An advertisement value evaluation method, device and medium based on a machine learning model

Abstract

The application discloses an advertisement value evaluation method and device based on a machine learning model and a medium, relates to the technical field of advertisement evaluation, and comprises the following steps: inputting a unified value evaluation input package into a multi-domain game perception machine learning model, constructing a user multi-touch time evolution journey through multi-touch time attribution calculation, quantifying the marginal contribution of each advertisement touch to the conversion time influence, simultaneously performing contribution reduction processing based on compliance constraint information in the unified value evaluation input package, outputting a journey-level incremental contribution summary result, coupling the journey-level incremental contribution summary result with market side bidding environment elements for coupled calculation, and generating a game-corrected transaction net value object; by constructing a unified value evaluation input package and inputting the same into a multi-domain game perception machine learning model, the application realizes joint embedding and coupled calculation of cross-domain elements, and solves the problem of scattered modeling of multi-domain features in the prior art.

Description

Technical Field

[0001] This invention relates to the field of data encryption technology, and in particular to a method, device and medium for evaluating advertising value based on a machine learning model. Background Technology

[0002] With the development of programmatic advertising and real-time bidding technologies, advertising transactions have gradually shifted from manual placement to an automated decision-making model driven by data. Existing advertising value assessment methods typically revolve around metrics such as click-through rate (CTR) prediction, conversion rate prediction, and estimated conversion value. They utilize logistic regression models, gradient boosting tree models, or deep neural network models to predict the effectiveness of single exposures or single touchpoint actions and generate bidding strategies. Regarding multi-touchpoint attribution, linear attribution, time decay attribution, and data-driven attribution models exist to allocate the contribution ratio of different advertising touchpoints to conversion results. Simultaneously, some technologies combine reinforcement learning or game theory optimization methods to dynamically adjust bidding strategies and budget allocation.

[0003] However, existing advertising value assessment methods still have room for improvement. First, existing technologies typically model advertising-side characteristics, user-side characteristics, and market-side bidding environment characteristics separately, lacking a unified data structure and cross-domain coupling mechanism. In terms of time dimension, most methods only predict whether a conversion will occur, without quantifying the impact of advertising touchpoints on the timing of conversion, making it difficult to construct a complete user time evolution path. In addition, in terms of compliance management, most methods adopt post-event filtering or rule restrictions, failing to incorporate compliance constraints into the core calculation process of value assessment, resulting in the value assessment results failing to reflect compliance risks simultaneously. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides an advertising value assessment method based on a machine learning model to address the problems of lacking a unified data structure and cross-domain coupling mechanism, and failing to incorporate compliance constraints into the core calculation process of value assessment.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides an advertising value assessment method based on a machine learning model, comprising,

[0008] When an ad request enters the ad transaction chain, the display opportunity is abstracted into a tradable value object, and compliance awareness feature encoding processing is performed on the ad-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements to generate a unified value assessment input package.

[0009] The unified value assessment input package is input into the multi-domain game perception machine learning model. The user's multi-touch time evolution journey is constructed through multi-touch time attribution calculation, and the marginal contribution of each advertising touchpoint to the conversion time is quantified. At the same time, contribution reduction processing is performed based on the compliance constraint information in the unified value assessment input package, and the journey-level incremental contribution summary result is output.

[0010] The summaries of journey-level incremental contributions are coupled with market-side bidding environment factors to generate a game-corrected net transaction value object.

[0011] Based on the game-theoretic corrected net value of the transaction, the system combines advertising-side placement factors with real-time user intent factors to perform decision optimization calculations, generate advertising value assessment results, and output corresponding bidding ranges, frequency control, and placement execution strategies.

[0012] As a preferred embodiment of the advertising value assessment method based on machine learning models described in this invention, the generation of the unified value assessment input package specifically includes:

[0013] Feature data is extracted from advertising-side placement elements, user-side real-time intent elements, and market-side bidding environment elements to form a multi-dimensional feature set. Compliance awareness feature encoding processing is then performed on the multi-dimensional feature set to generate compliance constraint information.

[0014] After encoding and processing the feature data and compliance constraint information, they are encapsulated to form a unified value assessment input package.

[0015] As a preferred embodiment of the advertising value evaluation method based on machine learning model described in this invention, the multi-domain game perception machine learning model includes a cross-domain feature embedding layer, a touchpoint time series modeling layer, a bidding game coupling layer, and a value decision output layer.

[0016] As a preferred embodiment of the advertising value evaluation method based on machine learning models described in this invention, the construction of the user's multi-touchpoint time evolution journey specifically includes:

[0017] The unified value assessment input package is input into the cross-domain feature embedding layer. The advertising-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements are respectively processed by the advertising-side feature mapping center, user-side feature mapping center, and bidding environment feature mapping center to output feature vector output results. The feature vector output results are then processed by vector concatenation according to the field index order to form a cross-domain joint feature vector.

