Cross-platform advertisement putting effect attribution analysis system

Abstract

The invention relates to the technical field of effect evaluation, in particular to a cross-platform advertisement putting effect attribution analysis system, which comprises a user behavior feature construction module for collecting behavior data of users in an advertisement putting platform and constructing a user feature map by using a user identity fuzzy matching algorithm; the user behavior feature analysis module is used for determining a user behavior path set after advertisement putting based on the constructed user feature map and a time sequence algorithm; the advertisement putting effect evaluation module is used for comprehensively evaluating the advertisement putting effect based on a hierarchical evaluation algorithm according to the determined user behavior path set; according to the method, a second-generation path set is generated by combining path disturbance through a path construction mechanism based on a time sequence, so that the effect of comprehensively restoring, optimizing and reconstructing a real conversion path of a user is realized; and the generated paths are screened by setting a constraint equation, so that the effects of removing the paths with unreasonable behavior structures and improving the effectiveness of the evaluation paths are achieved.

Description

Technical Field

[0001] This invention relates to the field of performance evaluation technology, and more specifically to a cross-platform advertising performance attribution analysis system. Background Technology

[0002] In the current technological landscape, advertising has become an integral part of people's lives, participating in all aspects of social life and changing people's lifestyles and consumption habits. How to effectively place advertisements has become a research goal for an increasing number of companies, as successfully targeting specific demographics can significantly enhance their results. However, with the ever-increasing amount of money spent on advertising, the effectiveness and conversion rate are often not guaranteed, and the following challenges remain:

[0003] The inability to accurately identify whether cross-platform behavior belongs to the same user leads to incomplete or distorted attribution results.

[0004] The failure to fully consider the synergistic effect of multiple behavioral nodes in the user behavior path, focusing only on the last click behavior, affected the accuracy of the assessment.

[0005] The lack of a tiered evaluation mechanism makes it impossible to refine the contribution of different behavioral types in the evaluation path to the final conversion. Summary of the Invention

[0006] In view of the above-mentioned problems, the present invention is proposed.

[0007] To address the aforementioned technical problems, this invention provides the following technical solution: a cross-platform advertising performance attribution analysis system, comprising a user behavior feature construction module, which collects user behavior data from the advertising platform and constructs a user feature map using a user identity fuzzy matching algorithm, specifically:

[0008] The corresponding fuzzy feature vectors are constructed using a fuzzy matching algorithm for user identity, and the probability score of two fuzzy feature vectors belonging to the same customer is calculated to determine the user's behavior records across all platforms, thereby constructing a user behavior feature map.

[0009] The user behavior feature analysis module, based on the constructed user feature map, determines the set of user behavior paths after ad delivery using a time-series algorithm. Specifically:

[0010] An initial set of behavioral paths is determined based on the temporal connection order of behavioral events in the user behavior feature map, and a first-generation set of behavioral paths is determined by weighted scoring.

[0011] Based on the path perturbation mechanism, the first-generation behavior path is partially reconstructed to determine the second-generation behavior path, and the user behavior path after the ad is delivered is determined by setting constraint equations.

[0012] The advertising performance evaluation module comprehensively evaluates the advertising performance based on a hierarchical evaluation algorithm, according to a determined set of user behavior paths.

[0013] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the specific steps of collecting user behavior data from the advertising platform are as follows:

[0014] By collecting user behavior data after ad delivery from the databases of multiple advertising platforms and constructing corresponding datasets, we can obtain the following:

[0015] U = {U1, U2, ... U} M}

[0016] U i ={b i,1 ,b i,2 ,...,b i.N}

[0017] Where U represents the set of user behavior data in the advertising platform database, U1 represents the user behavior data in the first advertising platform database, U M This represents user behavior data from the database of the Mth advertising platform, where M represents the total number of advertising platforms. i U represents user behavior data from the i-th advertising platform. i,1 U represents the first behavior record of user behavior data in the data of the i-th advertising platform. i,2 This represents the Nth behavior record in the user behavior data of the i-th advertising platform, where N represents the total number of user behavior types.

