Multi-source attribution analysis method and system for cross-platform heterogeneous data flow

Abstract

The application discloses a multi-source attribution analysis method and system for cross-platform heterogeneous data streams and relates to the technical field of cross-platform data attribution analysis. The application comprises a data acquisition unit, a user alignment unit, a touch sequence construction unit, a weight calculation unit and an attribution summary unit. Through a multi-dimensional weight model of platform confidence factors, time decay factors and channel basic weight factors, the application solves the noise interference and distribution deviation problems in cross-platform heterogeneous data attribution. A user alignment mechanism based on the degree of set overlap can flexibly cope with multi-identification systems and improve the coherence and accuracy of cross-platform user behavior journeys. An adaptive updating module is also introduced to dynamically feedback platform quality changes using a sliding window to continuously optimize the stability and timeliness of the attribution model. The design of sequential decay factors and cross-platform synergy factors depicts the synergy effect of dense touch redundancy and platform crossing, making the multi-touch attribution distribution more consistent with the real conversion path.

Description

Technical Field

[0001] This invention relates to the field of cross-platform data attribution analysis technology, specifically to a multi-source attribution analysis method and system for cross-platform heterogeneous data streams. Background Technology

[0002] In a multi-channel digital marketing environment, businesses typically advertise simultaneously on multiple heterogeneous data platforms such as search engines, social media, and e-commerce platforms. Users may be exposed to these ads multiple times across different platforms before completing a conversion. However, the field formats of event data vary across platforms, and the user identification systems used (such as device ID, email, and phone number) are not uniform, resulting in highly fragmented user behavior data. This makes it difficult to accurately associate cross-platform touchpoints with the same user. Existing cross-platform user alignment solutions often rely on precise matching of a single identifier. However, when faced with overlapping or missing identifier sets, this approach lacks effective quantitative assessment and threshold determination, making it difficult to balance matching accuracy and recall, directly hindering the reliability of subsequent attribution.

[0003] In conversion attribution, the core flaw of traditional rule-based models lies in their weighting mechanism's failure to effectively differentiate the reliability of different data sources and the inherent connections between touchpoints. First, they treat touchpoints from different data platforms as equal, failing to consider the differences in data quality across platforms. This results in touchpoints from platforms with severe field deficiencies or high matching error rates receiving the same attribution weight as high-quality touchpoints, easily introducing noise into the attribution results. Second, while some solutions introduce channel-based weights based on historical conversion statistics, these are mostly statically set and fail to reflect the actual changes in the contribution of each channel in a specific conversion path, and still do not incorporate the crucial dimension of platform credibility. Furthermore, existing methods do not adequately characterize the internal temporal structure of touchpoint sequences: they generally use decay functions that only consider the time interval between touchpoints and conversions, ignoring the potential redundancy when the time interval between adjacent touchpoints is too short, easily overestimating the independent contribution of dense exposure. In addition, users' behavior of moving from one platform to another usually reflects a stronger proactive intention; this cross-platform synergy has not been fully recognized and quantified in traditional attribution.

[0004] Therefore, there is an urgent need for a multi-source attribution scheme that can at least overcome the aforementioned fundamental defects, namely: to perform cross-platform user alignment based on identifier overlap, and to calculate multi-dimensional attribution weights by comprehensively considering platform data credibility, time decay, and channel historical contributions, so as to improve the accuracy and rationality of attribution allocation. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a multi-source attribution analysis method and system for cross-platform heterogeneous data streams, solving the problems mentioned in the background section.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a multi-source attribution analysis system for cross-platform heterogeneous data streams, comprising: The data acquisition unit is equipped with data adapters corresponding to the API interfaces of various data platforms. The data adapters have a built-in field mapping rule engine, which is used to dynamically map and convert the raw heterogeneous event data pushed by various data platforms with their own field formats into a unified standardized event data format for real-time acquisition. The data structure of the unified standardized event data format includes a set of identifiers for identifying users, event type, associated channel tags, and timestamp fields for the event occurrence. The user alignment unit is used to standardize event data from different data platforms and calculate the matching score of users based on the degree of overlap between the identifier sets of corresponding event data from different data platforms. Cross-platform events with matching scores greater than or equal to a preset matching threshold are determined to belong to the same user, and a unified user identifier is assigned to cross-platform events determined to belong to the same user. The touchpoint sequence construction unit is used to identify conversion events. A conversion event is event data that matches the preset set of conversion event types in the event type field. For each conversion event, all non-conversion events with the same unified user identifier as the conversion event within a preset time window are traced back from the time the conversion event occurred as touchpoints, and a corresponding cross-platform touchpoint sequence is constructed in chronological order. The weighting calculation unit is used to calculate the attribution weight for each touchpoint in the touchpoint sequence for each conversion event. The attribution weight is determined based on the product of at least the following three factors: The system includes a platform confidence factor reflecting the credibility of the data, a time decay factor that decays over time, and a channel base weight factor that is pre-set based on the historical conversion contribution of each channel. The calculated attribution weights are then normalized so that the sum of the attribution weights corresponding to the touchpoints of the same conversion event is 1. The attribution aggregation unit is used to allocate the preset attribution value associated with each conversion event to the corresponding channel based on the normalized attribution weight, and to aggregate the total attribution value obtained from each channel.