[0018] The cross-domain joint feature vector is used as the input data for the touchpoint time series modeling layer. At the same time, the advertising touchpoint log data in the unified value assessment input package is input into the touchpoint time series modeling layer. Touchpoint event identification and touchpoint type labeling are performed on the advertising touchpoint log data to form an advertising touchpoint sequence set.

[0019] The ad touchpoint sequence set is ordered according to the timestamp of the touchpoint occurrence, forming a time-ordered touchpoint sequence;

[0020] Based on the time interval between each advertising touchpoint and the conversion touchpoint in the time-ordered touchpoint sequence, a temporal position quantization calculation is performed to form a touchpoint temporal position weight sequence. The touchpoint temporal position weight sequence is then weighted and fused with the cross-domain joint feature vector to obtain the cross-domain corrected touchpoint temporal position weight sequence.

[0021] Based on the time interval between adjacent advertising touchpoints in the time-ordered touchpoint sequence, a time decay propagation calculation is performed, and the decay calculation result is combined with the cross-domain corrected touchpoint temporal position weight sequence to perform a touchpoint propagation decreasing calculation to form the touchpoint temporal influence result.

[0022] By combining the touchpoint type identification results of each ad touchpoint in the ad touchpoint sequence set, we perform influence duration interval modeling and encapsulate the modeling results into a journey structure to form a user multi-touchpoint time evolution journey.

[0023] As a preferred embodiment of the advertising value evaluation method based on machine learning models described in this invention, the output journey-level incremental contribution summary result specifically includes:

[0024] Extract the set of advertising touchpoint sequences from the user's multi-touchpoint time evolution journey, input them into the touchpoint time series modeling layer, and perform touchpoint effect interval analysis and conversion time correlation positioning processing on the set of advertising touchpoint sequences to form a set of touchpoint time impact calculations;

[0025] Based on the results of the touchpoint timing impact, the conversion time offset of each ad touchpoint in the touchpoint timing impact calculation set is quantitatively calculated, and the incremental effect value of each ad touchpoint on the conversion time is decomposed to form a touchpoint marginal contribution set.

[0026] The set of marginal contributions at touchpoints is associated with the compliance constraint information in the unified value assessment input package, and contribution reduction processing is performed on the set of marginal contributions at touchpoints. The reduced set of marginal contributions at touchpoints is then summarized and encapsulated to form a summary result of incremental contributions at the journey level.

[0027] As a preferred embodiment of the advertising value assessment method based on machine learning models described in this invention, wherein: the generation of the game-corrected transaction net value object specifically refers to:

[0028] Extract the touchpoint marginal contribution set from the trip-level incremental contribution summary result, input it into the bidding game coupling layer, and perform conversion revenue quantification mapping calculation on the touchpoint marginal contribution set to form the touchpoint revenue calculation set;

[0029] By integrating the revenue calculation set of touchpoints with market-side bidding environment factors, and performing bidding impact correction calculations, a revenue game adjustment set is formed.

[0030] Summarize the adjusted set of returns from the game and generate a game-corrected net asset value (NAV) object.

[0031] As a preferred embodiment of the advertising value assessment method based on machine learning models described in this invention, the generation of advertising value assessment results specifically includes:

[0032] Extract the set of return assessment parameters from the game-corrected net value of the transaction object, input it into the value decision output layer, perform value quantification calculation on the set of return assessment parameters, and form the advertising value calculation set.

[0033] The value calculation set is integrated with the budget constraint matching calculation of the advertising side delivery elements, and the delivery adaptation calculation is performed based on the real-time intent elements of the user side to form a value decision control set;

[0034] The results of the value decision control set are encapsulated and processed to generate advertising value assessment results.

[0035] As a preferred embodiment of the advertising value evaluation method based on machine learning models described in this invention, the step of forming a set of advertising touchpoint sequences specifically includes:

[0036] Perform log field parsing on the ad touchpoint log data to extract the touchpoint event identifiers and touchpoint timestamps corresponding to ad impression records, ad click records, and conversion feedback records.

[0037] Based on the touchpoint event identifiers, the ad exposure records, ad click records, and conversion feedback records are classified by touchpoint type to generate exposure touchpoint data, click touchpoint data, and conversion touchpoint data;

[0038] The exposure touchpoint data, click touchpoint data, and conversion touchpoint data are merged and ordered based on the timestamp of the touchpoint occurrence to form a set of advertising touchpoint sequences.

[0039] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the advertising value assessment method based on a machine learning model as described in the first aspect of the present invention.

[0040] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the advertising value assessment method based on a machine learning model as described in the first aspect of the present invention.

