertisement into a verified Bayesian network model, and generating a causal link of the current advertisement through the verified Bayesian network model. The application is beneficial to improving the acquisition efficiency of the causal link of the current advertisement.

Inventors

• LI YANPING

Assignees

• 优地网络有限公司

Dates

Publication Date

20260512

Application Date

20260113

Claims (10)

1. 1. An advertisement analysis method based on a bayesian network model, which is applied to an electronic device, the advertisement analysis method comprising: Splicing the features of the preset advertisement putting strategy, the features of the preset advertisement user behavior data and the features of the preset advertisement conversion result to obtain the multi-modal features of the preset advertisement, forming a sample by the multi-modal features of the preset advertisement and the causal links of the preset advertisement, dividing different samples into a training set and a verification set according to a preset dividing proportion, training a Bayesian network model by adopting the training set, and storing the trained Bayesian network model; Verifying the Bayesian network model through the verification set to obtain the current evaluation index of the trained Bayesian network model on the verification set; When the current evaluation index is larger than the preset evaluation index, storing the verified Bayesian network model, and acquiring the multi-mode data of the current advertisement, wherein the multi-mode data of the current advertisement comprises a putting strategy of the current advertisement, user behavior data of the current advertisement and a conversion result of the current advertisement; The method comprises the steps of adopting a feature extraction model to respectively extract features of a current advertisement putting strategy, current advertisement user behavior data and current advertisement conversion results to respectively obtain features of the current advertisement putting strategy, current advertisement user behavior data and current advertisement conversion results; And splicing the features of the putting strategy of the current advertisement, the features of the user behavior data of the current advertisement and the features of the conversion result of the current advertisement to obtain the multi-modal features of the current advertisement, inputting the multi-modal features of the current advertisement into a verified Bayesian network model, and generating a causal link of the current advertisement through the verified Bayesian network model.
2. 2. The advertisement analysis method according to claim 1, wherein the splicing the features of the preset advertisement placement strategy, the features of the preset advertisement user behavior data and the features of the preset advertisement conversion result to obtain the multi-modal features of the preset advertisement, forming a sample from the multi-modal features of the preset advertisement and the causal links of the preset advertisement, dividing the different samples into a training set and a verification set according to a preset dividing ratio, training the bayesian network model by using the training set, and storing the trained bayesian network model comprises: Acquiring multi-modal data of a preset advertisement, wherein the multi-modal data of the preset advertisement comprises a putting strategy of the preset advertisement, user behavior data of the preset advertisement and a conversion result of the preset advertisement; The method comprises the steps of adopting a feature extraction model to respectively conduct feature extraction on a preset advertisement putting strategy, preset advertisement user behavior data and preset advertisement conversion results to respectively obtain the features of the preset advertisement putting strategy, the preset advertisement user behavior data and the preset advertisement conversion results, splicing the features of the preset advertisement putting strategy, the preset advertisement user behavior data and the preset advertisement conversion results to obtain multi-mode features of the preset advertisement, and combining the multi-mode features of the preset advertisement and a causal link of the preset advertisement into one sample; Splicing the features of the preset advertisement putting strategy, the features of the preset advertisement user behavior data and the features of the preset advertisement conversion result to obtain the multi-modal features of the preset advertisement, forming one sample by the multi-modal features of the preset advertisement and the causal links of the preset advertisement, dividing different samples into a training set and a verification set according to a preset dividing proportion, training a Bayesian network model by adopting the training set, and storing the trained Bayesian network model.
3. 3. The advertisement analysis method according to claim 1, wherein when the current evaluation index is greater than the preset evaluation index, saving the verified bayesian network model, and obtaining multi-modal data of the current advertisement, where the multi-modal data of the current advertisement includes a delivery policy of the current advertisement, user behavior data of the current advertisement, and a conversion result of the current advertisement, and the method includes: Acquiring a current accuracy rate and a current accuracy rate from a current evaluation index, and acquiring a preset accuracy rate and a preset accuracy rate from a preset evaluation index; when the current accuracy is greater than the preset accuracy and the current accuracy is greater than the preset accuracy, And storing the verified Bayesian network model, and acquiring the multi-mode data of the current advertisement, wherein the multi-mode data of the current advertisement comprises the putting strategy of the current advertisement, the user behavior data of the current advertisement and the conversion result of the current advertisement.