[0007] As a further aspect of the present invention, the user alignment unit calculates the user's matching score in the following way: Select two data platforms, denoted as data platform a and data platform b. For event data from data platform a... Event data from data platform b User matching score ,in, and events and The set of user identifiers, || represents the number of elements in the set, and r and s are the event data indices in the two data platforms, respectively.

[0008] As a further aspect of this invention, the platform confidence factor The calculation method is as follows Where p = 1, 2, ..., P, P is the total number of data platforms. The basic quality score for data platform p is determined by the product of the field completeness rate and the historical event matching accuracy rate of the corresponding data on that data platform.

[0009] As a further aspect of this invention, the method also includes an adaptive update module, used to periodically and dynamically update the platform confidence factor and provide the updated platform confidence factor to the weight calculation unit. Specifically, the adaptive update module establishes a sliding window for conversion events. Whenever a predetermined number of conversion events are attributed, the quality score of each data platform is updated, and the platform confidence factor is also updated. The specific method is as follows: The average attribution weight U of the data platform p within the window is calculated. p And the proportion of direct conversion observed as the final point of contact G p ; Among them, the average attribution weight ratio U p The ratio of the sum of the attribution weights of all conversion events within the conversion event sliding window that are attributed to the data platform p to the total sum of the attribution weights of all touchpoints. Direct conversion observation percentage G p The proportion of the number of conversion events belonging to the data platform p at the last touchpoint within the conversion event sliding window to the total number of conversion events within the window; According to the formula Calculate the updated quality scores for all data platforms. ,in, The preset update step size, Quality scores before updates across all data platforms; Quality scores after updates across all data platforms. The platform confidence factor for each data platform was recalculated.

[0010] As a further aspect of the present invention, the attribution weight calculated by the weight calculation unit for touchpoint i is also based on the order decay factor. The attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor and the order decay factor, and then normalized. The sequential decay factor is set as follows: for the first contact in the contact sequence, the sequential decay factor is set to 1. For contacts with i ≥ 2, the sequential decay factor is a time decay function of the time interval between the current contact and the previous contact, such that the shorter the time interval, the smaller the value of the sequential decay factor.

[0011] As a further aspect of the present invention, the attribution weight calculated by the weight calculation unit for touch point i is also based on the cross-platform collaborative factor. The attribution weight is the product of the platform confidence factor, time decay factor, channel basic weight factor and cross-platform collaboration factor. When the attribution weight calculation of touchpoint i is also based on the sequential decay factor, the product also includes the sequential decay factor, and then normalization is performed. The cross-platform collaboration factor is set as follows: for the first contact in the contact sequence, the cross-platform collaboration factor is set to 1. For touchpoints with i≥2, if the data platform to which the touchpoint belongs is different from the data platform to which the previous touchpoint belongs, the cross-platform collaboration factor is greater than 1; if the data platforms are the same, the cross-platform collaboration factor is equal to 1.

[0012] As a further aspect of the present invention, the time decay function is a reciprocal decay function, specifically in the form of: ,in, The time interval during which the contact distance conversion occurs. This is the preset positive attenuation coefficient.

