Advertisement bidding method based on OCPX and related device

Abstract

The invention provides an advertisement bidding method based on OCPX and a related device. According to the advertisement bidding method based on the OCPX, through dynamic bidding adjustment driven by real-time data and high-precision advertisement conversion prediction, intelligence and high efficiency of advertisement putting are achieved, and the ROI and the overall effect of advertisement putting are remarkably improved.

Description

Technical Field

[0001] The present application relates to the technical field of Internet and e-commerce, and in particular to an advertisement bidding method and related devices based on OCPX. Background Art

[0002] Ad bidding is crucial in digital advertising, and it is directly related to whether advertisers can achieve their business goals at the best cost, such as acquiring new users, increasing sales, or enhancing brand awareness. A reasonable bidding strategy can ensure that ads are displayed to users who are most likely to take action at the most appropriate time, thereby maximizing advertising effects and improving return on investment (ROI). Efficient bidding can not only help advertisers stand out in the fierce market competition, but also reduce unnecessary advertising expenditures and resource waste by accurately targeting potential customer groups. However, existing advertising bidding strategies have certain drawbacks. Summary of the invention

[0003] In view of this, an embodiment of the present application provides an OCPX-based advertising bidding method and related devices.

[0004] In a first aspect, an embodiment of the present application provides an advertisement bidding method based on OCPX, the method comprising:

[0005] Acquire the advertisement conversion target and target cost of the advertiser; the advertisement conversion target refers to the conversion effect that the advertiser expects the target advertisement to achieve after the advertiser places the target advertisement; the target cost is the budget cost of the advertiser for placing the target advertisement;

[0006] Acquire the delivery data of the target advertisement in real time, and extract features based on the delivery data to obtain a feature vector for bid prediction;

[0007] Input the feature vector into an advertisement conversion prediction model to obtain a prediction probability; the prediction probability is the probability of reaching the advertisement conversion target after the target advertisement is released;

[0008] Based on the price adjustment algorithm, the basic bid is adjusted according to the predicted probability to obtain an advertisement delivery bid; the basic bid is the initial bid set for each click of the target advertisement; the advertisement delivery bid is the bid for a single click after the target advertisement is delivered;

[0009] The calculation formula of the basic bid is:

[0010] BaseBid=CPA target ×CR avg

[0011] Among them, BaseBid is the base bid, CPA targetCR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement;

[0012] The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm;

[0013] The linear price adjustment algorithm is:

[0014] Bid adjusted =BaseBid×CP

[0015] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability;

[0016] The piecewise linear adjustment algorithm is:

[0017]

[0018] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold;

[0019] The target advertisement delivery bid is made multiple times according to the advertisement delivery bid; the advertisement delivery bid is adjusted in real time during the process of delivering the target advertisement multiple times; the total advertisement delivery cost is less than or equal to the target cost; the total advertisement delivery cost is the sum of the advertisement delivery prices of each delivery of the target advertisement;

[0020] The training process of the advertising conversion prediction model is as follows:

[0021] Acquire multiple groups of historical data on advertising delivery and advertising conversion data corresponding to each group of historical data on advertising delivery; the historical data on advertising delivery includes advertising delivery data, user behavior data, competitive product data and market environment data; the advertising delivery data includes the content of the advertisement to be delivered, the advertising triggering keywords and the advertising display location; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of advertisements of the same type as the delivered advertisements; the market environment data includes data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of advertisements;

[0022] A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer;

[0023] Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing;

[0024] The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

[0025] In a possible implementation, the method further includes:

[0026] Use a generative adversarial network GAN or a SMOTE algorithm to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data.

[0027] In a possible implementation, inputting the feature vector into an advertisement conversion prediction model to obtain a prediction probability includes:

[0028] Combining a plurality of the feature vectors to generate an interactive feature, and introducing a time series into the interactive feature to obtain a target feature vector;

[0029] The target feature vector is input into the advertisement conversion prediction model.

[0030] In a possible implementation, the method further includes:

[0031] Collect user data; the user data includes historical behavior data, demographic data and interest preference data;

[0032] Processing the user data, extracting user features for building user portraits, and dividing the users into multiple groups based on a clustering algorithm;

[0033] Further segmenting the users in each group according to the user characteristics to obtain multiple segmented groups under each group;

[0034] Add feature labels to multiple segments;

[0035] The target advertisement is delivered to users in the segmented group having the same feature label as the target advertisement.