[0018] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the calculation of the probability score that two fuzzy feature vectors belong to the same customer is as follows:

[0019] Arbitrarily select two fuzzy feature vectors F i and F j And by calculating the feature similarity between the two fuzzy feature vectors, we have:

[0020]

[0021] Where d represents the number of feature dimensions, which is the number of feature values ​​in the fuzzy feature vector. In this embodiment, d = 5 is used for illustration. k F represents the weight coefficient of the k-th feature dimension. i (k) Represents the fuzzy feature vector F i The k-th feature dimension in Represents the fuzzy feature vector F j The k-th feature dimension in Sim(i,j) represents the cosine similarity between the k-th feature dimension of two fuzzy feature vectors, and Sim(i,j) represents the feature similarity between two fuzzy feature vectors.

[0022] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, wherein: based on the feature similarity between two fuzzy feature vectors, the probability score that the two fuzzy feature vectors belong to the same customer is determined, then...

[0023] If a feature similarity threshold is set, and a probability score is determined based on this threshold to indicate that two fuzzy feature vectors belong to the same customer, then...

[0024] If the feature similarity between two fuzzy feature vectors exceeds the set feature similarity threshold, it means that the two fuzzy feature vectors belong to the same customer; otherwise, it means that the two fuzzy feature vectors do not belong to the same customer.

[0025] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the determination of the user behavior path set after advertising delivery based on the time-series algorithm is as follows:

[0026] The temporal connection order of behavioral events in the user behavior feature graph is set as the initial set of behavioral paths;

[0027] For each behavior path in the initial set of behavior paths, a comprehensive score is calculated for each path using a weighted scoring method;

[0028] Based on weighted scoring, the comprehensive score of all behavioral paths in the initial behavioral path set is calculated sequentially. The comprehensive scores are then sorted in descending order, and the first m behavioral paths are selected to form a first-generation behavioral path set.

[0029] For the constructed first-generation behavior path set, the behavior path is partially reconstructed based on the path perturbation mechanism, and then the second-generation behavior path is determined.

[0030] The second-generation behavior path set is constrained and limited by constraint equations, and the customer behavior path after the advertisement is placed is determined.

[0031] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the initial set of behavioral paths is specifically as follows:

[0032]

[0033] Where P0 represents the initial set of behavioral paths constructed. This represents the first behavior path in the initial set of behaviors. This represents the nth behavior path in the initial set of behavior paths;

[0034] For the i-th behavior path in the initial set of behavior paths Where, represents the i-th behavior path in the initial behavior path set, and represents the user's behavior event from b1 to b. K The complete sequence of behavioral events.

[0035] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the local reconstruction of the behavior path based on the path perturbation mechanism is specifically as follows:

[0036] Based on each behavior path in the constructed first-generation behavior path set, arbitrarily select two non-terminating behavior events in the behavior path, and swap the order of the two selected behavior events in the original behavior path.

[0037] When the order of the two selected behavior events is swapped, any behavior event node in the behavior path will be replaced with the same behavior event node from other platforms.

[0038] Once the same behavior event on other platforms has been replaced, the behavior path sequence is truncated at the middle position, and the behavior path after the truncated node is replaced with the behavior path before the node of the next behavior path. The two are then spliced ​​together to form a new behavior path.

[0039] Once all paths in the first-generation behavior path set have undergone local reconstruction, the partially reconstructed behavior paths will be rebuilt into the second-generation behavior path set.

[0040] As a preferred embodiment of the cross-platform advertising performance attribution analysis system described in this invention, the constraint limitation on the second-generation behavioral path set is specifically as follows:

[0041]

[0042] Where P′ represents any behavior path in the second-generation behavior path set, P″ represents any behavior path in the first-generation behavior path set, L(P′) represents the path length of any behavior path in the second-generation behavior path set, L(P0) represents the path length of any behavior path in the initial behavior path set, and S core (P′) represents the comprehensive score corresponding to any behavior path in the second-generation behavior path set, S core (P″) represents the overall score corresponding to any behavioral path in the set of generational behavioral paths.

