An AI-based method and system for evaluating the effectiveness of advertising campaigns.

Abstract

This application provides an AI-based method and system for evaluating advertising performance. It acquires user interaction data and brand association data for advertising services within the evaluation period; extracts short-term performance indicators to characterize the direct conversion effect of the ads; generates short-term evaluation labels for the advertising performance based on a Shapley value attribution model and the short-term performance indicator features; extracts long-term value indicators to characterize user brand awareness and long-term relationships from the brand association data; and constructs a multilayer perceptron neural network as a fusion model, taking the short-term evaluation labels and long-term value indicators as input. Feature fusion and dynamic weight allocation are performed through a nonlinear mapping layer in the fusion model, and the output layer outputs a dynamic performance evaluation score. Using the scheme of this application, integrated feature fusion modeling based on short-term labels and long-term value indicators can be achieved, thereby improving the comprehensive accuracy of advertising performance evaluation across time scales.

Description

Technical Field

[0001] This application relates to the field of internet advertising service technology, and more specifically, to an artificial intelligence-based method and system for evaluating the effectiveness of advertising campaigns. Background Technology

[0002] In the current digital economy context, internet advertising services have become an important means of brand marketing. With the diversification of advertising channels and the increase in user touchpoints, advertisers can achieve precise targeting and real-time monitoring through online platforms. User interactions during advertising campaigns, such as exposure, clicks, dwell time, and conversions, are recorded in detail by the platform and used to evaluate advertising effectiveness. At the same time, users' long-term brand awareness and loyalty are also reflected through data such as search behavior, organic traffic conversations, and social media interactions. These data provide rich quantitative evidence for advertising effectiveness analysis. However, traditional advertising effectiveness evaluation methods mostly rely on single indicators or static analysis, making it difficult to fully grasp the comprehensive effect of advertising at different time scales.

[0003] Existing technologies have significant limitations in evaluating advertising effectiveness, particularly in integrating short-term conversion rates with long-term brand value. Current methods typically rely solely on short-term metrics such as conversion rate, click-through rate, or return on investment (ROI), failing to capture the potential impact of user brand awareness and long-term relationships on advertising performance. Furthermore, the lack of dynamic weighting and non-linear correlation modeling capabilities for different feature types results in insufficient accuracy across time scales. In complex multi-touchpoint, multi-channel advertising environments, single metrics or linear models struggle to capture the interactive effects and cumulative long-term value between advertising touchpoints, leading to significant biases and insufficient guidance in evaluation results. Therefore, achieving integrated feature fusion modeling based on short-term tags and long-term value metrics to improve the comprehensive accuracy of advertising effectiveness evaluation across time scales has become a challenging issue for the industry. Summary of the Invention

[0004] This application provides an artificial intelligence-based method and system for evaluating the effectiveness of advertising campaigns, which can achieve integrated feature fusion modeling based on short-term tags and long-term value indicators, thereby improving the overall accuracy of advertising campaign effectiveness evaluation across time scales.

[0005] Firstly, this application provides an artificial intelligence-based method for evaluating the effectiveness of advertising campaigns, including: In response to requests for evaluation of the effectiveness of internet advertising services, obtain user interaction data and brand association data of internet advertising services during the evaluation period; Based on the user interaction behavior data, short-term performance indicator features for characterizing the direct conversion effect of the advertisement are extracted, and then short-term evaluation labels for the campaign performance are generated based on the Shapley value attribution model and the short-term performance indicator features. Long-term value indicators used to characterize user brand awareness and long-term relationships are extracted from the brand association data. These long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content. A multilayer perceptron neural network is constructed as a fusion model. The short-term evaluation label and the long-term value index are taken as inputs. Feature fusion and dynamic weight allocation are performed through the nonlinear mapping layer in the fusion model, and a comprehensive dynamic effect evaluation score is output through the output layer.

[0006] Preferably, based on the user interaction behavior data, extracting short-term performance indicators to characterize the direct conversion effect of the advertisement specifically includes: Based on user interaction behavior data, identify advertising touchpoints related to ad delivery during the evaluation period; Calculate the ad content reach rate and user exposure frequency for each ad touchpoint; Statistics on the number of users who ultimately converted through each advertising touchpoint and the length of the conversion path; Based on the conversion behavior, calculate the click-through rate and conversion cost for each ad touchpoint; The ad content reach rate, user exposure frequency, number of users who converted, conversion path length, click pass rate, and conversion cost are integrated into a vector, and the vector is used as a short-term performance indicator to characterize the direct conversion effect of the ad.

[0007] Preferably, the short-term evaluation labels for the deployment effect generated based on the Shapley value attribution model and the aforementioned short-term effect indicator features specifically include: Each complete conversion path is defined as a cooperative game, in which each advertising touchpoint is regarded as a player in the game. Based on the Monte Carlo sampling method, the approximate Shapley value contribution of each ad touchpoint to the occurrence of the conversion in the conversion path is calculated. The Shapley value contribution of each advertising touchpoint is weighted and summed with the corresponding conversion cost in the short-term effect indicator features to generate a comprehensive conversion efficiency value. Based on a preset threshold range, the comprehensive conversion efficiency value is mapped to discrete short-term evaluation labels.

