Online advertisement analysis method and analysis system based on artificial intelligence

Abstract

The invention relates to the technical field of online advertisement analysis based on artificial intelligence, and discloses an online advertisement analysis method and system based on artificial intelligence. By setting an advertisement set and a standard record thereof, each element of advertisement data acquisition is determined, and the problems of data source inconsistency and data confusion are solved. The advertisement set comprises text description, image data, video data and display timestamps, and it is ensured that data formats are unified. A text value is calculated through a character counting function, so that text information can be quantified; the unique color type number is obtained through pixel-by-pixel scanning of image values, and the problem of image data quantification is solved; the video playing duration is analyzed from the file, and time information standardization is ensured. Finally, the values of the text, the image and the video data are summed to obtain an advertisement comprehensive description value, unified processing of multi-modal data is realized, and data integration and cross-modal analysis effects are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of online advertising analysis based on artificial intelligence, and in particular to an online advertising analysis method based on artificial intelligence. Background Art

[0002] With the rapid development of internet advertising and the widespread application of multimedia technologies, advertising delivery systems face significant challenges in collecting and analyzing massive amounts of advertising and user behavior data. Existing technologies primarily rely on decentralized data collection modules and poorly standardized preprocessing methods, making it difficult to achieve unified and accurate management of advertising and user behavior data. Traditional advertising data collection techniques typically focus on a single dimension within the text, image, or video data of advertisements. While methods such as character counting, image color statistics, and video duration extraction can achieve data quantification to a certain extent, they often suffer from inconsistent data formats and fragmented processing flows when merging multimodal data. Furthermore, existing technologies for collecting user behavior data also primarily rely on direct recording in log files and simple numerical encoding. Existing technologies generally use metrics such as click-through rate, pageviews, and conversion rate to measure user behavior. However, due to diverse data sources and varying recording formats, user behavior data often suffers from disorganized structure and inaccurate time series information. This leads to significant errors in subsequent analysis of advertising effectiveness and the establishment of causal relationships.

[0003] Furthermore, due to the lack of strict, unified standards, the digitization of advertising text, image, and video data during ad data preprocessing often relies on crude methods. For example, existing technologies typically quantitatively describe text data through simple character counts or word frequency statistics; image data often uses methods like histograms and color clustering to approximate color information; and video data primarily relies on extracting duration from file metadata. However, these methods struggle to meet high-precision requirements in terms of data consistency and accuracy. While these existing processing methods can meet basic requirements in certain application scenarios, when faced with cross-modal information fusion and subsequent complex causal effect calculations, data inconsistencies and ambiguity significantly impact overall analysis, reducing the accuracy of ad effectiveness evaluation. Regarding user behavior data collection, most current systems utilize logging technology, recording user actions such as clicks, browsing, and conversions as separate events, with fixed numerical codes for each action type. However, the lack of unified encoding standards and timestamp correction mechanisms across systems leads to format differences and data redundancy in user behavior data across multiple channels and platforms. In addition, existing technologies for preprocessing user behavior data often simply convert the raw data into a fixed format without deeply organizing and strictly standardizing the data, which makes it easy for errors and data confusion to occur in subsequent sorting, screening, quantitative comparison and other processes.

[0004] To this end, this case aims to propose an AI-based online advertising analysis method and system. By applying clear numerical methods to advertising text, image, and video data, combined with strict user behavior data recording standards, this method achieves standardized, unified, and precisely quantified data collection and preprocessing. This method ensures that all data within advertising data can be recorded in a stable and deterministic manner through unified standards and numerical processing. Furthermore, after preprocessing, user behavior data can be accurately matched with ad display data, providing a solid foundation for dynamic causal relationship modeling and subsequent multimodal data fusion. Summary of the Invention

[0005] The present invention provides an online advertising analysis method and analysis system based on artificial intelligence, which promotes solving the problems mentioned in the above background technology.

[0006] The present invention provides the following technical solution: an online advertising analysis method based on artificial intelligence, comprising: Set the ad set to ;in, It is the collection of all advertisements to be analyzed in this system; For the Ads, ; is the total number of advertisements; advertise Contains: text description , a string containing the ad copy content; image data , a two-dimensional matrix with each element being Color triplet, describing the ad image; video data , contains video file data, which includes playback duration information; display timestamp , a real number in seconds, indicating the time when the ad starts to be displayed; Set the string character count function to: ;in, The string whose characters are to be counted; is a positive integer; Computational Advertising The text value of , specifically: ;in, For advertising The number of characters in the text, in units of count; Set the image data digitization function to: ;in, is the image data to be digitized; For images A unique color value detected in ; For images The total number of different color categories detected in the image; this function scans the image pixel by pixel , and each unique The triples are recorded in a set, and the set is returned; Computational Advertising The image values ​​are: ;in, For advertising Medium Image The number of different colors in is an exact count of the elements in the collection; Set the function to get the video playback duration: ;in, The video file whose playback duration is to be obtained; is a positive real number; Computational Advertising The video duration is as follows: ;in, For advertising Medium video duration; Set up your ad The comprehensive description value of is: ;in, For advertising Comprehensive description of numerical values; Set up your ad The standard record is: .

[0007] Optionally, it also includes user behavior data collection and preprocessing, specifically: Set the user behavior collection to: ;in, For the User behavior records, ; is the total number of user behaviors; User behavior The collected data items include: behavior type , using numerical coding, defined as click as 1, browse as 2, conversion as 3; behavior timestamp , a real number in seconds; Setting User Behavior Recorded as: .

[0008] Optionally, it also includes initial causal model construction and hypothesis setting, specifically: Set the ad node collection to: ;in, For advertising The associated node identifies the advertising event; Set the user behavior node collection to: ;in, For user behavior associated nodes; Set the overall node collection to: ; For any advertising node User behavior nodes , set the time difference to: ;in, For advertising and user behavior The response time between the two, in seconds; Set the maximum allowed time difference between ad display and user behavior to ,set up ; Set time condition: Only when When, think Maybe Make an impact; Set the candidate causal edge set to: ; Set the static causal diagram to: .

[0009] Optionally, it also includes feature extraction and node attribute construction, specifically: For each advertising node , construct the attribute vector as: ; Set the context resonance attribute calculation function to: , ; in, is the set of all vectors consisting of three real numbers, that is, three-dimensional real space; function input parameters, is the number of text characters; is the number of image color types; is the video duration; output , for advertising The contextual resonance value of is a positive real number; Computational Advertising The contextual resonance value of is: ; For each user behavior node , construct the attribute vector, specifically: .

