Multi-contact marketing effect attribution method based on AI big data fusion

Abstract

The invention discloses a multi-contact marketing effect attribution method based on AI big data fusion, relates to the field of information technology and data science, and is used for solving the problems of relatively high calculation redundancy, large resource consumption and low execution efficiency of an overall model. Constructing a feature vector and calculating a contact weight coefficient, if the weight exceeds a set threshold value, preliminarily judging whether a causal relationship exists between the contacts by using a dynamic time warping algorithm, after judging that the causal relationship exists, collecting the interaction frequency and depth between the contacts, analyzing the correlation degree and calculating the correlation strength, and according to the correlation degree, calculating the correlation degree of the contacts; according to the method, a graph neural network algorithm is selected or association strength is applied to a random forest algorithm to detect a contact redundancy condition, the dynamic modeling capability of causal identification is improved, and adaptive redundancy identification and optimization of an attribution model are realized.

Description

Technical Field

[0001] The present invention relates to the fields of information technology and data science. More specifically, the present invention relates to a multi-touch marketing effect attribution method for AI big data integration. Background Art

[0002] With the rapid development of big data and artificial intelligence technologies, the multi-touch marketing effect attribution method has gradually become an important tool for enterprises to optimize market strategies and improve user conversion rates. However, the existing multi-touch marketing attribution methods still have certain limitations in aspects such as data integration, analysis accuracy, and real-time performance, and it is difficult to fully meet the needs of modern enterprises for efficient and precise marketing.

[0003] The existing technologies have the following deficiencies: Currently, the existing methods generally lack systematic feature screening strategies and graph structure optimization means, and fail to effectively eliminate the noise interference brought by low-value touchpoints to the attribution model. As a result, the overall model has a high calculation redundancy, large resource consumption, and low execution efficiency. Therefore, a multi-touch marketing effect attribution method for AI big data integration is proposed.

[0004] The above information disclosed in the background art section is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the embodiments of the present invention provide a multi-touch marketing effect attribution method for AI big data integration, which realizes accurate attribution of multi-touch marketing paths by integrating multi-source heterogeneous data and combining intelligent analysis models to solve the problems raised in the above background art.

[0006] To achieve the above object, the present invention provides the following technical solution, a multi-touch marketing effect attribution method for AI big data integration, including the following steps: Step S1: Collect user behavior data and transaction data of multiple touchpoints and preprocess them; Step S2: Construct feature vectors based on the preprocessed user behavior data and transaction data, and calculate the weight coefficients of each touchpoint; when the weight coefficient exceeds the set threshold, use the dynamic time warping algorithm to preliminarily judge whether there is a causal relationship between the touchpoints; Step S3: When it is preliminarily judged that there is a causal relationship between the touchpoints, collect the interaction frequency and interaction depth between the touchpoints, and comprehensively analyze the correlation degree between the touchpoints based on the interaction frequency and interaction depth; when the correlation degree is high, calculate the correlation strength between the touchpoints; Step S4: Select a graph neural network algorithm according to the correlation degree between the touchpoints or apply its correlation strength to a random forest algorithm to detect whether there is a redundancy problem between the touchpoints.

[0007] In a preferred embodiment, in step S1, the user behavior data of the touch points is represented by the calculation result obtained by hierarchically weighted summation of the number of clicks, residence duration, and page jump path of the user on different touch points. The transaction data of the touch points is obtained by detecting the number of orders and transaction amounts completed by the user on different touch points.

[0008] In a preferred embodiment, in step S1, the system sets multiple groups of touch point devices, respectively detects the user behavior data and transaction data of each group of touch points, and performs normalization processing to obtain multiple groups of behavior data coefficients and transaction data coefficients.

[0009] In a preferred embodiment, in step S2, the processed behavior data coefficients are merged into a behavior data set, and the transaction data coefficients are merged into a transaction data set. The specific steps for setting the weight threshold are as follows: For the two data sets, the same processing is performed. The data set is used as the analysis data set. The analysis data set is divided into two parts with the same number, and the average values in the two parts of the data sets are calculated and compared. The part with the larger average value is selected as the new round of analysis data set. Then the new round of analysis data set is divided into two parts with the same number to calculate the average value, and the part with the smaller average value is selected as the new round of analysis data set. Alternately taking the larger and the smaller values, repeat the operation on the data set until only the last data remains. The average value calculated from the data obtained by processing the two data sets is set as the weight threshold.

