Order rate prediction method, device and readable storage medium based on multi-channel

Abstract

The present application discloses a method, device and readable storage medium for predicting order rate based on multiple channels, wherein the method comprises: determining a target user unique identifier shared by a target user in multiple channels, and then based on the target user unique identifier, uniformly collecting multi-channel user behavior data in each of the channels, and then based on the multi-channel user behavior data and a preset order rate prediction model, performing a multi-channel order rate prediction for the target user to obtain a target order rate. The present application solves the technical problem of low computational efficiency in predicting order rate.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, device and readable storage medium for predicting order rates based on multiple channels. Background Art

[0002] With the continuous development of financial technology, especially Internet technology finance, more and more technologies (such as distributed, blockchain, artificial intelligence, etc.) are applied in the financial field, but the financial industry also puts forward higher requirements on technology, such as higher requirements on the distribution of corresponding to-do items in the financial industry.

[0003] With the continuous development of computer software and artificial intelligence, the application of neural networks is becoming more and more extensive. At present, after collecting user behavior data in the channel, the order rate is usually predicted based on the user behavior data of each channel. Among them, the order rate includes the click rate of a certain item, the probability of user loan, etc. However, a user usually generates user behavior in multiple channels, and the user identifiers of each channel are different. For example, when the channel is a WeChat applet, the user identifier is the WeChat account, and when the channel is a mobile phone APP, the user identifier is the user's mobile phone number, which leads to the inability to unify the user behavior data collected in various channels, resulting in a large amount of redundancy in user behavior data. When predicting the order rate of the user based on the collected user behavior data, the redundant part of the user behavior data will be repeatedly calculated multiple times, which increases the amount of calculation when predicting the order rate and reduces the calculation efficiency when predicting the order rate. Summary of the invention

[0004] The main purpose of the present application is to provide a method, device and readable storage medium for predicting order rate based on multiple channels, aiming to solve the technical problem of low computational efficiency in predicting order rate in the prior art.

[0005] To achieve the above-mentioned purpose, the present application provides a method for predicting an order rate based on multiple channels, the method for predicting an order rate based on multiple channels being applied to a device for predicting an order rate based on multiple channels, and the method for predicting an order rate based on multiple channels comprising:

[0006] Determine the unique identifier of the target user shared by multiple channels;

[0007] Based on the unique identifier of the target user, uniformly collect multi-channel user behavior data in each of the channels;

[0008] Based on the multi-channel user behavior data and a preset order rate prediction model, a multi-channel order rate prediction is performed on the target user to obtain a target order rate.

[0009] The present application also provides a method and device for predicting an order rate based on multiple channels, wherein the method and device for predicting an order rate based on multiple channels is a virtual device, and the method and device for predicting an order rate based on multiple channels is applied to an order rate prediction device based on multiple channels, and the method and device for predicting an order rate based on multiple channels comprises:

[0010] A determination module, used to determine a target user unique identifier shared by a target user in multiple channels;

[0011] A collection module, used for uniformly collecting multi-channel user behavior data in each of the channels based on the unique identifier of the target user;

[0012] The prediction module is used to perform multi-channel order rate prediction for the target user based on the multi-channel user behavior data and a preset order rate prediction model to obtain a target order rate.

[0013] The present application also provides an order rate prediction device based on multiple channels, which is a physical device. The order rate prediction device based on multiple channels includes: a memory, a processor, and a program of the order rate prediction method based on multiple channels stored in the memory and executable on the processor. When the program of the order rate prediction method based on multiple channels is executed by the processor, the steps of the order rate prediction method based on multiple channels as described above can be implemented.

[0014] The present application also provides a readable storage medium, on which is stored a program for implementing a multi-channel order rate prediction method. When the program for the multi-channel order rate prediction method is executed by a processor, the steps of the multi-channel order rate prediction method are implemented as described above.

[0015] The present application provides a method, device and readable storage medium for predicting an order rate based on multiple channels. Compared with the technical means of the prior art that predicts the order rate separately based on the user behavior data of each channel, the present application allocates a unique user identifier to the target user in multiple channels, and can uniformly collect the multi-channel user behavior data corresponding to the target user in each channel, thereby eliminating a large amount of redundancy in the user behavior data. Further, the order rate is predicted based on the multi-channel user behavior data, and the process of repeatedly calculating the redundant part of the user behavior data can be eliminated, thereby reducing the amount of calculation when predicting the order rate. The technical defect in the prior art that the user behavior data collected by each channel cannot be unified, resulting in a large amount of redundancy in the user behavior data, and then the redundant part of the user behavior data is repeatedly calculated when predicting the order rate is overcome, thereby improving the calculation efficiency when predicting the order rate, and thus solving the technical problem of low calculation efficiency when predicting the order rate. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0018] Figure 1 This is a flow chart of a first embodiment of a method for predicting order rates based on multiple channels of the present application;

[0019] Figure 2 The preset user information mapping library in the form of a table in the multi-channel order rate prediction method of the present application;

[0020] Figure 3 A schematic diagram of a user behavior path diagram in the multi-channel order rate prediction method of the present application;

[0021] Figure 4 This is a flow chart of a second embodiment of a method for predicting order rates based on multiple channels of the present application;

[0022] Figure 5 A schematic diagram of the target user behavior atlas described in the multi-channel order rate prediction method of this application;

[0023] Figure 6 This is a flow chart of a third embodiment of a method for predicting order rates based on multiple channels of the present application;

[0024] Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0025] The purpose, features and advantages of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0026] It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0027] The present application embodiment provides a method for predicting an order rate based on multiple channels. In the first embodiment of the method for predicting an order rate based on multiple channels of the present application, referring to Figure 1 , the multi-channel order rate prediction method includes:

[0028] Step S10, determining a target user unique identifier shared by the target user in multiple channels;

[0029] In this embodiment, it should be noted that the access channel is a channel for collecting user data, wherein the user data includes user behavior data, etc. The access channel includes WeChat applets, mobile phone APPs, etc.

[0030] In addition, it should be noted that due to the differences in user information between the various channels, for the same user, there will be a large amount of redundancy in the user behavior data collected in the various channels. For example, assume that user A generates user behavior data a and user behavior data b on the WeChat applet, and generates user behavior data a1 and user behavior data c on the video APP, where user behavior data a and user behavior data a1 are the same. Since the user information of the WeChat applet is the WeChat account and the user information of the video APP is the mobile phone number, the data collector cannot know that user behavior data a and user behavior data a1 correspond to the same user A. After the data collector collects the data, it is impossible to unify user behavior data a and user behavior data a1, which causes a large amount of redundancy in user behavior data. In particular, when the amount of data collected by the data collector is large, the redundancy of user behavior data will be more serious.

[0031] In addition, it should be noted that the target user unique identifier is the unique identity identifier of the target user, which is used to identify the unique identity of the target user, wherein the target user unique identifier includes a code, a mobile phone number, and an ID number, etc. The multi-channel-based order rate prediction device is the data collector, and there is a preset user information mapping library in the data collector, wherein the preset user information mapping library is a database that stores the user information mapping relationship between each channel, wherein the preset user information mapping library can be represented by a table, such as Figure 2 The preset user information mapping library in a tabular form is shown, wherein mini programs, public accounts, APPs, adding corporate WeChat through the advertising landing page H5, etc. are all the channel channels, WeChat account, mobile phone number, device ID and APP user ID, etc. are all the user information, and "√" indicates that the corresponding user information exists in the channel channel.

[0032] Determine the target user unique identifier shared by the target user in multiple access channels. Specifically, obtain the channel user information of the target user in a single access channel, and then query the channel user information in multiple access channels corresponding to the target user based on the user information mapping library. Further, query the target user unique identifier based on each of the channel user information. For example, assuming that the user information of user A in access channel a is mobile phone number m and WeChat ID n, and mobile phone number m and WeChat ID n have no corresponding target user unique identifier, but the user information of user A in access channel b is mobile phone number m and device ID h, and the target user unique identifier corresponding to mobile phone number m and device number h is recorded as (T, C, C), then it can be determined that the target user unique identifier corresponding to mobile phone number m and WeChat ID n is (T, C, C).

[0033] In another feasible scheme, after obtaining the channel user information of a certain access channel, the channel user information is input into a hash coding model, and the channel user information is hash-coded to determine the target user unique identifier corresponding to the target user, wherein it should be noted that the hash coding model is a neural network model optimized based on preset multi-channel category information, and the hash coding model can output the same hash coding value for the input channel user information belonging to the same data category, and then when training the hash coding model, the channel user information of the user in each access channel is obtained as each training data, and each training data is identified as the same data category, and then after the training is completed, the same polarized hash coding value can be output for the channel user information of the target user in each of the access channels, and then the polarized hash coding value is used as the unique identifier of the target user.

[0034] The step of determining the target user unique identifier shared by the target user in multiple access channels comprises:

[0035] Step S11, obtaining the channel user information of the target user in each of the channel channels;

[0036] In this embodiment, the channel user information of the target user in each of the channel channels is obtained. Specifically, the single-channel user information of the target user in a single channel channel is obtained, and it is determined whether the single-channel user information has a corresponding target user unique identifier. If so, the target user unique identifier is directly extracted. If not, based on the user information mapping library, other channel user information of the target user in other channel channels is queried to obtain the channel user information of the target user in each of the channel channels.

[0037] Step S12, determining whether the channel user information under each of the access channels contains the user identification of the target user;

[0038] In this embodiment, it should be noted that one access channel corresponds to one channel user information.

[0039] Step S13: if it exists, use the user ID as the target user unique ID;

[0040] In this embodiment, if it exists, the user identifier is used as the unique identifier of the target user. Specifically, if there is a channel user information in each of the channel channels that corresponds to the user identifier of the target user, wherein the user identifier is used to identify that the channel user information belongs to the target user, the user identifier is directly used as the unique identifier of the target user.