[0010] Optionally, it also includes constructing event timing and causal edge attributes, specifically: Record all ads By display time Sort in ascending order; Record all user behavior By time of occurrence Sort in ascending order; Will As a candidate edge time span; Setting up the function , used to assign time spans to edges, specifically: .

[0011] Optionally, it also includes computing dynamic causal relationship discovery and causal effects, specifically: Set up your ad The comprehensive description value of ; For each candidate edge , calculate the causal effect, specifically: ; in, For advertising User behavior The causal effect value of Set the fixed causal effect significance threshold, denoted as ,set up ; Set the valid causal edge set to: .

[0012] Optionally, it also includes verification of the causal diagram structure, specifically: Assume the number of remaining ads after preprocessing is , the remaining number of user behaviors that meet the time condition is , construct the causal effect matrix as: ; in, for OK A matrix of columns, where each element is from the set of real numbers ; ; For each valid edge , verify that strictly meets: and ; If there are records that do not meet the conditions, then in the causal effect matrix and the effective causal edge set Delete the corresponding item.

[0013] Optionally, it also includes an integrated display of the results, specifically: Establish a two-dimensional coordinate system, with the horizontal axis representing the ad display time , the vertical axis represents the causal effect value ; For each valid edge , draw a straight line in the graph: Starting point , the end point is ; When drawing, the line thickness and color are based on The values ​​use a predefined linear mapping relationship: the lower the value, the thicker the line and the darker the color; the higher the value, the thinner the line and the lighter the color; Each valid edge The detailed data is integrated into a data table, each row includes: Advertisement Number and corresponding in 、 、 and ; User behavior number and in and ; Calculated response time and causal effects .

[0014] A system for implementing the artificial intelligence-based online advertising analysis method, comprising: Advertisement content collection module: collects advertisement copy content, advertisement images, video file data and advertisement start time; User behavior data collection module: collects user behavior types and behavior timestamps; Computing module: used for data calculation.

[0015] The present invention has the following beneficial effects: 1. By establishing ad sets and their standard records, the solution clarifies the various elements involved in ad data collection, resolving issues of inconsistent data sources and data confusion. Specifically, by defining ad sets and explicitly describing each ad's text description, image data, video data, and display timestamp, the system ensures that each ad's information has a unified and clear source and format, addressing the issue of fragmented and non-standardized data and achieving structured data management. By establishing a string character count function and counting characters in ad text, the solution addresses the quantification of text data, ensuring that ad content can be represented with specific numerical values, making subsequent comparisons and calculations on text data feasible. This step transforms complex text information into simple numerical features, facilitating algorithmic processing and quantitative analysis. Furthermore, by defining an image data quantification function, the ad image is scanned pixel by pixel, and each unique color triplet is recorded in a set. A set counting function is then used to calculate the number of distinct color types in the image. This process addresses the challenges of abstracting and quantitatively representing image data. This allows image information to be objectively digitized in the description of overall ad content, enhancing data integration and cross-modal fusion. Similarly, a function for obtaining video playback duration is set up to parse the playback duration from the video file header, so that the video data can be represented by a unified and standard numerical value, solving the problem of inconsistent expression of time information in video data, thus laying the foundation for the subsequent comprehensive description of multimodal data. Finally, by simply summing the numerical values ​​of text, image and video data to obtain a comprehensive description value of the advertisement, the solution successfully integrates multiple data items into a unified indicator, solves the problem of difficulty in direct comparison of various data types, and achieves the effect of unified processing and convenient application of multimodal information. Overall, these steps ensure that in the entire process of advertising data collection and standardization, each link of the data has a clear definition and specific quantification method, thereby providing solid and clear basic data support for subsequent causal analysis, dynamic modeling and algorithm optimization.

[0016] 2. By setting up user behavior sets and defining user behavior records, the solution first standardizes and unifies the collection of user behavior data, thereby solving the problem of diverse data sources and inconsistent formats. The user behavior set is clearly defined in the steps. This approach ensures that all user behavior data has a clear identification and a unified indexing system, thereby avoiding data confusion and duplicate records. At the same time, by standardizing the data items collected by user behavior, such as encoding the behavior type with a numerical value, coding clicks as 1, browsing as 2, and conversions as 3, and specifying the behavior timestamp as a real number in seconds, the solution solves the problem of unstructured and non-standard behavior data before quantitative analysis. In this way, in each user behavior record, and The user's behavior category and occurrence time are clearly indicated, ensuring the comparability and accuracy of the data in subsequent analysis. The effect of this step is to convert the diverse user behavior data into standard records in a unified format after rigorous preprocessing, allowing the system to efficiently sort, filter, and quantitatively compare user behaviors. After data standardization, the system can quickly and accurately match the temporal correlation between ad display and user behavior when analyzing user reactions to ads, thereby providing a reliable time series basis for constructing dynamic causal relationship diagrams. Through this method, the system not only improves the efficiency of data processing, but also reduces the risk of errors caused by confusing data formats, laying a solid foundation for subsequent causal effect calculations and multimodal data fusion.

[0017] 3. By constructing an initial causal model and setting hypotheses, the system achieves preliminary modeling of the potential causal relationship between ad data and user behavior data, thereby resolving the issues of unclear temporal correlations and data matching difficulties between these two data types. First, by establishing an ad node set, a user behavior node set, and an overall node set, the system uniformly identifies each independent data source. The ad node set categorizes all ad-related events, while the user behavior node set identifies user behaviors after an ad is displayed. Together, the overall node set enables centralized data management and unified indexing. This step addresses the issues of fragmented data sources and chaotic node representation, providing a clear and unified foundation for subsequent causal relationship construction. Second, by defining the response time difference between ads and user behaviors, the system can quantify the time interval between ad display and user behavior, providing a direct basis for determining whether an ad has an impact on user behavior. Setting a maximum allowable time difference as a temporal condition further clarifies that an ad is considered to have an impact on user behavior only if the behavior occurs within a specific time range after the ad is displayed. This approach eliminates indirect impact behaviors that occur long after the ad is displayed, while retaining records of user behaviors that occur quickly after the ad is displayed and are causally related. Third, constructing a set of candidate causal edges is a critical step. Each edge in this set represents only the potential influence between ads and user behavior that meet the temporal conditions, without any additional attributes. This approach effectively eliminates data that does not conform to the causal assumption through a temporal filtering mechanism, thereby ensuring the accuracy and relevance of the data in subsequent causal model analysis. This process addresses the problem of spurious associations caused by excessively long time windows in traditional systems, while also reducing interference factors, allowing the system to focus on capturing direct and true causal relationships. Finally, constructing a static causal graph as the foundation for subsequent dynamic causal analysis not only provides a preliminary framework for establishing a network of relationships between advertising events and user behavior but also lays the data structure foundation for further introduction of dynamic updates and causal effect calculation. By constructing this preliminary model, the system can effectively distinguish between direct and indirect responses, ensuring that subsequent steps can more accurately quantify and verify the actual impact of ads on user behavior. Overall, through these steps, the system successfully overcomes the data management and causal relationship establishment challenges inherent in existing technologies, which are often caused by fragmented data sources, inconsistent node definitions, and inaccurate temporal matching. In terms of effect, this unified and rigorous causal model construction not only improves the accuracy and efficiency of data preprocessing, but also provides a solid foundation for subsequent complex dynamic causal analysis and multimodal data fusion, significantly enhancing the scientific nature and credibility of advertising effectiveness analysis.