4. 4. The advertisement analysis method according to claim 1, wherein the splicing the feature of the current advertisement placement strategy, the feature of the current advertisement user behavior data, and the feature of the current advertisement conversion result to obtain the multi-modal feature of the current advertisement, inputting the multi-modal feature of the current advertisement into the verified bayesian network model, and generating the causal link of the current advertisement through the verified bayesian network model comprises: Splicing the characteristics of the putting strategy of the current advertisement, the characteristics of the user behavior data of the current advertisement and the characteristics of the conversion result of the current advertisement to obtain the multi-mode characteristics of the current advertisement; Normalizing the multi-modal features of the current advertisement to obtain processed multi-modal features, inputting the processed multi-modal features into a verified Bayesian network model, and generating a causal link of the current advertisement through the verified Bayesian network model.
5. 5. The advertisement analysis method according to claim 1, wherein after the splicing the feature of the placement policy of the current advertisement, the feature of the user behavior data of the current advertisement, and the feature of the conversion result of the current advertisement to obtain the multi-modal feature of the current advertisement, the multi-modal feature of the current advertisement is input into the verified bayesian network model, and the causal link of the current advertisement is generated through the verified bayesian network model, the advertisement analysis method comprises: the causal link of the current advertisement is disassembled into a strategy node, a behavior node and a conversion node; and executing intervention operation on the strategy node, the behavior node and the conversion node to obtain a counter fact result of the current advertisement.
6. 6. The advertisement analysis method according to claim 5, wherein after the decomposing the causal link of the current advertisement into the policy node, the behavior node and the conversion node, performing an intervention operation on the policy node, the behavior node and the conversion node to obtain the anti-facts result of the current advertisement, the advertisement analysis method comprises: And acquiring a preset storage area, and storing the anti-facts result of the current advertisement into the storage area.
7. 7. The advertisement analysis method according to claim 6, wherein after the acquiring a preset storage area and saving the anti-facts result of the current advertisement in the storage area, the advertisement analysis method comprises: and creating a display window, and displaying the anti-facts result of the current advertisement through the display window.
8. 8. An advertisement analysis device based on a bayesian network model, which is applied to an electronic device, comprising: The dividing module is used for splicing the characteristics of the preset advertisement putting strategy, the characteristics of the user behavior data of the preset advertisement and the characteristics of the conversion result of the preset advertisement to obtain the multi-modal characteristics of the preset advertisement, forming a sample by the multi-modal characteristics of the preset advertisement and the causal links of the preset advertisement, dividing different samples into a training set and a verification set according to a preset dividing proportion, training a Bayesian network model by adopting the training set, and storing the trained Bayesian network model; The first acquisition module is used for verifying the Bayesian network model through the verification set, and acquiring the current evaluation index of the trained Bayesian network model on the verification set; The second acquisition module is used for storing the verified Bayesian network model when the current evaluation index is larger than the preset evaluation index, and acquiring the multi-mode data of the current advertisement, wherein the multi-mode data of the current advertisement comprises a putting strategy of the current advertisement, user behavior data of the current advertisement and a conversion result of the current advertisement; The extraction module is used for extracting the characteristics of the current advertisement putting strategy, the current advertisement user behavior data and the current advertisement conversion result by adopting the characteristic extraction model to obtain the characteristics of the current advertisement putting strategy, the current advertisement user behavior data and the current advertisement conversion result respectively; the analysis module is used for splicing the characteristics of the putting strategy of the current advertisement, the characteristics of the user behavior data of the current advertisement and the characteristics of the conversion result of the current advertisement to obtain the multi-modal characteristics of the current advertisement, inputting the multi-modal characteristics of the current advertisement into the verified Bayesian network model, and generating a causal link of the current advertisement through the verified Bayesian network model.
9. 9. An electronic device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the advertisement analysis method of any of claims 1 to 7 when the computer program is executed by the processor.
10. 10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the advertisement analysis method according to any one of claims 1 to 7.

