Target channel prediction method and device based on multiple channels and computer equipment

Abstract

The invention relates to the field of financial science and technology, in particular to a target channel prediction method and device based on multiple channels and computer equipment. The method comprises the following steps: inputting sample data of a plurality of original channels into a channel classification model to obtain a category attribution prediction result of each sample data in each original channel, which is output by the channel classification model and belongs to each known channel; determining the similarity between each original channel and the target channel according to the category attribution prediction result of each piece of sample data of each original channel; selecting at least one original channel of which the similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel; and inputting the to-be-predicted data of the target channel into a prediction model pre-constructed by a similar original channel to obtain a prediction result corresponding to the sample data of the target channel. The prediction effect of the prediction model selected based on the scheme is higher in accuracy compared with the prediction model obtained through training of the target channel.

Description

Technical Field

[0001] This specification relates to the field of financial technology, and in particular to a multi-channel based target channel prediction method, apparatus and computer equipment. Background Art

[0002] The user behavior prediction model in the field of financial technology has gone through a long period of development, from the early scoring card model to the machine learning model and deep learning model. With the continuous development of the financial market, the source of customer groups has become more and more diversified, and how to build a user identification model that adapts to multiple channels is becoming increasingly important. The user distribution of different channels varies greatly, and a single model cannot effectively identify users in each channel.

[0003] There are currently two ways to identify customers in each channel: 1. In the pre-business classification, pre-classify the channels to which users belong or user characteristics, and then use channel modeling to identify users, that is, use the channel model corresponding to each channel to identify users of the corresponding channel; 2. Use a unified model to identify user data, and correct the model identification results through post-strategies.

[0004] However, the inventors have discovered that there are problems with the above methods: both methods do not take into account multi-channel information, making it difficult to accurately identify multi-channel users, and thus making it difficult to provide and configure reasonable resources for multi-channel users. Summary of the invention

[0005] To solve the problems of the prior art, the embodiments of this specification provide a target channel prediction method, apparatus and computer equipment based on multiple channels.

[0006] An embodiment of the present specification provides a target channel prediction method based on multiple channels, the method comprising: inputting sample data of multiple original channels into a channel classification model, obtaining a category attribution prediction result of each sample data in each original channel output by the channel classification model, wherein the channel classification model is trained by known channel sample data and corresponding channel labels, the known channel sample data at least including target channel sample data; determining the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel; selecting at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel; inputting the data to be predicted of the target channel into a prediction model pre-constructed based on the similar original channel, and obtaining a prediction result corresponding to the sample data of the target channel.

[0007] According to one aspect of an embodiment of the present specification, determining the similarity between each original channel and the target channel based on the category attribution prediction result of each sample data of each original channel includes: determining the number of category attribution prediction results of the sample data of each original channel belonging to each type of known channels; constructing a similar sample quantity matrix based on the number of category attribution prediction results of the known channels, original channels and sample data of each original channel belonging to each type of known channels; determining the similarity between each original channel and the target channel based on the similar sample quantity matrix.

[0008] According to one aspect of an embodiment of the present specification, determining the similarity between each original channel and the target channel based on the similar sample quantity matrix includes: determining the proportion of similar samples that each original channel belongs to each type of target channel based on the number of category attribution prediction results of each original channel belonging to each type of known channels and the total amount of sample data of each original channel in the similar sample quantity matrix; calculating the average proportion of similar samples corresponding to the same classification result for all original channels based on the proportion of similar samples that each original channel belongs to each type of target channel; obtaining the approximate multiple of each original channel belonging to the same type of target channel based on the ratio of the similar sample proportion to the average proportion of similar samples; and determining the value of the approximate multiple as the similarity between the original channel and the target channel.

[0009] According to one aspect of an embodiment of the present specification, constructing a prediction model in the following manner includes: if the number of similar original channels is one, using sample data of the similar original channels and channel labels corresponding to the sample data, training an initial classification model, and constructing a first prediction model corresponding to the similar original channels; if there are multiple similar original channels, using sample data of any similar original channel and at least one of the other similar original channels and multiple channel labels of all sample data, training an initial classification model, and constructing a second prediction model corresponding to multiple original channels, wherein N is a positive integer greater than 1.

[0010] According to one aspect of the embodiments of the present specification, at least one original channel is selected according to a preset mode, and an initial classification model is trained using multiple channel labels of sample data of the original channel to construct second prediction models corresponding to multiple original channels, including: using sample data of each similar original channel and the channel label of the sample data to train the initial classification model to obtain a first number of second prediction models; arbitrarily selecting a preset number of original channels from N similar original channels, and using sample data of the arbitrarily selected preset number of original channels and the channel label of the sample data to train the initial classification model to obtain a second number of second prediction models, the first number of second prediction models and the second number of second prediction models constitute the second prediction model.

[0011] According to one aspect of an embodiment of the present specification, the channel classification model is trained in the following manner: obtaining a sample data set formed by a combination of sample data of known channels and a channel label corresponding to each sample data, wherein the known channels include a target channel, or a combination of a target channel and at least one original channel; using the sample data set to train an initial multi-classifier to obtain the channel classification model, and the channel classification model outputs a category attribution prediction result for each sample data.

[0012] According to one aspect of an embodiment of the present specification, if the known channel includes a target channel and at least one original channel, the method further includes: determining the classification prediction accuracy of each sample data in each original channel based on the category attribution prediction result of each sample data in each original channel and the sample channel label of the original channel; determining the prediction accuracy of the multiple classifiers for each original channel based on the classification prediction accuracy of each sample data in each original channel; and dividing the original channels into a first original channel and a second original channel based on the prediction accuracy and a preset accuracy threshold.

[0013] According to one aspect of an embodiment of the present specification, the method further includes: the method also includes: if the first similarity between the first original channel and each type of target channel is not higher than a preset threshold, then determining the second similarity between the second original channel and each type of target channel based on the category attribution prediction result of each sample data of the second original channel; selecting the second original channel whose second similarity is greater than the preset threshold as a similar original channel similar to each type of target channel.

[0014] The embodiment of the present specification also provides a target channel prediction device based on multiple channels, and the device includes: a category attribution prediction result determination unit, which is used to input sample data of multiple original channels into a channel classification model, and obtain category attribution prediction results of each sample data in each original channel output by the channel classification model belonging to each known channel, wherein the channel classification model is trained by known channel sample data and corresponding channel labels, and the known channel sample data at least includes target channel sample data; a similarity determination unit, which is used to determine the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel; a similar original channel determination unit, which is used to select at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel; a prediction result determination unit, which is used to input the data to be predicted of the target channel into a prediction model pre-constructed based on the similar original channel, and obtain the prediction result corresponding to the sample data of the target channel.