[0018] 4. Through feature extraction and node attribute construction, the system effectively addresses the issue of inconsistent multimodal information in advertising and user behavior data, making direct comparison difficult. First, by constructing an attribute vector for each ad node, the system converts ad text, image, and video data into clear numerical representations, with the number of text characters, number of image color types, and video duration as vector components. This step addresses the difficulty of uniformly processing the jumbled raw multimodal data, allowing all ad data to be expressed in a unified numerical space, laying the foundation for subsequent analysis. Second, the system introduces a contextual resonance attribute calculation function, defined in three-dimensional real space. The function takes as input the number of text characters, number of image color types, and video duration, and outputs the ad's contextual resonance value. This custom function combines different types of data into a single positive real number, reflecting the comprehensive performance of the ad content. This function effectively integrates information from different modalities, resolving the inconsistent data dimensions and difficulty in comprehensive comparison inherent in traditional methods. Furthermore, an attribute vector is constructed for each user behavior node, including behavior type and response time. This achieves a standardized representation of user behavior data, enabling comparison and correlation with the attribute vectors of ad nodes within the same mathematical space. This not only provides a precise quantitative basis for subsequent analysis of the causal relationship between advertising and user behavior, but also significantly improves data operability and analytical efficiency. Overall, through feature extraction and node attribute construction, the system addresses the issues of information loss and the lack of unified quantification in traditional advertising analysis caused by the clutter of multimodal data. This step extracts complex text, image, and video data into clear, unified numerical features and fuses these features into a comprehensive indicator using a contextual resonance function, allowing the comprehensive performance of advertising content to be reflected in a single, deterministic numerical form. Furthermore, after undergoing the same rigorous attribute construction, user behavior data can be directly compared with advertising data within the same space, providing high-precision data support for the construction of dynamic causal relationship models. The ultimate result is improved data processing accuracy and consistency, reduced errors caused by the difficulty of quantitative comparison between multimodal data, and significantly enhanced the scientific nature and reliability of advertising effectiveness analysis.

[0019] 5. By constructing event time series and causal edge attributes, the system successfully addresses the temporal disorder of ad and user behavior data and the lack of precise temporal characteristics for causal edges. First, the system sorts all ad records in ascending order by display time and all user behavior records in ascending order by occurrence time. This step ensures that ad and user behavior data are arranged in chronological order, allowing subsequent causal analysis to strictly adhere to temporal relationships and avoid misattribution caused by data misordering. Second, the system sets a time span as a temporal attribute for candidate edges. This span measures the time interval between user behavior and ad display. This definition of a time span ensures that the system adheres to temporal logic when constructing causal edges, establishing causal relationships only between ads and user behaviors that meet the temporal conditions. This design avoids the causal confusion that can occur in traditional ad analysis methods due to unclear time spans, thereby improving the accuracy of causal inference. Furthermore, the system implements a time span assignment function to assign the calculated time span to candidate causal edges, ensuring that each edge carries a clear temporal attribute. This function ensures that causal edges not only adhere to temporal ordering requirements but also possess quantifiable temporal characteristics. This enables subsequent causal analysis to be more refined and adjusted based on specific time spans. Compared to traditional methods that rely on manually defined time windows, this step provides an automated and precise temporal characterization mechanism, making the construction of causal edges more scientific and rational. Overall, by constructing event temporal ordering and causal edge attributes, the system addresses the issues of chaotic data temporal order and unclear temporal characteristics of causal edges in advertising analysis. After sorting advertising and user behavior data, causal analysis can be performed strictly according to temporal logic, avoiding erroneous causal inferences. Furthermore, by assigning precise time span attributes to causal edges, the system achieves standardized quantification of temporal information, enabling data-driven analysis of the causal relationship between advertising and user behavior rather than relying on subjective assumptions. This improvement enhances the reliability of causal analysis, making advertising effectiveness evaluation more accurate and laying a solid foundation for subsequent causal inference and model optimization.

[0020] 6. By calculating the steps of dynamic causal relationship discovery and causal effect, the system successfully solves the problems of ambiguous causal relationship between advertising and user behavior, difficulty in quantifying causal effects, and unclear handling of multi-user behaviors, making causal analysis more scientific and accurate. First, the system sets a comprehensive description value for the advertisement, and uniformly expresses multimodal information such as the number of text characters, number of image color types, and video length in the advertisement, providing a standardized measurement basis for subsequent causal analysis. In this way, the influence of advertising content can be expressed in a quantitative manner, avoiding the problem of difficulty in measuring advertising influence in traditional methods. Secondly, the system calculates the causal effect for each candidate edge and quantifies it using a formula. The design of this formula ensures that when the influence of the advertising content is determined, the shorter the user response time, the better. The smaller the value, the more significant the causal effect. This causal effect calculation method based on time decay can effectively avoid misjudgment of causal relationships, allowing the system to accurately identify the extent to which advertising affects user behavior. Compared to traditional causal analysis methods that rely on manual experience to set influence weights, this method provides an adaptive causal effect calculation method, which improves the objectivity and reliability of the analysis. In addition, the system sets a fixed causal effect significance threshold. , used to screen whether the causal relationship is significant. , then the advertisement is considered to have a clear causal effect on user behavior, otherwise it is discarded. Only advertisement-user behavior pairs with significant causal effects are allowed to pass the screening, making the candidate causal edge set more accurate and ensuring the high quality of the final causal analysis. The filter conditions have been relaxed, allowing more advertising-user behavior relationships to enter the scope of causal analysis, although the causal effects contained therein may be weaker. The system dynamically balances the accuracy and coverage of causal analysis to meet the needs of diverse applications. Finally, the system optimizes the handling of multiple user behaviors. If the same ad meets the time conditions for multiple user behaviors, the causal effect is calculated separately for each ad-user behavior combination. The calculations for each pair are performed independently and do not affect each other. This design ensures the independence of causal analysis and avoids errors caused by data interference during the causal effect calculation process. This allows the system to accurately assess the impact of ads on different user behaviors, providing data support for personalized advertising strategies. In summary, by calculating dynamic causal relationship discovery and causal effects, the system quantifies advertising influence, automatically calculates causal effects, and quantitatively screens causal significance, making advertising causal analysis more accurate and efficient. Compared to traditional methods that rely on empirical judgment or simple time window matching, this method provides a scientific causal effect calculation mechanism, improves the accuracy of advertising effectiveness evaluation, and provides a solid data foundation for personalized advertising delivery and optimization strategies.