[0036] In a second aspect, an embodiment of the present application provides an OCPX-based advertising bidding device, the device comprising:

[0037] The acquisition module is used to acquire the advertisement conversion target and target cost of the advertiser; the advertisement conversion target refers to the conversion effect that the advertiser expects the target advertisement to achieve after the target advertisement is released; the target cost is the budget cost of the advertiser for releasing the target advertisement; and is used to acquire the release data of the target advertisement in real time, and perform feature extraction based on the release data to obtain a feature vector for bid prediction;

[0038] A prediction module is used to input the feature vector into an advertisement conversion prediction model to obtain a prediction probability; the prediction probability is the probability of the target advertisement reaching the advertisement conversion target after delivery; and is used to adjust the basic bid according to the prediction probability based on a price adjustment algorithm to obtain an advertisement delivery bid; the basic bid is the initial bid set for each click of the target advertisement; the advertisement delivery bid is the bid for a single click of the target advertisement after delivery;

[0039] The calculation formula of the basic bid is:

[0040] BaseBid=CPA target ×CR avg

[0041] Among them, BaseBid is the base bid, CPA target CR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement;

[0042] The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm;

[0043] The linear price adjustment algorithm is:

[0044] Bid adjusted =BaseBid×CP

[0045] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability;

[0046] The piecewise linear adjustment algorithm is:

[0047]

[0048] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold;

[0049] A bidding module is used to make multiple bids for the target advertisement placement according to the advertisement placement bid; the advertisement placement bid is adjusted in real time during the process of placing the target advertisement multiple times; the total advertisement placement cost is less than or equal to the target cost; the total advertisement placement cost is the sum of the advertisement placement prices of each placement of the target advertisement;

[0050] A training module is used to obtain multiple groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data; the advertising delivery history data includes delivery advertising data, user behavior data, competitive product data and market environment data; the delivery advertising data includes the content of the advertisement to be delivered, the advertisement triggering keywords and the advertisement display location; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of the advertisement of the same type as the delivered advertisement; the market environment data includes the data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of the advertisement;

[0051] A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer;

[0052] Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing;

[0053] The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

[0054] In a possible implementation manner, the device further includes:

[0055] The optimization module is used to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data by using a generative adversarial network (GAN) or a SMOTE algorithm.

[0056] In a possible implementation, the prediction module is used to combine a plurality of the feature vectors to generate an interactive feature, and introduce a time series into the interactive feature to obtain a target feature vector; and input the target feature vector into the advertising conversion prediction model.

[0057] In a possible implementation manner, the device further includes:

[0058] A delivery module is used to collect user data; the user data includes historical behavior data, demographic data and interest preference data; process the user data, extract user features for building user portraits, and divide users into multiple groups based on a clustering algorithm; further segment users in each group according to the user features to obtain multiple sub-groups under each group; add feature labels to multiple sub-groups; and deliver the target advertisement to users in the sub-groups that have the same feature labels as the target advertisement.

[0059] In a third aspect, an embodiment of the present application provides a device, comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the OCPX-based advertising bidding method described in any one of the first aspects above.

[0060] In a fourth aspect, an embodiment of the present application provides a computer storage medium, in which a code is stored. When the code is executed, the device executing the code implements the OCPX-based advertising bidding method described in any one of the first aspects above.

[0061] The OCPX-based advertising bidding method provided in this application realizes the intelligence and efficiency of advertising delivery through real-time data-driven dynamic bidding adjustment and high-precision advertising conversion prediction, which significantly improves the ROI and overall effect of advertising delivery. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0063] Figure 1 A flowchart of an OCPX-based advertising bidding method provided in an embodiment of the present application;

[0064] Figure 2 A schematic diagram of the structure of an OCPX-based advertising bidding device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0065] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0066] like Figure 1 As shown, Figure 1 1 is a flow chart of an OCPX-based advertising bidding method provided by an embodiment of the present invention. The method comprises:

[0067] Step 1: Get the advertiser's ad conversion target and target cost.