[0013] A multi-source attribution analysis method for cross-platform heterogeneous data streams, implemented through a cross-platform heterogeneous data stream multi-source attribution analysis system, includes the following steps: S1. Collect event data streams in real time from at least two different data platforms. The event data from each data platform has its own field format, but all of them contain a set of identifiers for identifying users, event type, associated channel tags, and timestamps of the event occurrence. S2. Standardize the event data from different data platforms, and calculate the user matching score based on the degree of overlap between the identifier sets of events from different data platforms. Determine that cross-platform events with matching scores greater than or equal to the preset matching threshold belong to the same user, and assign a unified user identifier to cross-platform events determined to belong to the same user. S3. Identify conversion events. A conversion event is an event whose event type field matches a preset set of conversion event types. For each conversion event, trace back from the time the conversion event occurred to all non-conversion events within a preset time window that have the same unified user identifier as the conversion event as touchpoints, and construct a corresponding cross-platform touchpoint sequence in chronological order. S4. Calculate the product of the platform confidence factor, time decay factor, and channel basic weight factor, and normalize the calculation results so that the sum of the touchpoint weights of the same conversion event is 1, and calculate the attribution weight of each touchpoint in the touchpoint sequence of each conversion event. S5. Allocate the preset value associated with each conversion event to the corresponding channel according to the normalized weight, and summarize the total attribution value obtained from each channel.

[0014] As a further aspect of the present invention, the platform confidence factor is periodically and dynamically updated based on the real-time performance of the corresponding data on the data platform. The periodic dynamic update method is as follows: Establish a sliding window for conversion events. After a predetermined number of conversion events have been attributed, calculate the average attribution weight U of each data platform within the window. p And the proportion of direct conversion observed as the final point of contact G p ; According to the formula Calculate the updated quality scores for all data platforms. ,in, The preset update step size, Quality scores before updates across all data platforms; Quality scores after updates across all data platforms. The platform confidence factor for each data platform was recalculated.

[0015] As a further aspect of the present invention, when calculating the attribution weight for contact i, a sequential decay factor and / or a cross-platform synergy factor are also introduced into the product. When the sequential decay factor is introduced, the attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor and the sequential decay factor, and then normalized. When a cross-platform collaboration factor is introduced, the attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor, and the cross-platform collaboration factor, and then normalized. When both the sequential decay factor and the cross-platform collaboration factor are introduced, the attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor, the sequential decay factor, and the cross-platform collaboration factor, and then normalized. The sequential decay factor is configured as follows: the first contact in the contact sequence is 1, and the sequential decay factor of the contacts with i≥2 in the contact sequence is a monotonically decreasing function of the time interval between the contact and the previous contact. The cross-platform collaboration factor is configured as follows: the first touch point in the touch point sequence has a value of 1. For touch points with i≥2 in the touch point sequence, if their data platform is different from that of the previous touch point, the cross-platform collaboration factor is greater than 1; otherwise, it is equal to 1.

[0016] This invention provides a method and system for multi-source attribution analysis of cross-platform heterogeneous data streams. Compared with existing technologies, it has the following advantages: The data acquisition unit collects heterogeneous event data from multiple platforms in real time; the user alignment unit uses Jaccard similarity and threshold judgment to associate cross-platform events, enabling the attribution process to build a complete touchpoint sequence based on a unified user identifier, thereby overcoming the limitation of precise matching of a single identifier.

[0017] The weight calculation unit multiplies and normalizes the platform confidence factor, time decay factor, and channel basic weight factor. The platform confidence factor is determined by the field completeness rate and historical matching accuracy rate, so that low-quality platform touchpoints are assigned lower attribution weights, reducing noise interference.

[0018] The adaptive update module compares the average attribution weight ratio of the data platform with the direct conversion observation ratio through a sliding window, dynamically adjusts the platform quality score and confidence factor, and realizes online updates of the attribution model.

[0019] The sequential decay factor penalizes the time interval between adjacent touchpoints, suppressing over-attribution of short-duration repeated touchpoints; the cross-platform synergy factor assigns enhanced weights to adjacent touchpoints across platforms, reflecting the synergistic effect of multiple channels. The time decay function adopts a reciprocal form to ensure that early touchpoints still have a reasonable share. Attached Figure Description

[0020] Figure 1 This is a system block diagram of the multi-source attribution analysis method and system for cross-platform heterogeneous data streams of the present invention.