[0021] 7. Through the steps of causal graph structure verification, the system successfully solved the problems of abnormal data, low-credibility causal relationships, and erroneous causal edges that may exist in the causal analysis process, thereby ensuring the reliability and accuracy of the final causal analysis results. First, the system sets the number of advertisements remaining after preprocessing and the number of remaining user behaviors that meet the time conditions, and constructs a causal effect matrix based on this. By constructing this matrix, the system uniformly structures and manages all advertisement-user behavior causal relationships to ensure the efficiency of subsequent causal relationship verification and screening. Compared with traditional methods based on hash storage or separately recording causal relationships, this matrix structure can significantly improve data retrieval and computing efficiency, enabling the system to quickly perform batch verification and optimization. Secondly, the system performs strict numerical verification on each valid causal edge to ensure that two core conditions are met. and This step effectively avoids computational anomalies and data noise that may occur during the causal effect calculation process, improving the credibility and stability of causal analysis. Furthermore, the system implements an automatic removal mechanism for records that do not meet the above conditions. This mechanism simultaneously removes corresponding entries from the causal effect matrix and the set of valid causal edges, ensuring that the remaining causal relationship data conforms to predefined rules. This approach automatically selects high-quality causal edges, avoiding biased causal analysis results caused by erroneous or anomalous data. Compared to traditional methods that rely on manual filtering or simple time window screening, this step enables efficient causal relationship screening based on numerical verification, making the final causal graph structure more accurate and reliable. In summary, through the causal graph structure verification step, the system achieves structured storage of causal effects, rigorous numerical screening, and automatic removal of anomalous data, making the final causal analysis results more scientific and reliable. Compared to traditional methods, this solution provides an automated, efficient, and scalable causal relationship verification mechanism, providing a reliable data foundation for the precise construction of causal inference models, while reducing the error rate in the causal analysis process and providing higher-quality causal relationship data support for targeted advertising and user behavior prediction.

[0022] 8. By integrating and displaying results, the system successfully addresses issues such as difficulty visually presenting causal effect analysis results, difficulty comparing different causal relationships, and lack of structured data storage, making analysis results clearer, more intuitive, and more interpretable. First, the system establishes a two-dimensional coordinate system with ad display time as the horizontal axis and causal effect value as the vertical axis, allowing all valid causal edges to be visualized within a unified timeframe. Compared to the traditional method of listing numerical values ​​separately, this approach makes the distribution of causal effects over time more intuitive, allowing users to quickly identify ads with strong causal effects and their impact range. This visualization allows analysts to more easily observe causal trends after ad placement, thereby optimizing advertising strategies and improving the accuracy of advertising. Second, the system draws a straight line on the coordinate graph for each valid causal edge, starting at the ad display time and ending at the corresponding user behavior time. This approach clearly illustrates the span of the causal relationship along the timeline, visualizing the impact of advertising on user behavior. Compared to methods that display data using only numerical tables, this approach more intuitively reflects the time delay of causal effects, helping analysts optimize advertising time windows and improve advertising reach effectiveness. Furthermore, the system linearly maps causal effect values ​​using line thickness and color depth, making causal edges with stronger causal effects (low values) more prominent in the graph and those with weaker causal effects (high values) less prominent. This visualization strategy allows analysts to quickly identify the most influential advertising placements without having to examine each value individually, improving analysis efficiency and avoiding information overload caused by excessive data volumes. Finally, the system consolidates detailed data for all valid causal edges into a structured data table. Each row includes key information such as the ad ID, text description, image data, video data, user action ID, action type, action timestamp, calculated response time, and causal effect value. Compared to unstructured storage methods, this tabular management approach greatly improves data query and analysis convenience, facilitating subsequent use in machine learning models or optimization algorithms. This ensures the integrity and consistency of causal analysis data, providing solid data support for further ad optimization and placement strategies. In summary, through the integrated display of results, the system achieves visualization of causal effect analysis results, structured information storage, and efficient screening of key causal relationships, making causal analysis more intuitive, efficient, and interpretable. Compared to traditional methods, this solution not only improves data clarity but also greatly optimizes analysis efficiency, providing strong support for precise advertising decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0025] Example, see Figure 1 , an online advertising analysis method based on artificial intelligence, comprising: Set the ad set to ;in, It is the collection of all advertisements to be analyzed in this system; For the Ads, ; is the total number of advertisements; advertise Contains: text description , a string containing the ad copy content; image data , a two-dimensional matrix with each element being Color triplet, describing the ad image; video data , contains video file data, which includes playback duration information; display timestamp , a real number in seconds, indicating the time when the ad starts to be displayed; Set the string character count function to: ;in, The string whose characters are to be counted; Is a positive integer; this function takes an input string Count all the characters in and output a positive integer; Computational Advertising The text value of , specifically: ;in, For advertising The number of characters in the text, in units of count; Set the image data digitization function to: ;in, is the image data to be digitized; For images A unique color value detected in ; For images The total number of different color categories detected in the image; this function scans the image pixel by pixel , and each unique The triples are recorded in a set, and the set is returned; Computational Advertising The image values ​​are: ;in, For advertising Medium Image The number of different colors in is an exact count of the elements in the collection; Set the function to get the video playback duration: ;in, The video file whose playback duration is to be obtained; Is a positive real number; this function extracts Parse the file header and return the playback duration in seconds. The output is a positive real number. Computational Advertising The video duration is as follows: ;in, For advertising Medium video duration; Set up your ad The comprehensive description value of is: ;in, For advertising Comprehensive description of numerical values; For advertising The simple sum of the values ​​of each data item in , in "value units", is used for subsequent causal effect calculations; Set up your ad The standard record is: ;Record Contains all numerically determined variables, each with a clear definition and unit.