[0015] The embodiments of the present specification also provide a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the multi-channel-based target channel prediction method when executing the computer program.

[0016] The embodiments of the present specification also provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the multi-channel-based target channel prediction method is implemented.

[0017] In this specification, when a new target channel is added, the best prediction model can be quickly selected. The prediction effect of the prediction model selected based on this solution is more accurate than the prediction model trained by the target channel itself. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the embodiments of this specification 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, the drawings described below are only some embodiments of this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 It is a schematic diagram of a target channel prediction method based on multiple channels according to an embodiment of the present specification;

[0020] Figure 2 The flowchart of a method for determining the similarity between an original channel and a target channel according to an embodiment of the present specification is shown;

[0021] Figure 3 The flowchart of a method for determining the similarity between an original channel and a target channel according to an embodiment of the present specification is shown;

[0022] Figure 4 This is a flow chart of a method for constructing a prediction model according to an embodiment of this specification;

[0023] Figure 5 The figure is a flow chart of a method for constructing a channel classification model according to an embodiment of the present specification;

[0024] Figure 6 The figure shows a flow chart of a method for classifying original channels according to an embodiment of the present specification;

[0025] Figure 7 The figure is a flow chart of a method for determining similar original channels according to an embodiment of the present specification;

[0026] Figure 8It is a schematic diagram of the structure of a target channel prediction device based on multiple channels according to an embodiment of the present specification;

[0027] Fig.9A The figure shows a schematic diagram of a similar sample number matrix according to an embodiment of the present specification;

[0028] Fig. 9B The figure shows a schematic diagram of a similar sample ratio matrix according to an embodiment of the present specification;

[0029] Fig. 9C The figure shows a matrix diagram of a ratio mean value of an embodiment of this specification;

[0030] Fig.9D The figure shows a schematic diagram of a matrix of an approximate multiple matrix according to an embodiment of the present specification;

[0031] Fig.9E Shown is a schematic diagram of a similarity matrix according to an embodiment of this specification;

[0032] Fig.10 The figure shows a schematic diagram of the structure of a computer device according to an embodiment of the present specification.

[0033] Description of the accompanying symbols:

[0034] 801, a category attribution prediction result determination unit;

[0035] 802, similarity determination unit;

[0036] 803. Similar original channel determination unit;

[0037] 804. prediction result determination unit;

[0038] 1002. Computer equipment;

[0039] 1004. Processor;

[0040] 1006. Memory;

[0041] 1008. Driving mechanism;

[0042] 1010, input / output module;

[0043] 1012. Input device;

[0044] 1014. Output device;

[0045] 1016. Presentation equipment;

[0046] 1018. Graphical user interface;

[0047] 1020. Network interface;

[0048] 1022. Communication link;

[0049] 1024. Communication bus. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on the embodiments in this specification, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this specification.

[0051] It should be noted that the terms "first", "second", etc. in the description and claims of this specification and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of this specification described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, device, product or equipment that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0052] This specification provides method operation steps as described in the embodiments or flow charts, but more or fewer operation steps may be included based on routine or non-creative work. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the system or device product is executed in practice, it can be executed in the order of the method shown in the embodiments or the drawings or in parallel.

[0053] The information, data and signals involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data comply with the relevant laws, regulations and standards of relevant countries and regions. The acquisition, transmission, storage, use and processing of data in the technical solution of this application comply with the relevant provisions of national laws and regulations.

[0054] It should be noted that the multi-channel based target channel prediction method, device and computer equipment method of this specification can be used in the field of financial technology, and this specification does not limit the application field of the multi-channel based target channel prediction method, device and computer equipment.

[0055] In the first method of the prior art, the sub-channel modeling uses sample data from a single channel to build models corresponding to each channel during the model training process. Each sub-channel model loses multi-channel information synergy in samples and features. The sub-channel model lacks multi-channel generalization capabilities and cannot accurately identify new channel users or channel users with a small sample size. Therefore, it is difficult to better identify multi-channel users and cannot provide and allocate reasonable resources for multi-channel users. The second method of the prior art first uses a unified model to identify users, and then formulates strategies to adapt to multiple channels. However, multi-channel information is also not taken into account when building a unified model. The post-strategy has a lag, and it is also difficult to better identify multi-channel users and cannot provide and allocate reasonable resources for multi-channel users. Therefore, none of the prior art considers multi-channel information to identify multi-channel users.

[0056] This application considers information from multiple channels and constructs a prediction model suitable for users from each channel, thereby more accurately identifying multi-channel users.

[0057] Figure 1 The flowchart of a target channel prediction method based on multiple channels in an embodiment of this specification is shown. The execution subject of this method can be a server or a computer. In the embodiment of this specification, the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and big data and artificial intelligence platforms. This method specifically includes the following steps:

[0058] Step 101, input sample data of multiple original channels into a channel classification model, and obtain the prediction results of the category attribution of each sample data in each original channel output by the channel classification model to each known channel. The channel classification model is trained by known channel sample data and corresponding channel labels, and the known channel samples at least include target channel samples.

[0059] The purpose of this application is to use the channel classification model to predict the samples of the target channel, that is, to predict whether the user behavior corresponding to the samples of the target channel is risky. The target channel is a new unknown channel, and the sample data of the target channel does not yet have a corresponding risk label. Among them, the channel classification model is constructed from known channels, and the known channels at least include target channels. Specifically, the known channels include multiple target channels, or the known channels can be a combination of a target channel and at least one original channel. For example, the known channels include 5 target channels, or the known channels include 1 target channel and 4 original channels, etc.

[0060] In the embodiments of this specification, the original channel can be understood as an existing old channel. Specifically, the original channel is an old channel, and its sample data is analyzed data, and the sample data of the original channel corresponds to a marked channel label, so there is no need to predict the sample data of the original channel. Each original channel has sample data related to the user, including but not limited to: user information, user resource application behavior, user risk ranking and other data.

[0061] In some examples of this specification, the original channel may be a channel in the field of financial technology. The sample data of the original channel may include a channel label, such as the channel label corresponding to the sample data of the original channel a is a, the channel label corresponding to the sample data of the original channel e is e, etc. By combining the sample data of multiple original channels through channel labels, sample data sets of multiple original channels can be obtained, so that the sample data sets can be used to train the initial channel classification model. Among them, the number of sample data of each original channel is greater than 20,000.