[0041] Step S14: if it does not exist, then generate the target user unique identifier based on the common user information in the channel user information under each of the channel channels.

[0042] In this embodiment, if it does not exist, the unique identifier of the target user is generated based on the common user information in the channel user information under each of the channel channels. Specifically, if the channel user information in each of the channel channels does not correspond to the user identifier of the target user, a new user identifier is generated, and based on the common user information in the channel user information under each of the channel channels, the new user identifier is associated with the channel user information under each of the channel channels, and the preset user information mapping library is updated, and the new user identifier is used as the unique identifier of the target user.

[0043] The target user unique identifier includes a target hash code value.

[0044] The step of determining a target user unique identifier shared by a target user in a plurality of access channels comprises:

[0045] Step A10, obtaining channel user information corresponding to the target user, and inputting the channel user information into a hash coding model optimized based on preset multi-channel category information, performing polarized hash coding value on the channel user information, and obtaining a polarized hash coding value;

[0046] In this embodiment, it should be noted that the hash coding model includes a deep polarization network, wherein the deep polarization network is a deep learning model optimized based on preset multi-channel category information and a preset polarization loss function, and for input samples belonging to the same sample category, the hash coding model can output the same hash coding value, and the preset multi-channel category information is a set of preset hash coding values ​​corresponding to each sample category.

[0047] In addition, it should be noted that the deep polarization network includes a hidden layer and a hash layer, wherein the hidden layer is the data processing layer of the preset deep polarization network, which is used to perform data processing processes such as convolution processing and pooling processing, and the hidden layer is a one-layer or multi-layer neural network trained based on deep learning, and the hash layer is the output layer of the preset deep polarization network, which is used to perform polarized hashing on the data to be processed and output the corresponding hash result, and the hash layer is a one-layer or multi-layer neural network trained based on deep learning.

[0048] The channel user information corresponding to the target user is obtained, and the channel user information is input into a hash coding model optimized based on preset multi-channel category information, and a polarized hash coding value is performed on the channel user information to obtain a polarized hash coding value. Specifically, the channel user information of a single channel corresponding to the target user is obtained, and the data representation matrix corresponding to the channel user information is input into the hidden layer of the deep polarization network, and data processing is performed on the data representation matrix to be processed, wherein the data representation matrix is ​​a matrix representation form of the channel user information, and the data processing includes convolution, pooling, etc., and then a feature representation matrix corresponding to the data representation matrix is ​​obtained, that is, the category feature information corresponding to the channel user information is obtained, and then the feature representation matrix is ​​input into a preset hash layer, and the The feature representation matrix is ​​fully connected to obtain a fully connected vector, and then a corresponding polarization output channel is matched for each target bit in the fully connected vector, and based on each polarization output channel, the target bit corresponding to each polarization output channel is polarized, and a polarization identifier is assigned to each target bit, and then a hash vector corresponding to each polarized target bit is output, and the hash vector is used as the polarization hash code value, wherein the polarization output channel is a model output channel that assigns a polarization identifier to the target bit, and the polarization identifier is the positive and negative signs of the target bit. For example, assuming that the fully connected vector is (a, b), the hash vector is (a, -b), and then the polarization identifier corresponding to the target bit a is +1, and the polarization identifier corresponding to the target bit b is +1.

[0049] Step A20: Calculate the bit matching degree between the polarized hash code value and each preset hash code value, and determine the target hash code value from each preset hash code value based on each bit matching degree.

[0050] In this embodiment, it should be noted that each of the preset hash code values ​​is a unique user identifier corresponding to a different user.

[0051] The bit matching degree between the polarization hash code value and each preset hash code value is calculated, and based on each bit matching degree, the target hash code value is determined in each preset hash code value. Specifically, the bit difference degree between the polarization hash code value and each preset hash code is calculated, and then the bit matching degree corresponding to each bit difference degree is obtained. Further, the maximum target matching degree is determined in each bit matching degree, and then it is judged whether the target matching degree is greater than or equal to a preset matching degree threshold. If it is greater, the preset hash code value corresponding to the target matching degree is used as the target user unique identifier. If it is not greater, a new user identifier corresponding to the channel user information is generated, and based on the common user information in the channel user information under each channel channel, the new user identifier is associated with the channel user information under each channel channel, and then the associated new user identifier is used as the target user unique identifier.

[0052] The preset multi-channel category information includes a preset target hash code value.

[0053] Before the step of obtaining the channel user information corresponding to the target user, inputting the channel user information into a hash coding model optimized based on preset category information, performing polarized hash coding value on the channel user information, and obtaining the polarized hash coding value, the multi-channel order rate prediction method includes:

[0054] Step B10, obtaining training data and a hash coding model to be trained, and generating the preset target hash coding value corresponding to the training data based on a preset target hash coding method;

[0055] In this embodiment, it should be noted that the hash model to be trained is an untrained neural network model, the preset target hash coding method can be any pre-set hash coding method, the training data is a training sample used to train the hash model to be trained, and the preset target hash coding value is a hash coding value corresponding to the training sample category to which the training sample belongs, wherein one training sample category corresponds to one preset hash coding value, the training sample category includes at least one training sample, and one training sample category corresponds to one training user, and the training sample is the channel user information of the training user in a channel, that is, the training sample category includes the channel user information of the training user in each channel.

[0056] The training data and the hash coding model to be trained are obtained, and based on a preset target hash coding method, the preset target hash coding value corresponding to the training data is generated. Specifically, the training samples and the hash model to be trained are extracted from a preset training data storage database, and based on the preset target hash coding method, the training samples are hash-coded to generate a preset target hash coding value corresponding to the training sample category to which the training samples belong.

[0057] Wherein, in step B10, the preset target hash coding method includes a random target hash coding and an adaptive target hash coding value.

[0058] The step of generating the preset target hash code value corresponding to the training data based on the preset target hash code method includes:

[0059] Step B11, performing the random target hash coding on the training data to obtain the target preset hash coding value;

[0060] In this embodiment, it should be noted that if the preset target hash coding method is random target hash coding, the Hamming distance between the corresponding preset target hash coding values ​​belonging to different sample categories should be the target Hamming distance 2*K*p*(1-p), wherein K is the number of bits in the preset target hash coding value, and p is the probability that the sample corresponding to the preset target hash coding value belongs to the sample category.

[0061] The random target hash coding is performed on the training data to obtain the target preset hash coding value. Specifically, based on a preset Gaussian random method, a preset target hash coding value corresponding to the training sample is generated.

[0062] Step B12, performing the adaptive target hash coding on the training data to obtain the preset target hash coding value.

[0063] In this embodiment, the adaptive target hash coding is performed on the training data to obtain the preset target hash coding value. Specifically, multiple training samples of the training sample category corresponding to the training data are obtained to generate random hash coding values ​​corresponding to each of the training samples based on a preset Gaussian random method, wherein one training sample corresponds to one random hash coding value, and then the random hash coding value corresponding to each of the training samples in each of the training sample categories is determined, and the average value of each random hash coding value corresponding to each of the training sample categories is calculated to obtain the average hash coding value corresponding to each of the training sample categories, and then the average hash coding value is used as the preset hash coding value corresponding to all training samples in the corresponding training sample category.

[0064] Step B20, based on the preset target hash code value and the training data, iteratively train the hash code model to be trained to optimize the preset polarization loss function corresponding to the hash code model to be trained, until the hash code model to be trained reaches a preset iteration end condition, thereby obtaining the hash code model.

[0065] In this embodiment, it should be noted that the preset iteration end condition includes reaching a preset iteration number threshold, a preset polarization loss function convergence, etc., and the iterative training includes one or more rounds of training.

[0066] Based on the preset target hash code value and the training data, the hash code model to be trained is iteratively trained to optimize the preset polarization loss function corresponding to the hash code model to be trained, until the hash code model to be trained reaches a preset iteration end condition, and the hash code model is obtained. Specifically, the training data is input into the hash model to be trained, so as to train and update the hash model to be trained based on the preset hash code value, and determine whether the hash model to be trained after the training update meets the preset iteration end condition. If the hash model to be trained after the training update meets the preset iteration end condition, the hash model to be trained after the training update is used as the hash code model. If the hash model to be trained after the training update does not meet the preset iteration end condition, the initial hash code value output by the hash model to be trained based on the training data is obtained, and based on the initial hash code value and the preset target hash code value, the preset polarization loss function corresponding to each polarization output channel of the hash model to be trained is optimized, and the hash model to be trained is re-trained and updated until the hash model to be trained after the training update meets the preset iteration end condition.

[0067] Wherein, based on the preset target hash code value and the training data, the hash code model to be trained is iteratively trained to optimize the preset polarization loss function corresponding to the hash code model to be trained, until the hash code model to be trained reaches a preset iteration end condition, and the step of obtaining the hash code model includes:

[0068] Step B21, inputting the training data into the hash model to be trained, so as to perform hash coding on the training data based on the preset polarization loss function to obtain an initial hash coding value;

[0069] In this embodiment, the training data includes at least one training sample, and the initial hash code value includes an initial hash code value corresponding to each training sample.