[0068] Ad conversion goals refer to the conversion effects that advertisers expect to be achieved after placing target ads. Ad conversion goals can also be understood as specific effects or behaviors that advertisers expect to achieve after placing target ads. Ad conversion goals include sales conversions (users complete purchases after clicking on ads), registration conversions (users complete account registration after clicking on ads), download conversions (users download apps or files after clicking on ads), form submissions (users fill out and submit forms after clicking on ads), or click conversions (users click on other links or buttons after clicking on ads).

[0069] The target cost is the budget cost of the advertiser to place the target advertisement. The target cost can also be understood as the budget limit set by the advertiser when placing the target advertisement. Advertisers usually set a fixed cost cap based on their own marketing budget and expected results to control the total cost of advertising.

[0070] Before placing an ad, advertisers need to clarify their ad conversion goals through the ad management platform. Advertisers can select preset conversion goal types (such as sales, registrations, downloads, etc.) and set specific conversion quantity goals (such as "100 registrations" or "50 sales"). When setting an ad plan, advertisers need to enter a target cost as a budget limit. The target cost can be set based on the advertiser's market expectations, historical data, or industry benchmarks.

[0071] The advertising conversion goals and target costs set by advertisers are transmitted to the OCPX system through the API or user interface of the advertising management platform. In the OCPX system, these data are stored as the initial parameters of the advertising delivery plan for use in subsequent steps (such as feature extraction and bid adjustment).

[0072] The OCPX system will conduct preliminary verification of the conversion goals and target costs set by advertisers to ensure that the goals are reasonable and executable. For example, the OCPX system will check whether the target cost is lower than the industry average (if so, it may remind the advertiser to adjust it) or whether the conversion goal is too aggressive (if so, it may recommend adjustment).

[0073] Step 2: Obtain the delivery data of the target advertisement in real time, and perform feature extraction based on the delivery data to obtain a feature vector for bid prediction.

[0074] Delivery data refers to all relevant data generated during the advertising process, including but not limited to ad impressions, clicks, user behavior, competitor product performance, and market environment. Delivery data is used to evaluate the actual effectiveness of advertising and serve as a basis for subsequent bid adjustments.

[0075] Feature extraction refers to extracting feature information useful for ad conversion prediction from the original delivery data. The purpose of feature extraction is to convert complex data into concise and meaningful feature vectors for subsequent machine learning models.

[0076] During the advertising process, the system will collect advertising data in real time. This data can be obtained through the API of the advertising management platform, data monitoring tools or third-party advertising monitoring platforms. The original delivery data is usually multi-source, heterogeneous and may contain noise, so data cleaning and preprocessing are required. For example, data cleaning: remove invalid or abnormal data (such as data with too high or too low click-through rate). Data normalization: standardize data of different magnitudes to the same numerical range (such as mapping click-through rate from 0-100% to 0-1). Feature selection: filter out the most useful features for advertising conversion prediction and remove redundant or irrelevant features. Based on the preprocessed delivery data, the OCPX system uses feature extraction algorithms to extract features that are useful for advertising conversion prediction. The purpose of feature extraction is to convert complex data into concise and meaningful feature vectors. After feature extraction is completed, the system combines the extracted features into a multi-dimensional feature vector. For example: [click-through rate, conversion rate, user age, user interest, ad placement, competitor conversion rate, time period traffic trend]. The generated feature vector will be used as the input of the subsequent advertising conversion prediction model.

[0077] The generated feature vectors will be stored in the OCPX system database for use in subsequent steps (such as ad conversion prediction models and price adjustment algorithms). Feature vectors can also be transmitted to other modules of the OCPX system through real-time stream processing technologies (such as Kafka and Flink).

[0078] Step 3: Input the feature vector into the advertising conversion prediction model to obtain the predicted probability.

[0079] The predicted probability is the probability of reaching the advertising conversion target after the target advertisement is released. The predicted probability refers to a value output by the advertising conversion prediction model, which indicates the probability of reaching the conversion target set by the advertiser after the target advertisement is released. This probability value is usually a real number between 0 and 1, where the larger the value, the higher the possibility of conversion.