[0010] In a preferred embodiment, in step S2, the system calculates the behavior data coefficient and transaction data coefficient of the touch point to be measured and takes the average value as the touch point weight coefficient. When the touch point weight coefficient exceeds the weight threshold, the dynamic time warping algorithm is used to perform a preliminary match on the time series data between the touch points.

[0011] In a preferred embodiment, in step S3, the interaction frequency obtains the number of interactions between the touch points through the timestamp field in the access log table; the interaction depth is reflected by recording the number of interaction levels between the touch points within a set time period.

[0012] In a preferred embodiment, in step S3, the specific steps for using fuzzy inference when analyzing the correlation degree between the touch points are as follows: Define the interaction frequency and the number of interaction levels as input variables and divide them into fuzzy sets. Define the degree of association between contacts as the output variable and divide it into fuzzy sets; formulate fuzzy rules to describe the impact of interaction frequency and the number of interaction levels on the degree of association between contacts; Conduct fuzzy inference according to the fuzzy rules to determine the degree of association between contacts.

[0013] In a preferred embodiment, in step S3, when calculating the association strength between contacts, sum the interaction frequency and the number of interaction levels between contacts to obtain a normalization parameter, and calculate and obtain it using the linear normalization formula according to the normalization parameter.

[0014] In a preferred embodiment, in step S4, when the degree of association between contacts is low, the system uses the graph neural network algorithm to detect the relationship between contacts; When the degree of association between contacts is high, the system uses the random forest algorithm to detect the relationship between contacts; If the random forest algorithm detects a redundancy problem between contacts, give the redundancy strength between contacts; If no redundancy problem occurs, record the relationship between contacts.

[0015] Technical effects and advantages of the present invention: 1. The present invention collects user behavior data and transaction data of multiple contacts, preprocesses them, constructs feature vectors based on the preprocessed user behavior data and transaction data, calculates the weight coefficients of each contact, compares them with the set threshold, and when the weight coefficient exceeds the set threshold, uses the dynamic time warping algorithm to preliminarily judge whether there is a causal relationship between contacts. When it is preliminarily judged that there is a causal relationship between contacts, collect the interaction frequency and interaction depth between contacts, comprehensively analyze the degree of association between contacts based on the interaction frequency and interaction depth; when the degree of association is high, calculate the association strength between contacts, and select the graph neural network algorithm according to the degree of association between contacts or apply its association strength to the random forest algorithm to detect whether there is a redundancy problem between contacts, which improves the dynamic modeling ability of identifying causal relationships between multiple contacts and realizes the adaptive identification and optimization of the attribution model for contact redundancy problems. Brief Description of the Drawings

[0016] Figure 1 It is a schematic diagram of the overall process of the multi-contact marketing effect attribution method of AI big data fusion of the present invention.

[0017] Figure 2 It is a schematic diagram of the fuzzy inference process for analyzing the degree of association between contacts in the multi-contact marketing effect attribution method of AI big data fusion of the present invention. Detailed Embodiments

[0018] Next, in combination with the accompanying drawings in the embodiments of the present invention, the technical solutions in the embodiments of the present invention will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment 1: The present invention provides a multi-touch marketing effect attribution method for AI big data integration, and its specific implementation manner will be described in detail in combination with the appended Figure 1 and the appended Figure 2 drawings.

[0020] As Figure 1 shown, the entire method process is completed by the cooperation of multiple modules, including a data collection module, a preprocessing module, a feature vector construction module, a weight calculation module, a dynamic time warping module, an interaction frequency analysis module, an interaction depth analysis module, a fuzzy inference module, a correlation strength calculation module, and an algorithm selection module.

[0021] The following will describe the specific implementation steps of each module, as well as their connection relationships and cooperation methods, one by one in the order of the process.

