Attribution Analysis System, Training Method, Attribution Analysis Method, and Device

Abstract

An embodiment of the present invention discloses an attribution analysis system, a training method, an attribution analysis method, and a device. The system includes: a contact encoding module, configured to perform contact encoding on the behavior data of an analysis object to obtain contact encoding data; a contact analysis module, configured to perform contact contribution analysis on the analysis object data and the operation object data to obtain initial contact contribution data, wherein the analysis object data includes the behavior data of the analysis object and analysis object tags; and a data fusion module, configured to perform data fusion on the contact encoding data and the initial contact contribution data to obtain the contact contribution data of the analysis object. Through the technical solution disclosed in the embodiment of the present invention, the efficiency of attribution analysis is improved on the premise of ensuring the accuracy of the attribution result.

Description

Technical Field

[0001] Embodiments of the present invention relate to artificial intelligence technology, and in particular, to an attribution analysis system, a training method, an attribution analysis method, and a device. Background Art

[0002] In the process of Internet shopping, the process from a user searching for a product to successfully placing an order can be regarded as the process of user conversion. Usually, user conversion is caused by multiple factors, and each behavior has a certain contribution degree. Merchants and platforms hope to determine the contribution degree of each behavior to the final conversion through attribution analysis, that is, analyzing user behavior, so as to guide more efficient marketing strategies and resource allocation.

[0003] There are two common types of attribution analysis methods in the prior art. One is rule-based, and the other is a multi-touch attribution model.

[0004] In the implementation process of the prior art, when performing attribution based on the rule method, there is a problem of single feature consideration, resulting in inaccurate attribution results. The multi-touch attribution model can avoid the above defects by using a training model with multiple features such as positive and negative samples. However, a major problem with the existing multi-touch attribution model is its large computational complexity and time-consuming nature, resulting in low efficiency of attribution analysis. Summary of the Invention

[0005] The present invention provides an attribution analysis system, a training method, an attribution analysis method, and a device to improve the efficiency of attribution analysis while ensuring the accuracy of attribution results.

[0006] According to one aspect of the present invention, an attribution analysis system is provided. The system includes:

[0007] A touchpoint encoding module for performing touchpoint encoding on the behavior data of the analysis object to obtain touchpoint encoded data;

[0008] A touchpoint analysis module for performing touchpoint contribution analysis on the analysis object data and the operation object data to obtain initial touchpoint contribution data, where the analysis object data includes the behavior data of the analysis object and the analysis object label;

[0009] A data fusion module for fusing the touchpoint encoded data and the initial touchpoint contribution data to obtain the touchpoint contribution data of the analysis object.

[0010] According to another aspect of the present invention, an attribution analysis method is provided. The method includes:

[0011] Obtaining analysis object data and operation object data, and inputting the analysis object data and the operation object data into the attribution analysis system according to any one of the embodiments to obtain the touchpoint contribution data output by the attribution analysis system.

[0012] According to another aspect of the present invention, there is provided a training method for an attribution analysis system, the training method comprising:

[0013] Obtain sample data, wherein the sample data includes positive sample data with successful operations and negative sample data with failed operations, the analysis object data and the operation object data;

[0014] Based on the sample data, perform iterative training on the attribution analysis system to be trained until the training end condition is met, and obtain a trained attribution analysis system:

[0015] Input the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, wherein the contact contribution data includes first data of behavior contacts existing in the sample data and second data of behavior contacts not existing in the sample data;

[0016] Based on the first data and the second data, and their respectively corresponding standard data, determine a loss function, and adjust the parameters of the attribution analysis system based on the loss function.

[0017] According to another aspect of the present invention, there is provided a training device for an attribution analysis system, the mental device comprising:

[0018] A data acquisition module for obtaining sample data, wherein the sample data includes positive sample data with successful operations and negative sample data with failed operations, the analysis object data and the operation object data;

[0019] An attribution analysis system training module for performing iterative training on the attribution analysis system to be trained based on the sample data until the training end condition is met, and obtaining a trained attribution analysis system:

[0020] A contact contribution data acquisition module for inputting the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, wherein the contact contribution data includes first data of behavior contacts existing in the sample data and second data of behavior contacts not existing in the sample data;

[0021] An attribution analysis system adjustment module for determining a loss function based on the first data and the second data, and their respectively corresponding standard data, and adjusting the parameters of the attribution analysis system based on the loss function.

[0022] According to another aspect of the present invention, there is provided an electronic device, the electronic device comprising:

[0023] At least one processor; and

[0024] A memory communicatively connected to the at least one processor; wherein,

[0025] The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the at least one processor is enabled to execute the attribution analysis method according to any embodiment of the present invention; and / or, execute the training method of the attribution analysis system according to any embodiment.

[0026] According to another aspect of the present invention, there is provided a computer-readable storage medium storing computer instructions for causing a processor to implement the attribution analysis method according to any embodiment of the present invention when executed; and / or, execute the training method of the attribution analysis system according to any embodiment.

[0027] The technical solution of the embodiment of the present invention specifically includes: a contact coding module for performing contact coding on the behavior data of the analysis object to obtain contact coding data; a contact analysis module for performing contact contribution analysis on the analysis object data and the operation object data to obtain initial contact contribution data, wherein the analysis object data includes the behavior data of the analysis object and analysis object tags; a data fusion module for fusing the contact coding data and the initial contact contribution data to obtain the contact contribution data of the analysis object. The technical solution of the embodiment of the present invention directly inputs the behavior data into the above-mentioned attribution analysis system and directly obtains the contact contribution data of each behavior output by the attribution analysis system, achieving the improvement of the efficiency of attribution analysis while ensuring the accuracy of the attribution result.

[0028] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present invention, nor is it used to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0030] Figure 1 It is a structural diagram of an attribution analysis system provided in an embodiment of the present invention;

[0031] Figure 2 It is a structural diagram of another attribution analysis system provided in an embodiment of the present invention;;

[0032] Figure 3 is a flowchart of a method for attribution analysis provided in an embodiment of the present invention;

[0033] Figure 4 is a flowchart of a method for training an attribution analysis system provided in an embodiment of the present invention;

[0034] Figure 5 is a flowchart of another method for training an attribution analysis system provided in an embodiment of the present invention;

[0035] Figure 6 is a structural diagram of another attribution analysis system provided in an embodiment of the present invention;

[0036] Figure 7 is a structural diagram of a device for training an attribution analysis system provided in an embodiment of the present invention;

[0037] Figure 8 is a structural diagram of an electronic device provided in an embodiment of the present invention. Detailed implementation manners

[0038] In order to enable those skilled in the art to better understand the solution of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all 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.