[0080] The advertising conversion prediction model is a machine learning model that is used to predict the probability of achieving the conversion target set by the advertiser after the advertisement is released. By learning from the historical advertising data, the model can estimate the conversion possibility of the current advertisement based on its feature vector. Typical advertising conversion prediction models may include logistic regression, decision trees, random forests, neural networks, etc. These models can capture the relationship between advertising features and conversion results by learning patterns in historical data.

[0081] In one implementation, inputting the feature vector into an advertisement conversion prediction model to obtain a prediction probability includes:

[0082] Combining a plurality of the feature vectors to generate an interactive feature, and introducing a time series into the interactive feature to obtain a target feature vector;

[0083] The target feature vector is input into the advertisement conversion prediction model.

[0084] Feature vector combination refers to merging multiple independent feature vectors (i.e. features from different data sources or dimensions) to form a more comprehensive feature representation. For example, a feature vector represents the click-through rate of an ad, and another feature vector represents the user's interest preference. Combining these two feature vectors can generate a feature vector that contains more information.

[0085] Interactive features refer to new features generated by cross-combining or mathematically operating multiple features. Interactive features can capture the complex relationship between features, thereby improving the expressiveness of the model. For example, the intersection of user age (feature A) and ad type (feature B) can generate an interactive feature (age × ad type), which captures the preferences of users of different age groups for different ad types.

[0086] Time series information refers to data that changes over time during the advertising process, such as the trend of ad clicks over time, the time distribution of user behavior, etc. Introducing time series information allows the model to capture the dynamic changes of advertising delivery, thereby improving the accuracy of predictions.

[0087] The target feature vector refers to the feature vector used for advertising conversion prediction after combining features, generating interactive features, and introducing time series information based on the original feature vector. The target feature vector is usually more complex and richer than the original feature vector, and can better reflect the dynamics and complexity of advertising delivery.

[0088] Through this implementation, the feature vector is processed more finely, and feature interactions and time series information are introduced, so that the target feature vector can better reflect the dynamics and complexity of advertising. Ultimately, the ad conversion prediction model can output more accurate prediction probabilities based on these richer features, thus providing a more reliable basis for subsequent bid adjustments.

[0089] The training process of the advertising conversion prediction model is as follows:

[0090] Acquire multiple groups of historical data on advertising delivery and advertising conversion data corresponding to each group of historical data on advertising delivery; the historical data on advertising delivery includes advertising delivery data, user behavior data, competitive product data and market environment data; the advertising delivery data includes the content of the advertisement to be delivered, the advertising triggering keywords and the advertising display location; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of advertisements of the same type as the delivered advertisements; the market environment data includes data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of advertisements;

[0091] A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer;

[0092] Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing;

[0093] The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

[0094] In one embodiment, the method further includes: using a generative adversarial network (GAN) or a SMOTE algorithm to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data.

[0095] GAN is a deep learning model consisting of a generator and a discriminator. The task of the generator is to generate synthetic data similar to real data, while the discriminator determines whether the data is real or generated. Through adversarial training, the generator can generate high-quality synthetic data. SMOTE is an oversampling algorithm used to solve the problem of unbalanced data. It generates new synthetic samples by interpolating between minority class samples, thereby balancing the data distribution. By using the generative adversarial network (GAN) or SMOTE algorithm, more synthetic data can be generated based on existing advertising historical data and advertising conversion data, thereby expanding the training data set and improving the performance of the model.

[0096] Obtain multiple groups of advertising historical data and advertising conversion data corresponding to each group of advertising historical data. Perform feature standardization or normalization on these data. Input these data into the generator, which receives these data and randomly adds noise to obtain synthetic data. The task of the generator is to generate synthetic data similar to the real advertising historical data and advertising conversion data. The discriminator receives the real data and the synthetic data as input, and outputs a probability value, indicating the probability that the input data is the real data. The task of the discriminator is to distinguish between real data and generated data.

[0097] Step 4: Based on the price adjustment algorithm, the basic bid is adjusted according to the predicted probability to obtain an advertising bid.