[0041] The beneficial effects of this invention are as follows: By constructing a unified value assessment input package and inputting it into a multi-domain game-aware machine learning model, the joint embedding and coupled calculation of cross-domain elements are realized, solving the problem of scattered multi-domain feature modeling in the prior art; at the same time, by constructing the user's multi-touchpoint time evolution journey through multi-touchpoint time attribution calculation, the marginal contribution of each advertising touchpoint to the conversion time is quantified, expanding the value assessment from simply predicting whether a conversion will occur to a fine characterization of the impact on conversion time, enhancing the ability to express time series; furthermore, by embedding compliance constraint information and performing contribution reduction processing in the contribution calculation stage, compliance factors directly participate in the core value calculation process, thereby achieving the synchronous integration of effect assessment and risk control, improving the accuracy, time series awareness, and compliance reliability of advertising value assessment. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of an advertising value assessment method based on a machine learning model.

[0044] Figure 2 This is a schematic diagram of the structure of a multi-domain game perception machine learning model.

[0045] Figure 3 A schematic diagram illustrating the user's multi-touchpoint time evolution journey.

[0046] Figure 4 Flowchart for generating and optimizing transaction net asset value objects. Detailed Implementation

[0047] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0049] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0050] Reference Figures 1-4 This is one embodiment of the present invention, which provides an advertising value evaluation method based on a machine learning model, including the following steps:

[0051] S1: When an ad request enters the ad transaction chain, the display opportunity is abstracted into a tradable value object, and compliance awareness feature encoding processing is performed on the ad-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements to generate a unified value assessment input package.

[0052] S1.1: When an ad request enters the ad transaction chain, the ad transaction platform establishes a unique display opportunity identifier at the moment the display opportunity is generated to distinguish different ad display opportunities. The unique display opportunity identifier is then bound and registered with the ad slot number, media resource location information, and display timestamp to form a tradable value object identifier record.

[0053] The system synchronously retrieves data source content corresponding to advertising-side placement elements, user-side real-time intent elements, and market-side bidding environment elements, based on the identifier records of tradable value objects. Advertising-side placement elements include ad campaign number, bidding strategy type, target conversion event type, and budget consumption progress parameters. User-side real-time intent elements include user terminal device type, user historical browsing behavior tags, user real-time search keywords, and user conversion probability score parameters. Market-side bidding environment elements include the number of competing ads for the current ad slot, historical average transaction price, real-time bidding request density, and industry competition intensity level parameters.

[0054] S1.2: Perform field alignment and timestamp alignment on the ad-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements to form a multi-dimensional feature set under a unified time benchmark.

[0055] The compliance awareness feature encoding processing function calls a pre-built compliance rule library to perform compliance rule verification on each field of the multi-dimensional feature set. The compliance rule library includes advertising industry access rules, sensitive word rules for advertising content, geographic restriction rules, and user privacy compliance rules. The construction process of the pre-built compliance rule library is as follows: the advertising industry access rules, sensitive word list, geographic restriction list, and privacy compliance verification items are organized into rule entries and stored in the rule storage structure. At the same time, a corresponding rule number is assigned to each rule entry to complete the construction of the pre-built compliance rule library. For fields that pass verification, compliance pass identifier parameters are generated; for fields that are restricted verification, compliance restriction identifier parameters are generated. The compliance pass identifier parameters and compliance restriction identifier parameters together constitute the compliance constraint information.

[0056] The numerical fields in the multidimensional feature set are normalized linearly according to the range of field values. The category fields are numbered and encoded according to the order of category values. The time field is calculated based on the time difference between the display timestamp and the touch point occurrence timestamp and converted into second-level numerical encoding. The encoded feature data and compliance constraint information are then concatenated and encapsulated according to the field index order to generate a unified value assessment input package, providing a unified format of data input for the subsequent multi-domain game perception machine learning model.

[0057] S2: Input the unified value assessment input package into the multi-domain game perception machine learning model, construct the user's multi-touch time evolution journey through multi-touch time attribution calculation, quantify the marginal contribution of each advertising touchpoint to the conversion time, and perform contribution reduction processing based on the compliance constraint information in the unified value assessment input package, and output the journey-level incremental contribution summary result.

[0058] S2.1: The multi-domain game perception machine learning model includes a cross-domain feature embedding layer, a touchpoint time series modeling layer, a bidding game coupling layer, and a value decision output layer.

[0059] The cross-domain feature embedding layer includes an advertising-side feature mapping center, a user-side feature mapping center, and a bidding environment feature mapping center; each feature mapping center includes a feature input transformation structure and a feature vector output structure; the three feature vector output structures form a cross-domain joint feature vector concatenation structure.

[0060] The touch timing modeling layer includes a touch sequence input structure, a time-dependent propagation structure, and a touch effect calculation structure. The touch sequence input structure includes an exposure touch input channel, a click touch input channel, and a conversion touch input channel. The time-dependent propagation structure establishes touch time interval propagation calculation nodes.

[0061] The bidding game coupling layer includes a payoff propagation calculation structure and a competition intensity coupling structure; the payoff propagation calculation structure establishes payoff mapping calculation nodes; and the competition intensity coupling structure establishes bidding competition correction calculation nodes.