[0039] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above drawings are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0040] Figure 1The figure is a structural diagram of an attribution analysis system provided by an embodiment of the present invention. This embodiment is applicable to analyzing the contribution degree of each operation behavior to the completion of an operation by an analysis object during the process of completing a certain operation. In particular, it is applicable to the scenario of analyzing the contribution degree of each behavior of a user who successfully places an order before placing the order during the online shopping process; at the same time, it can also be applicable to other scenarios for analyzing the contribution degree of operation behaviors. Specifically, the system includes the following modules:

[0041] The contact point coding module 110 is used to perform contact point coding on the behavior data of the analysis object to obtain contact point coding data;

[0042] The contact point analysis module 120 is used to perform contact point contribution analysis on the analysis object data and the operation object data to obtain initial contact point contribution data, where the analysis object data includes the behavior data of the analysis object and the analysis object label;

[0043] The data fusion module 130 is used to fuse the contact point coding data and the initial contact point contribution data to obtain the contact point contribution data of the analysis object.

[0044] In the embodiment of the present invention, the analysis object can be understood as a behavior object that has executed at least one behavior contact point to achieve a preset execution result. The analysis object can include, but is not limited to, people and electronic devices, etc. In some embodiments, the analysis object can be the operating user who performs the operation. Generally, there are multiple influencing factors that lead to the achievement of the preset execution result, and each influencing factor makes a greater or lesser contribution. In order to analyze the contribution degree, that is, the contribution rate, of each influencing factor in the process of achieving the preset execution result, it is necessary to perform attribution analysis on each influencing factor. Specifically, contact point analysis can be performed on the analysis object, the behavior contact points executed by the analysis object, and the operation objects involved in the process of executing the behavior contact points, so as to obtain the contact point contribution data of the analysis object. In this embodiment, the contact point contribution data can include each contact point and the contribution data corresponding to each contact point.

[0045] It should be noted that in this embodiment, the contact point can be understood as an influencing factor that leads to the achievement of the preset execution result. Correspondingly, the contact point can be understood as the user behavior that affects the preset behavior result during the process of the user achieving the preset execution result, or the contact point can also be understood as an operation node during the task execution process. Exemplarily, taking user behavior as an example, if there are multiple user behaviors, there will be multiple behavior contact points accordingly; that is, each user behavior can be understood as a behavior contact point, and a behavior sequence can be generated in the order of execution during the execution of the user behavior.

[0046] Further, the process of analyzing the above behavior sequence to determine the contribution degree of each influencing factor can also be understood as the process of attributing each behavior contact point in the behavior sequence to determine the contact point contribution data of the analysis object.

[0047] Optionally, the method for obtaining the contact point contribution data of the analysis object may include obtaining the analysis object data of the analysis object and the operation object data of the operation object for which the analysis object executes the user behavior. Among them, the analysis object data includes the behavior data corresponding to the execution process of the user behavior and the analysis object label, and the above data is subjected to data analysis processing to obtain the contact point contribution data of the analysis object.

[0048] In this embodiment, during the execution process of the user behavior by the analysis object, the data generated by each user behavior collected by the system can be used as the behavior data. Among them, the storage of the behavior data can be sequentially stored based on the execution order of each user behavior, so that based on the behavior data, the execution order of the user behavior in the process of achieving the preset execution result can be determined, that is, the user behavior sequence is obtained. Specifically, different user behaviors will be executed according to different business scenarios, thereby generating different behavior data. For example, in the Internet shopping scenario, the analysis object may include the user who places an order. Correspondingly, the user behavior can be understood as behaviors such as searching for goods, opening the product details viewing page, and viewing the preferential activities set by the shopping platform or merchant. Another example is in the financial scenario, the analysis object may include the user who conducts financial management. Correspondingly, the user behavior can be understood as behaviors such as searching for financial products and viewing financial products.

[0049] Optionally, obtain the behavior data corresponding to each user behavior respectively, and perform contact point encoding on the behavior data based on the contact point encoding module 110 to obtain contact point encoding data. Through the contact point encoding data, the type of behavior contact point in the user behavior data can be determined, that is, the type of behavior contact point existing in the user behavior, so as to realize the synchronous analysis of multiple behavior contact points in the behavior data in one attribution analysis process, reduce the number of analyses of multiple behavior contact points, and improve the attribution analysis efficiency of multiple behavior contact points.

[0050] Specifically, the encoding data can be understood as a one-dimensional vector matrix composed of each behavior contact point and the behavior data corresponding to each behavior contact point. Contact point encoding can be understood as encoding and updating the encoding data corresponding to the behavior contact points in the initial encoding data based on the behavior contact points identified in the process of achieving the preset execution result, so as to obtain the encoding vector corresponding to each behavior data.

[0051] In this embodiment, the contact encoding module may be a binary encoder. Correspondingly, the method of performing contact encoding can be understood as performing binary encoding update on the encoding values corresponding to the behavior contacts included in the obtained behavior data, so as to determine which user behaviors are executed in the process of achieving the current preset behavior result, that is, to determine the user behaviors that contribute to the current preset behavior result.

[0052] In the embodiment of the present invention, contact encoding is performed on the behavior data of the analysis object based on the contact encoding module 110. Specifically, the contact encoding module is used to identify the types of behavior contacts in the behavior data; in the initial encoding data, the encoding values corresponding to the identified behavior contact types are set to preset encoding values, and the encoding values corresponding to the unrecognized behavior contact types are kept as the initial encoding values.

[0053] In this embodiment, in the process of generating behavior data based on each user behavior, a string of identifiers is generated as the behavior contact identifier of the user behavior when any user behavior is executed, and the behavior contact identifier of the user behavior is stored in the behavior data. Optionally, the behavior contact identifier may be a string generated based on the behavior execution time and the behavior type; of course, it may also be other forms of identifiers, which are not limited in this embodiment.