[0021] Figure 2 This is a flowchart illustrating the multi-source attribution analysis method and system for cross-platform heterogeneous data streams of the present invention. Detailed Implementation

[0022] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 and Figure 2 As shown, the embodiments of the present invention provide the following technical solutions: As an embodiment of the present invention: This invention provides a multi-source attribution analysis method for cross-platform heterogeneous data streams, the method comprising the following steps: S1. Platform Data Collection: Event data is collected in real time from three different data platforms: a search engine promotion platform, a social media advertising platform, and an e-commerce affiliate platform. The indexes of these data platforms are denoted as p, where p = 1, 2, ..., P, and P represents the total number of data platforms.

[0024] Each data platform pushes event data with its own defined field format, which includes at least the following core fields: a set of identifiers to identify users, an event type to represent user behavior, a channel tag to mark the source of the event, and a timestamp to record the time the event occurred.

[0025] In this embodiment, the event data of the search engine platform includes three identifiers: "device ID", "login email", and "encrypted phone number"; the event data of the social media platform includes "device ID", "social account ID", and "IP address"; and the event data of the e-commerce alliance platform includes "login email" and "alliance member ID".

[0026] S2. User Alignment and Uniform Identity Assignment: For event data from any two different data platforms a and b, let it be denoted as... and , Here, r and s are the event data indexes within the two data platforms, respectively.

[0027] Get event data and Their respective sets of user identifiers are denoted as follows: and .

[0028] The formula for calculating the user matching score is: , This represents the user matching score, where, This represents the intersection of two sets of identifiers. This represents the union of two sets of identifiers, and the symbol || indicates the number of elements in the set.

[0029] The matching score ranges from [0,1]. The higher the score, the greater the likelihood that the two events originated from the same user.

[0030] A preset matching threshold is extracted; in this embodiment, the matching threshold is set to 0.6.

[0031] if If the value is ≥0.6, then the event data is determined. and They belong to the same user.

[0032] At the same time, a globally unique unified user identifier is assigned to the event data, and a mapping relationship is established.

[0033] For scenarios involving multiple data platforms, all cross-platform events that meet the threshold conditions are grouped under the same unified user identifier by passing associations.

[0034] S3. Conversion Identification and Touchpoint Sequence Construction: Extract a preset set of conversion event types. In this embodiment, the set of conversion event types includes event types such as "order submission successful", "registration completed", and "paid consultation submission".

[0035] When any standardized event data has an event type field that exactly matches any item in the preset set of conversion event types, the event data is identified as a conversion event.

[0036] For each identified conversion event with a unique user identifier, the time T during which the conversion event occurs is used as the starting point. F Based on this, a preset time window is traced back, which in this embodiment is set to 30 days.

[0037] Within this time window, all other events that belong to the same unified user identifier and are not identified as conversion events are considered as touchpoints affecting this conversion.

[0038] Arranging these touchpoint events in ascending order of their timestamps creates a cross-platform touchpoint sequence corresponding to this conversion event: [T C，1 ,T C，2 ,...,T C，N ].

[0039] S4. Attribution weight calculation: For each contact T in the contact sequence C，i Calculate its attribution weight for conversion, where i = 1, 2, ..., N; i is the index of the touchpoint in the touchpoint sequence, N is the total number of touchpoints in the cross-platform touchpoint sequence, and the touchpoint i = 1 is the earliest touchpoint in the traceback time window, and the touchpoint i = N is the last touchpoint that is closest to but not equal to the conversion event.

[0040] The attribution weight is determined by the product of the confidence factor, time decay factor, and channel-based weight factor of the integrated platform. Contact T C，i Initial attribution weights The calculation formula is: ,in, For contact T C，iThe platform confidence factor of the data platform p(i) reflects the overall credibility of the data on that platform; its calculation method is as follows: , where Q p The base quality score for platform p is determined by the field completeness rate r of the platform's data. p Matching accuracy with historical events (a) p The product is determined, i.e., Q. p =r p ×a p Field completeness rate r p This refers to the percentage of events pushed by platform p where all core fields have no missing values; historical event matching accuracy a. p This is the proportion of platform events that are correctly matched to real users, based on historical data. (Denominator) It is the sum of the basic quality scores of all platforms, used for normalization. The time decay factor is a factor relating to the time interval. The monotonically decreasing function is used to indicate that the closer the contact is to the transition time, the greater its contribution. In this embodiment, the monotonically decreasing function adopts a reciprocal decay function, specifically in the form of: in, =T F -T i , indicating contact T C，i The time of occurrence T i Distance from the time T during the transformation F The time interval, μ is a positive decay coefficient, preset to 0.01, used to control the decay rate. The channel basic weight factor is a static weight value pre-set based on the historical conversion contribution of channel k(i), where k=1,2,...,K, and K is the total number of channels. Specifically, the channel basic weight factor is pre-generated and stored after statistical analysis of the historical conversion data of each channel. Historical conversion data, such as historical conversion volume, conversion rate, or offline transaction data, can also be manually configured by data analysts based on their business experience in this embodiment.