[0022] First of all, the data collection module is responsible for obtaining user behavior data and transaction data from multiple touch points. These touch points can be websites, mobile applications, social media platforms, or other digital channels.

[0023] In practical applications, the user behavior data of the touch points cannot be directly obtained. Therefore, the system performs hierarchical weighted summation by recording the number of clicks, the dwell time, and the page jump path of the user on different touch points to obtain a calculation result that can reflect the user behavior characteristics.

[0024] The transaction data is obtained by detecting the number of orders and the transaction amount completed by the user on different touch points.

[0025] To ensure the consistency and comparability of the data, the system sets multiple groups of touch point devices to detect the user behavior data and transaction data of each group of touch points respectively, and transmits them to the preprocessing module for normalization processing.

[0026] The specific operation of the normalization processing is to map the original data to a unified numerical range, thereby generating multiple groups of behavior data coefficients and transaction data coefficients.

[0027] These coefficients are used as the input data for the subsequent modules, providing a basis for the subsequent feature vector construction and weight calculation.

[0028] Next, the preprocessing module combines the processed multiple groups of behavior data coefficients and transaction data coefficients into a behavior data set and a transaction data set respectively, and transmits them to the feature vector construction module.

[0029] The feature vector construction module constructs feature vectors based on the behavior data set and the transaction data set, and simultaneously sets a weight threshold to evaluate the influence of each contact.

[0030] The process of setting the weight threshold is as follows: The same processing is performed on the two data sets. Each data set is divided into two parts with the same number, and the average values of the two parts of the data sets are calculated and compared. The part with the larger average value is selected as the new round of analysis data set. Then, the new round of analysis data set is divided into two parts with the same number, and after calculating the average value, the part with the smaller average value is selected as the new round of analysis data set. Alternately taking the larger and the smaller values and repeating the operation on the data set until only the last data remains. Calculate the average value of the data obtained by processing the two data sets as the final weight threshold.

[0031] Subsequently, the weight calculation module calculates the average value of the behavior data coefficient and the transaction data coefficient of the contact to be measured as the contact weight coefficient.

[0032] When the contact weight coefficient exceeds the weight threshold, the system determines that the contact has a high influence, and transmits the relevant data to the dynamic time warping module for further analysis.

[0033] The dynamic time warping module uses the dynamic time warping algorithm to perform a preliminary match on the time series data between the contacts to be measured.

[0034] The dynamic time warping algorithm is an algorithm used to measure the similarity of two time series, and determines whether there is a causal relationship by calculating the optimal alignment path between the two time series.

[0035] In practical applications, the system takes the time series data between contacts as input, calculates the distance matrix between them, and finds the optimal alignment path through the dynamic programming method.

[0036] If the cumulative distance of the optimal alignment path is less than the set threshold, the system determines that there is a causal relationship between the contacts, and transmits the relevant data to the interaction frequency analysis module and the interaction depth analysis module for further analysis.

[0037] The interaction frequency analysis module obtains the interaction times between contacts by accessing the timestamp field in the access log table and records the interaction frequency between contacts.

[0038] The interaction depth analysis module records the number of interaction levels between contacts within a set time period to reflect the interaction depth.

[0039] The interaction frequency and the number of interaction levels are transmitted to the fuzzy inference module as input variables for analyzing the degree of association between contacts.

[0040] As Figure 2 shown, the fuzzy inference module uses the fuzzy inference method to analyze the degree of association between contacts.

[0041] Specifically, the interaction frequency and the number of interaction levels are defined as input variables and divided into fuzzy sets; The degree of association between contacts is defined as the output variable and divided into fuzzy sets; Fuzzy rules are formulated to describe the influence of the interaction frequency and the number of interaction levels on the degree of association between contacts; Fuzzy inference is performed according to the fuzzy rules to determine the degree of association between contacts.

[0042] The result of the fuzzy inference is transmitted to the association strength calculation module for calculating the association strength between contacts.

[0043] The association strength calculation module sums the interaction frequency and the number of interaction levels between contacts to obtain a normalization parameter, and calculates and obtains the association strength between contacts using the linear normalization formula according to the normalization parameter.