[0054] Specifically, in the process of the contact encoding module 110 identifying the behavior data, if any behavior contact identifier is identified, the corresponding encoding value in the contact encoding is updated based on the behavior contact identifier. In this embodiment, the initial encoding data includes the initial encoding values corresponding to each behavior contact. For example, the initial contact encoding is a one-dimensional vector, and the one-dimensional vector includes the initial encoding values corresponding to multiple behavior contacts. Specifically, based on the identified behavior contact type, the encoding value corresponding to the behavior contact type in the initial contact encoding is updated; for example, the initial encoding value of the behavior contact type is updated to a preset encoding value. Further, if there are still unrecognized behavior contact types after the behavior data of the analysis object is identified, the corresponding encoding values are kept as the initial encoding values. It should be noted that the initial encoding value in this embodiment may be 0, and the preset encoding value may be 1; of course, the preset encoding value may be other values other than 0, which are not limited in this embodiment. By setting the encoding values corresponding to the unrecognized behavior contact types to 0, the error interference of non-zero contribution values of behavior contacts that do not exist in the behavior data is avoided.

[0055] In this embodiment, for the behavioral contacts determined based on the contact coding module, the initial contact contribution data is determined based on the processing of the user object data and the operation object data by the contact analysis module 120. The initial contact contribution data includes the initial contribution data of the behavioral contacts existing in the behavioral data and the initial contribution values ​​of the behavioral contacts not existing in the behavioral data. The final contact contribution data is determined by fusing the initial contact contribution data with the contact coding.

[0056] It should be noted that the analysis object data includes the behavior data of the analysis object and the analysis object label of the analysis object. The analysis object label can be understood as a label pre-set for the analysis object, and the label is related to the business scenario to which it is applied. For example, in the Internet shopping scenario, the analysis object label can include but is not limited to labels such as age, gender, region, and purchase preferences for placing orders. For another example, in the financial scenario, the analysis object label can include but is not limited to labels such as accounts and occupations of financial management users. The operation object can be understood as the object on which the analysis object performs user behavior. The operation object data may include attribute data of the operation object; for example, in the Internet shopping scenario, the operation object may be a commodity ordered by a user, and accordingly, the operation object data includes but is not limited to data such as brand quality and historical user evaluation of the commodity. For another example, in the financial scenario, the financial management product applied for by the user; accordingly, the operation object data includes but is not limited to data such as the income and suitable occupation of the product.

[0057] In some embodiments, the contact analysis module 120 may be a network structure module, and the network structure of the contact analysis module 120 may be a structure such as a convolutional neural network, a multi-layer perceptron, etc., which is not limited. The contact analysis module 120 is pre-trained and has a contribution data analysis function, and analyzes and processes the input user object data and operation object data to obtain initial contact contribution data.

[0058] In some embodiments, the contact analysis module 120 may include a pre-trained encoder and a contribution data analyzer. The encoder and the contribution data analyzer may be pre-trained network structures respectively, and the specific structures of the encoder and the contribution data analyzer are not limited. After acquiring the behavior data of the analysis object, the behavior features in the above behavior data may be extracted based on the pre-trained encoder. The encoder may be an encoder with a semantic extraction function. The contribution data analyzer performs contact contribution analysis based on the behavior features, the analysis object label, and the operation object data to obtain the initial contact contribution data. That is, it is possible to determine which user behaviors in the above behavior data have contributed to the process of achieving the preset execution result.

[0059] Further, after obtaining the initial contact contribution data, the contact encoding data and the initial contact contribution data are subjected to data fusion processing based on the data fusion module, so as to obtain the contact contribution data of the analysis object. Specifically, the contact encoding data and the initial contact contribution data may be subjected to an inner product to obtain the contact contribution data of the analysis object.

[0060] The attribution analysis system in this embodiment is an end-to-end processing system. Among them, the input ends of the contact encoding module and the contact analysis module are used as the input ends of the attribution analysis system. The input data corresponding to the contact encoding module is the behavior data of the analysis object. The input information of the contact analysis module includes the analysis object data and the operation object data. The attribution analysis system automatically processes the above input data and outputs the analysis result of the analysis object, that is, the contact contribution data. For the operating user, this end-to-end attribution analysis system has a simple processing method and reduces the skill requirements for the operating user.

[0061] In the technical solution of the embodiment of the present invention, the contact encoding module encodes the behavior data of the analysis object to obtain contact encoding data; the contact analysis module performs contact contribution analysis on the analysis object data and the operation object data to obtain initial contact contribution data, and through the data fusion module, the contact encoding data and the initial contact contribution data are fused to obtain the contact contribution data of the analysis object. The technical solution of the embodiment of the present invention extracts all behavior contacts in the behavior data, performs synchronous attribution analysis on multiple behavior contacts based on the analysis object data and the operation object data, and directly obtains the contact contribution data of each behavior contact output by the attribution analysis system, realizing the improvement of the attribution analysis efficiency on the premise of ensuring the accuracy of the attribution result.

[0062] Figure 2 Shown in the figure is a structural diagram of another attribution analysis system provided by an embodiment of the present invention. The embodiments of the present invention may be combined with each of the above optional solutions. On the basis of the above embodiments, optionally, the contact analysis module 120 includes: a semantic feature extraction unit 121, configured to perform semantic extraction on the behavior data of the analysis object to obtain behavior semantic features; a feature interaction unit 122, configured to process the behavior semantic features, the analysis object label, and the operation object data to obtain interaction features; and a contribution processing unit 123, configured to process the interaction features to obtain initial contact contribution data.

[0063] As Figure 2 shown, the embodiment of the present invention specifically includes the following content:

[0064] In this embodiment, the behavior semantics extraction unit 121 may include a pre-trained encoder; specifically, based on the pre-trained encoder, semantic extraction is performed on the behavior data to obtain the behavior semantic features in the behavior data. The feature interaction unit 122 may include a pre-trained multi-layer perceptron network; specifically, the multi-layer perceptron network is used to perform feature interaction processing on the extracted behavior semantic features and the obtained analysis object label and operation object data to obtain the interaction features of the analysis object.