[0026] By establishing ad sets and their standard records, the solution clarifies the various elements involved in ad data collection, resolving issues of inconsistent data sources and data disorganization. Specifically, by defining ad sets and explicitly describing each ad's text description, image data, video data, and display timestamp, the system ensures that each ad's information has a unified and clear source and format, addressing the issue of fragmented and non-standardized data and achieving structured data management. By establishing a string character count function and counting characters in ad text, the solution addresses the quantification of text data, ensuring that ad content can be represented with specific numerical values, making subsequent comparisons and calculations on text data feasible. This step transforms complex text information into simple numerical features, facilitating algorithmic processing and quantitative analysis. Furthermore, by defining an image data quantification function, the ad image is scanned pixel by pixel, and each unique color triplet is recorded in a set. A set counting function is then used to calculate the number of distinct color types in the image. This process addresses the challenges of abstracting and quantitatively representing image data. This allows image information to be objectively digitized in the description of overall ad content, enhancing data integration and cross-modal fusion. Similarly, a function for obtaining video playback duration is set up to parse the playback duration from the video file header, so that the video data can be represented by a unified and standard numerical value, solving the problem of inconsistent expression of time information in video data, thus laying the foundation for the subsequent comprehensive description of multimodal data. Finally, by simply summing the numerical values ​​of text, image and video data to obtain a comprehensive description value of the advertisement, the solution successfully integrates multiple data items into a unified indicator, solves the problem of difficulty in direct comparison of various data types, and achieves the effect of unified processing and convenient application of multimodal information. Overall, these steps ensure that in the entire process of advertising data collection and standardization, each link of the data has a clear definition and specific quantification method, thereby providing solid and clear basic data support for subsequent causal analysis, dynamic modeling and algorithm optimization.

[0027] It also includes user behavior data collection and preprocessing, specifically: Set the user behavior collection to: ;in, For the User behavior records, ; is the total number of user behaviors; User behavior The collected data items include: behavior type , using numerical coding, defined as click as 1, browse as 2, conversion as 3; behavior timestamp , a real number in seconds; Setting User Behavior Recorded as: .

[0028] By setting up user behavior sets and defining user behavior records, the solution first standardizes and unifies the collection of user behavior data, thereby solving the problem of diverse data sources and inconsistent formats. The user behavior set is clearly defined in the steps. This approach ensures that all user behavior data has a clear identification and a unified indexing system, thus avoiding data confusion and duplicate records. At the same time, by standardizing the data items collected by user behavior, such as encoding the behavior type with a numerical value, coding clicks as 1, browsing as 2, and conversions as 3, and specifying the behavior timestamp as a real number in seconds, the solution solves the problem of unstructured and non-standard behavior data before quantitative analysis. In this way, in each user behavior record, and The user's behavior category and occurrence time are clearly indicated, ensuring the comparability and accuracy of the data in subsequent analysis. The effect of this step is to convert the diverse user behavior data into standard records in a unified format after rigorous preprocessing, allowing the system to efficiently sort, filter, and quantitatively compare user behaviors. After data standardization, the system can quickly and accurately match the temporal correlation between ad display and user behavior when analyzing user reactions to ads, thereby providing a reliable time series basis for constructing dynamic causal relationship diagrams. Through this method, the system not only improves the efficiency of data processing, but also reduces the risk of errors caused by confusing data formats, laying a solid foundation for subsequent causal effect calculations and multimodal data fusion.

[0029] It also includes the construction of the initial causal model and the setting of assumptions, specifically: Set the ad node collection to: ;in, For advertising The associated node identifies the advertising event; Set the user behavior node collection to: ;in, For user behavior associated nodes; Set the overall node collection to: ; For any advertising node User behavior nodes , set the time difference to: ;in, For advertising and user behavior The response time between the two, in seconds; Set the maximum allowed time difference between ad display and user behavior to ,set up ;when When the value is large, the system will consider user behaviors that occur in a longer time range as potential responses to advertising influence, which may increase the number of candidate edges, but may also introduce user behaviors that are not directly affected by advertising; when When the value is small, the system only focuses on the user behavior that occurs immediately after the ad is displayed. The number of candidate edges may be reduced, but the direct causal relationship can be captured more accurately. Set time condition: Only when When, think Maybe Make an impact; Set the candidate causal edge set to: ; Each edge It only indicates that there is a potential influence relationship between advertisements that meet the conditions in time and user behavior, without assigning other additional attributes; Set the static causal diagram to: ;picture It is composed of all nodes and candidate causal edges and serves as the basis for subsequent dynamic causal analysis.

[0030] By constructing an initial causal model and setting hypotheses, the system achieves a preliminary modeling of the potential causal relationship between ad data and user behavior data, thereby resolving the issues of unclear temporal correlations and data matching difficulties between these two data types. First, by establishing an ad node set, a user behavior node set, and an overall node set, the system uniformly identifies each independent data source. The ad node set categorizes all ad-related events, while the user behavior node set identifies user behaviors after an ad is displayed. Together, the overall node set enables centralized data management and unified indexing. This step addresses the issues of fragmented data sources and chaotic node representation, providing a clear and unified foundation for subsequent causal relationship construction. Second, by defining the response time difference between ads and user behavior, the system can quantify the time interval between ad display and user behavior, providing a direct basis for determining whether an ad has an impact on user behavior. Setting a maximum allowable time difference as a temporal condition further clarifies that an ad is considered to have an impact on user behavior only if the behavior occurs within a specific time range after the ad is displayed. This approach eliminates indirect impact behaviors that occur long after the ad is displayed, while retaining records of user behaviors that occur quickly after the ad is displayed and are causally related. Third, constructing a set of candidate causal edges is a critical step. Each edge in this set represents only the potential influence between ads and user behavior that meet the temporal conditions, without any additional attributes. This approach effectively eliminates data that does not conform to the causal assumption through a temporal filtering mechanism, thereby ensuring the accuracy and relevance of the data in subsequent causal model analysis. This process addresses the problem of spurious associations caused by excessively long time windows in traditional systems, while also reducing interference factors, allowing the system to focus on capturing direct and true causal relationships. Finally, constructing a static causal graph as the foundation for subsequent dynamic causal analysis not only provides a preliminary framework for establishing a network of relationships between advertising events and user behavior but also lays the data structure foundation for further introduction of dynamic updates and causal effect calculation. By constructing this preliminary model, the system can effectively distinguish between direct and indirect responses, ensuring that subsequent steps can more accurately quantify and verify the actual impact of ads on user behavior. Overall, through these steps, the system successfully overcomes the data management and causal relationship establishment challenges inherent in existing technologies, which are often caused by fragmented data sources, inconsistent node definitions, and inaccurate temporal matching. In terms of effect, this unified and rigorous causal model construction not only improves the accuracy and efficiency of data preprocessing, but also provides a solid foundation for subsequent complex dynamic causal analysis and multimodal data fusion, significantly enhancing the scientific nature and credibility of advertising effectiveness analysis.