[0062] The value decision output layer includes a value assessment calculation structure and a decision control output structure; the decision control output structure includes a bid range generation center, a frequency control generation center, and a placement execution strategy generation center.

[0063] The training sample dataset is constructed from historical advertising transaction records; the historical advertising transaction records include data corresponding to the unified value assessment input package, conversion result data, and conversion timestamp data; the historical advertising transaction records are serialized and organized according to the user conversion path to form a multi-touchpoint training sequence.

[0064] The multi-touchpoint training sequence is input into the multi-domain game-aware machine learning model to perform forward propagation calculation, generating the touchpoint timing impact prediction results and transaction net value prediction results. The loss value is calculated based on the deviation between the touchpoint timing impact prediction results, transaction net value prediction results and supervision labels. The gradient of the loss value with respect to all trainable parameters in the cross-domain feature embedding layer, touchpoint timing modeling layer, bidding game coupling layer and value decision output layer is calculated through the backpropagation algorithm. Parameter update calculation is performed based on the gradient calculation results, and iterative updates are performed on the cross-domain joint feature vector mapping parameters, timing dependency propagation weight parameters, game payoff propagation parameters and value decision mapping parameters. Forward propagation calculation, loss value calculation, backpropagation calculation and parameter update calculation are performed repeatedly until the loss value reaches the convergence threshold, and the trained multi-domain game-aware machine learning model is obtained.

[0065] To further explain, the convergence threshold is set based on the stability of the decrease in the loss value during the training process. When the change in the loss value in two consecutive training iterations is lower than the convergence threshold, it is determined that the convergence state has been reached. The supervision label consists of conversion result data and conversion timestamp data from historical advertising transaction records (for example, if transaction net value prediction is required, "actual transaction price / settlement revenue" can be added).

[0066] S2.2: The unified value assessment input package is fed into the cross-domain feature embedding layer. The advertising-side placement elements, user-side real-time intent elements, and market-side bidding environment elements are respectively input into the advertising-side feature mapping center, user-side feature mapping center, and bidding environment feature mapping center. Each feature mapping center performs interval normalization numerical processing on the numerical fields according to the ratio between the original value and the maximum value of the field, and performs fixed numbering encoding processing on the category fields according to the field category order to form the corresponding feature vector output results. The three feature vector output layers perform vector concatenation processing on the feature vector output results from each feature mapping center according to the field index order to form a cross-domain joint feature vector. The cross-domain joint feature vector is input into the touch point timing modeling layer to perform cross-domain feature weighted correction calculation on the touch point timing position weight sequence and the touch point timing influence result.

[0067] S2.3: Advertising touchpoint log data input touchpoint timing modeling layer in the unified value assessment input package.

[0068] The ad touchpoint log data is parsed to extract the touchpoint event identifier, touchpoint timestamp, and unique display opportunity identifier corresponding to ad impression records, ad click records, and conversion feedback records. Touchpoint type classification is performed based on the event type code of the touchpoint event identifier: records coded as impressions are classified as impression touchpoint data, records coded as clicks are classified as click touchpoint data, and records coded as conversions are classified as conversion touchpoint data. Touchpoint merging is performed on the impression touchpoint data, click touchpoint data, and conversion touchpoint data using the unique display opportunity identifier as the merging key, and the data is arranged in the order of touchpoint timestamps to form an ad touchpoint sequence set.

[0069] The touchpoint timing modeling layer receives a set of ad touchpoint sequences, processes them according to the timestamps of the touchpoints, forming a time-ordered touchpoint sequence. Based on the time interval between each ad touchpoint and the conversion touchpoint in the time-ordered touchpoint sequence, it performs time-order position quantization calculation, assigns decreasing position coefficients according to the time interval from shortest to longest, and performs a weighted product calculation with the conversion probability score parameter in the real-time intent element on the user side to form a touchpoint timing position weight sequence. It then performs a weighted product calculation with the corresponding weight value of each ad touchpoint in the touchpoint timing position weight sequence and the feature value at the same field index position in the cross-domain joint feature vector, and performs numerical accumulation processing on the weighted result to obtain the cross-domain corrected touchpoint timing position weight sequence.

[0070] Time decay propagation calculation is performed based on the time interval between adjacent advertising touchpoints in a time-ordered touchpoint sequence: The time interval is converted to a 1-hour base unit and then divided to obtain the time interval in hours, n. Based on the preset time decay base parameter k (0 < k < 1), k is then used to calculate the time decay propagation.n Calculate the corresponding time decay coefficient, where the preset time decay benchmark parameter k is determined based on the statistical results of the average conversion contribution decay trend corresponding to different time intervals in the historical advertising conversion path data; perform a per-touch propagation and decreasing calculation on the time decay coefficient and the cross-domain corrected touch point time sequence position weight sequence to form the touch point time sequence influence result.