[0098] The base bid is the initial bid set for each click on the target ad; the base bid refers to the amount you are willing to pay for each ad display (impression) or click on the ad exchange. The bid determines the competitiveness of the ad in the auction. The higher the bid, the higher the probability that the ad will be displayed. Dynamic pricing refers to a strategy that adjusts ad bids in real time based on predicted probabilities and the current market environment. The goal is to maximize the conversion effect of ads while controlling costs. The dynamic pricing strategy needs to be implemented within the budget to ensure that the overall cost of the advertising campaign is controllable. The ad placement bid is the bid for a single click after the target ad is placed.

[0099] Get the predicted probability output by the ad conversion prediction model. The predicted probability value indicates the possibility of achieving the conversion target set by the advertiser after the current ad is released.

[0100] During the advertising process, the performance of the advertisement (such as click-through rate, conversion rate, cost, etc.) is monitored in real time, and the bidding strategy is adjusted dynamically according to the actual effect. The bidding strategy is adjusted in real time according to the monitoring indicators. For example, if the conversion rate is lower than expected, the bid can be lowered to save budget; if the conversion rate is higher than expected, the bid can be increased to obtain more conversions.

[0101] In one implementation, the basic bid is calculated as follows:

[0102] BaseBid=CAP target ×CR avg

[0103] Among them, BaseBid is the base bid, CPS target CR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement;

[0104] The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm;

[0105] The linear price adjustment algorithm is:

[0106] Bid adjusted =BaseBid×CP

[0107] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability;

[0108] The piecewise linear adjustment algorithm is:

[0109]

[0110] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold.

[0111] Step 5: Make multiple target ad placement bids based on the ad placement bids.

[0112] The advertising delivery bid is adjusted in real time during the process of delivering the target advertisement multiple times; the total advertising delivery cost is less than or equal to the target cost; the total advertising delivery cost is the sum of the advertising delivery prices of each delivery of the target advertisement;

[0113] The advertising price refers to the amount that advertisers are willing to pay for each ad display (impression) or click in the advertising trading platform. This price determines the competitiveness of the advertisement in the auction. The advertising bid refers to the bid for a single target advertisement.

[0114] Adjust the advertising price in real time based on the results of the previous advertising and the market environment. The basis for adjustment includes conversion effect, competition and budget control. Use the adjusted advertising price in the next advertising and participate in the subsequent advertising auction. During the advertising process, monitor the total cost of advertising in real time to ensure that the total cost does not exceed the target cost. If it is close to the target cost, you can appropriately lower the bid and reduce the frequency of subsequent advertising.

[0115] In one implementation, the method further includes:

[0116] Collect user data; the user data includes historical behavior data, demographic data and interest preference data;

[0117] Processing the user data, extracting user features for building user portraits, and dividing the users into multiple groups based on a clustering algorithm;

[0118] Further segmenting the users in each group according to the user characteristics to obtain multiple segmented groups under each group;

[0119] Add feature labels to multiple segments;

[0120] The target advertisement is delivered to users in the segmented group having the same feature label as the target advertisement.

[0121] Collect historical behavior data, demographic data, and interest preference data from multiple sources to provide comprehensive data support for the construction of user portraits. Preprocess and extract features from the collected user data, integrate them into user portraits, and comprehensively describe the characteristics and behavior patterns of users. Use clustering algorithms (such as K-Means, hierarchical clustering, etc.) to divide users into multiple groups, and users in each group have similar characteristics. Further segment each group to form more specific segment groups to better match advertising content and target audiences. Define and add feature tags for each segment group to identify the characteristics of the group for subsequent targeted advertising. According to the characteristics of the target advertisement, select the segment group that matches it, and target advertisements to users in these groups to ensure the accuracy and effectiveness of the advertisements. Through the above steps, this method can achieve accurate stratification of users and accurate delivery of advertisements, and improve the conversion rate of advertisements and user satisfaction.