[0070] The training data is input into the hash model to be trained, and the training data is hash-encoded based on the preset polarization loss function to obtain an initial hash code value. Specifically, the training matrix to be processed corresponding to the training sample is input into the hash model to be trained, wherein the training matrix to be processed is a matrix representation of the training sample, and then the training matrix to be processed is hashed to obtain a training hash vector, and then based on the preset polarization loss function, each bit of the training hash vector is forced to be polarized to obtain a training polarization vector corresponding to the training hash vector, and then based on the polarization identifier corresponding to each bit in the training polarization vector, the initial hash code value corresponding to the training sample is generated, wherein the preset polarization loss function is as follows,

[0071] L(v, t^c) = max(mv*t^c, 0)

[0072] Among them, L is the preset polarization loss function, m is the preset forced polarization parameter, v is the value of each hash vector bit in the training hash vector, and the absolute value of v is greater than m, t^c is the target hash value corresponding to the hash vector bit, the target hash value is the bit value of the preset target hash code value corresponding to the training sample, and t^c={-1,+1}, and the preset polarization loss function converges to 0. For example, assuming that m is 1, t^c is 1, and v is -1, then L=2. If the preset polarization loss function is to converge to 0, v needs to be forced to be polarized so that v is 1. At this time, L=0. Then when t^c is equal to 1, the training hash vector The value on the bit position will gradually move away from 0 in the positive direction. When t^c is equal to -1, the value on the bit position of the training hash vector will gradually move away from 0 in the negative direction. After the polarization is successful, the polarization identifier of each bit position in the obtained training polarization vector should be consistent with the corresponding target hash value. Furthermore, since the preset target hash code value of the same training sample category is the same, the polarization identifier on each bit position in the training polarization vector corresponding to each training sample belonging to the same training sample category is consistent, and then based on each polarization identifier, the obtained model output value is consistent, that is, for the model input samples belonging to the same sample category, the same hash code value can be obtained based on the trained hash model.

[0073] In addition, it should be noted that each bit in the training hash vector corresponds to a polarization output channel in the hash model to be trained, and the preset forced polarization parameters corresponding to each polarization output channel are obtained by training and optimizing the hash coding model, and thus the preset forced polarization parameters corresponding to each polarization output channel may be the same or different, wherein the polarization output channel is used to force the polarization of the value on the corresponding bit in the training hash vector based on the preset forced polarization parameter through the corresponding preset polarization loss function, and output the encoded value of the corresponding bit in the initial hash coding value.

[0074] Step B22, calculating a training Hamming distance between the initial hash code value and the preset target hash code value, and comparing the training Hamming distance with a preset Hamming distance threshold;

[0075] In this embodiment, a training Hamming distance between the initial hash code value and the preset target hash code value is calculated, and the training Hamming distance is compared with a preset Hamming distance threshold. Specifically, the value on each bit of the initial hash code value is compared with the value on each bit of the preset target hash code value, and the number of bits whose bit values ​​are different between the initial hash code value and the preset target hash code value is determined, and the number of bits is used as the training Hamming distance, and the training Hamming distance is compared with a preset Hamming distance threshold. For example, assuming that the initial hash code value is a vector (1, 1, 1, 1), and the target hash code result is a vector (-1, 1, 1, -1), the number of bits is 2, and the training Hamming distance is 2.

[0076] Step A23: if the training Hamming distance is greater than the preset Hamming distance threshold, it is determined that the hash layer to be trained has not reached the preset iteration end condition, and the preset polarization loss function is optimized based on the initial hash code value;

[0077] In this embodiment, if the training Hamming distance is greater than the preset Hamming distance threshold, it is determined that the hash layer to be trained has not reached the preset iteration end condition, and the preset polarization loss function is optimized based on the initial hash code value. Specifically, if the training Hamming distance is greater than the preset Hamming distance threshold, it is determined that the preset polarization loss function has not converged on all the polarization output channels, that is, the preset polarization loss function has not converged, and then it is determined that the hash model to be trained has not reached the preset iteration end condition, and then the initial hash code value and the preset hash code value are determined. One or more different bits between the values, and determine the unconverged polarization output channel corresponding to each of the different bits, and then adjust the preset forced polarization parameter in the preset polarization loss function corresponding to each of the unconverged polarization output channels, wherein the unconverged polarization output channel is a polarization output channel corresponding to the unconverged preset polarization loss function, wherein the hash model to be trained includes at least one polarization output channel, and the number of the polarization output channels is related to the number of bits in the training hash vector, that is, one bit in the training hash vector corresponds to one polarization output channel.

[0078] Step A24, retraining the hash model to be trained based on the optimized preset polarization loss function until the training Hamming distance is less than or equal to the preset Hamming distance threshold;

[0079] In this embodiment, based on the optimized preset polarization loss function, the to-be-trained hash model is re-trained until the training Hamming distance is less than or equal to the preset Hamming distance threshold. Specifically, the training data is re-acquired, and based on the re-acquired training data, the to-be-trained hash model corresponding to the optimized preset polarization loss function is re-iterated to continue optimizing the preset polarization loss function until the training Hamming distance is less than or equal to the preset Hamming distance threshold.

[0080] Step A25: If the training Hamming distance is less than or equal to the preset Hamming distance threshold, it is determined that the hash model to be trained meets the preset iteration end condition, and the hash model to be trained is used as the hash model.

[0081] In this embodiment, if the training Hamming distance is less than or equal to the preset Hamming distance threshold, it is determined that the hash model to be trained has reached the preset iteration end condition, and the hash model to be trained is used as the hash model. Specifically, if the training Hamming distance is less than or equal to the preset Hamming distance threshold, it is determined that the hash model to be trained has reached the preset iteration end condition, that is, the preset polarization loss function corresponding to each polarization output channel in the hash model to be trained converges, and then the hash model to be trained is used as the hash model.

[0082] In addition, it should be noted that the current hash coding method based on deep learning usually uses pairwise similarity labels as training targets, and constraints need to be added during training, which leads to more parameters that need to be optimized when training the hash coding model. When training the hash coding model in this embodiment, it is only necessary to optimize the hash coding model based on a simple polarization loss function, thereby avoiding the situation where pairwise similarity labels are used as training targets and constraints need to be added during training to train the hash coding model, thereby significantly improving the efficiency of hash coding model construction.

[0083] Step S20, based on the unique identifier of the target user, uniformly collect multi-channel user behavior data in each of the channels;

[0084] In this embodiment, based on the target user unique identifier, multi-channel user behavior data is uniformly collected in each channel. Specifically, single-channel user behavior data corresponding to the target user unique identifier is collected in each channel, and the target user behavior characteristics corresponding to each user behavior data are determined. Based on each target user behavior characteristic, each single-channel user behavior data is integrated to remove data redundancy between each single-channel user behavior data to obtain the multi-channel user behavior data.

[0085] The multi-channel user behavior data includes a user behavior path map,

[0086] The step of uniformly collecting multi-channel user behavior data in each of the channels based on the unique identifier of the target user includes:

[0087] Step S21, based on the unique identifier of the target user, extracting single-channel user behavior data in each of the channel channels;

[0088] In this embodiment, it should be noted that the single-channel user behavior data is the associated data of the user behavior of the target user in the corresponding channel.

[0089] Step S22, respectively generate behavior nodes corresponding to each of the single-channel user behavior data, and connect each of the behavior nodes according to the occurrence sequence of the user behaviors corresponding to each of the channel user behavior data to obtain the user behavior path diagram.

[0090] In this embodiment, it should be noted that the user behavior path diagram is a path diagram that records the target user behaviors occurring by the target user and the chronological order of occurrence of each target user behavior.

[0091] Generate behavior nodes corresponding to each of the single-channel user behavior data respectively, and connect each of the behavior nodes according to the occurrence order of the user behaviors corresponding to each of the channel user behavior data to obtain the user behavior path diagram. Specifically, generate a single-channel user behavior encoding vector corresponding to each of the single-channel user behavior data, wherein the single-channel user behavior encoding vector is a vector representation of the single-channel user behavior data. For example, assuming that the single-channel user behavior encoding vector is (i, j, k), then i is a feature value indicating that the target user has a target user behavior a, j is a feature value indicating that the target user has a target user behavior b, and k is a feature value indicating that the target user has a target user behavior c, and the target user first has user behavior a, then user behavior b, and finally user behavior c, and then generate behavior nodes corresponding to each of the single-channel user behavior data, wherein the behavior node is a data node that stores the corresponding single-channel user behavior encoding vector, and then the occurrence time order of the target user behavior of the target user is sequentially connected to obtain the user behavior path diagram, such as Figure 3 The above is a schematic diagram of the user behavior path map, wherein enterprise, business, portal, page, article, question, conversation, adding WeChat, making a phone call and placing an order are all names of user behaviors occurring in the behavior node. For example, conversation indicates that the user behavior occurring in the channel corresponding to the corresponding behavior node of the target user is a conversation.

[0092] Step S30, based on the multi-channel user behavior data and a preset order rate prediction model, a multi-channel order rate prediction is performed on the target user to obtain a target order rate.

[0093] In this embodiment, it should be noted that the preset order rate prediction model is a pre-trained order rate prediction model, which is used to predict the order rate caused by user behavior based on user behavior data corresponding to the user behavior of the user, for example, based on the web page information browsed by the user, predict the probability of the user purchasing a certain item, etc.

[0094] Based on the multi-channel user behavior data and the preset order rate prediction model, a multi-channel order rate prediction is performed on the target user to obtain a target order rate. Specifically, the multi-channel user behavior data is input into the preset order rate prediction model, the multi-channel user behavior data is classified, and the classification labels corresponding to the multi-channel user behavior data are obtained. Then, based on the classification labels, the target order rate is determined. For example, assuming that the classification label is a vector (1, 0.9), 1 represents the data category of the multi-channel user behavior data, and 0.9 is the target order rate.