[0071] The duration of touchpoint impact is determined by combining the touchpoint type identification results of each ad touchpoint in the ad touchpoint sequence set. Specifically, based on the statistical results of the time interval distribution between different touchpoint types and conversion touchpoints in the historical conversion path data of ad touchpoints, the upper limit of the time interval with the highest concentration of conversion contribution is selected as the basis for determining the duration of impact. The conversion time from exposure touchpoint to conversion touchpoint is mainly concentrated within 24 hours, and the conversion time from click touchpoint to conversion touchpoint is mainly concentrated within 72 hours. Therefore, the duration of impact of exposure touchpoint is set to 24 hours, the duration of impact of click touchpoint is set to 72 hours, and the duration of impact of conversion touchpoint is set to the time corresponding to the conversion timestamp. The time range of impact is expanded backward according to the touchpoint occurrence timestamp. On a unified time axis, the time interval expansion is performed on the touchpoint time sequence impact results corresponding to each ad touchpoint, and the numerical accumulation calculation is performed on the touchpoint time sequence impact results within the time overlap interval to form the continuous time distribution impact coverage modeling result. The touchpoint time sequence impact results and the impact coverage modeling results are encapsulated into a sequence structure to form the user's multi-touchpoint time evolution journey.

[0072] To further explain, the position coefficients are allocated in descending order of time as follows: position coefficients for time intervals within 1 hour are set to 1.0, position coefficients for time intervals between 1 and 6 hours are set to 0.8, position coefficients for time intervals between 6 and 24 hours are set to 0.6, and position coefficients for time intervals exceeding 24 hours are set to 0.4. The position coefficient values ​​are based on the statistical results of the average conversion contribution ratio corresponding to different time intervals in the historical conversion path data of the advertising touchpoints.

[0073] S2.4: Input the user's multi-touchpoint time evolution journey back into the touchpoint timing modeling layer to perform attribution quantification calculation: Extract the set of advertising touchpoint sequences in the user's multi-touchpoint time evolution journey; perform time difference calculation based on the touchpoint occurrence timestamps corresponding to each advertising touchpoint and the conversion occurrence timestamps corresponding to the conversion touchpoints in the advertising touchpoint sequence set to obtain the time interval values ​​from each advertising touchpoint to the conversion touchpoint; perform interval matching judgment processing based on the touchpoint influence duration interval and time interval values ​​of each advertising touchpoint in the user's multi-touchpoint time evolution journey, which is the interval inclusion relationship judgment. When the time interval value falls within the corresponding touchpoint influence duration interval, it is judged as a valid advertising touchpoint; establish a time difference correspondence between the touchpoint occurrence timestamps of valid advertising touchpoints and the corresponding conversion occurrence timestamps to form a touchpoint time influence calculation set.

[0074] The touchpoint timing modeling layer performs conversion time offset quantification calculation on each ad touchpoint in the touchpoint timing impact calculation set based on the touchpoint timing impact results: It calculates the conversion time offset by multiplying the time interval values ​​between each ad touchpoint and the conversion touchpoint with the corresponding touchpoint timing impact results to obtain the time advance or time delay of each ad touchpoint on the conversion time, forming a conversion time offset sequence. Based on the conversion time offset sequence, it performs touchpoint contribution decomposition calculation on each ad touchpoint. Specifically, it performs positive and negative impact judgment processing according to the conversion time offset values ​​corresponding to each ad touchpoint, marking the offset values ​​that advance the conversion time as positive incremental values ​​and the offset values ​​that delay the conversion time as negative incremental values. It then performs numerical quantification processing according to the absolute value of the offset to obtain the incremental value of each ad touchpoint on the change in conversion time. Finally, it performs set encapsulation processing on the incremental values ​​to form a touchpoint marginal contribution set.

[0075] S2.5: Synchronously introduce the compliance constraint information from the unified value assessment input package into the touchpoint marginal contribution set processing flow; based on the compliance pass identifier parameter and compliance restriction identifier parameter in the compliance constraint information, perform identifier matching judgment on each advertising touchpoint in the touchpoint marginal contribution set; when the advertising touchpoint corresponds to the compliance restriction identifier parameter, multiply the corresponding incremental action value with the preset reduction coefficient to obtain the reduced touchpoint marginal contribution value; when the advertising touchpoint only corresponds to the compliance pass identifier parameter, keep the incremental action value unchanged; perform summary encapsulation processing on the touchpoint marginal contribution value after the compliance reduction processing to form the journey-level incremental contribution summary result.

[0076] To further explain, the preset reduction factor is based on the statistical results of the ratio of the average actual conversion revenue of ad touchpoints with compliance restrictions to the average conversion revenue of compliant touchpoints in historical advertising transaction records. The preset value is 0.6. This is because when statistically analyzing historical advertising transaction records, ad touchpoints with compliance restriction labels are compared with ad touchpoints with only compliance approval labels. The average actual conversion revenue ratio of the two types of touchpoints is calculated. The statistical results show that the former is about 0.6 times that of the latter. Therefore, the preset reduction factor is set to 0.6.