[0065] The contribution processing unit 123 includes a processing layer 1231 and an activation function layer 1232. Among them, the value range of the activation function in the activation function layer is greater than 0, and is used to convert the processing result of the processing layer into a non-negative number. The processing layer 1231 can be understood as a pre-trained fully connected layer network; specifically, based on the fully connected layer network, feature processing is performed on the above-obtained interaction features, so as to obtain the processing result output by the contribution layer. The value range of the activation function in the activation function layer 1232 is greater than 0, and is used to convert the processing result of the processing layer into a non-negative number, so as to obtain the initial contact contribution data of the analysis object. Exemplarily, the activation function in the activation function layer 1232 may be a Sigmoid function, and the value range of the Sigmoid function is greater than 0 and less than 1, that is:

[0066]

[0067] The beneficial effect of using the above activation function layer 1232 to process the output result of the processing layer 1231 is that it can directly avoid negative values in the output result. Because usually, negative values indicate that the behavior contact plays a negative role in the process of achieving the preset execution result. Therefore, the above technical solution avoids the situation where the processing result has negative values and the interpretability is poor.

[0068] It should be noted that the implementation manners of the units in the contact analysis module 120 are only introduced as optional embodiments. In specific applications, other implementation manners may also be adopted according to actual situations to implement contact analysis, and the present embodiment does not limit the implementation manners of the above units.

[0069] The technical solution of the embodiment of the present invention extracts the semantics of the behavior data of the analysis object to obtain the behavior semantic features, processes the behavior semantic features, the analysis object tags, and the operation object data through the feature interaction unit to obtain the interaction features; processes the interaction features through the contribution processing unit to obtain the initial contact contribution data, and obtains the target contact contribution data by fusing the initial contact contribution data with the contact coding data. The technical solution of the embodiment of the present invention directly inputs the behavior data into the above attribution analysis system, and directly obtains the contact contribution data of each behavior output by the attribution analysis system. Moreover, the contribution processing unit in the attribution analysis system uses a specific activation function to ensure that all contributions are non-negative, avoiding the situation where negative values appear in the processing results and resulting in poor interpretability.

[0070] Based on the above embodiment, Figure 3 FIG. is a flowchart of an attribution analysis method provided by an embodiment of the present invention. This embodiment is applicable to analyzing the contribution degree of each operation behavior of an analysis object to the completion of an operation during the process of completing a certain operation, especially applicable to the scenario of analyzing the contribution degree of each behavior of a user who successfully places an order before placing the order during online shopping; at the same time, it can also be applicable to the scenario of analyzing the contribution degree of operation behaviors in other scenarios. This method can be executed by the above attribution analysis system, which can be implemented by software and / or hardware, and the attribution analysis system can be configured on an electronic computing device.

[0071] As Figure 3 shown, the method specifically includes the following steps:

[0072] S310. Obtain the analysis object data and the operation object data.

[0073] S320. Input the analysis object data and the operation object data into the attribution analysis system to obtain the contact contribution data output by the attribution analysis system.

[0074] In this embodiment, the analysis object data may include the behavior data of the analysis object and the analysis object tags. Specifically, the analysis object tags can be understood as user tags pre-set for the analysis object, and the tag is related to the business scenario to which it is applied. For example, in the Internet shopping scenario, the analysis object tags may include, but are not limited to, tags such as age, gender, region, and purchase preference for placing orders. Another example is in the financial scenario, the analysis object tags may include, but are not limited to, tags such as age, region, and occupation of wealth management users. The operation object can be understood as the object on which the analysis object performs user behavior. The operation object data may include the attribute data of the operation object; for example, in the Internet shopping scenario, the operation object may be the product ordered by the user. Correspondingly, the operation object data includes, but is not limited to, data such as the brand quality and historical user reviews of the product. Another example is in the financial scenario, the financial product applied for by the user; correspondingly, the operation object data includes, but is not limited to, data such as the income of the product and the suitable occupation.

[0075] Further, the obtained analysis object data and operation object data are input into the attribution analysis system, so as to obtain the contact point contribution data output by the attribution analysis system. In this embodiment, the contact point contribution data may include each contact point and the contribution data corresponding to each contact point. The attribution analysis system can be understood as any of the attribution analysis systems in the above embodiments, and will not be elaborated again in this embodiment. The beneficial effect of using the above attribution analysis system to determine the contact point contribution data is that after the input data is input into the system, the contact point contribution data corresponding to each behavior contact point of the input data can be directly obtained. In this way, for the input data of each analysis object, only by calculating once through the above system, the contribution data of each behavior contact point can be obtained; on the premise of ensuring the accuracy of the calculation result, the calculation amount of the system is reduced, thereby improving the analysis efficiency of the attribution analysis.

[0076] The technical solution of the embodiment of the present invention realizes the improvement of the attribution analysis efficiency on the premise of ensuring the accuracy of the attribution result by directly inputting the behavior data into the above attribution analysis system and directly obtaining the contact point contribution data of each behavior output by the attribution analysis system.

[0077] Based on the above embodiment, Figure 4The flowchart of a method for training an attribution analysis system provided by an embodiment of the present invention is applicable to the situation of analyzing the contribution degree of each operation behavior of an analysis object to the completion of an operation during the process of completing a certain operation, and is particularly applicable to the scenario of analyzing the contribution degree of each behavior of a user who successfully places an order before placing the order during the online shopping process; at the same time, it can also be applicable to the scenario of analyzing the contribution degree of operation behaviors in other scenarios. This method can be executed by the above-mentioned attribution analysis system training device, which can be implemented by software and / or hardware, and can be configured on an electronic computing device.

[0078] As Figure 4 shown, the method specifically includes the following steps:

[0079] S410. Obtain sample data, where the sample data includes positive sample data of successful operations, negative sample data of failed operations, analysis object data, and operation object data.

[0080] S420. Input the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, where the contact contribution data includes the first data of the behavior contacts existing in the sample data and the second data of the behavior contacts not existing in the sample data.

[0081] S430. Determine a loss function based on the first data, the second data, and their respectively corresponding standard data, and adjust the parameters of the attribution analysis system based on the loss function.

[0082] When the training end condition is satisfied, a trained attribution analysis system is obtained. When the training end condition is not satisfied, return to execute step S420 until the training end condition is satisfied.