[0008] Preferably, the long-term value indicators extracted from the brand association data to characterize user brand awareness and long-term relationships specifically include: Based on the brand association data, the relative change rate of brand search volume between the current evaluation period and the previous evaluation period is extracted, and the relative change rate is used as the month-on-month growth rate of brand search volume. Determine the growth rate of the number of natural traffic sessions in the current evaluation period relative to a preset baseline value, and use the growth rate as the incremental ratio of natural traffic sessions; Monitor user-generated content related to the brand on preset social media platforms during the current evaluation period, count the number of posts and positive interaction data identified by sentiment analysis, and calculate the proportion of user interaction content. The month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content were standardized, and the standardized results were combined to construct a long-term value indicator.

[0009] Preferably, constructing a multilayer perceptron neural network as a fusion model specifically includes: Determine the input layer dimension of the fusion model to match the total dimension of the short-term evaluation label and the long-term value indicator; At least one fully connected hidden layer is sequentially set as a nonlinear mapping layer after the input layer of the fusion model, and a Sigmoid or ReLU activation function is selected for each hidden layer; The output layer is connected after the last hidden layer, configured as a single neuron node, and the Sigmoid activation function is selected to limit the output value range to between 0 and 1. Initialize the connection weights and bias terms of all neurons in the input layer, hidden layer, and output layer; The fusion model is trained in a supervised manner using short-term evaluation tags, long-term value indicators, and corresponding labeling effect scores from historical advertising campaigns.

[0010] Preferably, the user interaction behavior data includes user interaction records such as exposure, clicks, dwell time, bounce rate, and conversions generated during the advertising campaign period.

[0011] Preferably, the brand association data includes brand-related search volume, organic traffic session data, and interactive statistics of user-generated content across platforms.

[0012] Secondly, this application provides an artificial intelligence-based advertising performance evaluation system, including: The acquisition module is used to respond to evaluation requests for the effectiveness of internet advertising services by acquiring user interaction data and brand association data of the internet advertising services during the evaluation period. The processing module is used to extract short-term performance indicator features that characterize the direct conversion effect of the advertisement based on the user interaction behavior data, and then generate short-term evaluation labels for the advertising performance based on the Shapley value attribution model and the short-term performance indicator features. The processing module is also used to extract long-term value indicators from the brand association data to characterize users’ brand awareness and long-term relationship. The long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of natural traffic sessions, and the proportion of user interaction content. The execution module is used to construct a multilayer perceptron neural network as a fusion model. It takes the short-term evaluation label and the long-term value index as input, performs feature fusion and dynamic weight allocation through the nonlinear mapping layer in the fusion model, and outputs a comprehensive dynamic effect evaluation score through the output layer.

[0013] Thirdly, this application provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described artificial intelligence-based advertising effectiveness evaluation method.

[0014] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned artificial intelligence-based advertising effectiveness evaluation method.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: This application proposes an AI-based method for evaluating advertising performance, achieving integrated feature fusion modeling of short-term conversion effectiveness and long-term brand value, thereby significantly improving the overall accuracy of advertising evaluation. First, short-term performance indicators are extracted to characterize the direct conversion effect of advertising. Then, based on the Shapley value attribution model and these short-term performance indicators, short-term evaluation tags for advertising performance are generated. These tags quantify the actual contribution of each advertising touchpoint in the user conversion path, accurately reflecting the direct conversion efficiency of advertising and providing a refined decision-making basis for short-term optimization. Second, long-term value indicators are extracted from brand-related data to characterize user brand awareness and long-term relationships. These indicators capture users' continuous attention and interaction with the brand, assessing the potential value of advertising in long-term brand building and user relationship maintenance, and providing data support for cross-cycle strategy adjustments. Finally, a multilayer perceptron neural network is constructed as... The fusion model takes the short-term evaluation labels and long-term value indicators as inputs, performs feature fusion and dynamic weight allocation through a nonlinear mapping layer, and outputs a comprehensive dynamic effect evaluation score through an output layer. This enables synergistic quantitative evaluation of short-term conversion effects and long-term brand value, automatically balances the weights of different features in the overall effect, generates continuous and quantifiable evaluation results, and improves prediction accuracy and decision reliability. Overall, this application's solution solves the problem of insufficient evaluation accuracy in multi-touchpoint and multi-channel advertising environments through collaborative modeling of short-term and long-term features, nonlinear fusion, and dynamic weight allocation. It achieves comprehensive evaluation across time scales and dimensions, improving the scientific nature and accuracy of advertising optimization decisions. Attached Figure Description

[0016] Figure 1 This is a schematic diagram illustrating an application scenario of an AI-based advertising effectiveness evaluation method according to some embodiments of this application; Figure 2 This is an exemplary flowchart of an AI-based advertising performance evaluation method according to some embodiments of this application; Figure 3 This is a schematic diagram of the process for generating short-term evaluation labels according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of an AI-based advertising performance evaluation system according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device that implements an artificial intelligence-based advertising performance evaluation method according to some embodiments of this application. Detailed Implementation

[0017] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] refer to Figure 1 This figure is a schematic diagram of an application scenario for an AI-based advertising performance evaluation method according to some embodiments of this application. The figure includes an advertising data source, a server, a communication network, and a terminal. The advertising data source communicates with the server through the network, and the terminal connects to the server through the communication network. The server obtains user interaction behavior data and brand association data provided by the advertising data source, extracts short-term performance indicators from the data, generates short-term evaluation labels based on Shapley values, extracts long-term value indicators, and outputs dynamic performance evaluation scores through a pre-trained neural network fusion model. When the server receives a performance evaluation request for a specific advertising campaign sent through the terminal, it feeds back the dynamic performance evaluation score and association analysis results to the terminal for advertising decision-makers to view and refer to.