[0095] The step of performing multi-channel order rate prediction on the target user based on the multi-channel user behavior data and a preset order rate prediction model to obtain a target order rate includes:

[0096] Step S31, determining a target user behavior feature combination corresponding to the multi-channel user behavior data;

[0097] In this embodiment, it should be noted that the target user behavior feature combination is a feature combination corresponding to each target user behavior feature corresponding to the multi-channel user behavior data, and the multi-channel user behavior data is the associated data of the current user behavior, wherein the current user behavior is the user behavior that has occurred at the current time point. For example, assuming that user A added WeChat to consult item A at the current time point and before the current time point, and then purchased item B associated with item A, etc., then the current user behavior is adding WeChat and purchasing associated item B, respectively, and the multi-channel user behavior data is the content of WeChat communication and the degree of association between associated item B and item A, etc.

[0098] Determine the target user behavior feature combination corresponding to the multi-channel user behavior data. Specifically, generate a target user behavior path map corresponding to the multi-channel user behavior data according to the chronological order of occurrence of each target user behavior corresponding to the multi-channel user behavior data, and then determine the target user behavior feature combination corresponding to the multi-channel user behavior data based on the target user behavior path map.

[0099] Step S32, based on the target user behavior feature combination, determining the preset order rate prediction model corresponding to the multi-channel user behavior data in the preset model set;

[0100] In this embodiment, it should be noted that the target user behavior feature combination is a combination of each of the target user behavior features, which can be represented by a feature combination vector composed of feature codes corresponding to each of the target user behavior features. The preset model set includes at least one prediction model, and each prediction model in the preset model set is a model trained based on different user behavior feature combinations. For example, assuming that the preset model set includes prediction model M and prediction model N, prediction model M is trained based on the target user behavior feature combination (x1, x2), and prediction model N is trained based on the target user behavior feature combination (x1, x2, x3), wherein x1, x2 and x3 are feature codes of target user behavior features.

[0101] Based on the target user behavior feature combination, the preset order rate prediction model corresponding to the multi-channel user behavior data is determined in the preset model set. Specifically, each preset feature combination corresponding to the preset model set is obtained, wherein the preset feature combination is a pre-set combination of different user behavior features, and then each preset feature combination is compared with the target user behavior feature combination, and the bit consistency rate corresponding to each preset feature combination and the target user behavior feature combination is obtained, and then the maximum target bit consistency rate is determined among the bit consistency rates, and the preset feature combination corresponding to the target bit consistency rate is used as the target The standard target feature combination corresponding to the user behavior feature combination, further, the prediction model corresponding to the standard target feature combination in the preset model set is used as the preset order rate prediction model. For example, assuming that the target user behavior feature combination is a vector (x1, x2, x3, x4), and each of the preset feature combinations is a vector A (x1, x2, x3), a vector B (x1, x2) and a vector C (x1, x3), then the bit consistency rate corresponding to vector A is 75%, the bit consistency rate corresponding to vector B is 50%, and the bit consistency rate corresponding to vector C is 50%, and thus the standard target feature combination is vector A (x1, x2, x3).

[0102] Step S33: Based on the preset order rate prediction model, the multi-channel user behavior data is classified to predict the order rate of the target user based on multiple channels to obtain the target order rate.

[0103] In this embodiment, based on the preset order rate prediction model, the multi-channel user behavior data is classified to perform a multi-channel order rate prediction for the target user to obtain the target order rate. Specifically, based on the standard target feature combination, standard multi-channel user behavior data is extracted from the multi-channel user behavior data, and the target feature representation matrix corresponding to the standard multi-channel user behavior data is input into the preset order rate prediction model, wherein each column of the target feature representation matrix corresponds to one of the target user behavior features. For example, assuming that a column of the user behavior feature representation matrix is ​​a vector D(a, b, c), wherein the vector (a, b, c) is a eigenvalue, if the target behavior feature corresponding to the vector D is telephone communication, then the vector (a, b, c) represents the voice data of the telephone communication. According to the present invention, the target feature representation matrix is ​​processed, wherein the data processing includes convolution, pooling and full connection, etc., so as to obtain a classification representation vector for the standard multi-channel user behavior data, and the first predicted order rate is determined based on the classification representation vector. For example, assuming that the multi-channel user behavior data corresponds to x1, x2 and x3, and the standard feature combination includes x1 and x2, then the standard multi-channel user behavior data is the user behavior data (A, B) corresponding to the feature x1 and the feature 2, wherein A is the vector representation of the user behavior data corresponding to the feature x1, and B is the vector representation of the user behavior data corresponding to the feature x2. Then based on the preset order rate prediction model, after data processing of the user behavior data (A, B), the classification label vector can be obtained, and then the target order rate can be obtained.

[0104] In addition, it should be noted that, since user behavior data is currently directly obtained, the characteristic dimensions of user behavior features corresponding to the user behavior data of each local user usually have large differences, that is, the number of user behavior features corresponding to each user behavior data is different, that is, the characteristic dimensions of the user behavior data of all users are in dynamic change, and the current neural network model is usually a model optimized based on user behavior data with specific characteristic dimensions, which results in the characteristic dimensions corresponding to the model and the characteristic dimensions corresponding to the user behavior data usually being in a mismatched state, which results in low accuracy of order rate prediction. Compared with the technical means of predicting the user's order rate using a neural network model with a specific characteristic dimension in the prior art, this embodiment pre-analyzes the target user behavior feature combination corresponding to the multi-channel user behavior data, and then matches the preset order rate prediction model that matches the characteristic dimension of the multi-channel user behavior data based on the target user behavior feature combination, thereby achieving accurate prediction of the target predicted order rate, overcoming the technical defect in the prior art that when the characteristic dimension of the user behavior data is in dynamic change, the order rate of all users cannot be accurately predicted due to the great difference in the characteristic dimensions corresponding to each user, thereby improving the accuracy of the order rate prediction.

[0105] The present embodiment provides a method, device and readable storage medium for predicting an order rate based on multiple channels. Compared with the prior art which predicts the order rate respectively based on the user behavior data of each channel directly, the present embodiment can uniformly collect the multi-channel user behavior data corresponding to the target user in each channel by assigning a unique user identifier to the target user, thereby eliminating a large amount of redundancy in the user behavior data. Further, the order rate is predicted based on the multi-channel user behavior data, and the process of repeatedly calculating the redundant part of the user behavior data can be eliminated, thereby reducing the amount of calculation when predicting the order rate. The technical defect in the prior art that the user behavior data collected by each channel cannot be unified, resulting in a large amount of redundancy in the user behavior data, and then the redundant part of the user behavior data is repeatedly calculated when predicting the order rate is overcome. The calculation efficiency of the order rate prediction is improved, thereby solving the technical problem of low calculation efficiency when predicting the order rate.

[0106] Further, refer to Figure 4 Based on the first embodiment of the present application, in another embodiment of the present application, the preset order rate prediction model includes a graph embedding prediction model.

[0107] The step of performing multi-channel order rate prediction on the target user based on the multi-channel user behavior data and the preset order rate prediction model to obtain the target order rate includes:

[0108] Step C10, generating a target user behavior graph corresponding to the multi-channel user behavior data, and converting the target user behavior graph into a user behavior graph embedding;

[0109] In this embodiment, it should be noted that the target user behavior graph is a structural graph that records user behaviors and the order in which user behaviors occur, wherein the target user behavior graph includes user behavior nodes, user behavior node connecting edges, and connecting edge weights corresponding to the user behavior node connecting edges, wherein the user behavior nodes are nodes used to represent user behaviors, one user behavior node corresponds to one target user behavior, the user behavior node connecting edges are directed curve segments connecting user behavior nodes, used to represent the process from one user behavior to another user behavior, and the direction of user behavior conversion is the direction of the directed curve segment, the connecting edge weight is used to represent the frequency of user behavior conversion from the user behavior node connecting edge corresponding to the connecting edge weight, wherein the connecting edge weight can be represented by the color of the user behavior node connecting edge or the path length, and multiple target user behavior graphs can constitute a target user behavior graph set, such as Figure 5 The figure shows a schematic diagram of the target user behavior atlas, wherein enterprise, business, portal, page, article, conversation, adding WeChat, calling, rating, question and order are all names of the user behavior nodes, and w is the connection edge weight.

[0110] In addition, it should be noted that the user behavior graph is embedded as a low-dimensional representation vector of the target user behavior graph, which is used to represent the graph structure features of the target user behavior graph, wherein the graph structure features include user behavior nodes, user behavior node connection edges, and connection edge weights corresponding to user behavior node connection edges, etc., wherein one target user behavior graph corresponds to one target user behavior graph.

[0111] Generate a target user behavior graph corresponding to the multi-channel user behavior data, and convert the target user behavior graph into a user behavior graph embedding. Specifically, determine each target user behavior corresponding to the multi-channel user behavior data, and determine the user behavior conversion relationship between each target user behavior, wherein the user behavior conversion relationship includes the conversion direction between each target user behavior and the conversion frequency between each target user behavior, and then generate the target user behavior graph based on each target user behavior and the user behavior conversion relationship, and then perform a low-dimensional vector space conversion on the target user behavior graph to map the target user behavior graph to a preset low-dimensional vector space to obtain the user behavior graph embedding, wherein the algorithm for converting the target behavior graph into the user behavior graph embedding includes a graph2vec algorithm, etc.

[0112] Wherein, the step of converting the target user behavior graph into a user behavior graph embedding comprises:

[0113] Step C11, determining each user behavior subgraph corresponding to the target user behavior graph, and based on each of the user behavior subgraphs, performing one-hot encoding on the target user behavior graph to obtain a one-hot encoding vector corresponding to the target user behavior graph;

[0114] In this embodiment, it should be noted that the user behavior subgraph is a subset of the target user behavior graph, and the number of user behavior nodes included in the user behavior subgraph is pre-set. For example, assuming that the target user behavior graph is abcde, the direction of the user behavior node connection edge is unidirectional, from a to e, and the number of user behavior nodes included in the user behavior subgraph is 4, then the user behavior subgraphs are abcd and bcde respectively.