[0043] As a preferred embodiment of the cross-platform advertising performance attribution analysis system of the present invention, the comprehensive evaluation of advertising performance based on the hierarchical evaluation algorithm is as follows:

[0044] Based on a defined set of user behavior paths, the number N(B) of all behavior events in the set is extracted, and the number N(b) of positive behavior events is determined from all extracted behavior events. pos The initial evaluation of advertising effectiveness is based on the proportion of positive behavioral events.

[0045] In addition, an in-depth evaluation of the effectiveness of advertising is conducted based on the proportion of different behavioral events in each behavioral path.

[0046] The beneficial effects of this invention are:

[0047] This invention introduces a fuzzy matching algorithm for user identity to construct a unified user behavior feature map, thereby achieving the effect of accurately identifying the same user behavior record across multiple advertising platforms.

[0048] By using a time-series-based path construction mechanism and combining path perturbation to generate a second-generation path set, the effect of fully restoring and optimizing the user's actual conversion path is achieved.

[0049] By setting constraint equations to filter the generated paths, the effects of removing paths with unreasonable behavioral structures and improving the effectiveness of path evaluation were achieved.

[0050] By introducing a hierarchical evaluation algorithm, multi-dimensional scoring is conducted from two dimensions: the number of behavioral events and the type of behavioral events, thus achieving a more accurate and comprehensive attribution evaluation of advertising effectiveness.

[0051] By combining path scoring and behavioral weighting in a comprehensive model, the collaborative value assessment of advertising across multiple user paths and behavioral touchpoints was achieved. Attached Figure Description

[0052] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. 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. Wherein:

[0053] Figure 1 This is a schematic diagram of the overall method steps of the cross-platform advertising performance attribution analysis system of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0055] Example 1

[0056] Reference Figure 1 The first embodiment of the present invention provides a cross-platform advertising performance attribution analysis system, including a user behavior feature construction module, a user behavior feature analysis module, and an advertising performance evaluation module.

[0057] Specifically, the user behavior feature construction module is used to collect user behavior data in the advertising platform and construct a user feature map based on the collected behavior data; the user behavior feature analysis module is used to reconstruct user behavior paths based on the constructed user feature map and determine the set of user behavior paths after advertising; the advertising effectiveness evaluation module is used to comprehensively evaluate the advertising effectiveness based on the determined set of user behavior paths.

[0058] Furthermore, the user behavior feature construction module collects user behavior data after ad delivery from multiple ad delivery platforms, and constructs a user behavior feature map based on the collected user behavior data using a fuzzy matching algorithm for user identity. The specific implementation is as follows:

[0059] By collecting user behavior data after ad delivery from the databases of multiple advertising platforms and constructing corresponding datasets, we can obtain the following:

[0060] U = {U1, U2, ... U} M}

[0061] U i ={b i,1 ,b i,2 ,...,b i.N}

[0062] Where U represents the set of user behavior data in the advertising platform database, U1 represents the user behavior data in the first advertising platform database, U M This represents user behavior data from the database of the Mth advertising platform, where M represents the total number of advertising platforms. i U represents user behavior data from the i-th advertising platform. i,1U represents the first behavior record of user behavior data in the data of the i-th advertising platform. i,2 This represents the Nth behavior record in the user behavior data of the i-th advertising platform, where N represents the total number of user behavior types.

[0063] It should be noted that because users use different accounts on different platforms, it is difficult to uniformly identify customers in the constructed user behavior data set. Therefore, a fuzzy matching algorithm for user identity is introduced to determine a unified identifier for user identity across different platforms. The specific implementation is as follows:

[0064] For each user behavior record in the user behavior dataset, constructing a corresponding fuzzy feature vector, we have:

[0065] F = [f dev ,f ip ,f cookie ,f geo ,f time ]

[0066] Where F represents the fuzzy feature vector corresponding to the user behavior record, f dev fi represents the device fingerprint in user behavior records. p fc represents the public IP address in the user behavior record. ookie f represents the local session identifier in the user behavior log. geo f represents the geographic location information in user behavior records. time This represents the timestamp of a user's behavior in their activity log.