[0122] The technical solution provided in this application clarifies the expected effect and budget limit of advertising by obtaining the advertiser's advertising conversion target and target cost, and provides clear guidance for subsequent bidding strategies. The delivery data of the target advertisement, including user behavior, advertising display location, competitive product data, etc., is collected in real time, and feature extraction is performed to generate feature vectors. The acquisition and processing of these real-time data ensure that the bidding strategy can be adjusted in time according to the actual delivery effect, and improve the dynamic and flexibility of the bid. By inputting the feature vector into the advertising conversion prediction model, the model uses a multi-layer neural network and a Sigmoid function to output the predicted probability and accurately evaluate the conversion effect after advertising delivery. The high accuracy of the prediction model ensures the scientificity and reliability of the bidding strategy. Based on the predicted probability, the price adjustment algorithm dynamically adjusts the basic bid to generate an advertising delivery bid. Through multiple delivery and real-time adjustment, it is ensured that the bid for each click can maximize the conversion effect and control the total cost within the target cost range. This dynamic price adjustment mechanism significantly improves the efficiency and cost-effectiveness of advertising delivery. The training process of the advertising conversion prediction model combines multiple sets of historical advertising delivery data, including advertising delivery data, user behavior data, competitive product data, and market environment data. Through multiple layers of linear transformation, activation function and Sigmoid function, the model can capture complex advertising conversion rules. Regularization and multiple rounds of training in the optimization process further improve the generalization ability and prediction accuracy of the model. By adjusting bids in real time and strictly controlling costs, this solution not only ensures that the total cost of advertising is within the budget, but also achieves higher conversion rates and return on investment (ROI) within a limited budget. This optimization mechanism effectively solves the problems of single bidding strategy, inaccurate conversion effect evaluation, and inflexible cost control in existing technologies.

[0123] The present application also provides an OCPX-based advertising bidding device, such as Figure 2 As shown, the device comprises:

[0124] The acquisition module 201 is used to acquire the advertisement conversion target and target cost of the advertiser; the advertisement conversion target refers to the conversion effect that the advertiser expects the target advertisement to achieve after the target advertisement is placed; the target cost is the budget cost of the advertiser for placing the target advertisement; and is used to acquire the delivery data of the target advertisement in real time, and perform feature extraction based on the delivery data to obtain a feature vector for bid prediction;

[0125] Prediction module 202 is used to input the feature vector into an advertisement conversion prediction model to obtain a prediction probability; the prediction probability is the probability of the target advertisement reaching the advertisement conversion target after delivery; and is used to adjust the basic bid according to the prediction probability based on a price adjustment algorithm to obtain an advertisement delivery bid; the basic bid is the initial bid set for each click of the target advertisement; the advertisement delivery bid is the bid for a single click of the target advertisement after delivery;

[0126] The calculation formula of the basic bid is:

[0127] BaseBid=CPA target ×CR avg

[0128] Among them, BaseBid is the base bid, CPA target CR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement;

[0129] The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm;

[0130] The linear price adjustment algorithm is:

[0131] Bid adjusted =BaseBid×CP

[0132] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability;

[0133] The piecewise linear adjustment algorithm is:

[0134]

[0135] Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold;

[0136] The bidding module 203 is used to bid for the target advertisement multiple times according to the advertisement delivery bid; the advertisement delivery bid is adjusted in real time during the process of delivering the target advertisement multiple times; the total advertisement delivery cost is less than or equal to the target cost; the total advertisement delivery cost is the sum of the advertisement delivery prices of each delivery of the target advertisement;

[0137] The training module 204 is used to obtain multiple groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data; the advertising delivery history data includes delivery advertising data, user behavior data, competitive product data and market environment data; the delivery advertising data includes the content of the advertisement to be delivered, the advertisement triggering keywords and the advertisement display position; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of the advertisement of the same type as the delivered advertisement; the market environment data includes the data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of the advertisement;

[0138] A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer;

[0139] Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing;

[0140] The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

[0141] In one embodiment, the device further comprises:

[0142] The optimization module is used to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data by using a generative adversarial network (GAN) or a SMOTE algorithm.

[0143] In one embodiment, the prediction module is used to combine multiple feature vectors to generate interactive features, and introduce time series into the interactive features to obtain a target feature vector; and input the target feature vector into the advertising conversion prediction model.

[0144] In one embodiment, the device further comprises:

[0145] A delivery module is used to collect user data; the user data includes historical behavior data, demographic data and interest preference data; process the user data, extract user features for building user portraits, and divide users into multiple groups based on a clustering algorithm; further segment users in each group according to the user features to obtain multiple sub-groups under each group; add feature labels to multiple sub-groups; and deliver the target advertisement to users in the sub-groups that have the same feature labels as the target advertisement.