[0041] Specifically, the channel tags, based on preset channel mapping rules, uniformly map the original channel tags of different platforms to standard channel identifiers. The channel mapping rules define the correspondence between original tags and standard channels. In this embodiment, the channel mapping rules are stored in the form of a two-dimensional mapping table. The table structure includes at least the data source platform, the original channel tag name, and the standard channel identifier. The data source platform is used to store data platform identifiers, such as "search engine platform," "social media platform," and "e-commerce alliance platform." The original channel tag name is used to store custom channel tags within each platform, such as "brand zone," "information flow advertisement," and "display network." The standard channel identifier is used to store the unified standard channel identifier after mapping. By looking up this mapping table, any original channel tag can be converted into a standard channel identifier, and its corresponding channel basic weight factor can be obtained.

[0042] In this embodiment, based on analysis of past marketing reports, the base weight for the "search engine brand keywords" channel can be set to 1.5, the base weight for the "social media feed" channel to 1.0, and the base weight for the "affiliate display ads" channel to 0.6. These weight values ​​are initialized and configured by business personnel based on experience or historical data statistical analysis results.

[0043] Calculate the initial weights of all N contacts in the contact sequence. Then, it needs to be normalized to obtain the final attribution weight. Ensure that the sum of contributions from all touchpoints is 1.

[0044] The normalization formula is , where j is the traversal index used for summation, j=1,2,...,N.

[0045] S5. Attribution Value Allocation and Summary: Each conversion event is associated with a preset attribution value. In this embodiment, for example, the conversion event corresponding to a successful order submission is associated with the transaction amount of that order, which serves as its attribution value.

[0046] Based on the normalized attribution weight W i The attribution value associated with the conversion event is then allocated to the corresponding channels for each touchpoint.

[0047] If the i-th contact point T C，i If the channel index is k, then the attribution value obtained by that channel from this conversion is... for ,in This is the pre-set total value of this conversion event.

[0048] Finally, the attribution values ​​allocated to each channel across all conversion events are summed up to obtain the total attribution value for each channel over a period of time.

[0049] As a second embodiment of the present invention: In specific implementation of this application, compared to Embodiment 1, the platform confidence factor α in Embodiment 1... p It is based on static calculations of historical data. The only difference between the technical solution of this embodiment and that of Embodiment 1 is that this embodiment incorporates an adaptive update mechanism, which enables it to make periodic dynamic adjustments based on the real-time performance of the platform data.

[0050] The periodic dynamic adjustment method is as follows: Step A: Set up a sliding window and trigger conditions. Set up a conversion event sliding window and set the window size to the M most recently completed conversion events, for example, M=500.

[0051] Each time the attribution processing of M new conversion events is completed, an update calculation of the quality score for each platform is triggered.

[0052] Step B: Calculate key metrics within the window. When an update is triggered, the system will calculate two metrics for each platform p based on the M conversion events and their touchpoint sequences within the sliding window: average attribution weight percentage U. p and the proportion of direct conversion observation G p .

[0053] Average attribution weight This reflects the overall contribution level of platform p under the current understanding of the model. Its calculation method is as follows: That is, the attribution weights assigned to all touchpoints belonging to platform p within the window. The sum of the values ​​is divided by the sum of the attribution weights of all touchpoints within the window. Here, I() is an indicator function that has a value of 1 when the condition is true and 0 otherwise.

[0054] Direct conversion observation percentage This reflects the observation proportion of platform p as the "final step," or the last point of contact. Its calculation method is as follows: This is calculated as the number of conversion events within the window where the last touchpoint in the touchpoint sequence originates from platform p, divided by the total number of conversion events M within the window. This metric is considered a relatively objective observation signal.