[0044] The calculation result of the association strength is transmitted to the algorithm selection module for selecting a suitable algorithm to detect the relationship between contacts.

[0045] The algorithm selection module selects different algorithms for detection according to the degree of association between contacts. When the degree of association between contacts is low, the system uses the graph neural network algorithm to detect the relationship between contacts; When the degree of association between contacts is high, the system uses the random forest algorithm to detect the relationship between contacts.

[0046] The random forest algorithm generates multiple decision tree models by sampling and training the data between contacts multiple times, and obtains the final detection result through a voting mechanism.

[0047] If the detection result shows that there is a redundancy problem between contacts, the system calculates the redundancy strength between contacts; if there is no redundancy problem, the system records and stores the relationship between contacts.

[0048] In the whole process, each module is tightly connected in the way of data flow.

[0049] The data acquisition module transmits the collected raw data to the preprocessing module for normalization processing, and the preprocessing module transmits the processed data to the feature vector construction module and the weight calculation module.

[0050] The weight calculation module determines whether a contact has a high influence based on the calculation result and transmits relevant data to the dynamic time warping module.

[0051] The matching result of the dynamic time warping module is transmitted to the interaction frequency analysis module and the interaction depth analysis module, and the analysis results of the interaction frequency analysis module and the interaction depth analysis module are transmitted to the fuzzy inference module.

[0052] The inference result of the fuzzy inference module is transmitted to the correlation strength calculation module, and the calculation result of the correlation strength calculation module is transmitted to the algorithm selection module.

[0053] The algorithm selection module selects a suitable algorithm according to the degree of correlation to detect the relationship between contacts and stores or feedbacks the detection result to other modules.

[0054] The above embodiments demonstrate the actual operation principle and process of the method of the present invention.

[0055] In an actual application scenario, for example, an e-commerce enterprise hopes to optimize its multi-contact marketing strategy. It can collect and analyze the user behavior data and transaction data of contacts such as its website, mobile application, and social media platform through this method.

[0056] First, the system obtains data such as the number of clicks, dwell time, page jump path, order quantity, and transaction amount of users on different contacts through the data collection module.

[0057] Then, the preprocessing module normalizes this data to generate a behavior data coefficient and a transaction data coefficient.

[0058] The feature vector construction module constructs a feature vector based on these coefficients and calculates the weight coefficient of each contact through the weight calculation module.

[0059] If the weight coefficient of a certain contact exceeds the set threshold, the dynamic time warping module matches the time series data between this contact and other contacts to determine whether there is a causal relationship.

[0060] If the matching is successful, the system further analyzes the interaction frequency and interaction depth between contacts, and calculates the degree of correlation and the correlation strength between contacts through the fuzzy inference module and the correlation strength calculation module.

[0061] Finally, the algorithm selection module selects a suitable algorithm according to the degree of correlation to detect the relationship between contacts and feedbacks the detection result to the enterprise to help the enterprise optimize the contact configuration and marketing strategy.

[0062] It can be seen from the above embodiments that the method of the present invention can effectively integrate multi-source heterogeneous data and combine an intelligent analysis model to achieve accurate attribution of the multi-contact marketing path.

[0063] In practical applications, this method can not only reduce complex computational requirements, but also improve the detection accuracy, clarify the relationship types between contacts, and provide a scientific decision-making basis for enterprises.

[0064] To enable relevant personnel in the technical field to better understand and implement the present invention, the following further supplements the specific implementation principle of the present invention in combination with a specific application scenario.

[0065] In practical applications, assume that an e-commerce enterprise hopes to optimize its multi-touch marketing strategy to improve user conversion rates and market competitiveness.

[0066] The enterprise collects and analyzes user behavior data and transaction data of contacts such as its website, mobile application, and social media platform through the method of the present invention.

[0067] The following are the specific operation steps and implementation principles of the method of the present invention in this scenario.

[0068] First, in the data collection module, the system obtains data such as the number of user clicks, page stay duration, jump path, order quantity, and transaction amount from multiple contacts by deploying sensors or calling API interfaces.