[0083] In the embodiment of the present invention, the sample data can be understood as the historical behavior data of the analysis object performing behavior contacts to achieve a preset execution result. For different business scenarios, the executed behavior contacts are different, and thus the obtained sample data is also different. Specifically, the sample data includes positive sample data of successful operations and negative sample data of failed operations. For example, in the Internet shopping scenario, the sample data can include the sample data of shopping users during the shopping order placement process. Correspondingly, the positive sample data of successful operations can be understood as the sample data of users who successfully place an order after browsing the goods. The negative sample data of failed operations can be understood as the sample data of users who do not successfully place an order after browsing the goods.

[0084] The data of the object to be analyzed may include the behavior data of the object to be analyzed and the tags of the object to be analyzed. Specifically, the tags of the object to be analyzed can be understood as user tags pre-set for the object to be analyzed, and the tag is related to the business scenario to which it is applied. For example, in the scenario of online shopping, the tags of the object to be analyzed may include, but are not limited to, tags such as age, gender, region, and purchase preference for placing orders. For another example, in the financial scenario, the tags of the object to be analyzed may include, but are not limited to, tags such as age, region, and occupation of wealth management users. The operation object can be understood as the object on which the object to be analyzed performs user behavior. The operation object data may include the attribute data of the operation object. For example, in the scenario of online shopping, the operation object may be the product ordered by the user. Correspondingly, the operation object data includes, but is not limited to, data such as the brand quality and historical user evaluations of the product. For another example, in the financial scenario, the wealth management product applied for by the user; correspondingly, the operation object data includes, but is not limited to, data such as the income of the product and the suitable occupation.

[0085] Further, obtain the attribution analysis system to be trained, and perform iterative training on the attribution analysis system to be trained based on the sample data to obtain the trained attribution analysis system. Specifically, the training process may include: inputting the sample data into the attribution analysis system to be trained to obtain the sample prediction result output by the system. The sample prediction result includes the contact contribution data of the object to be analyzed in the sample data, where the contact contribution data includes the first data of the behavior contacts existing in the sample data and the second data of the behavior contacts not existing in the sample data. The above technical solution can directly know the behavior contacts that contribute and the behavior contacts that do not contribute in the process of achieving the preset execution behavior by separately obtaining the first data of the behavior contacts existing in the sample data and the second data of the behavior contacts not existing in the sample data, and can more accurately determine the contact contribution data corresponding to each behavior contact.

[0086] Further, obtain the standard data corresponding to the first data and the second data respectively, and determine the loss function based on the first data, the second data, and their respective corresponding standard data, and adjust the parameters of the attribution analysis system based on the loss function. Among them, the standard data can be the pre-set target data, that is, the target data can be used as the label data to generate the loss function with the prediction data output by the model to be trained, so as to complete the parameter adjustment of the attribution analysis system. Exemplarily, for the positive sample data, the target data corresponding to the first data can be 1, and the second data can be 0; for the negative sample data, the target data corresponding to the first data can be 0, and the second data can be 0.

[0087] Specifically, the attribution analysis system to be trained can be repeatedly trained based on the determined loss function until the attribution analysis system converges during the training process, that is, the loss value of the attribution analysis system tends to zero or remains stable for a long time without changing with the increase in the number of training times. It is determined that the attribution analysis system at this time meets the training stop condition, that is, the model training is completed, and the trained attribution analysis system is obtained.

[0088] In the technical solution of this embodiment, by obtaining sample data, where the sample data includes positive sample data of successful operations and negative sample data of failed operations, the analysis object data and the operation object data; iteratively training the attribution analysis system to be trained based on the sample data until the training end condition is met, and obtaining the trained attribution analysis system: inputting the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, where the contact contribution data includes the first data of the behavior contacts existing in the sample data and the second data of the behavior contacts not existing in the sample data; determining the loss function based on the first data and the second data, and their corresponding standard data respectively, and adjusting the parameters of the attribution analysis system based on the loss function. The above technical solution trains the attribution analysis system by using the first data of the behavior contacts existing in the sample data and the second data of the behavior contacts not existing in the sample data to obtain the trained model, so as to directly know the behavior contacts that contribute and the behavior contacts that do not contribute during the process of achieving the preset execution behavior, and can more accurately determine the contact contribution data corresponding to each behavior contact, thereby improving the efficiency of attribution analysis on the premise of ensuring the accuracy of the attribution result.

[0089] Based on the above embodiment, Figure 5 It is a flowchart of another method for training an attribution analysis system provided by an embodiment of the present invention. The optional solutions in the embodiments of the present invention can be combined with each other. In the embodiment of the present invention, optionally, the semantic feature extraction unit in the attribution analysis system is a pre-trained encoder;

[0090] Correspondingly, adjusting the parameters of the attribution analysis system based on the loss function includes: adjusting the parameters of the feature interaction unit and the contribution processing unit in the attribution analysis system based on the loss function.

[0091] As Figure 5 shown, the method specifically includes the following steps:

[0092] S510. Obtain sample data, where the sample data includes positive sample data of successful operations and negative sample data of failed operations, the analysis object data and the operation object data;

[0093] S520. Input the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data. The contact contribution data includes the first data of the behavioral contacts existing in the sample data and the second data of the behavioral contacts not existing in the sample data.

[0094] S530. Based on the first data, the second data, and their respective corresponding standard data, determine the loss function, and adjust the parameters of the feature interaction unit and the contribution processing unit in the attribution analysis system based on the loss function.

[0095] When the training end condition is satisfied, obtain the trained attribution analysis system. When the training end condition is not satisfied, return to execute step S520 until the training end condition is satisfied.

[0096] In the process of the embodiment of the present invention, the attribution analysis system includes a contact analysis module, and the contact analysis module includes a semantic feature extraction unit, a feature interaction unit, and a contribution processing unit. Among them, the semantic feature extraction unit is a pre-trained encoder.

[0097] On the basis of the above embodiment, in the process of training the attribution analysis system in this embodiment, the pre-trained semantic feature extraction unit can be directly used to adjust the parameters of the feature interaction unit and the contribution processing unit in the contact analysis module, so as to obtain the trained attribution analysis system.