[0019] The advertising data source can be an advertising platform (such as a search engine marketing platform), website / application analytics tools, and social media data interfaces; the terminal can be, but is not limited to, a personal computer, a laptop, a tablet, or a dedicated management backend for advertisers; the server can be an independent physical server, a server cluster or distributed system consisting of multiple physical servers, or a cloud server that provides cloud computing services.

[0020] refer to Figure 2 The figure is an exemplary flowchart of an AI-based advertising performance evaluation method according to some embodiments of this application. The AI-based advertising performance evaluation method mainly includes the following steps: In step 101, in response to the evaluation request for the effectiveness of the internet advertising service, user interaction behavior data and brand association data of the internet advertising service during the evaluation period are obtained.

[0021] It should be noted that, in this application, responding to the evaluation request for the effectiveness of internet advertising services means that the entire evaluation process is initiated by specific triggering conditions. Specifically, the system receives active instructions from the advertiser's management backend through an application programming interface (API) or a scheduled task automatically generated according to a preset period, triggering an evaluation process. This request carries at least two parameters: the evaluation subject and the evaluation period. The system then initiates data calls to the advertising data platform, website analytics tools, and social media data warehouse based on these parameters to obtain corresponding user interaction behavior data and brand association data.

[0022] In its implementation, the system obtains user interaction data and brand-related data of the internet advertising service during the evaluation period by calling the application programming interface of the advertising service platform. It should be further noted that the user interaction data includes user interaction records such as exposure, clicks, dwell time, bounce rate, and conversion during the advertising period. The brand-related data includes brand-related search volume, organic traffic session data, and interactive statistics of user-generated content across platforms. The evaluation period refers to a pre-set, continuous data statistical time period with a clear start and end time, used to define the running time range and data collection boundaries of the advertising campaign targeted in this performance evaluation.

[0023] In step 102, based on the user interaction behavior data, short-term performance indicator features for characterizing the direct conversion effect of the advertisement are extracted, and then short-term evaluation labels for the campaign performance are generated based on the Shapley value attribution model and the short-term performance indicator features.

[0024] In some embodiments, extracting short-term performance metrics to characterize the direct conversion effect of advertising based on the user interaction behavior data can be achieved through the following steps: Based on user interaction behavior data, identify advertising touchpoints related to ad delivery during the evaluation period; Calculate the ad content reach rate and user exposure frequency for each ad touchpoint; Statistics on the number of users who ultimately converted through each advertising touchpoint and the length of the conversion path; Based on the conversion behavior, calculate the click-through rate and conversion cost for each ad touchpoint; The ad content reach rate, user exposure frequency, number of users who converted, conversion path length, click pass rate, and conversion cost are integrated into a vector, and the vector is used as a short-term performance indicator to characterize the direct conversion effect of the ad.

[0025] It should be noted that, in this application, an advertising touchpoint is a unique identifier used to identify a specific channel or scenario in which a user comes into contact with advertising content; the user exposure frequency is a statistical value used to quantify the average number of times a single user views an advertisement on the same advertising touchpoint; the conversion path length is a value used to measure the number of intermediate touchpoints experienced from the user's first advertising touchpoint to the final completion of the conversion behavior; the conversion cost is an indicator used to reflect the average cost invested in obtaining an effective conversion behavior on each advertising touchpoint; and the short-term effect indicator features are multi-dimensional feature vectors used to comprehensively characterize the direct conversion effect obtained by the advertisement within the evaluation period.

[0026] In practice, firstly, based on the user interaction behavior data, the system parses detailed logs containing timestamps, user anonymity identifiers, event types, and channel sources. By identifying records in the logs where the event type is ad display or ad click, and extracting their associated channel source and ad content identifier fields, each unique combination of channel source and ad content identifier is defined and identified as an independent ad touchpoint. Secondly, for each identified ad touchpoint, the system calculates its ad content reach rate and user exposure frequency: the ad content reach rate is calculated by counting the number of unique users who have at least one exposure record at that touchpoint. The number is then divided by the total target audience set for the advertising campaign within the evaluation period. User exposure frequency is obtained by counting the total number of exposure records occurring at that touchpoint and dividing that total number of exposure records by the number of unique users who had at least one exposure record at that touchpoint. Furthermore, the system counts the number of users who ultimately converted through each advertising touchpoint and the length of the conversion path: specifically, the system filters all log records with the event type of conversion and traces back to all ad display and ad click event sequences associated with the same anonymous user identifier to form a complete conversion path. For each advertising touchpoint appearing in the path... The system accumulates the number of all conversion paths that use the touchpoint as any link in the path (including the beginning, middle, or end), obtaining the number of users who ultimately converted via that touchpoint. Simultaneously, the system calculates the number of different ad touchpoints appearing in each conversion path, recording this value as the conversion path length. Then, based on the conversion behavior, the system calculates the click-through rate (CTR) and conversion cost for each ad touchpoint: the CTR is obtained by dividing the total number of ad clicks at that touchpoint by the total number of ad impressions; the conversion cost is obtained by querying the ad delivery management backend to obtain the total cost of conversion for that touchpoint during the evaluation period. The total advertising cost is calculated by dividing it by the number of users who ultimately converted through the touchpoint, as obtained from the above statistics. Finally, six values ​​for each advertising touchpoint—ad content reach rate, user exposure frequency, number of users who ultimately converted through the touchpoint, average length of all conversion paths related to the touchpoint, click-through rate, and conversion cost—are arranged in a preset order. Each value is then normalized to eliminate the influence of dimensions, thereby integrating them into a fixed-dimensional numerical vector. This vector is used as a short-term performance indicator to characterize the direct conversion effect of the advertisement.