[0069] These data reflect the behavior characteristics and transaction preferences of users on different contacts. Since the original data may have the problem of inconsistent dimensions, the system transmits this data to the preprocessing module.

[0070] In the preprocessing module, the system uses a normalization algorithm to process the original data and maps it to a unified numerical range.

[0071] For example, for the number of clicks and transaction amount, the system calculates their maximum and minimum values, and normalizes the data into a coefficient value between 0 and 1 through a linear transformation formula.

[0072] The normalized data is divided into a behavior data set and a transaction data set, and is respectively transmitted to the feature vector construction module.

[0073] Subsequently, the feature vector construction module generates feature vectors based on the behavior data set and the transaction data set, and sets weight thresholds to evaluate the influence of each contact.

[0074] The process of setting the weight threshold is as follows: The system uses the behavior data set and the transaction data set as analysis data sets respectively, and divides each data set into two parts with the same quantity.

[0075] The system calculates the average values of these two parts of data, compares their sizes, and screens out the part with the larger average value as the new round of analysis data set.

[0076] Next, the system divides the new analysis data set into two parts again, calculates the average value, and filters out the part with the smaller average value as the analysis data set for the next round.

[0077] In this way, taking the larger and the smaller values alternately, the operation is repeated until only one data remains in each data set. The system calculates the average value of the data finally obtained from the two data sets as the weight threshold.

[0078] The weight calculation module calculates the average value based on the behavior data coefficient and transaction data coefficient of the contact as the contact weight coefficient.

[0079] If the weight coefficient of a certain contact exceeds the set threshold, the system determines that the contact has a high influence and transmits the relevant data to the dynamic time warping module.

[0080] The dynamic time warping module uses the dynamic time warping algorithm to match the time series data between the contacts to be measured.

[0081] Specifically, the system converts the behavior data and transaction data between contacts into time series form and calculates the distance matrix between the two. Through the dynamic programming method, the system finds the optimal alignment path and records the cumulative distance.

[0082] If the cumulative distance of the optimal alignment path is less than the set threshold, the system determines that there is a causal relationship between the contacts.

[0083] For example, when a user browses a product on a social media platform and then visits an e-commerce website and completes a purchase within a short time, the system can identify this causal relationship and transmit the relevant data to the interaction frequency analysis module and the interaction depth analysis module.

[0084] The interaction frequency analysis module counts the interaction times between contacts through the timestamp field in the access log table and records the interaction frequency.

[0085] For example, the system can count the number of times a user jumps from a social media platform to an e-commerce platform. At the same time, the interaction depth analysis module records the number of interaction levels between contacts through a set time period to reflect the interaction depth.

[0086] For example, the system can record the number of intermediate contacts passed by a user from the initial contact to the final contact. The interaction frequency and the number of interaction levels are transmitted to the fuzzy inference module as input variables.

[0087] As Figure 2 shown, the fuzzy inference module uses the fuzzy inference method to analyze the degree of association between contacts. The system defines the interaction frequency and the number of interaction levels as input variables and divides them into fuzzy sets.

[0088] For example, the interaction frequency can be divided into three fuzzy sets: "low", "medium", and "high", and the number of interaction levels can be divided into three fuzzy sets: "shallow", "medium", and "deep".

[0089] The degree of association between contacts is defined as the output variable and is also divided into fuzzy sets. The system formulates fuzzy rules to describe the influence of interaction frequency and the number of interaction levels on the degree of association between contacts.

[0090] For example, "when the interaction frequency is high and the number of interaction levels is deep, the degree of association between contacts is strong". After fuzzy reasoning based on the fuzzy rules, the system determines the degree of association between contacts and transmits the result to the association strength calculation module.

[0091] The association strength calculation module sums the interaction frequency and the number of interaction levels between contacts to obtain a normalization parameter, and uses a linear normalization formula to calculate the association strength between contacts.

[0092] For example, assume that the interaction frequency of a certain two contacts is 10 times and the number of interaction levels is 3 levels. The system sums these two values to obtain a normalization parameter of 13, and maps it to an association strength value between 0 and 1 through the normalization formula.