[0098] It should be noted that before training the attribution analysis system, since the trained semantic feature extraction unit is directly adopted, it is necessary to pre-train the semantic feature extraction unit in advance.

[0099] Optionally, the training method of the semantic feature extraction unit includes: obtaining the behavioral data of the analysis object in the sample data, and extracting the behavioral sequence in the behavioral data; determining the training input data and the standard output data based on the behavioral sequence, where the training input data is a local continuous subsequence in the behavioral data, and the standard output data is a local continuous subsequence that is one position behind the training input data; inputting the training input data into the training model to obtain the training output data output by the training model, where the training model includes an encoder and a decoder; adjusting the parameters of the training model based on the standard output data and the training output data, and iteratively executing the training process of the training model until the training completion condition is satisfied, and determining the trained encoder as the semantic feature extraction unit.

[0100] Specifically, the method for obtaining the behavior sequence in the behavior data may include: extracting the contact data of each behavior contact according to the execution order of each behavior contact to obtain the behavior sequence in the behavior data. Exemplarily, in the Internet shopping scenario, users perform a series of interactive behaviors such as browsing, clicking, and inputting, and finally place an order to purchase. In the behavior data corresponding to the above user behaviors, the obtained behavior sequence may include search - recommendation - advertisement - product detail page - order placement. It can represent the execution order that the user first searches, then browses the advertisement, reaches the product detail page, and finally places an order.

[0101] Furthermore, determine the training input data and the standard output data based on the obtained behavior sequence. Optionally, the obtained behavior sequence can be encoded by sequence to obtain each subsequence with a temporal relationship. Among them, the prerequisite relationship between the pre - behavior and the post - behavior of the analysis object can be represented between each subsequence. For example, in the above Internet shopping scenario, search is the prerequisite behavior for recommendation, and vice versa, recommendation is the post - behavior of search. Thus, the obtained training model can more accurately extract the semantic features of the behavior data.

[0102] It should be explained that the training input data is the local continuous subsequence in the behavior data; the standard output data is the local continuous subsequence that is one position later than the training input data.

[0103] Specifically, the local continuous subsequence can be understood as the continuous partial sequence in the behavior sequence. For example, in the above Internet shopping scenario, the behavior sequence in the obtained behavior data includes search - recommendation - advertisement - product detail page - order placement. Then the local continuous subsequences may include continuous subsequences such as search - recommendation, search - recommendation - advertisement, and recommendation - advertisement - product detail page. However, it should be noted that in order to obtain the standard output data that is one position later, the local continuous subsequence in the training input data does not include the last behavior in the behavior sequence. Further, if it is determined that the current training input data is recommendation - advertisement - product detail page, the local continuous subsequence of the standard output may be advertisement - product detail page - order placement.

[0104] Furthermore, input the training input data into the training model to obtain the training output data output by the training model. Specifically, use the training output data as the sample prediction result, and use the above - mentioned standard output result based on the sample subsequence as the label data. Then generate the loss function of the training model based on the sample prediction result and the label data, and adjust the parameters of the training model based on the loss function to obtain the trained training model.

[0105] It should be noted that the training model includes an encoder and a decoder. Specifically, both the encoder and the decoder adopt unidirectional rather than bidirectional recurrent neural networks, and the effect is that the temporal relationship between subsequences in the above-mentioned behavior sequence can be learned more accurately.

[0106] In the technical solution of this embodiment, by pre-training the semantic feature extraction unit in the attribution analysis system and using the trained semantic feature extraction unit to train other units in the attribution analysis system, the training speed of the system is improved, thereby improving the efficiency of using the system for attribution analysis.

[0107] On the basis of the above technical solution, this embodiment also provides a preferred embodiment, which specifically introduces the use of the above attribution analysis system to perform attribution analysis on the behavior touchpoints of the analysis object in the Internet shopping scenario, so as to obtain the touchpoint contribution data of the analysis object. Specifically, see Figure 6 , and the technical solution includes the following contents:

[0108] Based on Figure 6 As can be seen, the figure specifically includes three parts: the first part is the encoder training module, the second part is the attribution analysis system and its corresponding training module. Among them, the attribution analysis system only updates the system parameters during training and is only used to directly obtain the output touchpoint contribution data after inputting the data during prediction. In this embodiment, the attribution analysis system is an end-to-end system, which is mainly reflected in that after inputting the data, the contribution degrees of each touchpoint in the output result can be directly obtained without repeatedly calling the system for calculation. In this embodiment, the contribution processing layer contains one-to-one corresponding behavior touchpoints and touchpoint contribution data. In this way, for each group of input data, only by calculating once through the above attribution analysis system, the contribution degrees of each behavior touchpoint in the group of input data can be obtained.

[0109] Optionally, before training the attribution analysis system, first train the encoder in the system, that is, train the encoder and decoder in the first part based on the user behavior sequence. Among them, the user behavior in the figure can be understood as the behavior data of the analysis object. Specifically, encode the user behavior sequence to obtain the training input data and output sequence for training the encoder; further, input the obtained training input data into the encoder and decoder to be trained, obtain the training output result output by the decoder, and use this training output result as the sample prediction data to generate a loss function with the output sequence as the label data, and adjust the parameters in the encoder and decoder to obtain the trained encoder and decoder.

[0110] Further, after the encoder is trained, the current attribution analysis system is trained based on the trained decoder. Among them, the input data of the attribution analysis system includes three types of data, including: the behavior data of the analysis object, the analysis object label in the analysis object data of the analysis object, and the operation object data of the analysis object. Among them, the behavior data of the analysis object can be understood as the user behavior sequence, the analysis object label in the analysis object data of the analysis object can be understood as the user portrait label, and the operation object data of the analysis object can be understood as the category brand characteristics. Specifically, the user behavior sequence refers to a sequence such as search - recommendation - advertisement - order placement, which constitutes a shopping path, indicating that the user first searches, then is recommended and exposed, then is exposed to advertisements, and then places an order to purchase. The user portrait label refers to the label attached to the user and is related to purchases, such as the user's age, gender, region, purchase preferences, etc. The category brand characteristics refer to which categories and brands of products the user has browsed, clicked on, and purchased.