[0067] Based on the constructed fuzzy feature vectors, the probability score of two fuzzy feature vectors belonging to the same customer is determined, specifically as follows:

[0068] Arbitrarily select two fuzzy feature vectors F i and F j， And by calculating the feature similarity between the two fuzzy feature vectors, we have:

[0069]

[0070] Where d represents the number of feature dimensions, which is the number of feature values ​​in the fuzzy feature vector. In this embodiment, d = 5 is used for illustration. k The weight coefficient for the k-th feature dimension is set by the implementer based on the actual application scenario. F i (k) Represents the fuzzy feature vector F i The k-th feature dimension in Represents the fuzzy feature vector F j The k-th feature dimension in Sim(i,j) represents the cosine similarity between the k-th feature dimension of two fuzzy feature vectors, and represents the feature similarity between the two fuzzy feature vectors. It is used to determine the probability score that the two fuzzy feature vectors belong to the same customer. Specifically:

[0071] Set the feature similarity threshold Sim T Given (i,j), and based on the set feature similarity threshold, determine the probability score that two fuzzy feature vectors belong to the same customer, then we have,

[0072] If the feature similarity between two fuzzy feature vectors satisfies the formula Sim(i,j)≥Sim T When (i,j), it means that the two fuzzy feature vectors belong to the same customer; otherwise, it means that the two fuzzy feature vectors do not belong to the same customer.

[0073] It should be noted that by traversing all fuzzy feature vectors and determining each user's behavior records across all platforms, and then arranging all user behavior records in ascending chronological order to determine the user behavior sequence, and simultaneously constructing a user behavior feature map, we have:

[0074] G = (B, E)

[0075] Where G represents the constructed user behavior feature map, B represents user behavior events, and E represents the temporal connection order of user behavior events in the user behavior feature map.

[0076] It should be noted that user behavior event B is a behavior event determined from all platforms for the current user; the temporal sequence E of the behavior events is the temporal sequence of the behavior events triggered after the user is targeted by the advertisement.

[0077] Furthermore, the user behavior feature analysis module reconstructs the entire path of user behavior records based on the constructed user feature map and in chronological order, and applies conditional constraints to the reconstructed behavior path based on constraint equations. The specific implementation is as follows:

[0078] If we define the temporal connection order of behavioral events in the user behavior feature graph as the initial set of behavioral paths, then we have:

[0079]

[0080] Where P0 represents the initial set of behavioral paths constructed. This represents the first behavior path in the initial set of behaviors. This represents the nth behavior path in the initial set of behavior paths;

[0081] For the i-th behavior path in the initial set of behavior paths Where, represents the i-th behavior path in the initial behavior path set, and represents the user's behavior event from b1 to b. K The complete sequence of behavioral events.

[0082] For each behavior path in the initial set of behavior paths, a comprehensive score is calculated for each path using a weighted scoring method. Then, we have...

[0083]

[0084] Where j represents the sequence number of the behavior event in each behavior path, K represents the total number of behavior events in each behavior path, and α j S represents the weight coefficient corresponding to the j-th behavior event in the behavior path, which is set by the implementer according to the actual application scenario. core (P) represents the overall score for each behavioral path;

[0085] Based on weighted scoring, the comprehensive score of all behavioral paths in the initial behavioral path set is calculated sequentially. The comprehensive scores are then sorted in descending order, and the first m behavioral paths are selected to form the first generation behavioral path set P1.

[0086] For the constructed first-generation behavior path set, a path perturbation mechanism is used to partially reconstruct the behavior paths, thereby determining the second-generation behavior paths, specifically:

[0087] Based on each behavior path in the constructed first-generation behavior path set P1, arbitrarily select two non-terminating behavior events in the behavior path, and swap the order of the two selected behavior events in the original behavior path.

[0088] When the order of the two selected behavior events is swapped, any behavior event node in the behavior path will be replaced with the same behavior event node from other platforms.