[0055] Step C: Update the platform quality score criteria and The difference is used to correct the quality score of platform p.

[0056] Intuitively, if a platform's actual observed conversion rate... The attribution weight percentage is consistently higher than the average attribution weight percentage calculated by its model. This indicates that the platform's contribution may be underestimated, and its quality score should be improved, or vice versa.

[0057] The updated formula is: in, and These are the quality scores of platform p before and after the update, respectively. η is the preset update step size; in this embodiment, η is set to 0.05 to control the magnitude of each update.

[0058] Step D: Update the platform confidence factor based on the calculated updated quality scores for all platforms. Where q=1,2,...,P, the platform confidence factor for each platform is recalculated. The formula is And the newly calculated dynamic platform confidence factor Replace with the original .

[0059] As an embodiment of the present invention: In specific implementation, compared with Embodiment 1 and Embodiment 2, the technical solution of this embodiment is to combine the solutions of Embodiment 1 and Embodiment 2. The difference between the technical solution of this embodiment and Embodiment 1 and Embodiment 2 is only that in this embodiment, when there are multiple repeated touchpoints generated by users on the same platform or channel in a very short period of time, the independent contribution of each of these dense touchpoints is usually weaker than that of touchpoints with longer time intervals.

[0060] This embodiment also incorporates a sequence decay factor in the attribution weight calculation, as detailed below. For the touch sequence [T] corresponding to the conversion event C，1 ,T C，2 ,...,T C，N Contact T C，i The attribution weight calculation not only includes the platform confidence factor, time decay factor, and channel basic weight factor in the product, but also multiplies by the order decay factor. .

[0061] Sequential decay factor The setting rules are as follows: For the first contact T in the contact sequence C，1 The sequential decay factor is 1, that is =1.

[0062] For contact T in the contact sequence where i≥2 C，i Sequential decay factor For this contact and its previous contact T C，i-1 The time interval δ between i It is a monotonically decreasing function. Where the time interval δ i =T i -T i-1 .

[0063] This embodiment uses a reciprocal decay function to model the sequential decay factor, so that the shorter the time interval between adjacent contacts, the smaller the value of the sequential decay factor, thereby effectively suppressing the weight of short-duration dense contacts. The calculation formula is as follows: Wherein, γ is a positive half-life interval parameter. In this embodiment, when the time interval between adjacent contacts is equal to the preset half-life interval parameter γ, the value of γ is 0.5 hours.

[0064] when When =γ, =0.5; when When it approaches 0, It also tends to 0; when When it is very large, Approaching 1.

[0065] Therefore, in this embodiment, contact T C，i Initial attribution weights The calculation formula has been upgraded to The subsequent normalization process is also performed, that is... .

[0066] As an embodiment of the present invention: In its specific implementation, compared to Embodiments 1, 2, and 3, the difference between this embodiment and Embodiments 1, 2, and 3 lies only in that, in this embodiment, considering the characteristics of user cross-platform behavior paths, a cross-platform collaboration factor λ is also incorporated. i Attribution weights are calculated, and the cross-platform collaborative factor λ is calculated. i The setting rules are as follows: For the first contact T in the contact sequence C，1 The cross-platform collaboration factor is 1, i.e., λ1=1.

[0067] For contacts in the contact sequence where i ≥ 2 TC，i Compare its data platform p(i) with the previous touchpoint TC, i-1 belongs to data platform p(i-1).

[0068] If p(i)≠p(i-1), meaning a cross-platform transition has occurred, then the cross-platform collaboration factor is greater than 1 and is set to a fixed value, such as λi=1.5.

[0069] If p(i) = p(i-1), meaning it is still within the same platform, then the cross-platform collaboration factor is equal to 1, i.e., λi = 1.

[0070] When this embodiment is combined with Embodiment 3, that is, when both the sequential decay factor and the cross-platform synergy factor are introduced, contact T C，i Initial attribution weights The calculation formula achieves the most complete form: Final attribution weight Still through the Normalization is performed to obtain the result.