[0115] Determine each user behavior subgraph corresponding to the target user behavior graph, and based on each of the user behavior subgraphs, perform one-hot encoding on the target user behavior graph to obtain a one-hot encoding vector corresponding to the target user behavior graph. Specifically, based on a preset subgraph node number threshold, determine each user behavior subgraph corresponding to the target user behavior graph, wherein the preset subgraph node number threshold is the number of user behavior nodes in the user behavior subgraph that is preset, and then based on each of the user behavior subgraphs and all preset local user behavior subgraphs, perform one-hot encoding on the target user behavior graph to obtain a one-hot encoding vector corresponding to the target user behavior graph, wherein each of the preset local user behavior subgraphs is the All user behavior subgraphs corresponding to the target user behavior graph set, wherein the target user behavior graph set is a set of user behavior graphs of all local users. For example, assuming that the target user behavior graph set includes 5 target user behavior graphs, and all target user behavior graphs correspond to 5 user behavior subgraphs a, b, c, d, and e. If the target user behavior graph A in the target user behavior graph set includes user behavior subgraphs a and c, then the unique hot encoding vector corresponding to the target user behavior graph A is (1, 0, 1, 0, 0), wherein 1 indicates that the target user behavior graph A includes the user behavior subgraph corresponding to the bit corresponding to 1, and 0 indicates that the target user behavior graph A does not include the user behavior subgraph corresponding to the bit corresponding to 0.

[0116] Step C12, performing dimensionality reduction processing on the one-hot encoded vector to obtain the user behavior graph embedding.

[0117] In this embodiment, the one-hot encoded vector is subjected to dimensionality reduction processing to obtain the target user behavior graph representation vector. Specifically, the one-hot encoded vector is input into a preset hidden layer network, and data processing is performed on the one-hot encoded vector, wherein the data processing includes convolution, pooling, full connection, etc., so as to map the one-hot encoded vector to a preset low-dimensional vector space, obtain the low-dimensional output vector of the preset hidden layer network, and embed the low-dimensional space vector as the user behavior graph.

[0118] Step C20, embedding the user behavior graph into the graph embedding prediction model, classifying the user behavior graph embedding, so as to predict the order rate of the target user based on multiple channels, and obtain the target order rate.

[0119] In this embodiment, the user behavior graph embedding is input into the graph embedding prediction model, and the user behavior graph embedding is classified to perform a multi-channel order rate prediction for the target user to obtain the target order rate. Specifically, the user behavior graph embedding is input into a preset order rate prediction model, and the user behavior graph embedding is classified to obtain a graph embedding classification signature vector corresponding to the user behavior graph embedding, and then based on the graph embedding classification signature vector, the target order rate is determined, wherein the graph embedding classification signature vector is a vector that identifies the category information of the user behavior graph embedding, and one category information of the user behavior graph embedding corresponds to one target order rate. For example, assuming that the graph embedding classification signature vector is (1, 1, 0, 0), the category of the user behavior graph embedding corresponding to the graph embedding classification signature vector is category A, wherein the target order rate corresponding to category A is 90%.

[0120] This embodiment generates a target user behavior graph corresponding to the multi-channel user behavior data, converts the target user behavior graph into a user behavior graph embedding, and then inputs the user behavior graph embedding into the graph embedding prediction model, and classifies the user behavior graph embedding to predict the order rate of the target user based on multiple channels, thereby obtaining the target order rate. That is, after obtaining the multi-channel user behavior data, this embodiment first generates a target user behavior graph corresponding to the multi-channel user behavior data, and then converts the target user behavior graph into a user behavior graph embedding of a specific dimension, and then inputs the user behavior graph embedding into a preset target prediction model, and predicts the order rate of the target user corresponding to the user behavior graph embedding, thereby obtaining the target prediction result, and thus this embodiment realizes the purpose of mapping the user behavior data to the user behavior graph embedding of a specific feature dimension, and predicting the order rate based on the user behavior graph embedding of the specific feature dimension, and then even if the feature dimensions of the user behavior data of each local user are different, by converting the user behavior data into the user behavior graph embedding, the order rate corresponding to each user can still be accurately predicted based on the user behavior graph embedding of the same feature dimension, thereby improving the accuracy of the order rate prediction.

[0121] Further, refer to Figure 6 Based on the first embodiment and the second embodiment of the present application, in another embodiment of the present application, the multi-channel order rate prediction method is applied to a first device, and the multi-channel order rate prediction is performed on the target user based on the multi-channel user behavior data and the preset order rate prediction model, and the step of obtaining the target order rate includes:

[0122] Step D10, sending the prediction request corresponding to the target user to the second device, so that the second device can determine the second multi-channel user behavior data;

[0123] In this embodiment, a prediction request corresponding to the target user is sent to a second device so that the second device can determine the second multi-channel user behavior data. Specifically, the sample ID of the target user is sent to the second device as the prediction request so that the second device can determine the target user based on the sample ID and obtain the second multi-channel user behavior data corresponding to the target user, wherein the sample ID is the identity of the target user, and the second multi-channel user behavior data is a collection of user behavior data collected based on multiple data channels at the second device, wherein the first prediction task corresponding to the multi-channel user behavior data and the second prediction task corresponding to the second multi-channel user behavior data may be inconsistent. For example, if the user intends to buy an insurance product with a loan, the first prediction task may be to predict the probability of the user purchasing the insurance product, and the second prediction task may be to predict the probability of the user taking out a loan.

[0124] Step D20, based on the multi-channel user behavior data and the preset order rate prediction model, perform vertical federated prediction with the second device to combine the second multi-channel user behavior data to predict the target order rate.

[0125] In this embodiment, it should be noted that before performing the vertical federated prediction, the first device and the second device need to perform vertical federated learning modeling to construct a vertical federated model jointly held by the first device and the second device, wherein the vertical federated model includes a preset first-part prediction model and a preset second-part vertical federated model, wherein the preset first-part prediction model is the part of the vertical federated model held by the first device, and the preset second-part prediction model is the part of the vertical federated model held by the second device, and the number of the second devices is greater than or equal to 1, that is, the vertical federated learning modeling can be multi-party federated learning modeling or two-party federated learning modeling, wherein the preset order rate prediction model is the preset first-part vertical federated model.

[0126] Based on the multi-channel user behavior data and the preset order rate prediction model, a vertical federated prediction is performed with the second device to combine the second multi-channel user behavior data to predict the target order rate. Specifically, based on the multi-channel user behavior data and the preset first-part vertical federated model, a first intermediate result corresponding to the multi-channel user behavior data is determined, and a vertical federated prediction is performed with the second device, and a second intermediate result determined by the second device based on the second multi-channel user behavior data and the preset second-part prediction model is received, and then the first intermediate result and the second intermediate result are aggregated to obtain the target order rate. For example, assuming that the vertical federated model is a linear model Y=A1X1+A2X2+A3X3+A4X4, the preset first-part prediction model is Y1=A1X1+A2X2, and the preset second-part prediction model is Y2=A3X3+A4X4, then the first intermediate result is Y1, the second intermediate result is Y2, and the target order rate is Y=Y1+Y2.

[0127] In an implementable solution, the vertical federation model is a preset multi-party joint classification decision tree, and the step of performing vertical federation prediction with the second device based on the multi-channel user behavior data and the preset order rate prediction model to combine the second multi-channel user behavior data to predict the target order rate includes:

[0128] Step D21, inputting the multi-channel user behavior data into a preset multi-party joint classification decision tree, and traversing the preset multi-party joint classification decision tree to determine the ownership of the root node of the preset multi-party joint classification decision tree;

[0129] In this embodiment, it should be noted that the preset multi-party joint classification decision tree is a vertical GBDT (Gradient Boosting Decision A gradient boosting decision tree) model is provided, wherein the preset multi-party joint classification decision tree includes at least one tree node, and the topmost tree node is a root node, wherein each tree node corresponds to a node tag for identifying the ownership of the tree node, and each leaf node type of the tree node stores leaf information, wherein the leaf information includes the target order rate, and each non-leaf node type of the tree node corresponds to a feature splitting value, and the feature splitting value is used to split the tree node into feature values ​​of left and right child nodes, and the feature splitting value corresponding to the tree node is only held by the owner of the tree node. For example, assuming that the feature corresponding to the root node is age, the feature value range is 15 to 35, and the feature splitting value is 20, then the feature value range of the left child node of the root node is 15 to 20, and the feature value range of the right child node of the root node is 20 to 35, and then the samples corresponding to the feature value range of 15 to 20 all fall on the left child node, and the samples corresponding to the feature value range of 20 to 35 all fall on the right child node, and then the root node can be divided into left and right child nodes.

[0130] In addition, it should be noted that each of the tree nodes corresponds to a target user behavior feature, and thus at each of the tree nodes, a classification decision can be made for the target user based on the feature splitting value corresponding to the target user behavior feature, and thus for the preset multi-party joint classification decision tree, a classification decision can be made for the target user based on the target user behavior features of all parties, and thus a classification decision can be made for the target user based on a feature combination with higher feature richness, thereby improving the prediction accuracy for the target user.

[0131] The multi-channel user behavior data is input based on a preset multi-party joint classification decision tree, and the preset multi-party joint classification decision tree is traversed to determine the root node ownership of the preset multi-party joint classification decision tree. Specifically, the multi-channel user behavior data is input based on a preset multi-party joint classification decision tree, and the preset multi-party joint classification decision tree is started to be traversed to obtain the root node tag of the root node of the preset multi-party joint classification decision tree, and based on the root node tag, the root node ownership of the root node of the preset multi-party joint classification decision tree is determined.