[0062] In an embodiment of the present specification, a channel classification model is used to learn the differences in customer groups reflected by sample data of the original channel and sample data of the target channel. The channel classification model is trained and constructed by known channel sample data and corresponding channel labels. Among them, the known channel samples at least include target channel samples. The known channel samples may include multiple target channel samples, or may include at least one target channel sample and at least one original channel sample. In the present specification, the target channel is a newly added channel in the financial field, and it is necessary to analyze the users of the newly added target channel.

[0063] In an embodiment of the present specification, if there is one newly added target channel, sample data of multiple original channels and sample data of the newly added target channel are used, and channel labels corresponding to the sample data of the target channel and the sample data of the original channel are used as labels for training the initial channel classification model to construct a channel classification model.

[0064] In an embodiment of the present specification, if there are three newly added target channels, sample data from multiple original channels and sample data from the three newly added target channels are used, and channel labels corresponding to the sample data of the target channels and the sample data of the original channels are used as labels for training the initial channel classification model to construct a channel classification model.

[0065] After the channel classification model is constructed, the sample data of multiple original channels are input into the channel classification model to obtain the category attribution prediction results of each sample data in each original channel belonging to each known channel. Among them, when constructing the channel classification model, the known channels used will have corresponding category attribution prediction results. For example, there are 5 known channels used to train the channel classification model. After the sample data of multiple original channels are input into the channel classification model, there are 5 category attribution prediction results of the sample data of the original channels belonging to each known channel. In an example, there are 5 known channels used to train the channel classification model: channel 1, channel 2, channel 3, channel 4, channel 5. After the sample data 1 and sample data 2 of the original channel 1 are input into the channel classification model, the category attribution prediction results of sample data 1 belonging to channels 1-5 are obtained as one of channels 1 to channel 5 (such as the category attribution prediction result is that sample data 1 belongs to channel 2), and the category attribution prediction results of sample data 2 belonging to channels 1-5 are obtained as one of channels 2 to channel 5 (such as the category attribution prediction result is that sample data 2 belongs to channel 5).

[0066] In the embodiments of this specification, compared with the method of constructing a channel classification model using only original channels, this solution uses known channels including at least the target channel to construct a channel classification model, which can better learn the differences in sample data between the target channel and the original channel. The channel classification model thus constructed has higher classification accuracy in the application stage and better classification prediction effect.

[0067] Step 102: Determine the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel.

[0068] After the channel classification model is constructed in the above steps, the sample data of the original channel is input into the channel classification model to obtain the category attribution prediction result of each sample data output by the channel classification model. Based on the category attribution prediction result, a similarity matrix between the original channel and the known channel is constructed, and the similarity between the original channel and the target channel can be further calculated. For a description of the scheme for determining the similarity between each original channel and the target channel in some examples, please refer to the following section on Figure 3 Description.

[0069] Step 103, select at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel. In the embodiment of this specification, the higher the similarity, the more similar the original channel is to the target channel, and such an original channel can be regarded as a channel similar to the target channel.

[0070] In the embodiments of this specification, one original channel with high similarity can be selected as the original channel similar to the target channel; or multiple original channels with high similarity can be selected as the original channels similar to the target channel. This specification does not limit the number of original channels similar to the target channel.

[0071] Step 104: input the to-be-predicted data of the target channel into a prediction model pre-built based on the similar original channel to obtain a prediction result corresponding to the sample data of the target channel.

[0072] In this step, compared with the large amount of sample data of the original channel, the amount of sample data of the target channel is relatively small. If the prediction model is directly constructed using the sample data of the target channel, the prediction effect of the obtained prediction model on the target channel's to-be-predicted data is not ideal. Therefore, a prediction model constructed using an original channel similar to the target channel can be used to perform predictions to obtain an ideal prediction effect. In some examples, before performing step 104, a prediction model is constructed based on the similar original channels that have been determined in step 103. In other examples, a prediction model can also be constructed in step 104. In other embodiments, the prediction model can be constructed in step 104. Figure 1 Before the method was proposed, prediction models corresponding to each original channel were constructed based on the sample data of each original channel.

[0073] In the embodiments of this specification, the prediction model is pre-built through the original channels, and the prediction model can be built using some or all of the original channels. For example, the prediction model can be built using the sample data of each original channel, or multiple original channels can be combined to obtain sample data sets to jointly build the prediction model.

[0074] It is worth noting that the sample data of the original channel has its corresponding risk label. Therefore, the prediction model can be trained better through the sample data and risk labels of the original channel.

[0075] For example, if there are 20 original channels, then the sample data of the 20 original channels and the risk labels corresponding to each sample data are used to train the initial prediction models respectively, and finally 20 prediction models are obtained respectively; for another example, multiple original channels among the 20 channels are combined each time according to the preset number of models, and the initial prediction model is trained with the combined sample data, and finally a prediction model constructed according to the preset number of models is obtained. The preset number of models can be 2, 3, etc. For example, if there are 14 original channels and the preset number of models is 2, any 2 original channels among the 14 original channels are combined in pairs to construct 91 prediction models. For details on the specific method of using original channels to construct prediction models in this manual, please refer to Figure 4 describe.

[0076] In the embodiment of this specification, at least one original channel similar to the target channel is selected to construct a prediction model, and a prediction model is constructed. The prediction model is pre-constructed by the original channel, integrates a large amount of sample data from multiple channels, and has the ability to generalize to multiple channels. When a new channel is introduced, the prediction model corresponding to the new channel can be quickly matched, ensuring the prediction effect on the target channel and the target channel users. Figure 2 The flowchart of a method for determining the similarity between an original channel and a target channel according to an embodiment of the present specification is shown, which specifically includes the following steps:

[0077] Step 201, determining the number of category attribution prediction results of sample data of each original channel belonging to each category of known channels.