[0093] This result is transmitted to the algorithm selection module.

[0094] The algorithm selection module selects a suitable algorithm for detection according to the degree of association between contacts. When the degree of association between contacts is low, the system uses the graph neural network algorithm to detect the relationship between contacts.

[0095] The graph neural network algorithm analyzes the connection pattern between nodes by constructing a graph structure model between contacts, so as to identify whether there is a potential connection between contacts.

[0096] When the degree of association between contacts is high, the system uses the random forest algorithm for detection.

[0097] The random forest algorithm generates multiple decision tree models by sampling and training the data between contacts multiple times, and obtains the final detection result through a voting mechanism.

[0098] If the detection result shows that there is a redundancy problem between contacts, the system will further calculate the redundancy strength between contacts; If there is no redundancy problem, the system records and stores the relationship between contacts.

[0099] From the above steps, it can be seen that the method of the present invention can effectively integrate multi-source heterogeneous data and combine an intelligent analysis model to achieve accurate attribution of multi-contact marketing paths.

[0100] In practical applications, this method can not only reduce complex computational requirements, but also improve the detection accuracy, clarify the relationship types between contacts, and provide a scientific decision-making basis for enterprises.

[0101] For example, by analyzing and finding that the correlation strength between certain contacts is low and there are redundancy problems, enterprises can adjust resource allocation, reduce investment in redundant contacts, thereby optimizing marketing strategies and enhancing market competitiveness.

[0102] Specifically, the above content describes the multi-touch marketing effect attribution method of the AI big data fusion of the present invention from the perspective of a module system. Further, the multi-touch marketing effect attribution method of the AI big data fusion includes the following steps: Step S1: Collect user behavior data and transaction data of multiple contacts and preprocess them; Step S2: Construct feature vectors according to the preprocessed user behavior data and transaction data, and calculate the weight coefficients of each contact; when the weight coefficient exceeds the set threshold, use the dynamic time warping algorithm to preliminarily judge whether there is a causal relationship between contacts; Step S3: When it is preliminarily judged that there is a causal relationship between contacts, collect the interaction frequency and interaction depth between contacts, and comprehensively analyze the correlation degree between contacts based on the interaction frequency and interaction depth; when the correlation degree is high, calculate the correlation strength between contacts; Step S4: Select the graph neural network algorithm according to the correlation degree between contacts or apply its correlation strength to the random forest algorithm to detect whether there are redundancy problems between contacts.

[0103] The above formulas are all dimensionless and take their numerical values for calculation. The formulas are obtained by collecting a large amount of data for software simulation to obtain a formula that is closest to the actual situation. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0104] The above embodiments can be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented using software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center in a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or a data center that contains a collection of one or more available media. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, or a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state drive.

[0105] It should be understood that in various embodiments of the present application, the order numbers of the above processes do not mean the order of execution, and the order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0106] Those of ordinary skill in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be implemented by electronic hardware, or by a combination of computer software and electronic hardware. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Professional technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present application.

[0107] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the systems, devices, and units described above can refer to the corresponding processes in the foregoing method embodiments and will not be described herein again.

[0108] In several embodiments provided in this application, it should be understood that the disclosed systems, devices, and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling, direct coupling, or communication connection to each other can be through some interfaces. The indirect coupling or communication connection of the devices or units can be in electrical, mechanical, or other forms.

[0109] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0110] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit.

[0111] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art or part of this 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 enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of this application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical discs that can store program codes.

[0112] As described above, the above are only specific implementation manners of this application, but the protection scope of this application is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in this application and should be covered by the protection scope of this application. Therefore, the protection scope of this application should be subject to the protection scope of the claims.