[0132] Step D22: if the root node belongs to the second device, receiving the intermediate classification result fed back by the second device based on the second multi-channel user behavior data, so as to determine the child node corresponding to the target user based on the intermediate classification result;

[0133] In this embodiment, if the root node belongs to the second device, the classification intermediate result fed back by the second device based on the second multi-channel user behavior data is received to determine the child node corresponding to the target user based on the classification intermediate result. Specifically, if the root node belongs to the second device, a prediction request is sent to the second device, wherein the prediction request includes a sample ID and a target feature code, wherein the sample ID is the identity of the target user, and the target feature code is the identity corresponding to the target user behavior feature corresponding to the tree node. After receiving the sample ID and the feature code, the second device can obtain the corresponding locally stored feature splitting value based on the sample ID and the target feature code to determine the classification intermediate result of the target user, wherein the classification intermediate result is the result of determining whether the target user belongs to a left child node or a right child node, and then based on the classification intermediate result, determine the child node to which the target user belongs.

[0134] Step D23, based on the corresponding child node, determining the target leaf node corresponding to the target user to obtain the target order rate;

[0135] In this embodiment, based on the child node, the target leaf node corresponding to the target user is determined to obtain the target order rate. Specifically, it is determined whether the child node belongs to a leaf node type, wherein a tree node belonging to the leaf node type has no child node. If the child node belongs to a leaf node type, the child node is the target leaf node of the target user, and the leaf information recorded in the child node is obtained, wherein the leaf information is the classification information of the target user, and then the target order rate is obtained from the leaf information. For example, it is assumed that the leaf information is a vector (a, 0.9), wherein a is the category code of the target user, and 0.9 is the target order rate. If the child node does not belong to a leaf node type, the preset multi-party joint classification decision tree is continued to be traversed until the target leaf node corresponding to the target user is determined to obtain the target order rate.

[0136] The step of determining the target leaf node corresponding to the target user based on the child node to obtain the target order rate includes:

[0137] Step D231, determining whether the child node is a leaf node type, and if the child node is a leaf node type, determining the target order rate based on the child node;

[0138] In this embodiment, it is determined whether the child node belongs to the leaf node type. If the child node belongs to the leaf node type, the target order rate is determined based on the child node. Specifically, it is determined whether the child node belongs to the leaf node type. If the child node belongs to the leaf node type, the child node is the target leaf node corresponding to the target user, that is, the classification decision of the target user is completed, and then the leaf information in the child node is extracted, and the target order rate is obtained from the leaf information.

[0139] Step D232, if the child node does not belong to the leaf node type, continue to traverse the preset multi-party joint classification decision tree until the target leaf node corresponding to the multi-channel user behavior data is determined to obtain the target order rate.

[0140] In this embodiment, if the child node does not belong to the leaf node type, the preset multi-party joint classification decision tree is continued to be traversed until the target leaf node corresponding to the multi-channel user behavior data is determined to obtain the target order rate. Specifically, if the child node does not belong to the leaf node type, the preset multi-party joint classification decision tree is continued to be traversed to obtain the node tag of the child node to determine the attribution of the child node. If the child node belongs to the first device, the second-layer feature splitting value corresponding to the child node is locally obtained, wherein the second-layer feature splitting value is the feature splitting value corresponding to the child node of the root node, and then based on the second-layer feature splitting value, the target child node of the child node corresponding to the target user is continued to be determined until the target user falls into a tree node of a certain leaf node type, the target leaf node corresponding to the target user is determined, and the target order rate is obtained. If the child node belongs to a certain second device, the sample ID and the feature code of the target user behavior target feature corresponding to the child node are sent to the second device corresponding to the child node to determine the corresponding target leaf node and obtain the target order rate.

[0141] Step D24, if the root node belongs to the first device, then based on the feature splitting value in the first device, determine the child node of the target user to obtain the target order rate.

[0142] In this embodiment, if the root node belongs to the first device, the child node of the target user is determined based on the feature splitting value in the first device to obtain the target order rate. Specifically, if the root node belongs to the first device, the child node corresponding to the target user is determined among all child nodes of the root node based on the feature splitting value corresponding to the root node in the first device. If the child node is a leaf node type, the target order rate can be determined based on the leaf information of the child node. If the child node is not a leaf node type, the node mark of the child node is obtained, and the preset multi-party joint classification decision tree is continued to be traversed until the target leaf node corresponding to the target user is determined in the preset multi-party joint classification decision tree to obtain the target order rate.

[0143] Furthermore, in an implementable solution, the method for constructing the preset multi-party joint classification decision tree is as follows:

[0144] Step E10, obtaining first sample data, and calculating feature splitting gain data based on the first sample data by performing federated interaction with each second device;

[0145] In this embodiment, it should be noted that the second device can be communicatively connected with the first device, and the first device and each of the second devices can perform vertical federated learning, and the first device is the leading party of the vertical federated learning, and each of the second devices is a participant in the vertical federated learning.

[0146] In addition, it should be noted that all samples corresponding to the first sample data or identifiers corresponding to the samples are at the initial root node of the preset multi-party joint classification decision tree to be constructed when modeling is just started, and the initial root node is used as the current node to be split, wherein the current node to be split includes sample data, and the sample data includes first sample data belonging to the first device and second sample data belonging to the second device, and the feature splitting gain data includes a feature splitting gain histogram, wherein the feature splitting gain histogram is used to find the maximum gain splitting point in the feature splitting gain histogram based on a preset gain calculation formula.

[0147] Acquire first sample data, and based on the first sample data, calculate feature splitting gain data by performing federal interaction with each second device. Specifically, perform sample alignment with each second device to determine the first sample data, further perform feature binning on the first sample data to discretize the first sample data to obtain multiple sample data bins, further calculate a first-order gradient set and a second-order gradient set corresponding to the first sample data, and then calculate the feature splitting gain data based on the first-order gradient set and the second-order gradient set by performing federal interaction with each second device.

[0148] The step of calculating feature splitting gain data based on the first sample data by performing federated interaction with each second device includes:

[0149] Step E11, performing feature binning on the first sample data to obtain first feature binning data;

[0150] In this embodiment, the first sample data is feature binned to obtain first feature binned data. Specifically, the first sample data is feature binned to discretize the first sample data to obtain first feature binned data corresponding to the first sample data, wherein the first feature binned data includes each of the sample data bins, and a key pair required for this federated learning is generated through a preset key generation module.

[0151] Step E12, obtaining a model residual and a data label corresponding to the first feature binned data, and calculating a first-order gradient set and a second-order gradient set corresponding to the first feature binned data based on the model residual and the data label;

[0152] In this embodiment, it should be noted that the first-order gradient set includes the first-order gradient corresponding to each training sample in the first sample data, and the second-order gradient includes the second-order gradient corresponding to each sample in the first feature binning data, wherein a training sample corresponds to the user behavior data of a user, and each of the samples corresponds to a data label, a sample ID and a feature code, wherein the sample ID is the identity label of the training sample, for example, data such as user telephone number and user ID number can be set as the sample ID, the feature code identifies the feature type of the training sample, and the data label is the sample label of the training sample, which is used to identify the type of the training sample. For example, assuming that the first sample data is customer loan repayment record data, the data label identifies the customer's creditworthiness based on the customer loan repayment record data, that is, identifies the type of the customer, for example, the customer is a good customer or a bad customer, etc. The model residual is the model residual of the previous round of federation performed before this round of federation, which can be extracted from the local database of the first device.

[0153] Obtain the model residual and the data label corresponding to the first feature binned data, and based on the model residual and the data label, calculate the first-order gradient set and the second-order gradient set corresponding to the first feature binned data. Specifically, obtain the model residual and the data label corresponding to each sample in the first feature binned data, and then establish a target loss function with respect to the model residual and the data label based on the model residual and the data label, and then obtain the first-order derivative and the second-order derivative of the target function with respect to the data label, obtain the first-order derivative and the second-order derivative corresponding to each of the training samples, and then obtain the first-order gradient set and the second-order gradient set, wherein the first-order gradient set includes the first-order gradient of each of the training samples, and the second-order gradient set includes the second-order gradient of each of the training samples.

[0154] Step E13, based on the first-order gradient set and the second-order gradient set, calculate the feature splitting gain data by performing federated interaction with each of the second devices to perform longitudinal federated learning.

[0155] In this embodiment, it should be noted that the feature splitting gain data includes a first feature splitting gain histogram and a second feature splitting gain histogram, wherein the first feature splitting gain histogram is a histogram of the gain after the reaction feature splitting calculated locally by the first device, and the second feature splitting gain histogram is a histogram of the gain after the reaction feature splitting calculated jointly by the first device and each second device.

[0156] Based on the first-order gradient set and the second-order gradient set, the feature splitting gain data is calculated by performing federated interaction with each of the second devices to perform vertical federated learning. Specifically, based on the first-order gradient set and the second-order gradient set, a first feature splitting histogram corresponding to the first sample data is calculated by a preset gradient histogram algorithm, and based on the public key in the public-private key, the first-order gradient set and the second-order gradient set are encrypted and sent to each of the second devices respectively, wherein the encryption method is homomorphic encryption, so that each of the second devices calculates an encrypted second feature splitting gain histogram based on the encrypted first-order gradient set and the encrypted second-order gradient set, and feeds the encrypted second feature splitting gain histogram back to the first device, and then the first device receives the encrypted second feature splitting gain histograms respectively fed back by each of the second devices based on the first-order gradient set and the second-order gradient set, and then decrypts the encrypted second feature splitting gain histograms based on the private key in the key pair to obtain each of the second feature splitting gain histograms.