[0078] In the embodiments of the present specification, since each original channel has a large amount of sample data, and each sample data of the same original channel has a corresponding category attribution prediction result, there will be a situation where multiple sample data belong to the same category attribution prediction result. Therefore, it is necessary to integrate the category attribution prediction results of all sample data according to the channel classification model predicted for each sample data of the original channel, and obtain a similar sample quantity matrix based on the integrated category attribution prediction results. Among them, the elements in the similar sample quantity matrix are the integrated category attribution prediction results, and the integrated category attribution prediction results are used to characterize the number of sample data of the original channel that belong to each category attribution prediction result. For example, Fig.9A As shown in the figure, the total number of sample data of the original channel a is 38852. The 38852 sample data are predicted according to the channel classification model, and 38852 category attribution prediction results are obtained. Then, the same category attribution prediction results are integrated and statistically analyzed, and the sample data belonging to the same category attribution prediction result are classified into one category, so as to determine how many sample data of the original channel a belong to each category attribution prediction result. The category attribution prediction results corresponding to the sample data are filled into the similar sample quantity matrix, so as to obtain each element in the similar sample quantity matrix. In this way, the similar sample quantity matrix is ​​constructed, as shown in FIG. Fig.9A shown. Fig.9A The figure shows the prediction results of the channel classification model for each sample data of a certain original channel. For example, the number of samples of the original channel a predicted to belong to the target channel o is 2111, the number of samples belonging to the target channel z is 5479, the number of samples belonging to the target channel c is 29181, and so on.

[0079] For example, a channel classification model is constructed from known channels, which include 5 target channels, such as Fig.9AIn the first row, columns 2 to 6 are respectively: target channel o, target channel z, target channel d, target channel u and target channel c. The first column in the figure shows that the original channels used to build the channel classification model are channel a, channel e, channel f, channel h...channel x and channel y, a total of 14 original channels.

[0080] The channel classification model predicts the category attribution prediction results of each sample data of the original channel, that is, predicts which one of channel o, channel d, channel a, channel e or channel f each sample data should belong to. For example, the original channel a has 38852 sample data. After the channel classification model prediction, 2111 sample data in the original channel a are predicted to belong to the target channel o, 5479 sample data are predicted to belong to the target channel z, 652 sample data are predicted to belong to the target channel d, 1429 sample data are predicted to belong to the target channel u, and 29181 sample data are predicted to belong to the target channel c. It can be seen that all sample data in the original channel are predicted to belong to different target channels, among which most of the sample data in the original channel are predicted to belong to the target channel c.

[0081] In the same way, the sample data of other multiple original channels are predicted respectively to obtain the number of sample data of other original channels belonging to each target channel.

[0082] Step 202, constructing a similar sample quantity matrix according to the known channels, original channels and the number of category attribution prediction results of the sample data of each original channel belonging to each category of known channels.

[0083] In this specification, the known channels, original channels and channel classification models are used to predict the category of sample data from the original channels to construct a similar sample quantity matrix. Fig.9A The first row of ), the original channel is used as the ordinate of the similar sample number matrix (such as Fig.9A The first column of ). The sample data of each original channel belongs to the number of each known channel, which is each element in the similar sample quantity matrix. Fig.9A As shown, Fig.9A The elements from the second to the fifteenth row and the second to the sixth column in represent the number of sample data from the 14 original channels, namely, original channel a, original channel e…original channel x and original channel y, belonging to the five known channels, namely, known channel o, known channel z, known channel d, known channel u and known channel c, respectively.

[0084] Step 203: Determine the similarity between each original channel and the target channel according to the similar sample quantity matrix.

[0085] By processing each element in the similar sample quantity matrix, the similarity between each original channel and the target channel can be obtained.

[0086] In the embodiment of this specification, the first row of the similar sample quantity matrix represents the known channels, and each of the remaining rows represents the prediction results of the categories of the sample data of each original channel belonging to each known channel. This specification aims to determine the similarity between the original channel and the target channel. If the known channel includes the target channel and the original channel, when calculating the similarity, the prediction results of the categories of the target channels in each known channel that the sample data of each original channel belongs to can be selected for calculation. If the known channel only includes the target channel, the similarity is directly calculated based on each element in the similar sample quantity matrix without screening.

[0087] In this specification, by determining the similarity between the original channel and the target channel, it is convenient to select the original channel that is close to or similar to the target channel. Because the user behaviors in similar channels are also similar, in the subsequent steps, the prediction model is constructed by a large amount of sample data of similar original channels, which can be approximated to the prediction model constructed for the target channel, so as to avoid the problem that the prediction effect of the prediction model is not good after the prediction model is constructed due to the small amount of sample data of the target channel, thereby ensuring the prediction effect of the target channel and improving the efficiency of building the prediction model.

[0088] Figure 3 The flowchart of a method for determining the similarity between an original channel and a target channel according to an embodiment of the present specification is shown, which specifically includes the following steps:

[0089] Step 301, based on the number of category attribution prediction results of sample data of each original channel belonging to each category of known channels and the total amount of sample data of each original channel in the similar sample quantity matrix, determine the proportion of similar samples of sample data of each original channel belonging to each category of target channels.

[0090] Based on the above, the known channels at least include the target channels, and the known channels can be the target channels, or a combination of the target channels and the original channels. Therefore, it is necessary to first determine the composition of the known channels, so as to further determine whether the sample data of each type of original channels belongs to the category attribution prediction result of each type of known channels, and whether the sample data of each original channel belongs to the category attribution prediction result of the target channel.

[0091] In the embodiment of this specification, the number of category attribution prediction results of each original channel belonging to each type of target channel is divided by the total amount of sample data of the original channel to obtain the proportion of similar samples of the original channel corresponding to each type of target channel. Fig. 9BAs shown, the number of samples of original channel a predicted to belong to target channel o is 2111, and the total sample data of original channel a is 38852. Ratioing the number of samples of original channel a predicted to belong to target channel o with the total sample data of original channel a, it is obtained that the proportion of similar samples of original channel a belonging to target channel o is 5.43%.

[0092] Step 302, based on the proportion of similar samples of each original channel belonging to each type of target channel, calculate the average proportion of similar samples corresponding to the same category attribution prediction result of all original channels.

[0093] After determining the proportion of similar samples of each original channel belonging to each type of target channel, the proportion of similar samples of all original channels under each category attribution prediction result of the channel classification model is averaged. Fig. 9B As shown in , the average of the similar sample proportions of multiple original channels corresponding to each target channel category of the channel classification model is calculated. For example, the average of the similar sample proportions of the 14 original channels corresponding to the target channel o of the channel classification model is calculated, and the average of the similar sample proportions corresponding to the target channel o is 12.56%, as shown in Fig. 9C In the same way, the average of the similar sample proportions of multiple original channels corresponding to each other target channel can be calculated respectively.

[0094] In this way, the proportion of similar samples corresponding to each category of the channel classification model can be calculated, and a similar sample proportion matrix can be constructed.

[0095] Step 303: According to the ratio of the similar sample proportion to the average similar sample proportion, the approximate multiple of each original channel belonging to the same type of target channel is obtained.