[0027] In some embodiments, reference Figure 3 As shown in the figure, this is a schematic diagram of the process for generating short-term evaluation labels in some embodiments of this application. In this embodiment, the generation of short-term evaluation labels for the deployment effect based on the Shapley value attribution model and the short-term effect indicator features can be achieved by the following steps: In step 1021, each complete conversion path is defined as a cooperative game, where each advertising touchpoint is regarded as a game participant; In step 1022, based on the Monte Carlo sampling method, the approximate Shapley value contribution of each advertising touchpoint to the occurrence of the conversion in the conversion path is calculated; In step 1023, the contribution of the Shapley value of each advertising touchpoint is weighted and summed with the corresponding conversion cost in the short-term effect indicator features to generate a comprehensive conversion efficiency value; In step 124, the comprehensive conversion efficiency value is mapped to discrete short-term evaluation labels according to a preset threshold range.

[0028] It should be noted that the game participants in this application are abstract entities representing advertising touchpoints and participating in contribution allocation in a cooperative game model; the approximate Shapley value contribution is an estimate of the magnitude of each advertising touchpoint's role in driving a conversion in a single conversion path, based on cooperative game theory and sampling approximation methods; the comprehensive conversion efficiency value is a scalar value used to integrate the Shapley value contribution of advertising touchpoints with their conversion cost information to comprehensively evaluate the overall conversion efficiency of the advertising campaign within the evaluation period; and the short-term evaluation label is a qualitative evaluation identifier used to intuitively and discretely classify the short-term effects of the advertising campaign based on the threshold range of the comprehensive conversion efficiency value.

[0029] In practical implementation, firstly, each complete conversion path is defined as a cooperative game, where each ad touchpoint is considered a player. Specifically, the system treats each user path from initial exposure to final conversion as an independent game instance, with all different ad touchpoints in the path constituting the set of players for this game. The value of this conversion is defined as the total payoff of this game. Secondly, based on the Monte Carlo sampling method, the approximate Shapley value contribution of each ad touchpoint to the conversion in the conversion path is calculated. Specifically, for a conversion path containing n ad touchpoints... The conversion path is calculated by the system through the following steps: (a) Define a feature function whose input is a subset of touchpoints and whose output is the conversion value of the subset in the context of this specific path. In a preferred embodiment, the feature function is defined as follows: if the current subset contains the last touchpoint on the path that led to the conversion, the value is 1; otherwise, it is 0. (b) Perform K rounds of Monte Carlo sampling. In each round of sampling, a random arrangement of the n touchpoints is generated to simulate the process of touchpoints joining the alliance sequentially. (c) For each ad touchpoint, the value is calculated according to the arrangement of the current round. First, the value change of the touchpoint before and after joining the alliance is calculated using the aforementioned feature function to obtain its marginal contribution in this round of sampling; (d) after completing K rounds of sampling, the marginal contribution of each touchpoint in all rounds is averaged to obtain its approximate Shapley value contribution on this path. The sampling rounds K are a preset integer greater than zero, such as 1000 or 10000 rounds, to ensure the accuracy of the approximate calculation; further, the calculation results of all conversion paths within the evaluation period are aggregated. For each ad touchpoint, the approximate Shapley value calculated on all relevant conversion paths is used to calculate its contribution. The Shapley value contribution is weighted and averaged, with the weights being the conversion value of each path, ultimately yielding the overall Shapley value contribution of that touchpoint within the evaluation period. Then, the Shapley value contribution of each ad touchpoint is weighted and summed with the corresponding conversion cost from the short-term performance indicator features to generate a comprehensive conversion efficiency value. Specifically, the overall Shapley value contribution and corresponding conversion cost of each ad touchpoint are read, and a predefined weighted summation formula is applied for calculation. This formula can be: Comprehensive Conversion Efficiency Value = Σ(α * Shapley Value Contribution_i) - β*conversion cost_i), where the summation iterates through all ad touchpoints. α and β are pre-set positive weighting coefficients used to balance the importance of contribution and cost in the overall evaluation. The calculated result is the comprehensive conversion efficiency value representing the overall conversion efficiency of this advertising campaign. It should be further noted that the specific setting method of the positive weighting coefficients α and β can be determined based on historical experience of business objectives or optimization experiments: a simple implementation method is for experts to pre-set a set of empirical values ​​based on the core objectives of the current advertising campaign. For example, when focusing on measuring conversion contribution, α=0.7 and β=0 are set.3. When focusing on cost efficiency, α=0.4 and β=0.6 are set. Another more systematic approach is to collect a certain amount of historical advertising campaign data, whose overall conversion efficiency value has been manually scored by experts based on actual results. Further, using this manual score as the target, and minimizing the error between the calculated result of the weighted summation formula and the target, a set of optimal α and β values ​​is obtained through linear regression or grid search methods. In actual system deployment, this set of coefficients is determined as fixed parameters and embedded in the weighted summation formula. Finally, based on a preset threshold range, the overall conversion efficiency is... The overall conversion efficiency value is mapped to discrete short-term evaluation labels. Specifically, a set of ordered numerical threshold intervals is pre-configured. For example, threshold points 0.3 and 0.7 divide the range of the overall conversion efficiency value into three sub-intervals: [0, 0.3), [0.3, 0.7), and [0.7, 1.0]. Each sub-interval is assigned a discrete numerical level label, corresponding to level 1, level 2, and level 3, respectively. The system compares the calculated overall conversion efficiency value with these threshold intervals to determine its corresponding interval and uses the numerical level label corresponding to that interval as the short-term evaluation label for this evaluation.