[0111] Specifically, the trained encoder is used to encode the user behavior sequence, and a 128 - dimensional vector is output. The user portrait label uses a discrete - value vector, and the category brand characteristics both use 64 - dimensional embedding vectors. These three types of data are input into the feature interaction layer, that is, the feature interaction unit in the attribution analysis system. The feature interaction layer uses a simple multi - layer perceptron network, and this layer is used to analyze the interaction behaviors of different users on different categories. The touchpoint contribution output layer, that is, the contribution processing layer, is used to output the initial contribution degrees of each touchpoint, and the number of output results is equal to the number of touchpoints to be attributed. To make the credit of each touchpoint non - negative, we use the activation function Sigmoid:

[0112]

[0113] Among them, 0 < S(x), so S(x) is suitable for representing the contribution degree and successfully avoids negative values.

[0114] Furthermore, the binary encoder in this embodiment is the core that ensures a one-to-one correspondence between the contribution values and the contacts in the contact contribution layer. The purpose of the binary encoder is to binarize the user behavior sequence, making the positions of the contacts that appear 1 and those that do not appear 0. Thus, it can represent which contacts appear in the user behavior sequence. For example, if there are 100 contacts in the current user behavior sequence and the types of the user behavior sequence include search - recommendation - advertisement - search, a 100 - dimensional zero - valued vector (0, 0, 0, …, 0) can be generated. Further, if search, recommendation, and advertisement respectively correspond to the first three indices of the vector, then set these three index positions to 1 and the others to 0, obtaining the vector b = (1, 1, 1, …, 0), which is the output of the binary encoder. If the output vector of the contact contribution layer is y = (a_1, a_2, …, a_100), then the dot product of the vectors b·y = a_1 + a_2 + a_3 represents the sum of the contributions of the contacts in the sequence. Thus, the sum of the contributions of the contacts not in the sequence (1 - b)·y = a_4 + a_5 + (… + a)_100 can be calculated. This prepares for constructing the optimization objective of the attribution analysis system.

[0115] Furthermore, the goal of the attribution analysis system is that all the contacts that appear in the behavior sequence share the credit, and the contacts that do not appear have no credit. That is, assume a certain sample contains contacts a1, a2, …, a m , and the contacts not included are b1, b2, …, b n , m + n = 100 (assuming the total number of contacts is 100), and the label of this sample is 1. Then the goal can be expressed as:

[0116] a1 + a2 + … + a m = 1 and b1 + b2 + … + b n = 0.

[0117] Among them, a i , b j represent both the contacts and the contributions of the contacts. The above two goals are respectively obtained by taking the dot product with the binary vectors b and 1 - b.

[0118] Furthermore, based on the goal and the sample data, the attribution analysis system is trained. After obtaining the trained attribution analysis system, only the attribution analysis system needs to be used for attribution analysis. Among them, the input can be the user behavior sequence, the user portrait label, and the category brand characteristics, and the output result can be the contribution degree of each contact. In practical applications, the sum of the contributions of each contact is not equal to 1, so normalization is required, that is, the contributions of each contact are normalized. Specifically, the normalization process can adopt existing normalization methods, and this embodiment does not limit the normalization method.

[0119] Figure 7The figure shows the structure diagram of a training device for an attribution analysis system provided by an embodiment of the present invention. This embodiment is applicable to analyzing the contribution degree of each operation behavior of an analysis object to the completion of an operation during the process of completing a certain operation, especially applicable to the scenario of analyzing the contribution degree of each behavior of a user who successfully places an order before placing the order during the online shopping process; at the same time, it can also be applicable to the scenario of analyzing the contribution degree of operation behaviors in other scenarios. Refer to Figure 7 , the specific structure of the training device of the attribution analysis system includes:

[0120] A data acquisition module 610, configured to acquire sample data, where the sample data includes positive sample data of successful operations, negative sample data of failed operations, the analysis object data, and the operation object data;

[0121] A contact contribution data acquisition module 620, configured to input the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, where the contact contribution data includes first data of behavior contacts existing in the sample data and second data of behavior contacts not existing in the sample data;

[0122] An attribution analysis system adjustment module 630, configured to determine a loss function based on the first data, the second data, and their respectively corresponding standard data, and adjust the parameters of the attribution analysis system based on the loss function.

[0123] When the training end condition is satisfied, a trained attribution analysis system is obtained. When the training end condition is not satisfied, return to execute the contact contribution data acquisition module 620 until the training end condition is satisfied.

[0124] Optionally, the semantic feature extraction unit in the attribution analysis system is a pre-trained encoder;

[0125] Correspondingly, the attribution analysis system adjustment module 630 includes:

[0126] An attribution analysis system adjustment unit, configured to adjust the parameters of the feature interaction unit and the contribution processing unit in the attribution analysis system based on the loss function.

[0127] Optionally, the training method of the semantic feature extraction unit includes:

[0128] Obtain the behavior data of the analysis object in the sample data, and extract the behavior sequence in the behavior data;

[0129] Determine training input data and standard output data based on the behavior sequence, where the training input data is a local continuous subsequence in the behavior data, and the standard output data is a local continuous subsequence that is one position behind the training input data;

[0130] Input the training input data into a training model to obtain training output data output by the training model, where the training model includes an encoder and a decoder;

[0131] Adjust the parameters of the training model based on the standard output data and the training output data, and iteratively execute the training process of the training model. When the training completion condition is met, determine the trained encoder as the semantic feature extraction unit.

[0132] The training device of the attribution analysis system provided by the embodiments of the present invention can execute the training method of the attribution analysis system provided by any embodiment of the present invention, and has corresponding functional modules and beneficial effects for executing the method.

[0133] It should be noted that in the embodiments of the training device of the above attribution analysis system, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be realized; in addition, the specific names of the functional units are only for the convenience of mutual distinction and do not limit the protection scope of the present invention.

[0134] Figure 8 The structure diagram of an electronic device 10 that can be used to implement the embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are only examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0135] As Figure 8As shown, the electronic device 10 includes at least one processor 11 and a memory communicatively connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc. The memory stores a computer program executable by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0136] Multiple components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disc, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0137] The processor 11 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the attribution analysis method and / or the attribution analysis training method.