[0096] In this step, for Fig. 9B For each element in the similar sample ratio matrix constructed in , divide the value of the similar sample ratio of each element by Fig. 9C The mean of the proportion of similar samples shown in , minus 1, to get the new value of each element. Use the new values ​​of the elements to construct a matrix to get an approximate multiple matrix. Fig.9D As shown in the figure, the proportion of similar samples of the original channel a belonging to the target channel o is 5.43%, which is divided by the average proportion of similar samples corresponding to the target channel o, which is 12.56%. Then minus 1, we get -0.57, which is the approximate multiple value of the original channel belonging to the target channel o, which is Fig.9D The element value in the second row and second column of .

[0097] Step 304: The value of the approximate multiple is used as the similarity between the original channel and the target channel.

[0098] In an embodiment of the present specification, for each column in the approximate multiple matrix (i.e., each category of the channel classification model) obtained according to the above steps, the maximum value of the "approximate multiple" on each original channel is calculated respectively. For example, the maximum value of the approximate multiple of the second column of the approximate multiple matrix on the 14 original channels is 1.27; the maximum value of the approximate multiple of the third column of the approximate multiple matrix on the 14 original channels is 0.81. The ratio of the "approximate multiple" of each original channel in each category of the channel classification model to the maximum value of the approximate multiple is calculated to obtain the similarity of each original channel in each category of the channel classification model. This similarity can represent the similarity between the original channel and the target channel. For example, the ratio of the approximate multiple value -0.57 of the original channel a belonging to the target channel o to the maximum value of the approximate multiple (1.27) of the target channel o is calculated, and -0.45 is obtained, which indicates that the similarity of the original channel to the target channel o is -0.45. Using the same method, we calculated the similarities between the original channel a and the other four target channels in the channel classification model, which are -0.88, -0.15, -0.28 and 1 respectively. Fig.9E .

[0099] In this specification, by constructing a similar sample quantity matrix, the proportion of similar samples of the original channel belonging to the target channel and the approximate multiples of the original channel belonging to the same type of target channel are determined, and finally the similarity between the original channel and the target channel is determined, thereby ensuring the prediction effect on the target channel.

[0100] Figure 4 The flowchart of a method for constructing a prediction model in an embodiment of this specification is shown. This specification provides method operation steps as described in the embodiment or flowchart, but more or fewer operation steps may be included based on conventional or non-creative labor. The order of steps listed in the embodiment is only one way of executing the steps among many, and does not represent the only execution order. Figure 4 When the steps in the actual system or device product are executed, they can be executed in parallel as shown in the embodiments or drawings. In this specification, the risk label corresponding to each sample of the original channel is known, so the sample data of the original channel can be directly used to build a risk prediction model. Specifically, the following steps are included:

[0101] Step 401, if the number of similar original channels is one, use the sample data of the similar original channels and the channel labels corresponding to the sample data to train the initial classification model, and construct a first prediction model corresponding to the similar original channel. In the embodiment of the present specification, the target channel is the channel to be predicted, but the amount of sample data of the target channel is not large. If the prediction model is directly constructed with the sample data of the target channel, the prediction effect of the prediction model is not ideal. Therefore, this step uses the original channels similar to the target channel to train the initial classification model, so as to construct a prediction model. Among them, the initial classification model is also called the base model (Base Model), which is a model obtained through pre-training of massive data. The prediction model is used to predict the risk ranking corresponding to the sample data of the input channel.

[0102] In an embodiment of the present specification, if there is only one similar original channel similar to the target channel, the sample data of the similar original channel and the channel label corresponding to the sample data of the similar original channel are used to train the initial classification model, and finally a first prediction model corresponding to the similar original channel is obtained.

[0103] In the embodiments of this specification, the initial classification model or base model may be a binary classification model, for example, one or a combination of lightgbm, catboost, or xgboost. The output result of the base model may be: the sample data is risky, or the sample data is risk-free.

[0104] According to the above method, if there are multiple target channels, and the number of similar original channels of each target channel is one, these similar original channels can be used respectively to construct prediction models corresponding to each similar original channel. For example, if there are 5 target channels, and according to the above steps, the number of similar original channels of these 5 target channels is only one, then the 5 similar original channels similar to these 5 target channels can be used respectively to train the initial models respectively, and finally 5 prediction models can be constructed.

[0105] Step 402: If there are N similar original channels, select at least one original channel according to a preset mode, use multiple channel labels of sample data of the original channel to train an initial classification model, and construct a second prediction model corresponding to the multiple original channels, where N is a positive integer greater than 1.

[0106] In the embodiment of this specification, step 401 and step 402 are executed in parallel, and are not limited to a sequential execution relationship.

[0107] Specifically, on the one hand, for N similar original channels, the sample data of each similar original channel and the channel label of the sample data are used to train the initial classification model to obtain a first number of second prediction models. That is to say, a second prediction model is trained using a single similar original channel, and N second prediction models are trained using N similar original channels.

[0108] On the other hand, a preset number of original channels are arbitrarily selected from N similar original channels, and the sample data of the arbitrarily selected preset number of original channels and the channel labels of the sample data are used to train the initial classification model to obtain the second number of prediction models. In this step, the preset mode is: N similar original channels are combined in different quantities, for example, any 2 similar original channels are combined, or any 3 similar original channels are combined. The sample data of the combined multiple similar original channels and the channel labels corresponding to the sample data are used to train the channel classification model to construct a second prediction model corresponding to the similar original channels. The first number of second prediction models and the second number of second prediction models together constitute a second prediction model for predicting the sample data of the target channel. For example, there are N clear similar original channels, and N+N*(N-1) / 2 binary classification models can be made together to obtain a second prediction model. Therefore, in terms of quantity, the number of second prediction models is greater than the number of first prediction models.

[0109] In some other embodiments of this specification, a plurality of original channels can be used to pre-build a prediction model without waiting for the similarity between the original channel and the target channel to be calculated before building the prediction model. For example, in the initial state, there are 14 original channels, sample data corresponding to the 14 original channels, and risk labels corresponding to each sample data. First, based on the sample data and risk labels of each of the 14 original channels, 14 base models are constructed; secondly, the 14 original channels are combined in pairs, and the sample data sets composed of the sample data of the original channels combined in pairs are used to construct a total of 91 base models, thereby constructing 105 base models. In the embodiments of this specification, it is also possible to use 14 original channels, and every 3 original channels are combined, or every 4 original channels are combined, so as to build a prediction model. This specification does not limit the number of original channel combinations used to build a prediction model. Thus, the target channel can quickly select a candidate model for risk prediction on the target channel.