[0146] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0147] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the OCPX-based advertising bidding method described in any embodiment of the present application.

[0148] The computer storage medium stores codes, and when the codes are executed, the device executing the codes implements the OCPX-based advertising bidding method described in any embodiment of the present application.

[0149] The "first" and "second" (if any) in the names such as "first" and "second" mentioned in the embodiments of the present application are only used as name identifiers and do not represent the first or second in order.

[0150] The above description is merely an exemplary embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. An advertisement bidding method based on OCPX, characterized in that: include: Obtain the advertiser's advertising conversion target and target cost; The advertisement conversion target refers to the conversion effect that the advertiser expects the target advertisement to achieve after placing the target advertisement; the target cost refers to the budget cost of the advertiser for placing the target advertisement; Acquire the delivery data of the target advertisement in real time, and extract features based on the delivery data to obtain a feature vector for bid prediction; Input the feature vector into an advertisement conversion prediction model to obtain a prediction probability; the prediction probability is the probability of reaching the advertisement conversion target after the target advertisement is released; Based on the price adjustment algorithm, the basic bid is adjusted according to the predicted probability to obtain an advertisement delivery bid; the basic bid is the initial bid set for each click of the target advertisement; the advertisement delivery bid is the bid for a single click after the target advertisement is delivered; The calculation formula of the basic bid is: BaseBid=CPA target ×CR avg Among them, BaseBid is the base bid, CPA target CR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement; The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm; The linear price adjustment algorithm is: Bid adjusted =BaseBid×CP Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability; The piecewise linear adjustment algorithm is: Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold; The target advertisement delivery bid is made multiple times according to the advertisement delivery bid; the advertisement delivery bid is adjusted in real time during the process of delivering the target advertisement multiple times; the total advertisement delivery cost is less than or equal to the target cost; the total advertisement delivery cost is the sum of the advertisement delivery prices of each delivery of the target advertisement; The training process of the advertising conversion prediction model is as follows: Acquire multiple groups of historical data on advertising delivery and advertising conversion data corresponding to each group of historical data on advertising delivery; the historical data on advertising delivery includes advertising delivery data, user behavior data, competitive product data and market environment data; the advertising delivery data includes the content of the advertisement to be delivered, the advertising triggering keywords and the advertising display location; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of advertisements of the same type as the delivered advertisements; the market environment data includes data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of advertisements; A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer; Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing; The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

2. The method according to claim 1, characterized in that: The method further comprises: Use a generative adversarial network GAN or a SMOTE algorithm to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data.

3. The method according to claim 1, characterized in that The step of inputting the feature vector into an advertisement conversion prediction model to obtain a prediction probability includes: Combining a plurality of the feature vectors to generate an interactive feature, and introducing a time series into the interactive feature to obtain a target feature vector; The target feature vector is input into the advertisement conversion prediction model.

4. The method according to claim 1, characterized in that: The method further comprises: Collect user data; the user data includes historical behavior data, demographic data and interest preference data; Processing the user data, extracting user features for building user portraits, and dividing the users into multiple groups based on a clustering algorithm; Further segmenting the users in each group according to the user characteristics to obtain multiple segmented groups under each group; Add feature labels to multiple segments; The target advertisement is delivered to users in the segmented group having the same feature label as the target advertisement.