[0030] In step 103, long-term value indicators that characterize user brand awareness and long-term relationships are extracted from the brand association data. The long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content.

[0031] In some embodiments, extracting long-term value indicators characterizing user brand awareness and long-term relationships from the brand association data can be achieved through the following steps: Based on the brand association data, the relative change rate of brand search volume between the current evaluation period and the previous evaluation period is extracted, and the relative change rate is used as the month-on-month growth rate of brand search volume. Determine the growth rate of the number of natural traffic sessions in the current evaluation period relative to a preset baseline value, and use the growth rate as the incremental ratio of natural traffic sessions; Monitor user-generated content related to the brand on preset social media platforms during the current evaluation period, count the number of posts and positive interaction data identified by sentiment analysis, and calculate the proportion of user interaction content. The month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content were standardized, and the standardized results were combined to construct a long-term value indicator.

[0032] It should be noted that the month-on-month growth rate in this application is a rate indicator used to dynamically characterize the changing trend of a brand's active search attention within adjacent evaluation periods; the incremental ratio is a proportional indicator used to quantify the degree of free growth of organic traffic relative to historical baseline levels within the evaluation period; the user interaction content ratio is an influence indicator used to reflect the proportion of brand-related content spontaneously generated by users with positive emotional tendencies in public social contexts to the total brand-related voice volume; and the long-term value indicator is a multi-dimensional indicator used to comprehensively quantify and evaluate the depth of user brand awareness and the health of long-term relationships.

[0033] In practice, firstly, the system obtains the total number of search queries related to the brand's core keywords and main derivative terms within the current evaluation period (T) and the previous evaluation period (T-1) from the search engine data interface. If the total search volume_T-1 is greater than zero, the relative change rate of the search volume is calculated using the formula: (Total Search Volume_T - Total Search Volume_T-1) / Total Search Volume_T-1. If the total search volume_T-1 is equal to zero, the relative change rate is set to a preset default value, such as 0, and this calculation result is used as the month-on-month growth rate of the brand's search volume. Secondly, the preset baseline value is the average number of organic traffic sessions over the past N historical periods, where N is a preset integer greater than 1, such as N=4. The system extracts the total number of unique sessions originating from organic search and direct access within the current evaluation period from the website analytics tool, and calculates the relative change rate using the formula: (Current Period Organic Traffic Sessions - Preset Baseline Value) / The growth rate is calculated based on a preset baseline value and used as the incremental proportion of organic traffic sessions. Then, through configured data collection rules, original posts, videos, and comments containing brand official account mentions, brand keywords, or related hashtags are crawled from the public application programming interfaces of a preset set of social media platforms or through authorized data streams. These are identified as user-generated content, and a pre-trained sentiment analysis model is used to determine the sentiment orientation of the content. The total number of user-generated content posts judged as positive sentiment is counted, and the total number of likes, comments, reposts, and other interactive behaviors received is summarized. The proportion of user-generated content interaction is calculated using the formula: Total number of positive sentiment user-generated content interactions / (Total number of positive sentiment user-generated content interactions + ... The total number of interactions with official brand content is calculated, which is synchronously obtained from the brand's official account data. Finally, the system performs Z-score standardization on the calculated month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content. That is, each indicator value is subtracted from the mean of the indicator on the historical dataset, and then divided by its standard deviation. The three standardized values ​​are combined into a multi-dimensional vector in a predetermined order, and this vector is used as a long-term value indicator to represent users' brand awareness and long-term relationship.

[0034] In step 104, a multilayer perceptron neural network is constructed as a fusion model. The short-term evaluation label and the long-term value index are used as inputs. Feature fusion and dynamic weight allocation are performed through the nonlinear mapping layer in the fusion model, and a comprehensive dynamic effect evaluation score is output through the output layer.