[0110] In some embodiments of the present specification, after constructing the first prediction model or the second prediction model using the similar original channel of the target channel, the first prediction model or the second prediction model can be applied to the new channel, so that the newly added target channel in the new channel can select the similar original channel with a higher similarity to the newly added target channel based on the similarity, and then obtain the corresponding prediction model from the original channel training base model. Thus, the trained prediction model can be used to predict the users of the newly added target channel, and the newly added target channel can have a corresponding adapted prediction model. For example, there are 5 target channels in total, and the target channel o selects the original channel s training base model to construct the corresponding prediction model; the target channel z selects the original channels x and y combined training base model to construct the corresponding prediction model; the target channel d selects the original channel i training base model to construct the corresponding prediction model; the target channel u selects the original channels s and w combined training base model to construct the corresponding prediction model; the target channel c selects the original channels a and j combined training base model to construct the corresponding prediction model.

[0111] The prediction effect of the prediction model selected based on this solution is better than the prediction effect of the prediction model obtained by training the target channel itself. For example, the prediction models of the five target channels O, Z, D, U, and C obtained by training their own channels have aucs of 0.648, 0.688, 0.668, 0.880, and 0.790, respectively, while the prediction effect aucs of the prediction model selected based on this solution are 0.695, 0.702, 0.757, 0.773, and 0.774, respectively. It can be seen that three of the five channels can improve the prediction effect of the model by adopting this solution, and the highest auc improvement is from 0.668 to 0.757, an increase of 13%. In this manual, prediction models corresponding to similar original channels are constructed according to different numbers of similar original channels. The constructed prediction models are numerous and in many forms, and for the target channels, there are more prediction models that can be selected and adapted. Using similar original channels to construct prediction models has better prediction effects than constructing prediction models by the target channels themselves.

[0112] Figure 5 The flowchart of a method for constructing a channel classification model according to an embodiment of the present specification is shown, which specifically includes the following steps:

[0113] Step 501, obtaining a sample data set formed by combining sample data of known channels and a channel label corresponding to each sample data, wherein the known channels include a target channel, or a combination of a target channel and at least one original channel.

[0114] In the embodiments of the present specification, a known channel is a channel with known sample data and channel labels, which can be one of a target channel and an original channel or a combination thereof. Specifically, when there is only one target channel, in order to achieve multi-classification of the sample data of the target channel, the amount of sample data of this target channel alone is relatively small, for example, dozens or hundreds. Compared with the sample data of no less than 20,000 pieces of the original channel, the amount of sample data of the target channel is small and the model cannot be well constructed. Therefore, this target channel needs to use the sample data of other original channels to jointly construct a channel classification model. When there are multiple target channels, and the amount of sample data corresponding to each of the multiple target channels is sufficient (for example, the amount of sample data is not less than 20,000 pieces), the sample data of multiple target channels can be used to jointly train the channel classification model.

[0115] Whether it is a combination of one target channel and multiple original channels, or a combination of multiple target channels, each channel has corresponding sample data, and each sample data has a corresponding channel label. The sample data of all channels together constitute the sample data set.

[0116] Step 502: Use the sample data set to train an initial multi-classifier to obtain the channel classification model, which outputs a prediction result of the category attribution of each sample data. In the embodiment of this specification, the process of training the initial multi-classifier is the process of building a multi-classification model.

[0117] In the embodiments of the present specification, an initial multi-classifier is selected as the basic model for constructing a channel classification model, and the process of training the initial multi-classifier is the process of constructing a channel classification model. The number of categories of the category attribution prediction results output by the initial multi-classifier is equal to the number of target channels. The target channels are the channels for the risks to be predicted, and the classification results of the multi-classification model are determined based on the number of target channels. For example, if there are 5 target channels, the channel classification model to be constructed is a five-classification model. Thus, the channel classification model can predict the input sample data and obtain the category attribution prediction results of each sample data.

[0118] In this specification, the known channels are determined according to the number of target channels. When the sample data of the target channels is small, a large amount of sample data of the original channels is used to jointly train and construct the channel classification model. Compared with constructing a channel classification model using only sample channels of the target channels, the channel classification model constructed in this application has a better prediction effect.

[0119] Figure 6 The flowchart of a method for classifying original channels in an embodiment of the present specification is shown, which specifically includes the following steps:

[0120] Step 601 , determining the classification prediction accuracy of each sample data in each original channel according to the category attribution prediction result of each sample data in each original channel and the sample channel label of the original channel.

[0121] In this step, after the channel classification model is constructed, the sample data of each original channel is input into the channel classification model for prediction, and the category attribution prediction result output by the channel classification model is obtained, that is, the channel classification model predicts which channel each sample data of each original channel belongs to. Furthermore, each sample data also has a corresponding channel label, that is, the real channel to which each sample data belongs. Therefore, according to the category attribution prediction result predicted by the channel classification model and the channel label of the sample data, the classification prediction accuracy of each sample data is calculated.

[0122] For example, the channel classification model is a five-category model, the target channels used to construct the channel classification model are channel o and channel d, and the original channels used to construct the channel classification model are channel a, channel e and channel f. Therefore, after predicting each input data (sample data of the original channel), the constructed channel classification model predicts which of the input data the input data should belong to: channel o, channel d, channel a, channel e or channel f. For example, the channel classification model predicts that the category attribution prediction result of the sample data of a certain original channel is channel o, then the category attribution prediction result is compared with the channel label of the sample data to determine the accuracy of the classification prediction of the sample data. Specifically, the accuracy of the classification prediction is: the ratio of the number of samples correctly predicted by the channel classification model for the target channel to the total number of samples of the target channel.

[0123] Step 602: Determine the prediction accuracy of the multiple classifiers for each original channel according to the classification prediction accuracy of each sample data in each original channel.

[0124] In the embodiment of this specification, the number of sample data of an original channel may be more than 20,000 or even more. According to the classification prediction accuracy of each sample data in an original channel, the prediction accuracy of the multi-classifier for the original channel as a whole is determined. Among them, according to the prediction results of different categories of different sample data of an original channel, the distribution of different users of the original channel can be determined.

[0125] Step 603, according to the prediction accuracy and the preset accuracy threshold, the original channel is divided into a first original channel and a second original channel. In this specification, a channel whose prediction accuracy is higher than the preset accuracy threshold can be determined as the first original channel; a channel whose prediction accuracy is lower than or equal to the preset accuracy threshold can be determined as the second original channel.