[0138] In some embodiments, the attribution analysis method and the training method of the attribution analysis system can be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed onto the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the attribution analysis method and / or the attribution analysis training method described above can be executed. Alternatively, in other embodiments, the processor 11 can be configured to execute the attribution analysis method and / or the attribution analysis training method by any other appropriate means (e.g., by means of firmware).

[0139] The various embodiments of the systems and techniques described above in this specification can be implemented in digital electronic circuitry, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor that receives data and instructions from, and transmits data and instructions to, a storage system, at least one input device, and at least one output device.

[0140] The computer programs for implementing the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the computer programs, when executed by the processor, cause the functions / operations specified in the flowchart and / or block diagram to be implemented. The computer programs can be executed entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

[0141] In the context of the present invention, a computer-readable storage medium can be a tangible medium that can contain, or store a computer program for use by or in connection with an instruction execution system, apparatus, or device. The computer-readable storage medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, the computer-readable storage medium can be a machine-readable signal medium. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0142] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and a pointing device (e.g., a mouse or a trackball) through which the user can provide input to the electronic device. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and the input received from the user can be in any form (including acoustic input, voice input, or tactile input).

[0143] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected to each other by digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include: local area network (LAN), wide area network (WAN), blockchain network, and the Internet.

[0144] The computing system can include a client and a server. The client and the server are generally far from each other and usually interact through a communication network. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system and solves the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services.

[0145] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps recited in the present invention can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved, and no limitation is imposed herein.

[0146] The above specific embodiments do not constitute a limitation on the protection scope of the present invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

[0147] Note that the above are only the preferred embodiments of the present invention and the technical principles applied. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, re-adjustments and substitutions can be made by those skilled in the art without departing from the protection scope of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments only. Without departing from the concept of the present invention, more other equivalent embodiments can be included, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. An attribution analysis system, characterized in that, Including: A contact encoding module, configured to perform contact encoding on the behavior data of an analysis object to obtain contact encoding data; A contact analysis module, configured to perform contact contribution analysis on the analysis object data and the operation object data to obtain initial contact contribution data, wherein the analysis object data includes the behavior data of the analysis object and analysis object labels; A data fusion module, configured to fuse the contact encoding data and the initial contact contribution data to obtain the contact contribution data of the analysis object; Wherein, the contact analysis module includes: A semantic feature extraction unit, configured to perform semantic extraction on the behavior data of the analysis object to obtain behavior semantic features; A feature interaction unit, configured to process the behavior semantic features, analysis object labels, and operation object data to obtain interaction features; A contribution processing unit, configured to process the interaction features to obtain initial contact contribution data.

2. The attribution analysis system according to claim 1, characterized in that, The contact encoding module is configured to identify the behavior contact types in the behavior data, and in the initial encoding data, set the encoding values corresponding to the identified behavior contact types to preset encoding values, and keep the encoding values corresponding to the unrecognized behavior contact types as the initial encoding values.

3. The attribution analysis system according to claim 1, characterized in that, The contribution processing unit includes a processing layer and an activation function layer, wherein the value range of the activation function in the activation function layer is greater than 0, and is used to convert the processing result of the processing layer into a non-negative number.

4. An attribution analysis method, characterized in that, Including: Obtain analysis object data and operation object data, input the analysis object data and operation object data into the attribution analysis system according to any one of claims 1-3, and obtain the contact contribution data output by the attribution analysis system.

5. A training method for an attribution analysis system according to any one of claims 1 - 3, characterized in that, Including: Obtain sample data, wherein the sample data includes positive sample data with successful operations and negative sample data with failed operations, the analysis object data and the operation object data; Iteratively train the attribution analysis system to be trained based on the sample data until the training end condition is met, and obtain the trained attribution analysis system: Input the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, wherein the contact contribution data includes first data of behavior contacts existing in the sample data and second data of behavior contacts not existing in the sample data; Determine a loss function based on the first data, the second data, and their respective corresponding standard data, and adjust the parameters of the attribution analysis system based on the loss function.

6. The method according to claim 5, characterized in that, The semantic feature extraction unit in the attribution analysis system is a pre-trained encoder; Correspondingly, the adjusting the parameters of the attribution analysis system based on the loss function includes: Adjusting the parameters of the feature interaction unit and the contribution processing unit in the attribution analysis system based on the loss function.

7. The method according to claim 6, characterized in that, The training method of the semantic feature extraction unit includes: Obtain the behavior data of the analysis object in the sample data, and extract the behavior sequence in the behavior data; Determine training input data and standard output data based on the behavior sequence, where the training input data is a local continuous subsequence in the behavior data, and the standard output data is a local continuous subsequence that is one position behind the training input data; Input the training input data into the training model to obtain the training output data output by the training model, where the training model includes an encoder and a decoder; Adjust the parameters of the training model based on the standard output data and the training output data, and iteratively execute the training process of the training model. When the training completion condition is met, determine the trained encoder as the semantic feature extraction unit.

8. A training device for an attribution analysis system according to any one of claims 1 - 3, characterized in that, Comprising: A data acquisition module for acquiring sample data, where the sample data includes positive sample data with successful operations, negative sample data with failed operations, the analysis object data, and the operation object data; An attribution analysis system training module for iteratively training the attribution analysis system to be trained based on the sample data until the training end condition is met, obtaining a trained attribution analysis system: A contact contribution data acquisition module for inputting the sample data into the attribution analysis system to obtain the contact contribution data of the analysis object in the sample data, where the contact contribution data includes first data of the behavior contacts existing in the sample data and second data of the behavior contacts not existing in the sample data; An attribution analysis system adjustment module for determining a loss function based on the first data and the second data, and their respectively corresponding standard data, and adjusting the parameters of the attribution analysis system based on the loss function.

9. An electronic device, characterized in that,The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; where The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can execute the attribution analysis method according to claim 4; and / or, execute the training method of the attribution analysis system according to any one of claims 5-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing a processor to implement the attribution analysis method according to claim 4 when executed; and / or, execute the training method of the attribution analysis system according to any one of claims 5-7.

11. A computer program product, characterized in that, The computer program product includes a computer program that implements the attribution analysis method according to claim 4 when executed by a processor; and / or, executes the training method of the attribution analysis system according to any one of claims 5-7.

Images