5. An advertisement bidding device based on OCPX, characterized in that: include: The acquisition module is used to obtain the advertising conversion target and target cost of the advertiser; The advertisement conversion target refers to the conversion effect that the advertiser expects the target advertisement to achieve after placing the target advertisement; the target cost is the budget cost of the advertiser for placing the target advertisement; and is used to obtain the delivery data of the target advertisement in real time, and perform feature extraction based on the delivery data to obtain a feature vector for bid prediction; A prediction module is used to input the feature vector into an advertisement conversion prediction model to obtain a prediction probability; the prediction probability is the probability of the target advertisement reaching the advertisement conversion target after delivery; and is used to adjust the basic bid according to the prediction probability based on a price adjustment algorithm to obtain an advertisement delivery bid; the basic bid is the initial bid set for each click of the target advertisement; the advertisement delivery bid is the bid for a single click of the target advertisement after delivery; The calculation formula of the basic bid is: BaseBid=CPA target ×CR avg Among them, BaseBid is the base bid, CPA target CR is the conversion cost per ad click. avg The historical average conversion rate of other advertisements of the same type as the target advertisement; The price adjustment algorithm includes a linear price adjustment algorithm, a piecewise linear price adjustment algorithm and a nonlinear price adjustment algorithm; The linear price adjustment algorithm is: Bid adjusted =BaseBid×CP Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, and CP is the predicted probability; The piecewise linear adjustment algorithm is: Among them, Bid adjusted is the advertisement delivery bid, BaseBid is the base bid, CP is the predicted probability, Min_Factor is the bid adjustment factor, and Threshold is the predicted probability threshold; A bidding module is used to make multiple bids for the target advertisement placement according to the advertisement placement bid; the advertisement placement bid is adjusted in real time during the process of placing the target advertisement multiple times; the total advertisement placement cost is less than or equal to the target cost; the total advertisement placement cost is the sum of the advertisement placement prices of each placement of the target advertisement; A training module is used to obtain multiple groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data; the advertising delivery history data includes delivery advertising data, user behavior data, competitive product data and market environment data; the delivery advertising data includes the content of the advertisement to be delivered, the advertisement triggering keywords and the advertisement display location; the user behavior data refers to the behavior of the user after clicking on the advertisement, and the user behavior data includes the user's search keywords, user browsing time, the number of pages browsed by the user and the user's conversion behavior; the competitive product data is the delivery data of the advertisement of the same type as the delivered advertisement; the market environment data includes the data on the impact of each advertising delivery time period, each region and various types of equipment on the delivery of the advertisement; A set of the advertising delivery history data is input into the model to be trained, and the model to be trained predicts the delivery conversion probability of the advertisement corresponding to the advertising delivery history data according to the advertising delivery history data. After the advertising delivery history data is input into the model to be trained, it is processed by multiple layers of linear transformation and activation function, and the predicted conversion probability is obtained through the Sigmoid function of the output layer; Comparing the predicted conversion probability with the advertisement conversion data corresponding to the advertisement delivery history data, calculating the value of the loss function, and starting from the output layer, using the chain rule to calculate the gradient of the loss function to the parameters of each layer layer by layer, optimizing the model to be trained according to the calculated gradient to obtain the optimized model to be trained; the optimization process of the model to be trained includes adjusting the number of hidden layers, adjusting neuron data, replacing activation functions, introducing attention mechanisms, and performing regularization processing; The optimized model to be trained is further trained and optimized using another set of advertising delivery history data and corresponding advertising conversion data until the error between the predicted conversion probability output by the optimized model to be trained and the corresponding advertising conversion data is less than a preset threshold, and the optimized model to be trained is used as the advertising conversion prediction model.

6. The device according to claim 5, characterized in that The device also includes: The optimization module is used to generate more groups of advertising delivery history data and advertising conversion data corresponding to each group of advertising delivery history data based on the multiple groups of advertising delivery history data and the advertising conversion data corresponding to each group of advertising delivery history data by using a generative adversarial network (GAN) or a SMOTE algorithm.

7. The device according to claim 5, characterized in that The prediction module is used to combine multiple feature vectors to generate interactive features, and introduce time series into the interactive features to obtain a target feature vector; and input the target feature vector into the advertising conversion prediction model.

8. The device according to claim 5, characterized in that The device also includes: A delivery module is used to collect user data; the user data includes historical behavior data, demographic data and interest preference data; process the user data, extract user features for building user portraits, and divide users into multiple groups based on a clustering algorithm; further segment users in each group according to the user features to obtain multiple sub-groups under each group; add feature labels to multiple sub-groups; and deliver the target advertisement to users in the sub-groups that have the same feature labels as the target advertisement.

9. A computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the OCPX-based advertising bidding method as described in any one of claims 1 to 4 is implemented.

10. A computer storage medium, characterized in that: The computer storage medium stores instructions, and when the instructions are executed on the terminal device, the terminal device executes the OCPX-based advertising bidding method as described in any one of claims 1-4.

Images