[0089] Once the same behavior event on other platforms has been replaced, the behavior path sequence is truncated at the middle position, and the behavior path after the truncated node is replaced with the behavior path before the node of the next behavior path. The two are then spliced ​​together to form a new behavior path.

[0090] Once all paths in the first-generation behavior path set have undergone local reconstruction, the partially reconstructed behavior paths will be rebuilt into the second-generation behavior path set P2.

[0091] It should be noted that, in order to ensure that the second-generation behavior path set is superior to the first-generation behavior path set, constraint equations are used to constrain and limit the second-generation behavior path set. The specific constraint equations are as follows:

[0092]

[0093] Where P′ represents any behavior path in the second-generation behavior path set, P″ represents any behavior path in the first-generation behavior path set, L(P′) represents the path length of any behavior path in the second-generation behavior path set, L(P0) represents the path length of any behavior path in the initial behavior path set, and S core (P′) represents the comprehensive score corresponding to any behavior path in the second-generation behavior path set, S core (P″) represents the overall score corresponding to any behavioral path in the set of generational behavioral paths.

[0094] It should be noted that the second-generation behavior path determined under the constraints of the constraint equation is the user behavior path after the ad is placed, and the set of second-generation behavior paths is sent to the ad performance evaluation module for comprehensive evaluation of the ad performance.

[0095] Furthermore, the advertising performance evaluation module, based on a determined set of user behavior paths and a hierarchical evaluation algorithm, evaluates advertising performance at different levels to achieve a comprehensive assessment of advertising effectiveness. The specific evaluation is as follows:

[0096] Based on the characteristics of behavioral events in the defined user behavior path set P2, a comprehensive evaluation of the advertising effectiveness is conducted, specifically as follows:

[0097] Based on the expected advertising results, the behavioral events in each path of the user behavior path set are divided into positive behavioral events b. pos Reverse behavior event b rev and events with no impact on behavior b imp ;

[0098] Based on the characteristics of behavioral events in the user behavior path, a tiered evaluation mechanism is used to comprehensively evaluate the effectiveness of ad placement. The specific evaluation is as follows:

[0099] The initial evaluation of ad performance is based on the proportion of positive behavioral events in the entire user behavior path set, specifically:

[0100] Based on a defined set of user behavior paths, the number N(B) of all behavior events in the set is extracted, and the number N(b) of positive behavior events is determined from all extracted behavior events. pos Based on the proportion of positive behavioral events, the initial evaluation of advertising effectiveness is as follows:

[0101] After setting the ad placement criteria, the baseline percentage threshold P for positive behavioral events is determined. T (b posAnd calculate the percentage of positive behavior events in the determined set of user behavior paths, then we have,

[0102]

[0103] Where N(B) represents the total number of behavioral events in the user behavior path set, N(b pos P(b) represents the number of positive behavior events in the user behavior path set. pos This represents the percentage of positive behavioral events in the user behavior path set, used for the initial evaluation of advertising effectiveness.

[0104] If the ratio of positive behavior events in the calculated user behavior path set satisfies the formula P(b) compared to the set baseline threshold for positive behavior events... pos ) < P T (b pos If the current ad performance is lower than the baseline expected performance, the ad delivery time and frequency will be readjusted, and the user behavior path will be redefined until the proportion of positive behavior events in the user behavior path set exceeds the set baseline threshold for positive behavior events. The proportion of positive behavior events exceeding the set baseline threshold will be used as the initial evaluation result of the ad performance.

[0105] Based on the proportion of positive behavioral events in each behavioral path, an in-depth evaluation of the advertising effectiveness is conducted, specifically:

[0106] Calculate the percentage of positive events P(b) in each behavior path of the user behavior path set. pos ), the proportion of reverse behavior events P(b) rev ) and the percentage of behaviors with no impact P(b) imp );

[0107] Based on the proportion of different behavioral events in each behavioral path, an in-depth evaluation of the advertising effectiveness is conducted, resulting in:

[0108] For each behavioral path, comparing the percentages of positive events, negative events, and events with no impact, we find that...

[0109] If the highest percentage of behavioral events in a behavioral path is the percentage of positive behavioral events, then the current path is a positive behavioral path.