[0126] In this specification, the original channels are distinguished according to the classification prediction accuracy of the sample data of the original channels to obtain different types of original channels. This is conducive to selecting some original channels from different types of original channels in the subsequent steps to calculate the similarity with the target channel and further construct a prediction model to improve the prediction effect of the prediction model on the target channel.

[0127] Figure 7 The flowchart of a method for determining similar original channels according to an embodiment of the present specification is shown, which specifically includes the following steps:

[0128] Step 701, if the first similarity between the first original channel and each type of target channel is not higher than the preset threshold, the second similarity between the second original channel and each type of target channel is determined according to the category attribution prediction result of each sample data of the second original channel.

[0129] based on Figure 6 Description: After distinguishing the first original channel from the second original channel, this specification can give priority to using the channel classification model to predict the first original channel and determine the first similarity between the first original channel and the target channel. However, if the first similarity between the first original channel and each type of target channel is lower than the preset threshold, it means that the target channel cannot find a similar original channel from the first original channel. It is impossible to further use the prediction model constructed by the similar original channel to predict the sample data of the target channel.

[0130] Therefore, the similarity between the second original channel and the target channel is calculated in the above manner, and the similarity between the second original channel and the target channel is called the second similarity.

[0131] Step 702, selecting a second original channel whose second similarity is greater than a preset threshold as a similar original channel similar to each type of target channel. In this step, a second original channel whose second similarity is greater than a preset threshold is selected, and it is determined that the second original channel has a high similarity to the target channel. Such a second original channel is a similar original channel similar to the target channel.

[0132] In the embodiments of this specification, institutions that operate different businesses or have different business attributes can become channels described in this specification. For example, there are 10 existing institutions (also understood as 10 channels), of which 3 institutions have mortgage loan business as their main business, 3 institutions have credit card business as their main business, 3 institutions have consumer loan business as their main business, and another institution has cash loan business as its main business. In this specification, sample data of 10 channels and channel labels corresponding to the sample data can be obtained to construct a channel classification model. Further, several channels (for example, 6 channels are selected) can be determined from the 10 channels as original channels, and sample data of the original channels are obtained respectively. The sample data of the original channels are predicted using the channel classification model to obtain the category attribution prediction results of each sample data belonging to each known channel. When a new channel or a target channel needs to be predicted, the similarity between each original channel and the target channel or the new channel is further determined; the original channel similar to the target channel is selected according to the similarity.

[0133] A prediction model composed of original channels similar to the target channel is further determined, and the prediction model is used to predict the data to be predicted of the target channel to obtain the prediction result corresponding to the sample data of the target channel.

[0134] The embodiments of this specification can adapt to identify the user classification of new channels and allocate corresponding reasonable virtual resources to users of different classifications. For example, sample data 1 of new channel 1 is the data of user A. According to the method provided in the above embodiment, when the category attribution prediction result of sample data 1 is that it belongs to channel a, virtual resources corresponding to channel a are allocated to user A. For another example, sample data 2 of new channel 2 is the data of user B. According to the method provided in the above embodiment, when the category attribution prediction result of sample data 2 is that it belongs to channel b, virtual resources corresponding to channel b are allocated to user B.

[0135] like Figure 8 The figure shows a schematic diagram of the structure of a target channel prediction device based on multiple channels according to an embodiment of the present specification. The figure describes the basic structure of the target channel prediction device based on multiple channels, wherein the functional units and modules can be implemented in software, or a general chip or a specific chip can be used to implement the target channel prediction based on multiple channels. The device specifically includes:

[0136] The category attribution prediction result determination unit 801 is used to input sample data of multiple original channels into the channel classification model, and obtain the category attribution prediction results of each sample data in each original channel output by the channel classification model, wherein the channel classification model is trained by the known channel sample data and the corresponding channel labels, and the known channel sample data at least includes the target channel sample data;

[0137] A similarity determination unit 802 is used to determine the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel;

[0138] A similar original channel determining unit 803 is used to select at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel;

[0139] The prediction result determination unit 804 is used to input the to-be-predicted data of the target channel into a prediction model pre-built based on the similar original channel to obtain a prediction result corresponding to the sample data of the target channel.

[0140] The embodiment of the specification also includes a similarity determination device, the device including: determining the number of category attribution prediction results of sample data of each original channel belonging to each category of known channels;

[0141] A similar sample quantity matrix construction unit is used to construct a similar sample quantity matrix according to the known channels, the original channels and the number of category attribution prediction results of the sample data of each original channel belonging to each category of known channels;

[0142] The similarity determination unit is used to determine the similarity between each original channel and the target channel according to the similar sample quantity matrix.

[0143] The embodiment of this specification also includes another similarity determination device, the device comprising:

[0144] A similar sample ratio determination unit is used to determine the similar sample ratio of each original channel's sample data belonging to each type of target channel according to the number of category attribution prediction results of each type of known channel and the total amount of sample data of each original channel in the similar sample quantity matrix;

[0145] A similar sample ratio mean calculation unit is used to calculate the mean ratio of similar samples corresponding to all original channels of the same category attribution prediction result according to the ratio of similar samples belonging to each category of target channels of each original channel;

[0146] An approximate multiple determination unit, used to obtain an approximate multiple of each original channel belonging to the same type of target channel according to a ratio of the similar sample proportion to an average of the similar sample proportions;

[0147] The similarity determination unit is used to use the value of the approximate multiple as the similarity between the original channel and the target channel.

[0148] The embodiment of this specification also includes a prediction model construction device, the device comprising:

[0149] A first prediction model building unit is used to, if the number of the similar original channels is one, use the sample data of the similar original channels and the channel labels corresponding to the sample data to train a channel classification model, and to build a first prediction model corresponding to the similar original channels;

[0150] The embodiment of this specification includes a second prediction model framework device, including:

[0151] A first number of second prediction model construction units, used to respectively use sample data of each similar original channel and the channel label of the sample data to train an initial classification model to obtain a first number of second prediction models;

[0152] The second number of second prediction model construction units are used to arbitrarily select a preset number of original channels from N similar original channels, use the sample data of the arbitrarily selected preset number of original channels and the channel labels of the sample data to train the initial classification model to obtain the second number of second prediction models, and the first number of second prediction models and the second number of second prediction models constitute the second prediction model.

[0153] The embodiment of this specification includes a channel classification model training device, including:

[0154] An acquisition unit, configured to acquire a sample data set formed by combining sample data of known channels and a channel label corresponding to each sample data, wherein the known channels include a target channel or a combination of a target channel and at least one original channel;

[0155] The category attribution prediction result determination unit is used to train an initial multi-classifier using the sample data set to obtain a category attribution prediction result for each sample data output by the initial multi-classifier.