[0031] It also includes feature extraction and node attribute construction, specifically: For each advertising node , construct the attribute vector as: ; Set the context resonance attribute calculation function to: , ; in, is the set of all vectors consisting of three real numbers, that is, three-dimensional real number space; function input parameters, is the number of text characters; is the number of image color types; is the video duration; output , for advertising The contextual resonance value of is a positive real number; Computational Advertising The contextual resonance value of is: ; For each user behavior node , construct the attribute vector, specifically: .

[0032] Through feature extraction and node attribute construction, the system effectively addresses the inconsistency and difficulty in directly comparing multimodal information between advertising and user behavior data. First, by constructing an attribute vector for each ad node, the system converts ad text, image, and video data into clear numerical representations, with the number of text characters, number of image color types, and video duration as vector components. This step addresses the difficulty of unifying the jumbled raw multimodal data, allowing all ad data to be represented in a unified numerical space, laying the foundation for subsequent analysis. Second, the system introduces a contextual resonance attribute calculation function, defined in three-dimensional real space. The function takes as input the number of text characters, number of image color types, and video duration, and outputs the ad's contextual resonance value. This custom function combines different types of data into a single positive real number, reflecting the comprehensive performance of the ad content. This function effectively integrates information from different modalities, resolving the inconsistent data dimensions and difficulty in comprehensive comparison inherent in traditional methods. Furthermore, an attribute vector is constructed for each user behavior node, including behavior type and response time. This achieves a standardized representation of user behavior data, enabling comparison and correlation with the attribute vectors of ad nodes within the same mathematical space. This not only provides a precise quantitative basis for subsequent analysis of the causal relationship between advertising and user behavior, but also significantly improves data operability and analytical efficiency. Overall, through feature extraction and node attribute construction, the system addresses the issues of information loss and the lack of unified quantification in traditional advertising analysis caused by the clutter of multimodal data. This step extracts complex text, image, and video data into clear, unified numerical features and fuses these features into a comprehensive indicator using a contextual resonance function, allowing the comprehensive performance of advertising content to be reflected in a single, deterministic numerical form. Furthermore, after undergoing the same rigorous attribute construction, user behavior data can be directly compared with advertising data within the same space, providing high-precision data support for the construction of dynamic causal relationship models. The ultimate result is improved data processing accuracy and consistency, reduced errors caused by the difficulty of quantitative comparison between multimodal data, and significantly enhanced the scientific nature and reliability of advertising effectiveness analysis.

[0033] It also includes the construction of event timing and causal edge attributes, specifically: Record all ads By display time Sort in ascending order; Record all user behavior By time of occurrence Sort in ascending order; Will As a candidate edge time span; Setting up the function , used to assign time spans to edges, specifically: ;function Only the calculated As an edge The attribute ensures that the edge carries clear and quantifiable information.

[0034] By constructing event temporal ordering and causal edge attributes, the system successfully addresses the temporal disorder of ad and user behavior data and the lack of precise temporal characteristics in causal edges. First, the system sorts all ad records in ascending order by display time and all user behavior records in ascending order by occurrence time. This step ensures that ad and user behavior data are arranged in chronological order, allowing subsequent causal analysis to strictly adhere to temporal relationships and avoid misattribution caused by data misordering. Second, the system sets a time span as a temporal attribute for candidate edges. This span measures the time interval between user behavior and ad display. This definition of a time span ensures that the system adheres to temporal logic when constructing causal edges, establishing causal relationships only between ads and user behaviors that meet the temporal conditions. This design avoids the causal confusion that can occur in traditional ad analysis methods due to unclear time spans, thereby improving the accuracy of causal inference. Furthermore, the system implements a time span assignment function to assign the calculated time span to candidate causal edges, ensuring that each edge carries a clear temporal attribute. This function ensures that causal edges not only adhere to temporal ordering requirements but also possess quantifiable temporal characteristics. This enables subsequent causal analysis to be more refined and adjusted based on specific time spans. Compared to traditional methods that rely on manually defined time windows, this step provides an automated and precise temporal characterization mechanism, making the construction of causal edges more scientific and rational. Overall, by constructing event temporal ordering and causal edge attributes, the system addresses the issues of chaotic data temporal order and unclear temporal characteristics of causal edges in advertising analysis. After sorting advertising and user behavior data, causal analysis can be performed strictly according to temporal logic, avoiding erroneous causal inferences. Furthermore, by assigning precise time span attributes to causal edges, the system achieves standardized quantification of temporal information, enabling data-driven analysis of the causal relationship between advertising and user behavior rather than relying on subjective assumptions. This improvement enhances the reliability of causal analysis, making advertising effectiveness evaluation more accurate and laying a solid foundation for subsequent causal inference and model optimization.

[0035] It also includes computational dynamic causal relationship discovery and causal effects, specifically: Set up your ad The comprehensive description value of ; For each candidate edge , calculate the causal effect, specifically: ; in, For advertising User behavior The design of this formula ensures that under the condition that the value of the advertising content is determined, the shorter the response time, the The smaller it is, the more significant the causal effect is; Set the fixed causal effect significance threshold, denoted as ,set up ; When the calculated When , the causal effect represented by the candidate edge is considered significant and clear; smaller The value requires that the response time between advertising and user behavior is shorter than the value of the comprehensive description of the advertising, so that only relationships with very fast responses have significant causal effects, resulting in fewer and stricter candidate edges. The value of 1 relaxes this condition, and more candidate edges will be considered as valid causal relationships, which may introduce more edges, but the causal effects contained in them may not be as good as those of low 1. As obvious as when Set the valid causal edge set to: ;gather Each edge Additional attributes They are all deterministic values ​​calculated by the causal effect quantification function; If the same advertisement Behavior for multiple users If both meet the time condition, then calculate each pair of combinations separately , each of them is independent and the parameters are directly substituted into the calculation.