[0077] Preferably, compared to existing evaluation methods, this method integrates advertising, user, and market-side elements through a multi-domain game-theoretic machine learning model and introduces multi-touchpoint time attribution calculation to accurately construct the user's time evolution journey, quantify the marginal contribution of each touchpoint to conversion time, and embed compliance constraints to reduce contributions, thereby achieving game-corrected transaction net value evaluation and decision optimization. This method significantly improves the accuracy of value assessment, time-series awareness, and compliance security, optimizes bidding, frequency, and delivery strategies, enhances the overall efficiency and sustainability of the advertising transaction chain, and reduces invalid impressions and clicks.

[0078] S3: Couple the summary results of journey-level incremental contributions with market-side bidding environment factors to generate a game-corrected net transaction value object.

[0079] S3.1: Extract the touchpoint marginal contribution set from the trip-level incremental contribution summary results, and perform conversion revenue quantification mapping calculation on the touchpoint marginal contribution set to form a touchpoint revenue calculation set. Specifically, the incremental contribution value of each ad touchpoint in the touchpoint marginal contribution set is multiplied by a preset conversion revenue weight parameter to obtain the touchpoint revenue value corresponding to each ad touchpoint. The conversion revenue weight parameter is derived from the average conversion value ratio corresponding to the same touchpoint type in historical ad transaction records. The touchpoint revenue values ​​of all ad touchpoints are then encapsulated in a set according to the touchpoint occurrence timestamp order to form the touchpoint revenue calculation set.

[0080] S3.2: Integrate the touchpoint revenue calculation set with market-side bidding environment factors, perform bidding impact correction calculations, and form a revenue game adjustment set. The specific process is as follows: Extract the current number of competing ads for the ad slot, historical average transaction price, real-time bidding request density, and industry competition intensity level parameters from the market-side bidding environment factors; multiply each touchpoint revenue value in the touchpoint revenue calculation set with the competition correction factor. The expression for the competition correction factor is:

[0081] ;

[0082] in, Indicates a competing correction factor. This represents the historical average transaction price. Indicates the real-time bidding request density. This parameter indicates the level of industry competition intensity.

[0083] When the density of real-time bidding requests increases or the industry competition intensity level parameter increases, the competition correction factor decreases, thereby reducing the touch point revenue value to reflect the market saturation effect; all corrected touch point revenue values ​​are summed to form a revenue game adjustment set.

[0084] The total adjusted revenue value is obtained by summing all the corrected touchpoint revenue values ​​in the revenue game adjustment set. The total adjusted revenue value is then multiplied by the expected single conversion value corresponding to the target conversion event type of the advertising-side placement element in the unified value assessment input package to obtain the game-corrected net transaction value object. The game-corrected net transaction value object represents the expected net transaction revenue after time-series attribution and compliance reduction adjustments under the current market bidding environment.

[0085] S4: Based on the game-theoretic corrected net transaction value object, the system combines advertising-side placement elements and user-side real-time intent elements to perform decision optimization calculations, generate advertising value assessment results, and output corresponding bidding ranges, frequency control, and placement execution strategies.

[0086] S4.1: Extract the set of return assessment parameters from the game-corrected net value object, perform value quantification calculation on the set of return assessment parameters, and form an advertising value calculation set, specifically:

[0087] The revenue assessment parameter set includes the total adjusted revenue value and the expected single conversion value; the total adjusted revenue value and the expected single conversion value are multiplied to obtain the advertising value value; the advertising value value is multiplied by the user conversion probability score parameter of the user-side real-time intent element in the unified value assessment input package to obtain the final advertising value score; the final advertising value score is encapsulated into a set to form the advertising value calculation set.

[0088] S4.2: Integrate the advertising value calculation set with the budget constraint matching calculation of advertising-side delivery elements, and perform delivery adaptation calculation based on real-time intent elements on the user side to form a value decision control set, specifically:

[0089] Extract the budget consumption progress parameter from the advertising-side delivery elements in the unified value assessment input package, and calculate the remaining budget percentage using the following formula:

[0090] ;

[0091] in, Indicates the remaining percentage of the budget. This indicates the total budget amount. This indicates the amount of budget already used.

[0092] By analyzing historical advertising transaction records and analyzing the average conversion completion rate corresponding to different budget remaining percentage ranges, it was determined that the average conversion completion rate begins to decline when the budget remaining percentage is below 0.2. Therefore, 0.2 is set as the budget control threshold. When the value is less than 0.2, the advertising value score in the advertising value calculation set is multiplied by the budget adjustment coefficient of 0.7 to obtain the advertising value score after budget constraint adjustment; when the remaining budget ratio is... When the value is ≥0.2, the advertising value score remains unchanged, and the budget-constrained advertising value score is obtained.