[0071] As a fifth embodiment of the present invention: In its specific implementation, compared to Embodiments 1, 2, 3, and 4, the technical solution of this embodiment lies in combining the solutions of Embodiments 1, 2, 3, and 4. This embodiment also provides a multi-source attribution analysis system for cross-platform heterogeneous data streams. This system is used to implement a multi-source attribution analysis method for cross-platform heterogeneous data streams, which includes the following steps: The data acquisition unit is used to collect event data in real time from at least two different data platforms. The event data from each platform has its own field format, but all of them contain a set of identifiers for identifying users, event type, associated channel tags, and timestamps of the event occurrence. The user alignment unit is used to standardize event data from different data platforms and calculate the matching score of users based on the degree of overlap between the identifier sets of corresponding event data from different data platforms. Cross-platform events with matching scores greater than or equal to a preset matching threshold are determined to belong to the same user, and a unified user identifier is assigned to cross-platform events determined to belong to the same user. The touchpoint sequence construction unit is used to identify conversion events. A conversion event is event data that matches the preset set of conversion event types in the event type field. For each conversion event, all non-conversion events with the same unified user identifier as the conversion event within a preset time window are traced back from the time the conversion event occurred as touchpoints, and a corresponding cross-platform touchpoint sequence is constructed in chronological order. The weighting calculation unit is used to calculate the attribution weight for each touchpoint in the touchpoint sequence for each conversion event. The attribution weight is determined based on the product of at least the following three factors: The platform confidence factor reflects the credibility of the platform data, the time decay factor decays over time, and the channel basic weight factor is pre-set based on the historical conversion contribution of each channel; and the calculated attribution weights are normalized so that the sum of the attribution weights corresponding to the touchpoints of the same conversion event is 1. The attribution aggregation unit is used to allocate the preset attribution value associated with each conversion event to the corresponding channel based on the normalized attribution weight, and to aggregate the total attribution value obtained from each channel.

[0072] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0073] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0074] All the above formulas use dimensionless numerical calculations, that is, by introducing a reference benchmark, the original physical quantities are transformed into dimensionless relative values, eliminating the influence of units and retaining the relative size relationship of physical quantities; the formula is the closest to the real situation obtained by software simulation of a large amount of data, and the preset parameters and threshold selection in the formula are set by those skilled in the art according to the actual situation.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0076] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A multi-source attribution analysis system for cross-platform heterogeneous data streams, characterized in that, include: The data acquisition unit is used to collect event data from various data platforms in real time. The event data from each data platform has its own field format, and each of them contains a set of identifiers for identifying users, event type, associated channel tags, and timestamp of the event occurrence. The user alignment unit is used to standardize event data from different data platforms and calculate the matching score of users based on the degree of overlap between the identifier sets of corresponding event data from different data platforms. Cross-platform events with matching scores greater than or equal to a preset matching threshold are determined to belong to the same user, and a unified user identifier is assigned to cross-platform events determined to belong to the same user. The touchpoint sequence construction unit is used to identify conversion events. A conversion event refers to event data whose event type field matches a preset set of conversion event types. For each conversion event, all non-conversion events with the same unified user identifier as the conversion event within a preset time window are traced back from the time the conversion event occurs as touchpoints, and a corresponding cross-platform touchpoint sequence is constructed in chronological order. The weight calculation unit is used to calculate the attribution weight of each touchpoint in the touchpoint sequence of each conversion event. The attribution weight is determined based on at least the product of the platform confidence factor, the time decay factor and the preset channel basic weight factor. The result is normalized so that the sum of the attribution weights corresponding to the touchpoints of the same conversion event is 1. The attribution aggregation unit is used to allocate the preset attribution value associated with each conversion event to the corresponding channel based on the normalized attribution weight, and to aggregate the total attribution value obtained from each channel.

2. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 1, characterized in that, The user alignment unit calculates the user's matching score as follows: Select two data platforms, denoted as data platform a and data platform b. For event data from data platform a... Event data from data platform b User matching score ,in, and events and The set of user identifiers, || represents the number of elements in the set, r is the event data index of data platform a, and s is the event data index of data platform b.

3. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 1, characterized in that, Platform confidence factor The calculation method is as follows Where p = 1, 2, ..., P, P is the total number of platforms. The basic quality score for platform p is determined by the product of the field completeness rate and the historical event matching accuracy rate of the platform's data.

4. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 1, characterized in that, It also includes an adaptive update module, which is used to periodically and dynamically update the platform confidence factor and provide the updated platform confidence factor to the weight calculation unit. Specifically, the adaptive update module is used to set up a conversion event sliding window. Whenever a predetermined number of conversion events are attributed, the quality score of each platform is updated and the platform confidence factor is updated.

5. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 4, characterized in that, The specific method of the adaptive update module is as follows: The average attribution weight U of platform p within the window. p And the proportion of direct conversion observed as the final point of contact G p ; Among them, the average attribution weight ratio U p The percentage of direct conversion observations G is the sum of the attribution weights of all conversion events attributable to platform p within the conversion event sliding window, relative to the total sum of the attribution weights of all touchpoints. p The proportion of the number of conversion events belonging to platform p at the last touchpoint within the conversion event sliding window to the total number of conversion events within the window; According to the formula Calculate the updated quality scores for all platforms. ,in, The preset update step size, Quality score before all platforms update; Quality score after all platforms update The platform confidence factor for each platform was recalculated.

6. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 1, characterized in that, The attribution weight calculated for touchpoint i by the weight calculation unit is also based on the order decay factor. The attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor and the order decay factor, and then normalized. The sequential decay factor is set as follows: for the first contact in the contact sequence, the sequential decay factor is 1; for contacts i≥2 in the contact sequence, the sequential decay factor is the time decay function of the time interval between the contact and the previous contact.

7. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 6, characterized in that, The attribution weight calculated for touchpoint i by the weight calculation unit is also based on cross-platform collaborative factors; The attribution weight is the product of the platform confidence factor, time decay factor, channel basic weight factor and cross-platform collaboration factor. When the attribution weight calculation of touchpoint i is also based on the sequential decay factor, the product also includes the sequential decay factor, and then normalization is performed. The cross-platform collaboration factor is set as follows: for the first contact in the contact sequence, the cross-platform collaboration factor is 1; for contacts i≥2 in the contact sequence, if the data platform to which the contact belongs is different from the data platform to which the previous contact belongs, the cross-platform collaboration factor is greater than 1; if the data platforms are the same, the cross-platform collaboration factor is equal to 1.

8. The multi-source attribution analysis system for cross-platform heterogeneous data streams according to claim 1, characterized in that, The time decay function is a reciprocal decay function, specifically in the form of: ,in, The time interval during which the contact distance conversion occurs. This is the preset positive attenuation coefficient.

9. A multi-source attribution analysis method for cross-platform heterogeneous data streams, wherein the method is implemented using the multi-source attribution analysis system for cross-platform heterogeneous data streams as described in any one of claims 1-8, characterized in that, The method includes the following steps: S1. Collect event data streams in real time from at least two different data platforms. The event data from each platform has its own field format, but all contain a set of identifiers for identifying users, event type, associated channel tags, and timestamps of the event occurrence. S2. Standardize the event data from different platforms, and calculate the user matching score based on the degree of overlap between the identifier sets of events from different platforms. Determine that cross-platform events with matching scores greater than or equal to the preset matching threshold belong to the same user, and assign a unified user identifier to cross-platform events determined to belong to the same user. S3. Identify conversion events. A conversion event is an event whose event type field matches a preset set of conversion event types. For each conversion event, trace back from the time the conversion event occurred to all non-conversion events within a preset time window that have the same unified user identifier as the conversion event as touchpoints, and construct a corresponding cross-platform touchpoint sequence in chronological order. S4. Calculate the product of the platform confidence factor, time decay factor, and channel basic weight factor, and normalize the calculation results so that the sum of the touchpoint weights of the same conversion event is 1, and calculate the attribution weight of each touchpoint in the touchpoint sequence of each conversion event. S5. Allocate the preset value associated with each conversion event to the corresponding channel according to the normalized weight, and summarize the total attribution value obtained from each channel.

10. The multi-source attribution analysis method for cross-platform heterogeneous data streams according to claim 9, characterized in that: When calculating the attribution weight for touch point i, a sequential decay factor and / or a cross-platform synergy factor are also introduced into the product; When the sequential decay factor is introduced, the attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor and the sequential decay factor, and then normalized. When a cross-platform collaboration factor is introduced, the attribution weight is the product of the platform confidence factor, the time decay factor, the channel basic weight factor, and the cross-platform collaboration factor, and then normalized. When both sequential decay factor and cross-platform collaboration factor are introduced, the attribution weight is the product of platform confidence factor, time decay factor, channel basic weight factor, sequential decay factor and cross-platform collaboration factor, and then normalized.

Images