[0157] Step E20: Based on the feature splitting gain data, a target gain splitting point set that meets a preset feature splitting gain condition is determined to construct the preset multi-party joint classification decision tree.

[0158] In this embodiment, based on the feature splitting gain data, a target gain splitting point set that meets the preset feature splitting gain conditions is determined to construct the preset multi-party joint classification decision tree. Specifically, based on the preset gain calculation formula, the first feature splitting gain histogram and the first maximum gain splitting point in each of the second feature splitting gain histograms are found, wherein the first maximum gain splitting point is the maximum gain splitting point of the initial root node corresponding to the first sample data, and then based on the first maximum gain splitting point, the initial root node is split to obtain the initial child node of the initial root node, and then it is determined whether the initial child node reaches the preset stop splitting condition. If so, then Stop splitting, and use the initial child node as the leaf node of the current decision tree obtained in this round of federation. If it is not reached, split the initial child node until all nodes of the current decision tree being constructed can no longer be split, and obtain the maximum gain splitting point corresponding to each non-leaf node type of tree node, that is, obtain the target gain splitting point set, and then obtain the current decision tree. If the current decision tree meets the preset federation end condition, use the current decision tree as the preset multi-party joint classification decision tree. If the current decision tree does not meet the preset federation end condition, reconstruct the current decision tree until the current decision tree meets the preset federation end condition.

[0159] The target gain splitting point set includes a first maximum gain splitting point and a second maximum gain splitting point, the feature splitting gain data includes a first feature splitting gain histogram and a second feature splitting gain histogram,

[0160] The step of determining a target gain splitting point set that meets a preset feature splitting gain condition based on the feature splitting gain data comprises:

[0161] Step E21, based on the first feature splitting gain histogram and each of the second feature splitting histograms, determining the first maximum gain splitting point of the initial root node corresponding to the first sample data by a preset gain calculation formula;

[0162] In this embodiment, based on the first feature splitting gain histogram and each of the second feature splitting histograms, the first maximum gain splitting point of the initial root node corresponding to the first sample data is determined by a preset gain calculation formula. Specifically, each splittable point corresponding to the first feature splitting gain histogram is determined, and then the sum of the first-order gradients of the left child nodes and the sum of the second-order gradients of the left child nodes of all samples of the left child nodes corresponding to each of the splittable points, as well as the sum of the first-order gradients of the right child nodes and the sum of the second-order gradients of the right child nodes of all samples of the corresponding right child nodes are determined. It should be noted that based on the splittable points, the first sample of the splittable points is The initial root node corresponding to the data is obtained, and the left child node and the right child node corresponding to the initial root node are obtained, and then the calculation gain corresponding to each splittable point is calculated based on the preset gain calculation formula and the sum of the first-order gradient of the left child node corresponding to each splittable point, the sum of the second-order gradient of the corresponding left child node, the sum of the first-order gradient of the corresponding right child node, and the sum of the second-order gradient of the right child node. Similarly, the calculation gain corresponding to each second feature splitting histogram can be calculated, and then the maximum gain is selected from all the calculation gains, and the splittable point corresponding to the maximum gain is used as the first maximum gain splitting point, wherein the preset gain calculation formula is as follows:

[0163]

[0164] Wherein, M is the calculation gain, and G L is the sum of the first-order gradients of the samples in the left child node, G R is the sum of all first-order gradients in the right child node, the H L is the sum of the second-order gradients of the samples in the left child node, H Ris the sum of all second-order gradients in the right child node, G is the sum of the first-order gradients of the samples in the parent node, H is the sum of the second-order gradients of the samples in the parent node, and is an adjustment parameter used to adjust the value range of the gain. Therefore, when performing the calculation, it should be noted that the first sample and the second sample have a value range, and there are multiple splittable points within the value range. Based on each splittable point, the first sample data or the second sample data can be split to obtain left child sample data and right child sample data, wherein the left child sample data corresponds to the left child node after the parent node is split, and the right child sample data corresponds to the right child node after the parent node is split, and then the gain, that is, the value of M is calculated, and then the splitting point corresponding to the largest M value is taken as the maximum gain splitting point.

[0165] Step E22: splitting the initial root node corresponding to the first sample data based on the first maximum gain splitting point to obtain an initial child node corresponding to the initial root node;

[0166] In this embodiment, based on the first maximum gain splitting point, the initial root node corresponding to the first sample data is split to obtain the initial child node corresponding to the initial root node. Specifically, based on the first maximum gain splitting point, the initial root node is split to divide the first feature bin data into a left sample set belonging to the left child node of the initial root node and a right sample set belonging to the right child node of the initial root node, thereby obtaining the initial child node, wherein the initial child node is the left child node and the right child node of the initial root node.

[0167] Step E23, determining whether the initial child node reaches a preset stop splitting condition, if the initial child node reaches the preset stop splitting condition, determining that the initial child node belongs to a leaf node type, and using the first maximum gain splitting point as the target gain splitting point set;

[0168] In this embodiment, it should be noted that the preset stop splitting condition is a condition for determining whether a tree node can be split again. The preset stop splitting condition includes that the number of samples in the tree node reaches a preset minimum sample number threshold and there are no splitting points in the sample set corresponding to the tree node.

[0169] Determine whether the initial child node reaches a preset stop splitting condition. If the initial child node reaches the preset stop splitting condition, determine that the initial child node belongs to a preset leaf node type, and use the first maximum gain splitting point as the target gain splitting point set. Specifically, determine whether both the left child node and the right child node of the initial root node reach the preset stop splitting condition. If both reach the preset stop splitting condition, the left child node and the right child node of the initial root node are both preset leaf node types, and then use the first maximum gain splitting point as the target gain splitting point set, that is, the target gain splitting point set is the first maximum gain splitting point.

[0170] Step E24: if the initial child node does not meet the preset stop splitting condition, determine the second maximum gain splitting point corresponding to the initial child node based on the preset gain calculation formula;

[0171] In this embodiment, if the initial child node does not meet the preset stop splitting condition, the second maximum gain splitting point corresponding to the initial child node is determined based on the preset gain calculation formula. Specifically, if there are splittable child nodes in the left child node and the right child node of the initial root node that do not meet the preset stop splitting condition, the second maximum gain splitting point corresponding to the splittable child node is determined based on the preset gain calculation formula.

[0172] Step E25: based on the second maximum gain splitting point, the initial child node is split again until each splitting node corresponding to the initial child node reaches the preset stop splitting condition, thereby obtaining the target gain splitting point set.

[0173] In this embodiment, based on the second maximum gain splitting point, the initial child node is re-split until each split node corresponding to the initial child node reaches the preset stop splitting node, and the target gain splitting point set is obtained. Specifically, based on the second maximum gain splitting point, the splittable child node is re-split to obtain the re-split child node of the splittable child node, and then the re-split child node is re-judged whether it reaches the preset stop splitting condition until the tree nodes obtained based on the initial root node can no longer be split, and the maximum gain splitting point corresponding to each non-leaf node type tree node is obtained, wherein each of the maximum gain splitting points includes the first maximum gain splitting point and the second maximum gain splitting point, that is, the target gain splitting point set is obtained.

[0174] This embodiment sends a prediction request corresponding to the target user to a second device so that the second device can determine second multi-channel user behavior data, and then performs vertical federated prediction with the second device based on the multi-channel user behavior data and the preset order rate prediction model to combine the second multi-channel user behavior data to predict the target order rate. That is, this embodiment provides a method for predicting an order rate by combining multi-channel user behavior data of a first device and second multi-channel user behavior data of a second device, that is, the first device and the second device first collect user behavior data locally, i.e., through each data channel, and then the first device and the second device first expand the feature richness of the sample data locally, and further, the first device and the second device perform vertical federal prediction, thereby achieving the purpose of predicting the order rate by combining user behavior data collected through multiple data channels of multiple parties, further expanding the feature richness of the sample data, and thus achieving the purpose of predicting the order rate based on sample data with higher feature richness. Compared with the technical means of the prior art that uses a neural network model with a specific feature dimension to predict the user's order rate, the present application obtains user behavior data from multiple data channels, and performs vertical federal prediction with the second device based on the user behavior data from multiple data channels, thereby expanding the richness of the sample data in the local data acquisition process and the order rate prediction process, respectively, and overcomes the technical defect of the prior art that the order rate prediction accuracy is low due to the low feature richness of the sample data, thereby improving the prediction accuracy of the order rate.

[0175] Reference Figure 7 , Figure 7 It is a schematic diagram of the device structure of the hardware operating environment involved in the embodiment of the present application.

[0176] like Figure 7 As shown, the multi-channel order rate prediction device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to realize the connection and communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0177] Optionally, the multi-channel order rate prediction device may also include a rectangular user interface, a network interface, a camera, an RF (Radio Frequency) circuit, a sensor, an audio circuit, a WiFi module, etc. The rectangular user interface may include a display screen (Display), an input submodule such as a keyboard (Keyboard), and the optional rectangular user interface may also include a standard wired interface and a wireless interface. The network interface may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0178] Those skilled in the art will understand that Figure 7 The structure of the multi-channel order rate prediction device shown in the figure does not constitute a limitation on the multi-channel order rate prediction device, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0179] like Figure 7 As shown, the memory 1005 as a computer storage medium may include an operating system, a network communication module, and a multi-channel order rate prediction method program. The operating system is a program that manages and controls the hardware and software resources of the multi-channel order rate prediction device, and supports the operation of the multi-channel order rate prediction method program and other software and / or programs. The network communication module is used to realize the communication between the components inside the memory 1005, and the communication with other hardware and software in the multi-channel order rate prediction method system.

[0180] exist Figure 7 In the multi-channel order rate prediction device shown, the processor 1001 is used to execute the multi-channel order rate prediction method program stored in the memory 1005 to implement any of the steps of the multi-channel order rate prediction method described above.