[0093] To further explain, the budget adjustment factor of 0.7 is determined based on the statistical results of the average conversion revenue in the range where the remaining budget ratio is less than 0.2 in historical advertising transaction records, relative to the average conversion revenue in the range where the budget is sufficient. Since the statistical result of the ratio is approximately 0.7, the budget adjustment factor is set to 0.7.

[0094] The advertising value score adjusted by budget constraints is converted into the target conversion cost parameter corresponding to the bidding strategy type in the advertising placement elements by the execution ratio to generate a bidding range. The upper limit of the bidding range is the advertising value score adjusted by budget constraints, and the lower limit is 0.8 times the advertising value score adjusted by budget constraints. The value of 0.8 is the lower quartile of the transaction price distribution in the historical advertising transaction records.

[0095] The user conversion probability score parameter is extracted from the real-time intent elements on the user side of the unified value assessment input package. The ad value score adjusted for budget constraints is multiplied with the user conversion probability score parameter to calculate the placement priority score. The placement priority score is compared with the median of 0.5 in the distribution of conversion probability scores in historical ad transaction records. When the placement priority score is greater than 0.5, the frequency control is set to 3 times / 24 hours. When the placement priority score is less than or equal to 0.5, the frequency control is set to 1 time / 24 hours. This is determined based on the statistical results of the marginal conversion improvement rate corresponding to different placement priority score intervals in historical ad transaction records. This setting can improve conversion efficiency in high conversion probability user scenarios and reduce invalid exposure in low conversion probability user scenarios, thereby balancing conversion effect and budget consumption.

[0096] The bidding range, frequency control parameters, and corresponding delivery execution strategy parameters are encapsulated in the order of field index to form a value decision control set.

[0097] S4.3: The bid range, frequency control, and delivery execution strategy in the value decision control set are structured and encapsulated according to the field index order to form an advertising value assessment result record. The advertising value assessment result record includes the advertising plan number, the upper limit of the bid range, the lower limit of the bid range, the number of frequency control times, and the delivery execution strategy type. The advertising value score after budget constraint adjustment is written into the value score field of the advertising value assessment result record to obtain the advertising value assessment result.

[0098] This embodiment also provides a computer device applicable to the advertising value assessment method based on machine learning models, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the advertising value assessment method based on machine learning models as proposed in the above embodiment.

[0099] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0100] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the advertising value assessment method based on a machine learning model as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0101] In summary, this invention addresses the problem of fragmented multi-domain feature modeling in existing technologies by constructing a unified value assessment input package and inputting it into a multi-domain game-aware machine learning model to achieve joint embedding and coupled computation of cross-domain elements. Simultaneously, it constructs a user's multi-touchpoint time evolution journey through multi-touchpoint time attribution computation, quantifying the marginal contribution of each advertising touchpoint to the conversion time, thus expanding value assessment from simply predicting whether a conversion will occur to a fine-grained characterization of the impact on conversion time, enhancing its temporal expression capabilities. Furthermore, it embeds compliance constraint information and performs contribution reduction processing during the contribution calculation stage, allowing compliance factors to directly participate in the core value calculation process, thereby achieving the synchronous integration of effect evaluation and risk control, and improving the accuracy, temporal awareness, and compliance reliability of advertising value assessment.

[0102] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating advertising value based on a machine learning model, characterized in that: include, When an ad request enters the ad transaction chain, the display opportunity is abstracted into a tradable value object, and compliance awareness feature encoding processing is performed on the ad-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements to generate a unified value assessment input package. The unified value assessment input package is input into the multi-domain game perception machine learning model. The user's multi-touch time evolution journey is constructed through multi-touch time attribution calculation, and the marginal contribution of each advertising touchpoint to the conversion time is quantified. At the same time, contribution reduction processing is performed based on the compliance constraint information in the unified value assessment input package, and the journey-level incremental contribution summary result is output. The multi-domain game perception machine learning model includes a cross-domain feature embedding layer, a touchpoint time series modeling layer, a bidding game coupling layer, and a value decision output layer. The construction of the user's multi-touchpoint time evolution journey specifically includes: The unified value assessment input package is input into the cross-domain feature embedding layer. The advertising-side delivery elements, user-side real-time intent elements, and market-side bidding environment elements are respectively processed by the advertising-side feature mapping center, user-side feature mapping center, and bidding environment feature mapping center to output feature vector output results. The feature vector output results are then processed by vector concatenation according to the field index order to form a cross-domain joint feature vector. The cross-domain joint feature vector is used as the input data for the touchpoint time series modeling layer. At the same time, the advertising touchpoint log data in the unified value assessment input package is input into the touchpoint time series modeling layer. Touchpoint event identification and touchpoint type labeling are performed on the advertising touchpoint log data to form an advertising touchpoint sequence set. The ad touchpoint sequence set is ordered according to the timestamp of the touchpoint occurrence, forming a time-ordered touchpoint sequence; Based on the time interval between each advertising touchpoint and the conversion touchpoint in the time-ordered touchpoint sequence, a temporal position quantization calculation is performed to form a touchpoint temporal position weight sequence. The touchpoint temporal position weight sequence is then weighted and fused with the cross-domain joint feature vector to obtain the cross-domain corrected touchpoint temporal position weight sequence. Based on the time interval between adjacent advertising touchpoints in the time-ordered touchpoint sequence, a time decay propagation calculation is performed, and the decay calculation result is combined with the cross-domain corrected touchpoint temporal position weight sequence to perform a touchpoint propagation decreasing calculation to form the touchpoint temporal influence result. The influence duration interval modeling process is performed by combining the touchpoint type labeling results of each ad touchpoint in the ad touchpoint sequence set, and the modeling results are encapsulated into a journey structure to form a user multi-touchpoint time evolution journey. The summaries of journey-level incremental contributions are coupled with market-side bidding environment factors to generate a game-corrected net transaction value object. Based on the game-theoretic corrected net value of the transaction, the system combines advertising-side placement factors with real-time user intent factors to perform decision optimization calculations, generate advertising value assessment results, and output corresponding bidding ranges, frequency control, and placement execution strategies.