[0110] If the highest percentage of behavioral events in a behavioral path is the percentage of reverse behavioral events, then the current path is a reverse behavioral path.

[0111] If the highest percentage of behavioral events in a behavioral path is the percentage of insignificant behavioral events, then the current path is an insignificant behavioral path.

[0112] Based on the defined behavioral event paths, an in-depth evaluation of the advertising effectiveness is conducted, specifically:

[0113] For the user behavior path set, extract the behavior path after ad delivery, and perform in-depth evaluation based on the behavior path after ad delivery. Then, we have...

[0114] If the user's behavior path after the ad is delivered is a positive behavior path, then the ad delivery effect is positive.

[0115] If, after an ad is delivered, the user's action path is the reverse action path, then the ad delivery effect is the opposite.

[0116] If the user's behavior path after the ad is displayed is an inconsequential behavior path, then the ad display effect is considered to be non-impactful.

[0117] It should be noted that positive behavioral events are user-triggered behavioral events after the ad placement node, which are ad-guided effect events, including clicking on the ad-related product and purchasing the ad-related product; negative behavioral events are user-triggered behavioral events after the ad placement node, which are ad-reverse guided effect events, including leaving the ad-related product page and canceling the purchase of the ad-related product; and non-impact behavioral events are user-triggered behavioral events after the ad placement node, which are ad-ineffective events, including maintaining the original page browsing and opening a new page unrelated to the ad.

[0118] Furthermore, if the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0119] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0120] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0121] 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 the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cross-platform advertising performance attribution analysis system, characterized in that: This includes a user behavior feature construction module, which collects user behavior data from the advertising platform and uses a fuzzy matching algorithm to construct a user feature map, specifically: The corresponding fuzzy feature vectors are constructed using a fuzzy matching algorithm for user identity, and the probability score of two fuzzy feature vectors belonging to the same customer is calculated to determine the user's behavior records across all platforms, thereby constructing a user behavior feature map. The user behavior feature analysis module, based on the constructed user feature map, determines the set of user behavior paths after ad delivery using a time-series algorithm. Specifically: An initial set of behavioral paths is determined based on the temporal connection order of behavioral events in the user behavior feature map, and a first-generation set of behavioral paths is determined by weighted scoring. Based on the path perturbation mechanism, the first-generation behavior path is partially reconstructed to determine the second-generation behavior path, and the user behavior path after the ad is delivered is determined by setting constraint equations. The ad performance evaluation module comprehensively evaluates ad performance based on a hierarchical evaluation algorithm, using a defined set of user behavior paths.

2. The cross-platform advertising performance attribution analysis system according to claim 1, characterized in that, The specific details of collecting user behavior data from the advertising platform are as follows: By collecting user behavior data after ad delivery from the databases of multiple advertising platforms and constructing corresponding datasets, we can obtain the following: U={U1,U2,...U M } U i ={b i,1 ,b i,2 ,...,b i.N } Where U represents the set of user behavior data in the advertising platform database, U1 represents the user behavior data in the first advertising platform database, U M This represents user behavior data from the database of the Mth advertising platform, where M represents the total number of advertising platforms. i U represents user behavior data from the i-th advertising platform. i,1 U represents the first behavior record of user behavior data in the data of the i-th advertising platform. i,2 This represents the Nth behavior record in the user behavior data of the i-th advertising platform, where N represents the total number of user behavior types.

3. The cross-platform advertising performance attribution analysis system according to claim 2, characterized in that, The specific calculation of the probability score that two fuzzy feature vectors belong to the same customer is as follows: Arbitrarily select two fuzzy feature vectors F i and F j And by calculating the feature similarity between the two fuzzy feature vectors, we have: Where d represents the number of feature dimensions, which is the number of feature values ​​in the fuzzy feature vector. In this embodiment, d = 5 is used for illustration. k F represents the weight coefficient of the k-th feature dimension. i (k) Represents the fuzzy feature vector F i The k-th feature dimension in Represents the fuzzy feature vector F j The k-th feature dimension in Sim(i,j) represents the cosine similarity between the k-th feature dimension of two fuzzy feature vectors, and Sim(i,j) represents the feature similarity between two fuzzy feature vectors.