[0035] In some embodiments, constructing a multilayer perceptron neural network as a fusion model can be achieved by the following steps: Determine the input layer dimension of the fusion model to match the total dimension of the short-term evaluation label and the long-term value indicator; At least one fully connected hidden layer is sequentially set as a nonlinear mapping layer after the input layer of the fusion model, and a Sigmoid or ReLU activation function is selected for each hidden layer; The output layer is connected after the last hidden layer, configured as a single neuron node, and the Sigmoid activation function is selected to limit the output value range to between 0 and 1. Initialize the connection weights and bias terms of all neurons in the input layer, hidden layer, and output layer; The fusion model is trained in a supervised manner using short-term evaluation tags, long-term value indicators, and corresponding labeling effect scores from historical advertising campaigns.

[0036] It should be noted that the nonlinear mapping layer in this application is used to perform nonlinear transformation on the input features to learn the complex relationships between them and to provide a basic neural network layer for subsequent dynamic weight allocation; a single neuron node is used to aggregate and map the complex feature representation finally learned by the fusion model into a single, continuous prediction value output unit; the connection weights and bias terms are learnable model parameters used to adjust the signal transmission strength and activation threshold between each neuron node during the forward propagation of the neural network.

[0037] In specific implementation, firstly, the short-term evaluation label is converted into a fixed-length numerical vector through one-hot encoding, and the long-term value index is directly used as the numerical input. The lengths of these two vectors are then added together, and the sum is used to determine the number of neurons required for the input layer, ensuring that the input layer can fully receive all features. Secondly, the number of hidden layers and the number of neurons in each layer can be set according to the problem complexity. For example, two hidden layers can be set, with 64 nodes in the first layer and 32 nodes in the second layer. Each neuron in each hidden layer is connected to all nodes in the previous layer (fully connected), and the ReLU function is used as its activation function to introduce non-linearity and alleviate the gradient vanishing problem. Then, an output layer is connected after the last hidden layer. The output layer is configured as a single neuron, and the Sigmoid activation function is selected to limit the output value range to between 0 and 1. This output layer node receives the output of all nodes in the last hidden layer and converts it into a value between 0 and 1 using the Sigmoid function. This value represents the dynamic effect evaluation score. Finally, the connection weights and biases of all neurons in the input, hidden, and output layers are initialized. Here, a method called Xavier initialization is typically used. Based on the number of input and output neurons in each layer, values ​​are randomly sampled from a standard normal distribution with a mean of 0 and a variance calculated using a specific method, to initially assign values ​​to all weights and biases. Finally, short-term evaluation tags, long-term value indicators, and corresponding labeling performance scores from historical advertising campaigns are used as the training sample set to perform supervised training on the fusion model. The specific training process is as follows: the data sample set is divided into a training set and a validation set proportionally, and training parameters, such as the learning rate (e.g., 0.001), are set. The batch size is set to 32, and the number of training epochs is set to 100. Mean squared error is used as the loss function to measure the difference between the model's predicted value and the actual labeled score. An adaptive moment estimation optimizer is used to calculate the gradient of the loss function with respect to all weights and biases of the model through backpropagation. These parameters are then iteratively updated based on the gradient direction and learning rate. After each training epoch, the model performance is evaluated using a validation set. Training stops when the validation set loss no longer decreases significantly for several consecutive epochs, and the model at this parameter state is saved as the final fusion model that has been trained and can be used for performance evaluation.

[0038] In some embodiments, taking the short-term evaluation label and the long-term value index as input, performing feature fusion and dynamic weight allocation through the nonlinear mapping layer in the fusion model, and outputting a comprehensive dynamic performance evaluation score through the output layer can be achieved through the following steps: The short-term evaluation labels and the long-term value indicators are normalized and then input into the input layer of the fusion model. The input features are nonlinearly transformed by at least one fully connected hidden layer in the fusion model, wherein the neuron nodes of each layer dynamically weight and combine the input according to their connection weights. In the last hidden layer, the fused features, after undergoing multi-level nonlinear transformation and dynamic weight allocation, are passed to a single neuron node in the output layer. In the output layer of the fusion model, the neuron node calculates and outputs an output value between 0 and 1 based on its input, connection weights, bias terms and activation function, and then uses this output value as the dynamic effect evaluation score.

[0039] It should be noted that the fusion feature in this application is a high-level feature representation that is generated after being abstracted and integrated through multiple hidden layers of a neural network, and contains a complex relationship between short-term effects and long-term value; the dynamic effect evaluation score is a continuous numerical result that is calculated and output by the fusion model to comprehensively and quantitatively reflect the overall effect of a specific advertising campaign within the evaluation period.