[0036] By calculating the steps of dynamic causal relationship discovery and causal effect, the system successfully solves the problems of ambiguous causal relationship between advertising and user behavior, difficulty in quantifying causal effects, and unclear handling of multi-user behavior, making causal analysis more scientific and accurate. First, the system sets a comprehensive description value for the advertisement, and uniformly expresses multimodal information such as the number of text characters, number of image color types, and video length in the advertisement, providing a standardized measurement basis for subsequent causal analysis. In this way, the influence of advertising content can be expressed in a quantitative way, avoiding the problem of difficulty in measuring advertising influence in traditional methods. Secondly, the system calculates the causal effect for each candidate edge and quantifies it using a formula. The design of this formula ensures that when the influence of the advertising content is determined, the shorter the user response time, the better. The smaller the value, the more significant the causal effect. This causal effect calculation method based on time decay can effectively avoid misjudgment of causal relationships, allowing the system to accurately identify the extent to which advertising affects user behavior. Compared to traditional causal analysis methods that rely on manual experience to set influence weights, this method provides an adaptive causal effect calculation method, which improves the objectivity and reliability of the analysis. In addition, the system sets a fixed causal effect significance threshold. , used to screen whether the causal relationship is significant. , then the advertisement is considered to have a clear causal effect on user behavior, otherwise it is discarded. Only advertisement-user behavior pairs with significant causal effects are allowed to pass the screening, making the candidate causal edge set more accurate and ensuring the high quality of the final causal analysis. The filter conditions have been relaxed, allowing more advertising-user behavior relationships to enter the scope of causal analysis, although the causal effects contained therein may be weaker. The system dynamically balances the accuracy and coverage of causal analysis to meet the needs of diverse applications. Finally, the system optimizes the handling of multiple user behaviors. If the same ad meets the time conditions for multiple user behaviors, the causal effect is calculated separately for each ad-user behavior combination. The calculations for each pair are performed independently and do not affect each other. This design ensures the independence of causal analysis and avoids errors caused by data interference during the causal effect calculation process. This allows the system to accurately assess the impact of ads on different user behaviors, providing data support for personalized advertising strategies. In summary, by calculating dynamic causal relationship discovery and causal effects, the system quantifies advertising influence, automatically calculates causal effects, and quantitatively screens causal significance, making advertising causal analysis more accurate and efficient. Compared to traditional methods that rely on empirical judgment or simple time window matching, this method provides a scientific causal effect calculation mechanism, improves the accuracy of advertising effectiveness evaluation, and provides a solid data foundation for personalized advertising delivery and optimization strategies.

[0037] It also includes verification of the causal graph structure, specifically: Assume the number of remaining ads after preprocessing is , the remaining number of user behaviors that meet the time condition is , construct the causal effect matrix as: ; in, for OK A matrix of columns, where each element is from the set of real numbers ; ;matrix Chinese elements Fully quantified advertising and user behavior causal effects between For each valid edge , verify that strictly meets: and ; If there are records that do not meet the conditions, then in the causal effect matrix and the effective causal edge set Delete the corresponding items to ensure that the retained records are valid data that meets the specified requirements.

[0038] Through the steps of causal graph structure verification, the system successfully solved the problems of abnormal data, low-credibility causal relationships, and erroneous causal edges that may exist in the causal analysis process, thereby ensuring the reliability and accuracy of the final causal analysis results. First, the system sets the number of advertisements remaining after preprocessing and the number of remaining user behaviors that meet the time conditions, and constructs a causal effect matrix based on this. By constructing this matrix, the system uniformly structures and manages all advertisement-user behavior causal relationships to ensure the efficiency of subsequent causal relationship verification and screening. Compared with traditional methods based on hash storage or separate recording of causal relationships, this matrix structure can significantly improve data retrieval and computing efficiency, enabling the system to quickly perform batch verification and optimization. Secondly, the system performs strict numerical verification on each valid causal edge to ensure that two core conditions are met and This step effectively avoids computational anomalies and data noise that may occur during the causal effect calculation process, improving the credibility and stability of causal analysis. Furthermore, the system implements an automatic removal mechanism for records that do not meet the above conditions. This mechanism simultaneously removes corresponding entries from the causal effect matrix and the set of valid causal edges, ensuring that the remaining causal relationship data conforms to predefined rules. This approach automatically selects high-quality causal edges, avoiding biased causal analysis results caused by erroneous or anomalous data. Compared to traditional methods that rely on manual filtering or simple time window screening, this step enables efficient causal relationship screening based on numerical verification, making the final causal graph structure more accurate and reliable. In summary, through the causal graph structure verification step, the system achieves structured storage of causal effects, rigorous numerical screening, and automatic removal of anomalous data, making the final causal analysis results more scientific and reliable. Compared to traditional methods, this solution provides an automated, efficient, and scalable causal relationship verification mechanism, providing a reliable data foundation for the precise construction of causal inference models, while reducing the error rate in the causal analysis process and providing higher-quality causal relationship data support for targeted advertising and user behavior prediction.

[0039] It also includes the integrated display of results, specifically: Establish a two-dimensional coordinate system, with the horizontal axis representing the ad display time , the vertical axis represents the causal effect value ; For each valid edge , draw a straight line in the graph: Starting point , the end point is , on the horizontal axis Map the moment when user behavior occurs; When drawing, the line thickness and color are based on The values ​​use a predefined linear mapping relationship: the lower the value (the more significant the causal effect), the thicker the line and the darker the color; the higher the value, the thinner the line and the lighter the color; Each valid edge The detailed data is integrated into a data table, each row includes: Advertisement Number and corresponding in 、 、 and ; User behavior number and in and ; Calculated response time and causal effects .