Description

Advertisement analysis method, device, equipment and medium based on Bayesian network model Technical Field The application relates to the technical field of big data analysis and artificial intelligence, in particular to an advertisement analysis method, device, equipment and medium based on a Bayesian network model. Background In the whole flow of advertisement delivery and effect optimization, advertisement analysis is a core link, and the core target is to provide scientific basis for strategy adjustment by observing the actual effect of the delivery strategy and positioning the key path and bottleneck converted by the user. However, the existing advertisement analysis method is difficult to acquire the causal link of the current advertisement, which is unfavorable for improving the acquisition efficiency of the causal link. The method is characterized in that the existing advertisement analysis method adopts a manual acquisition mode to acquire the causal link of the current advertisement, and adopts a manual acquisition mode to screen massive advertisement delivery data one by one and manually verify causal association among nodes one by one, so that the whole process is time-consuming and tedious and is easily influenced by manual intervention, and the acquisition efficiency of the causal link is not facilitated to be improved. Disclosure of Invention The embodiment of the application provides an advertisement analysis method, device, equipment and medium based on a Bayesian network model, which are used for solving the technical problems that the existing advertisement analysis method is difficult to acquire a causal link of a current advertisement and is unfavorable for improving the acquisition efficiency of the causal link. In a first aspect, an embodiment of the present application provides an advertisement analysis method based on a bayesian network model, which is applied to an electronic device, and the advertisement analysis method includes: Splicing the features of the preset advertisement putting strategy, the features of the preset advertisement user behavior data and the features of the preset advertisement conversion result to obtain the multi-modal features of the preset advertisement, forming a sample by the multi-modal features of the preset advertisement and the causal links of the preset advertisement, dividing different samples into a training set and a verification set according to a preset dividing proportion, training a Bayesian network model by adopting the training set, and storing the trained Bayesian network model; Verifying the Bayesian network model through the verification set to obtain the current evaluation index of the trained Bayesian network model on the verification set; When the current evaluation index is larger than the preset evaluation index, storing the verified Bayesian network model, and acquiring the multi-mode data of the current advertisement, wherein the multi-mode data of the current advertisement comprises a putting strategy of the current advertisement, user behavior data of the current advertisement and a conversion result of the current advertisement; The method comprises the steps of adopting a feature extraction model to respectively extract features of a current advertisement putting strategy, current advertisement user behavior data and current advertisement conversion results to respectively obtain features of the current advertisement putting strategy, current advertisement user behavior data and current advertisement conversion results; And splicing the features of the putting strategy of the current advertisement, the features of the user behavior data of the current advertisement and the features of the conversion result of the current advertisement to obtain the multi-modal features of the current advertisement, inputting the multi-modal features of the current advertisement into a verified Bayesian network model, and generating a causal link of the current advertisement through the verified Bayesian network model. In one possible implementation manner of the first aspect, the splicing the feature of the preset advertisement placement policy, the feature of the preset advertisement user behavior data and the feature of the preset advertisement conversion result to obtain the multi-modal feature of the preset advertisement, forming a sample by the multi-modal feature of the preset advertisement and the causal link of the preset advertisement, dividing different samples into a training set and a verification set according to a preset dividing ratio, training the bayesian network model by using the training set, and storing the trained bayesian network model includes: Acquiring multi-modal data of a preset advertisement, wherein the multi-modal data of the preset advertisement comprises a putting strategy of the preset advertisement, user behavior data of the preset advertisement and a conversion result of the preset advertisement; The metho