[0040] In specific implementation, firstly, the short-term evaluation labels and the long-term value indicators are normalized and input into the input layer of the fusion model. Specifically, the discrete short-term evaluation labels are first converted into numerical form, for example, levels 1, 2, and 3 are mapped to 0.2, 0.5, and 0.8, respectively. Then, each value in the long-term value indicator is standardized by subtracting its historical mean and dividing by its historical standard deviation, so that the numerical ranges of all input features are within a similar order of magnitude. Finally, these processed values ​​are arranged into a one-dimensional feature vector in a predetermined order, and this vector is input into the fusion model as a whole. The input layer receives a value for one corresponding dimension of the vector from each neuron node. Next, at least one fully connected hidden layer in the fusion model performs a non-linear transformation on the input features. Specifically, each neuron node dynamically weights and combines the input based on its connection weights. The output of the input layer serves as the input to the first hidden layer. Each neuron node in the first hidden layer multiplies all received input values ​​by their corresponding connection weights, which have been learned and determined during model training. The summation of all products is then performed with a similarly learned bias term. Finally, the summation is passed through its activation... The liveness function (such as the ReLU function) undergoes a nonlinear transformation, and the output is used as the input to the next hidden layer. This process, known as dynamic weight allocation and nonlinear transformation, enables the model to automatically learn and fuse complex patterns between short-term and long-term features. Then, in the last hidden layer, the fused features, after multiple levels of nonlinear transformation and dynamic weight allocation, are passed to a single neuron node in the output layer. The output values ​​of all neurons in the last hidden layer constitute the final high-level fused feature vector, which is then passed to the single neuron node in the output layer. Finally, in the output layer of the fused model, the neurons... The node calculates and outputs an output value between 0 and 1 based on its input, connection weights, bias term, and activation function. This output value is then used as the dynamic effect evaluation score. Specifically, the output layer neuron multiplies all the input values ​​it receives by the corresponding pre-trained connection weights and sums them, then adds a pre-trained bias term. Finally, the summation result is mapped through the Sigmoid activation function, which compresses the value to the range of 0 to 1. The system directly uses this final output value as the dynamic effect evaluation score for this advertising campaign.

[0041] It should also be noted that the proposed solution constructs a multilayer perceptron neural network, takes short-term evaluation labels and long-term value indicators as inputs, and uses a nonlinear mapping layer for feature fusion and dynamic weight allocation. This enables the capture of the complex nonlinear relationship between short-term behavioral features and long-term value indicators, achieving a comprehensive quantitative evaluation of effects across time scales. The dynamic effect evaluation score generated by the output layer not only provides continuous and quantifiable effect evaluation results, but also facilitates direct comparison and optimization with business objectives.

[0042] On the other hand, in some embodiments, this application provides an artificial intelligence-based advertising performance evaluation system, for reference... Figure 4 The figure is a schematic diagram of the structure of an AI-based advertising performance evaluation system according to some embodiments of this application. The AI-based advertising performance evaluation system 400 includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to respond to the evaluation request of the effectiveness of the Internet advertising service and to acquire user interaction behavior data and brand association data of the Internet advertising service during the evaluation period. Processing module 402, in this application, is used to extract short-term effect indicator features that characterize the direct conversion effect of the advertisement based on the user interaction behavior data, and then generate short-term evaluation labels for the advertising effect based on the Shapley value attribution model and the short-term effect indicator features. In this application, the processing module 402 is also used to extract long-term value indicators from the brand association data to characterize users’ brand awareness and long-term relationship. The long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of natural traffic sessions, and the proportion of user interaction content. The execution module 403 in this application is mainly used to construct a multilayer perceptron neural network as a fusion model, take the short-term evaluation label and the long-term value index as input, perform feature fusion and dynamic weight allocation through the nonlinear mapping layer in the fusion model, and output a comprehensive dynamic effect evaluation score through the output layer.

[0043] In addition, this application also provides a computer device, which includes a memory and a processor. The memory stores code, and the processor is configured to acquire the code and execute the above-described artificial intelligence-based advertising effectiveness evaluation method.

[0044] In some embodiments, reference Figure 5The figure is a schematic diagram of the structure of a computer device implementing an AI-based advertising performance evaluation method according to some embodiments of this application. The AI-based advertising performance evaluation method in the above embodiments can... Figure 5 The computer device 500 shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0045] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0046] The communication bus 502 can be used to transmit information between the aforementioned components.

[0047] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0048] The memory 503 stores program code for executing the solution of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. In the above embodiments, the artificial intelligence-based advertising effectiveness evaluation method can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0049] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0050] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0051] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0052] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-mentioned artificial intelligence-based advertising effectiveness evaluation method.