[0040] By integrating and displaying results, the system successfully addresses issues such as the difficulty in visually presenting causal effect analysis results, the difficulty in comparing different causal relationships, and the lack of structured data storage, making the analysis results clearer, more intuitive, and more interpretable. First, the system establishes a two-dimensional coordinate system with ad display time as the horizontal axis and causal effect value as the vertical axis, allowing all valid causal edges to be visualized within a unified timeframe. Compared to the traditional method of listing numerical values ​​separately, this approach makes the distribution of causal effects over time more intuitive, allowing users to quickly identify ads with strong causal effects and their impact areas. This visualization allows analysts to more easily observe causal trends after ad placement, thereby optimizing advertising strategies and improving the accuracy of advertising. Second, the system draws a straight line on the coordinate graph for each valid causal edge, starting at the ad display time and ending at the corresponding user behavior time. This approach clearly illustrates the span of the causal relationship along the timeline, visualizing the process of advertising's impact on user behavior. Compared to methods that display data using only numerical tables, this approach more intuitively reflects the time delay of causal effects, helping analysts optimize advertising delivery windows and improve advertising reach effectiveness. Furthermore, the system linearly maps causal effect values ​​using line thickness and color depth, making causal edges with stronger causal effects (low values) more prominent in the graph and those with weaker causal effects (high values) less prominent. This visualization strategy allows analysts to quickly identify the most influential advertising placements without having to examine each value individually, improving analysis efficiency and avoiding information overload caused by excessive data volumes. Finally, the system consolidates detailed data for all valid causal edges into a structured data table. Each row includes key information such as the ad ID, text description, image data, video data, user action ID, action type, action timestamp, calculated response time, and causal effect value. Compared to unstructured storage methods, this tabular management approach greatly improves data query and analysis convenience, facilitating subsequent use in machine learning models or optimization algorithms. This ensures the integrity and consistency of causal analysis data, providing solid data support for further ad optimization and placement strategies. In summary, through the integrated display of results, the system achieves visualization of causal effect analysis results, structured information storage, and efficient screening of key causal relationships, making causal analysis more intuitive, efficient, and interpretable. Compared to traditional methods, this solution not only improves data clarity but also greatly optimizes analysis efficiency, providing strong support for precise advertising decision-making.

[0041] This embodiment further provides a system for an online advertising analysis method based on artificial intelligence, including: Advertisement content collection module: collects advertisement copy content, advertisement images, video file data and advertisement start time; User behavior data collection module: collects user behavior types and behavior timestamps; Computing module: used for data calculation.

[0042] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0043] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present invention. These improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. An online advertising analysis method based on artificial intelligence, characterized in that: include: Set the ad set to ;in, It is the collection of all advertisements to be analyzed in this system; For the Ads, ; is the total number of advertisements; advertise Contains: text description , a string containing the ad copy content; image data , a two-dimensional matrix with each element being Color triplet, describing the ad image; video data , contains video file data, which includes playback duration information; display timestamp , a real number in seconds, indicating the time when the ad starts to be displayed; Set the string character count function to: ;in, The string whose characters are to be counted; is a positive integer; Computational Advertising The text value of , specifically: ;in, For advertising The number of characters in the text, in units of count; Set the image data digitization function to: ;in, is the image data to be digitized; For images A unique color value detected in ; For images The total number of different color categories detected in the image; this function scans the image pixel by pixel , and each unique The triples are recorded in a set, and the set is returned; Computational Advertising The image values ​​are: ;in, For advertising Medium Image The number of different colors in is an exact count of the elements in the collection; Set the function to get the video playback duration: ;in, The video file whose playback duration is to be obtained; is a positive real number; Computational Advertising The video duration is as follows: ;in, For advertising Medium video duration; Set up your ad The comprehensive description value of is: ;in, For advertising Comprehensive description of numerical values; Set up your ad The standard record is: .

2. The online advertising analysis method based on artificial intelligence according to claim 1, characterized in that: It also includes user behavior data collection and preprocessing, specifically: Set the user behavior collection to: ;in, For the User behavior records, ; is the total number of user behaviors; User behavior The collected data items include: behavior type , using numerical coding, defined as click as 1, browse as 2, conversion as 3; behavior timestamp , a real number in seconds; Setting User Behavior Recorded as: .

3. The online advertising analysis method based on artificial intelligence according to claim 2, characterized in that: It also includes the construction of the initial causal model and the setting of assumptions, specifically: Set the ad node collection to: ;in, For advertising The associated node identifies the advertising event; Set the user behavior node collection to: ;in, For user behavior associated nodes; Set the overall node collection to: ; For any advertising node User behavior nodes , set the time difference to: ;in, For advertising and user behavior The response time between the two, in seconds; Set the maximum allowed time difference between ad display and user behavior to ,set up ; Set time condition: Only when When, think Maybe Make an impact; Set the candidate causal edge set to: ； Set the static causal diagram to: .

4. The online advertising analysis method based on artificial intelligence according to claim 3, characterized in that: It also includes feature extraction and node attribute construction, specifically: For each advertising node , construct the attribute vector as: ; Set the context resonance attribute calculation function to: , ; in, is the set of all vectors consisting of three real numbers, that is, three-dimensional real space; function input parameters, is the number of text characters; is the number of image color types; is the video duration; output , for advertising The contextual resonance value of is a positive real number; Computational Advertising The contextual resonance value of is: ; For each user behavior node , construct the attribute vector, specifically: 。 5. The online advertising analysis method based on artificial intelligence according to claim 4, characterized in that: It also includes the construction of event timing and causal edge attributes, specifically: Record all ads By display time Sort in ascending order; Record all user behavior By time of occurrence Sort in ascending order; Will As a candidate edge time span; Setting up the function , used to assign time spans to edges, specifically: .

6. The online advertising analysis method based on artificial intelligence according to claim 5, characterized in that: It also includes computational dynamic causal relationship discovery and causal effects, specifically: Set up your ad The comprehensive description value of ; For each candidate edge , calculate the causal effect, specifically: ; in, For advertising User behavior The causal effect value of Set the fixed causal effect significance threshold, denoted as ,set up ; Set the valid causal edge set to: .

7. The online advertising analysis method based on artificial intelligence according to claim 6, characterized in that: It also includes verification of the causal graph structure, specifically: Assume the number of remaining ads after preprocessing is , the remaining number of user behaviors that meet the time condition is , construct the causal effect matrix as: ; in, for OK A matrix of columns, where each element is from the set of real numbers ; ; For each valid edge , verify that strictly meets: and ; If there are records that do not meet the conditions, then in the causal effect matrix and the effective causal edge set Delete the corresponding item.

8. The online advertising analysis method based on artificial intelligence according to claim 7, characterized in that: It also includes the integrated display of results, specifically: Establish a two-dimensional coordinate system, with the horizontal axis representing the ad display time , the vertical axis represents the causal effect value ; For each valid edge , draw a straight line in the graph: Starting point , the end point is ; When drawing, the line thickness and color are based on The values ​​use a predefined linear mapping relationship: the lower the value, the thicker the line and the darker the color; the higher the value, the thinner the line and the lighter the color; Each valid edge The detailed data is integrated into a data table, each row includes: Advertisement Number and corresponding in 、 、 and ; User behavior number and in and ; Calculated response time and causal effects .

9. A system using the artificial intelligence-based online advertising analysis method according to claim 8, characterized in that: include: Advertisement content collection module: collects advertisement copy content, advertisement images, video file data and advertisement start time; User behavior data collection module: collects user behavior types and behavior timestamps; Computing module: used for data calculation.

Images