[0181] The specific implementation of the multi-channel order rate prediction device of the present application is basically the same as the various embodiments of the multi-channel order rate prediction method described above, and will not be repeated here.

[0182] The embodiment of the present application further provides a method and apparatus for predicting an order rate based on multiple channels. The method and apparatus for predicting an order rate based on multiple channels is applied to a device for predicting an order rate based on multiple channels. The method and apparatus for predicting an order rate based on multiple channels includes:

[0183] A determination module, used to determine a target user unique identifier shared by a target user in multiple channels;

[0184] A collection module, used for uniformly collecting multi-channel user behavior data in each of the channels based on the unique identifier of the target user;

[0185] The prediction module is used to perform multi-channel order rate prediction for the target user based on the multi-channel user behavior data and a preset order rate prediction model to obtain a target order rate.

[0186] Optionally, the determining module includes:

[0187] An acquisition unit, used for acquiring channel user information of the target user under each of the access channels;

[0188] A judgment unit, used to determine whether the channel user information under each of the access channels contains the user identification of the target user;

[0189] A first determination unit, configured to use the user identifier as the target user unique identifier if the target user identifier exists;

[0190] The second determination unit is configured to generate the target user unique identifier based on the common user information in the channel user information under each of the channel channels if the target user unique identifier does not exist.

[0191] Optionally, the determining module further includes:

[0192] a polarization hash coding unit, configured to obtain channel user information corresponding to the target user, input the channel user information into a hash coding model optimized based on preset multi-channel category information, perform polarization hash coding on the channel user information, and obtain a polarization hash coding value;

[0193] A calculation unit is used to calculate the bit matching degree between the polarized hash code value and each preset hash code value, and determine the target hash code value from each preset hash code value based on each bit matching degree.

[0194] Optionally, the multi-channel order rate prediction device further includes:

[0195] A pre-coding module, used to obtain training data and a hash coding model to be trained, and generate the preset target hash coding value corresponding to the training data based on a preset target hash coding method;

[0196] An iterative training module is used to iteratively train the hash coding model to be trained based on the preset target hash coding value and the training data to optimize the preset polarization loss function corresponding to the hash coding model to be trained until the hash coding model to be trained reaches a preset iteration end condition, thereby obtaining the hash coding model.

[0197] Optionally, the acquisition module includes:

[0198] An extraction unit, configured to extract single-channel user behavior data from each of the channel channels based on the unique identifier of the target user;

[0199] The first generating unit is used to respectively generate behavior nodes corresponding to each of the single-channel user behavior data, and connect each of the behavior nodes according to the occurrence sequence of user behaviors corresponding to each of the channel user behavior data to obtain the user behavior path diagram.

[0200] Optionally, the prediction module includes:

[0201] A second generating unit, configured to generate a target user behavior graph corresponding to the multi-channel user behavior data, and convert the target user behavior graph into a user behavior graph for embedding;

[0202] The first prediction unit is used to embed the user behavior graph into the graph embedding prediction model, classify the user behavior graph embedding, so as to predict the order rate of the target user based on multiple channels to obtain the target order rate.

[0203] Optionally, the generating unit includes:

[0204] An encoding subunit, configured to determine each user behavior subgraph corresponding to the target user behavior graph, and based on each of the user behavior subgraphs, perform one-hot encoding on the target user behavior graph to obtain a one-hot encoding vector corresponding to the target user behavior graph;

[0205] The dimensionality reduction processing subunit is used to perform dimensionality reduction processing on the one-hot encoded vector to obtain the user behavior graph embedding.

[0206] Optionally, the prediction module further includes:

[0207] A first determining unit, configured to determine a target user behavior feature combination corresponding to the multi-channel user behavior data;

[0208] A second determining unit, configured to determine, based on the target user behavior feature combination, the preset order rate prediction model corresponding to the multi-channel user behavior data in a preset model set;

[0209] The second prediction unit is used to classify the multi-channel user behavior data based on the preset order rate prediction model to predict the order rate of the target user based on multiple channels to obtain the target order rate.

[0210] Optionally, the prediction module further includes:

[0211] a sending unit, configured to send the prediction request corresponding to the target user to a second device, so that the second device can determine second multi-channel user behavior data;

[0212] A vertical federation prediction unit is used to perform vertical federation prediction with the second device based on the multi-channel user behavior data and the preset order rate prediction model, so as to combine the second multi-channel user behavior data to predict the target order rate.

[0213] The specific implementation of the multi-channel order rate prediction method device of the present application is basically the same as the above-mentioned multi-channel order rate prediction method embodiments, and will not be repeated here.

[0214] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent processing scope of the present application.

Claims

1. A method for predicting order rate based on multiple channels, characterized in that: The multi-channel order rate prediction method includes: Determine the unique identifier of the target user shared by multiple channels; Based on the unique identifier of the target user, uniformly collect multi-channel user behavior data in each of the channels; Based on the multi-channel user behavior data and a preset order rate prediction model, a multi-channel order rate prediction is performed on the target user to obtain a target order rate; The target user unique identifier includes a target hash code value, and the step of determining the target user unique identifier shared by the target user in multiple access channels includes: Acquire channel user information corresponding to the target user, input the channel user information into a hash coding model optimized based on preset multi-channel category information, perform polarized hash coding value on the channel user information, and obtain a polarized hash coding value; The bit matching degree between the polarized hash code value and each preset hash code value is calculated, and based on each of the bit matching degrees, the target hash code value is determined from each of the preset hash code values.

2. The method for predicting order rate based on multiple channels as claimed in claim 1, characterized in that: The preset multi-channel category information includes a preset target hash code value, Before the step of obtaining the channel user information corresponding to the target user, inputting the channel user information into a hash coding model optimized based on preset category information, performing polarized hash coding value on the channel user information, and obtaining the polarized hash coding value, the multi-channel-based order rate prediction method further includes: Acquire training data and a hash coding model to be trained, and generate the preset target hash coding value corresponding to the training data based on a preset target hash coding method; Based on the preset target hash code value and the training data, the hash code model to be trained is iteratively trained to optimize a preset polarization loss function corresponding to the hash code model to be trained, until the hash code model to be trained reaches a preset iteration end condition, thereby obtaining the hash code model.

3. The method for predicting order rate based on multiple channels as claimed in claim 1, characterized in that: The multi-channel user behavior data includes a user behavior path map, The step of uniformly collecting multi-channel user behavior data in each of the channels based on the unique identifier of the target user includes: Based on the unique identifier of the target user, extracting single-channel user behavior data in each of the channel; Behavior nodes corresponding to each of the single-channel user behavior data are generated respectively, and according to the occurrence sequence of user behaviors corresponding to each of the channel user behavior data, each of the behavior nodes is connected to obtain the user behavior path diagram.

4. The method for predicting order rate based on multiple channels as claimed in claim 1, characterized in that: The preset order rate prediction model includes a graph embedding prediction model, The step of performing multi-channel order rate prediction on the target user based on the multi-channel user behavior data and the preset order rate prediction model to obtain the target order rate includes: Generating a target user behavior graph corresponding to the multi-channel user behavior data, and converting the target user behavior graph into a user behavior graph for embedding; The user behavior graph is embedded into the graph embedding prediction model, and the user behavior graph embedding is classified to predict the order rate of the target user based on multiple channels to obtain the target order rate.

5. The method for predicting order rate based on multiple channels as claimed in claim 4, characterized in that: The step of converting the target user behavior graph into a user behavior graph and embedding it comprises: Determine each user behavior subgraph corresponding to the target user behavior graph, and based on each of the user behavior subgraphs, perform one-hot encoding on the target user behavior graph to obtain a one-hot encoding vector corresponding to the target user behavior graph; Perform dimensionality reduction processing on the one-hot encoded vector to obtain the user behavior graph embedding.

6. The method for predicting order rate based on multiple channels as claimed in claim 1, characterized in that: The step of performing multi-channel order rate prediction on the target user based on the multi-channel user behavior data and the preset order rate prediction model to obtain the target order rate includes: Determine a target user behavior feature combination corresponding to the multi-channel user behavior data; Based on the target user behavior feature combination, determining the preset order rate prediction model corresponding to the multi-channel user behavior data in a preset model set; Based on the preset order rate prediction model, the multi-channel user behavior data is classified to perform multi-channel-based order rate prediction for the target user to obtain the target order rate.

7. The method for predicting order rate based on multiple channels as claimed in claim 1, characterized in that: The multi-channel order rate prediction method is applied to the first device, and the step of performing multi-channel order rate prediction on the target user based on the multi-channel user behavior data and the preset order rate prediction model to obtain the target order rate includes: Sending a prediction request corresponding to the target user to a second device, so that the second device can determine second multi-channel user behavior data; Based on the multi-channel user behavior data and the preset order rate prediction model, a vertical federated prediction is performed with the second device to combine the second multi-channel user behavior data to predict the target order rate.

8. A multi-channel order rate prediction device, characterized in that: The multi-channel order rate prediction device includes: a memory, a processor, and a program stored in the memory for implementing the multi-channel order rate prediction method. The memory is used to store a program for implementing a method for predicting an order rate based on multiple channels; The processor is used to execute a program for implementing the multi-channel order placement rate prediction method to implement the steps of the multi-channel order placement rate prediction method as described in any one of claims 1 to 7.

9. A readable storage medium, characterized in that: The readable storage medium stores a program for implementing a method for predicting an order rate based on multiple channels, and the program for implementing a method for predicting an order rate based on multiple channels is executed by a processor to implement the steps of the method for predicting an order rate based on multiple channels as described in any one of claims 1 to 7.

Images