[0053] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0054] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for evaluating the effectiveness of advertising placement based on artificial intelligence, characterized in that, include: In response to requests for evaluation of the effectiveness of internet advertising services, obtain user interaction data and brand association data of internet advertising services during the evaluation period; Based on the user interaction behavior data, short-term performance indicator features for characterizing the direct conversion effect of the advertisement are extracted, and then short-term evaluation labels for the campaign performance are generated based on the Shapley value attribution model and the short-term performance indicator features. Specifically, the short-term evaluation labels for the deployment effect generated based on the Shapley value attribution model and the aforementioned short-term effect indicator features include: The conversion cost corresponding to each advertising touchpoint in the short-term effect indicator features is normalized to obtain the normalized conversion cost corresponding to each advertising touchpoint. Each complete conversion path from first exposure to final conversion obtained based on the user interaction behavior data is defined as an independent cooperative game, where each advertising touchpoint is regarded as a game participant. Based on the Monte Carlo sampling method, the approximate Shapley value contribution of each ad touchpoint to the conversion in the conversion path is calculated, and the approximate Shapley value contribution of each ad touchpoint on each relevant conversion path within the evaluation period is weighted and averaged to obtain the overall Shapley value contribution of the ad touchpoint. Based on pre-set positive weighting coefficients, the weighted difference between the overall Shapley value contribution of each ad touchpoint and the corresponding normalized conversion cost is summed, and the comprehensive conversion efficiency value is generated according to the following formula: Comprehensive conversion efficiency value = Σ_i (α × overall Shapley value contribution_i - β × normalized conversion cost_i); where i represents the ad touchpoint, Σ_i represents the summation over all ad touchpoints, and α and β are pre-set positive weighting coefficients; Based on a preset threshold range, the comprehensive conversion efficiency value is mapped to discrete short-term evaluation labels; Long-term value indicators used to characterize user brand awareness and long-term relationships are extracted from the brand association data. These long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content. Specifically, the long-term value indicators extracted from the brand association data to characterize user brand awareness and long-term relationships include: Based on the brand association data, the relative change rate of brand search volume between the current evaluation period and the previous evaluation period is extracted, and the relative change rate is used as the month-on-month growth rate of brand search volume. Determine the growth rate of the number of natural traffic sessions in the current evaluation period relative to a preset baseline value, and use the growth rate as the incremental ratio of natural traffic sessions; Monitor user-generated content related to the brand within the preset social media platform set during the current evaluation period, count the number of posts and positive interaction data identified by sentiment analysis, and calculate the proportion of user interaction content; The month-on-month growth rate of brand search volume, the incremental proportion of organic traffic sessions, and the proportion of user interaction content are standardized respectively, and the standardized results are combined to construct the long-term value indicators. The discrete short-term evaluation labels are converted into numerical forms, and the numericalized short-term evaluation labels and the standardized results that make up the long-term value index are arranged in a predetermined order to form a one-dimensional input feature vector. A multilayer perceptron neural network is constructed as a fusion model. The one-dimensional input feature vector is input into the fusion model, and feature fusion and dynamic weight allocation are performed through the nonlinear mapping layer in the fusion model. Finally, a comprehensive dynamic performance evaluation score is output through the output layer.

2. The method as described in claim 1, characterized in that, Based on the user interaction behavior data, the following short-term performance metrics are extracted to characterize the direct conversion effect of the advertisement: Based on user interaction behavior data, identify advertising touchpoints related to ad delivery during the evaluation period; Calculate the ad content reach rate and user exposure frequency for each ad touchpoint; Statistics on the number of users who ultimately converted through each advertising touchpoint and the length of the conversion path; Based on the conversion behavior, calculate the click-through rate and conversion cost for each ad touchpoint; The ad content reach rate, user exposure frequency, number of users who converted, conversion path length, click pass rate, and conversion cost are integrated into a vector, and the vector is used as a short-term performance indicator to characterize the direct conversion effect of the ad.

3. The method as described in claim 1, characterized in that, Constructing a multilayer perceptron neural network as a fusion model specifically includes: Determine the input layer dimension of the fusion model to match the total dimension of the short-term evaluation label and the long-term value indicator; At least one fully connected hidden layer is sequentially set as a nonlinear mapping layer after the input layer of the fusion model, and a Sigmoid or ReLU activation function is selected for each hidden layer; The output layer is connected after the last hidden layer, configured as a single neuron node, and the Sigmoid activation function is selected to limit the output value range to between 0 and 1. Initialize the connection weights and bias terms of all neurons in the input layer, hidden layer, and output layer; The fusion model is trained in a supervised manner using short-term evaluation tags, long-term value indicators, and corresponding labeling effect scores from historical advertising campaigns.

4. The method as described in claim 1, characterized in that, The user interaction data includes user interaction records such as exposure, clicks, dwell time, bounce rate, and conversions during the advertising campaign period.

5. The method as described in claim 1, characterized in that, The brand association data includes brand-related search volume, organic traffic session data, and interactive statistics of user-generated content across platforms.

6. An artificial intelligence-based advertising effectiveness evaluation system, the system being configured to perform the method according to any one of claims 1 to 5 to evaluate advertising effectiveness, characterized in that, The system includes: The acquisition module is used to respond to evaluation requests for the effectiveness of internet advertising services by acquiring user interaction data and brand association data of the internet advertising services during the evaluation period. The processing module is used to extract short-term performance indicator features that characterize the direct conversion effect of the advertisement based on the user interaction behavior data, and then generate short-term evaluation labels for the advertising performance based on the Shapley value attribution model and the short-term performance indicator features. The processing module is also used to extract long-term value indicators from the brand association data to characterize users’ brand awareness and long-term relationship. The long-term value indicators include at least the month-on-month growth rate of brand search volume, the incremental proportion of natural traffic sessions, and the proportion of user interaction content. The execution module is used to construct a multilayer perceptron neural network as a fusion model. It takes the short-term evaluation label and the long-term value index as input, performs feature fusion and dynamic weight allocation through the nonlinear mapping layer in the fusion model, and outputs a comprehensive dynamic effect evaluation score through the output layer.

7. A computer device comprising a memory and a processor, the memory storing code, characterized in that, The processor is configured to acquire the code and execute the AI-based advertising performance evaluation method as described in any one of claims 1 to 5.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the AI-based advertising performance evaluation method as described in any one of claims 1 to 5.

Images