4. The cross-platform advertising performance attribution analysis system according to claim 3, characterized in that, Based on the feature similarity between two fuzzy feature vectors, the probability score of the two fuzzy feature vectors belonging to the same customer is determined, then we have: If a feature similarity threshold is set, and a probability score is determined based on this threshold to indicate that two fuzzy feature vectors belong to the same customer, then... If the feature similarity between two fuzzy feature vectors exceeds the set feature similarity threshold, it means that the two fuzzy feature vectors belong to the same customer; otherwise, it means that the two fuzzy feature vectors do not belong to the same customer.

5. The cross-platform advertising performance attribution analysis system according to claim 4, characterized in that, The specific steps for determining the set of user behavior paths after ad delivery based on the time-series algorithm are as follows: The temporal connection order of behavioral events in the user behavior feature graph is set as the initial set of behavioral paths; For each behavior path in the initial set of behavior paths, a comprehensive score is calculated for each path using a weighted scoring method; Based on weighted scoring, the comprehensive score of all behavioral paths in the initial behavioral path set is calculated sequentially. The comprehensive scores are then sorted in descending order, and the first m behavioral paths are selected to form a first-generation behavioral path set. For the constructed first-generation behavior path set, the behavior path is partially reconstructed based on the path perturbation mechanism, and then the second-generation behavior path is determined. The second-generation behavior path set is constrained and limited by constraint equations, and the customer behavior path after the advertisement is placed is determined.

6. The cross-platform advertising performance attribution analysis system according to claim 5, characterized in that, The initial set of behavioral paths is as follows: Where P0 represents the initial set of behavioral paths constructed. This represents the first behavior path in the initial set of behaviors. This represents the nth behavior path in the initial set of behavior paths; For the i-th behavior path in the initial set of behavior paths Where, represents the i-th behavior path in the initial behavior path set, and represents the user's behavior event from b1 to b. K The complete sequence of behavioral events.

7. The cross-platform advertising performance attribution analysis system according to claim 6, characterized in that, The local reconstruction of the behavior path based on the path perturbation mechanism is as follows: Based on each behavior path in the constructed first-generation behavior path set, arbitrarily select two non-terminating behavior events in the behavior path, and swap the order of the two selected behavior events in the original behavior path. When the order of the two selected behavior events is swapped, any behavior event node in the behavior path will be replaced with the same behavior event node from other platforms. Once the same behavior event on other platforms has been replaced, the behavior path sequence is truncated at the middle position, and the behavior path after the truncated node is replaced with the behavior path before the node of the next behavior path. The two are then spliced ​​together to form a new behavior path. Once all paths in the first-generation behavior path set have undergone local reconstruction, the partially reconstructed behavior paths will be rebuilt into the second-generation behavior path set.

8. The cross-platform advertising performance attribution analysis system according to claim 7, characterized in that, The specific constraints on the second-generation behavior path set are as follows: Where P′ represents any behavior path in the second-generation behavior path set, P″ represents any behavior path in the first-generation behavior path set, L(P′) represents the path length of any behavior path in the second-generation behavior path set, L(P0) represents the path length of any behavior path in the initial behavior path set, and S core (P′) represents the comprehensive score corresponding to any behavior path in the second-generation behavior path set, S core (P″) represents the overall score corresponding to any behavioral path in the set of generational behavioral paths.

9. The cross-platform advertising performance attribution analysis system according to claim 8, characterized in that, The specific details of the comprehensive evaluation of advertising effectiveness based on the hierarchical evaluation algorithm are as follows: Based on a defined set of user behavior paths, the number N(B) of all behavior events in the set is extracted, and the number N(b) of positive behavior events is determined from all extracted behavior events. pos The initial evaluation of advertising effectiveness is based on the proportion of positive behavioral events. In addition, an in-depth evaluation of the effectiveness of advertising is conducted based on the proportion of different behavioral events in each behavioral path.

Images