Claims

1. A multi-touch marketing effect attribution method integrating AI and big data, characterized in that: It includes the following steps: Step S1: Collect the user behavior data and transaction data of multiple touchpoints, and preprocess them; Step S2: Construct feature vectors based on the preprocessed user behavior data and transaction data, and calculate the weight coefficients of each touchpoint; when the weight coefficient exceeds the set threshold, use the dynamic time warping algorithm to preliminarily judge whether there is a causal relationship between touchpoints; Step S3: When it is preliminarily judged that there is a causal relationship between touchpoints, collect the interaction frequency and interaction depth between touchpoints, and comprehensively analyze the association degree between touchpoints based on the interaction frequency and interaction depth; when the association degree is high, calculate the association strength between touchpoints; Step S4: Select a graph neural network algorithm according to the association degree between touchpoints or apply its association strength to a random forest algorithm to detect whether there are redundant problems between touchpoints.

2. The multi-touchpoint marketing effect attribution method based on AI big data fusion according to claim 1, wherein: In step S1, the user behavior data of the touchpoint is represented by the calculation result obtained by grading and weighted summation through recording the number of clicks, residence duration, and page jump path of the user on different touchpoints; The transaction data of the touchpoint is obtained by detecting the number of orders completed and the transaction amount of the user on different touchpoints.

3. The multi-touchpoint marketing effect attribution method based on AI big data fusion according to claim 2, wherein: In step S1, the system sets multiple groups of touchpoint devices, respectively detects the user behavior data and transaction data of each group of touchpoints, and performs normalization processing to obtain multiple groups of behavior data coefficients and transaction data coefficients.

4. The multi-touchpoint marketing effect attribution method based on AI big data fusion according to claim 3, wherein: In step S2, the processed behavior data coefficients are merged into a behavior data set, and the transaction data coefficients are merged into a transaction data set; The specific steps for setting the weight threshold are as follows: Make the same processing for the two data sets, take the data set as the analysis data set, divide the analysis data set into two parts with the same number, calculate the average values in the two parts of the data set respectively for comparison, and select the part with the larger average value as the new round of analysis data set; Then divide the new round of analysis data set into two parts with the same number to calculate the average value, and select the part with the smaller average value as the new round of analysis data set; Alternately take the larger and smaller values and repeat the operation on the data set until the last data remains; Calculate the average value of the data obtained by processing the two data sets and set it as the weight threshold.

5. The multi-touchpoint marketing effect attribution method based on AI big data fusion according to claim 4, wherein: In step S2, the system calculates the behavior data coefficient and transaction data coefficient of the touchpoint to be measured and takes the average value as the touchpoint weight coefficient; When the touchpoint weight coefficient exceeds the weight threshold, use the dynamic time warping algorithm to perform preliminary matching on the time series data between touchpoints.

6. The multi-touchpoint marketing effect attribution method based on AI big data fusion according to claim 1, wherein: In step S3, the interaction frequency obtains the interaction times between touchpoints through the timestamp field in the access log table; the interaction depth is reflected by recording the number of interaction levels between touchpoints within a set time period.

7. The multi-touch marketing effect attribution method of AI big data fusion according to claim 5, characterized in that: In step S3, the specific steps of using fuzzy inference when analyzing the correlation degree between contacts are as follows: Define the interaction frequency and the number of interaction levels as input variables and divide them into fuzzy sets; Define the correlation degree between contacts as the output variable and divide it into fuzzy sets; formulate fuzzy rules to describe the influence of the interaction frequency and the number of interaction levels on the correlation degree between contacts; Perform fuzzy inference according to the fuzzy rules to determine the correlation degree between contacts.

8. The multi-touch marketing effect attribution method of AI big data fusion according to claim 7, characterized in that: In step S3, when calculating the correlation strength between contacts, the interaction frequency and the number of interaction levels between contacts are summed to obtain a normalization parameter, and the linear normalization formula is used for calculation and acquisition according to the normalization parameter.

9. The multi-touch marketing effect attribution method of AI big data fusion according to claim 8, characterized in that: In step S4, when the correlation degree between contacts is low, the system uses the graph neural network algorithm to detect the relationship between contacts; When the correlation degree between contacts is high, the system uses the random forest algorithm to detect the relationship between contacts; When using the random forest algorithm to detect redundant problems between contacts, the redundancy strength between contacts is given; When no redundant problems occur, the relationship between contacts is recorded.

Images