2. The advertising value assessment method based on a machine learning model as described in claim 1, characterized in that: The generation of the unified value assessment input package specifically includes: Feature data is extracted from advertising-side placement elements, user-side real-time intent elements, and market-side bidding environment elements to form a multi-dimensional feature set. Compliance awareness feature encoding processing is then performed on the multi-dimensional feature set to generate compliance constraint information. After encoding and processing the feature data and compliance constraint information, they are encapsulated to form a unified value assessment input package.

3. The advertising value assessment method based on a machine learning model as described in claim 1, characterized in that: The output journey-level incremental contribution summary result is as follows: Extract the set of advertising touchpoint sequences from the user's multi-touchpoint time evolution journey, input them into the touchpoint time series modeling layer, and perform touchpoint effect interval analysis and conversion time correlation positioning processing on the set of advertising touchpoint sequences to form a set of touchpoint time impact calculations; Based on the results of the touchpoint timing impact, the conversion time offset of each ad touchpoint in the touchpoint timing impact calculation set is quantitatively calculated, and the incremental effect value of each ad touchpoint on the conversion time is decomposed to form a touchpoint marginal contribution set. The set of marginal contributions at touchpoints is associated with the compliance constraint information in the unified value assessment input package, and contribution reduction processing is performed on the set of marginal contributions at touchpoints. The reduced set of marginal contributions at touchpoints is then summarized and encapsulated to form a summary result of incremental contributions at the journey level.

4. The advertising value assessment method based on a machine learning model as described in claim 1, characterized in that: The generated game-corrected net value object is specifically: Extract the touchpoint marginal contribution set from the trip-level incremental contribution summary result, input it into the bidding game coupling layer, and perform conversion revenue quantification mapping calculation on the touchpoint marginal contribution set to form the touchpoint revenue calculation set; By integrating the revenue calculation set of touchpoints with market-side bidding environment factors, and performing bidding impact correction calculations, a revenue game adjustment set is formed. Summarize the adjusted set of returns from the game and generate a game-corrected net asset value (NAV) object.

5. The advertising value assessment method based on a machine learning model as described in claim 1, characterized in that: The generated advertising value assessment results are specifically as follows: Extract the set of return assessment parameters from the game-corrected net value of the transaction object, input it into the value decision output layer, perform value quantification calculation on the set of return assessment parameters, and form the advertising value calculation set. The value calculation set is integrated with the budget constraint matching calculation of the advertising side delivery elements, and the delivery adaptation calculation is performed based on the real-time intent elements of the user side to form a value decision control set; The results of the value decision control set are encapsulated and processed to generate advertising value assessment results.

6. The advertising value assessment method based on a machine learning model as described in claim 1, characterized in that: The formation of the advertising touchpoint sequence set specifically includes: Perform log field parsing on the ad touchpoint log data to extract the touchpoint event identifiers and touchpoint timestamps corresponding to ad impression records, ad click records, and conversion feedback records. Based on the touchpoint event identifiers, the ad exposure records, ad click records, and conversion feedback records are classified by touchpoint type to generate exposure touchpoint data, click touchpoint data, and conversion touchpoint data; The exposure touchpoint data, click touchpoint data, and conversion touchpoint data are merged and ordered based on the timestamp of the touchpoint occurrence to form a set of advertising touchpoint sequences.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the advertising value evaluation method based on the machine learning model as described in any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the advertising value evaluation method based on the machine learning model as described in any one of claims 1 to 6.

Images