[0156] The embodiment of this specification provides an original channel division device, the device comprising:

[0157] A sample data classification prediction accuracy determination unit, used to determine the classification prediction accuracy of each sample data in each original channel according to the category attribution prediction result of each sample data in each original channel and the sample channel label of the original channel;

[0158] An original channel prediction accuracy determination unit, used to determine the prediction accuracy of the multiple classifiers for each original channel according to the classification prediction accuracy of each sample data in each original channel;

[0159] The division unit is used to divide the original channel into a first original channel and a second original channel according to the prediction accuracy and a preset accuracy threshold.

[0160] The embodiment of the present specification also includes a similar original drive determination device, including:

[0161] A second similarity determination unit is used to determine a second similarity between the second original channel and each type of target channel according to a category attribution prediction result of each sample data of the second original channel if the first similarity between the first original channel and each type of target channel is not higher than a preset threshold;

[0162] The similar original channel determination unit is used to select a second original channel whose second similarity is greater than a preset threshold as a similar original channel similar to each type of target channel.

[0163] like Fig.9A Shown is a schematic diagram of a similar sample quantity matrix according to an embodiment of this specification.

[0164] The first row in the similar sample quantity matrix is ​​the known channel, and the first column in the similar sample quantity matrix is ​​the original channel. The sample data of each original channel belongs to the number of each known channel, which is each element in the similar sample quantity matrix. Fig.9A is an i*j "similar sample quantity" matrix, i represents the number of original channels, and j represents the number of target channels. For each target channel, the samples of the i original channels can be predicted by the channel classification model respectively, and the samples of the i original channels are predicted as the number of samples of each category of the channel classification model, which is recorded as the "similar sample quantity" matrix.

[0165] Fig. 9B The figure shows a schematic diagram of a similar sample ratio matrix of an embodiment of the present specification. Specifically, according to the number of category attribution prediction results of sample data of each original channel belonging to each category of known channels and the total amount of sample data of each original channel in the similar sample quantity matrix, the similar sample ratio of sample data of each original channel belonging to each category of target channels is determined, and the similar sample ratio of sample data of each original channel belonging to each category of target channels is obtained. Fig. 9B Matrix showing the proportion of similar samples.

[0166] Fig. 9C The figure shows a schematic diagram of a ratio mean matrix of an embodiment of this specification. According to the proportion of similar samples of each original channel belonging to each type of target channel, the average proportion of similar samples corresponding to the same category attribution prediction result of all original channels is calculated, and the result is as follows: Fig. 9C The matrix of proportional means shown.

[0167] Fig.9D The figure shows a schematic diagram of an approximate multiple matrix of an embodiment of the present specification. According to the ratio of the similar sample proportion to the average of the similar sample proportion, the approximate multiple of each original channel belonging to the same type of target channel is obtained by subtracting 1. Fig.9D The approximate multiples matrix shown.

[0168] Fig.9E The diagram is a schematic diagram of a similarity matrix according to an embodiment of the present specification. Fig.9D The approximate multiple matrix of the original channel is obtained by determining the value of the approximate multiple, which is the similarity between the original channel and the target channel. Fig.9E The similarity matrix shown. Fig.9E It is the i*j “similarity index” matrix, where i is 14, i represents the number of original channels, and j is 5, j represents the number of target channels.

[0169] In one embodiment of the present specification, it is necessary to predict the prediction results corresponding to the to-be-predicted data of the target channel. The target channel includes Fig.9A Target channel o, target channel z, target channel d, target channel u and target channel c in. First, determine the sample data of 14 known original channels, namely original channel a, original channel e, original channel f, original channel h, original channel i, original channel j, original channel p, original channel q, original channel r, original channel s, original channel t, original channel w, original channel x, original channel y. Each original channel has tens of thousands of sample data.

[0170] The channel classification model predicts the category attribution prediction results of a large number of sample data from each original channel. For example, the total amount of sample data from original channel a is 38,852, the number of samples of original channel a predicted to belong to target channel o is 2,111, the number of samples of original channel a predicted to belong to target channel z is 5,479, the number of samples of original channel a predicted to belong to target channel d is 652, the number of samples of original channel a predicted to belong to target channel u is 1,429, and the number of samples of original channel a predicted to belong to target channel c is 29,181. According to the above method, the category attribution prediction results corresponding to the sample data of each original channel are determined, and then the following is constructed. Fig. 9B As shown in the similar sample quantity matrix, for example, it is determined that the similar samples of original channel a belonging to target channel o account for 5.43%, and the similar samples of original channel a belonging to target channel c account for 75.1%. Fig. 9B shown.

[0171] according to Fig. 9B For each element in the similar sample ratio matrix, divide the value of the similar sample ratio of each element by Fig. 9C The mean of the proportion of similar samples in , minus 1, is used to construct an approximate multiple matrix, such as Fig.9DAs shown in the figure, the proportion of similar samples of the original channel a belonging to the target channel o is 5.43%, divided by the average proportion of similar samples corresponding to the target channel o is 12.56%, and then minus 1, we get -0.57, which is the approximate multiple value of the original channel belonging to the target channel o.

[0172] According to the similar sample proportions and approximate multiples of the 14 original channels, select the original channels that are similar to the target channels. Fig.9E As shown, the similarity between each original channel and each target channel can be determined. For example, the target channel o selects the original channel s as a similar original channel training base model to obtain the corresponding prediction model; the target channel z selects the original channels x and y as similar channel training base models to obtain the corresponding prediction model. In other words, the original channel s is similar to the target channel o, so the original channel s training base model can be used to construct a prediction model for predicting the sample data of the target channel; similarly, the original channels x and y are combined to train the base model to construct a prediction model, which is used to predict the data to be predicted of the target channel z, that is, to predict the user-related data in the target channel.

[0173] like Fig.10 As shown, it is a schematic diagram of a computer device provided by an embodiment of this specification. The target channel prediction method based on multiple channels described in this application can be applied to the computer device. The computer device 1002 may include one or more processors 1004, such as one or more central processing units (CPUs), and each processing unit may implement one or more hardware threads. The computer device 1002 may also include any memory 1006, which is used to store any kind of information such as code, settings, data, etc. Non-limiting, for example, the memory 1006 may include any one or more combinations of the following: any type of RAM, any type of ROM, flash memory device, hard disk, optical disk, etc. More generally, any memory can use any technology to store information. Further, any memory can provide volatile or non-volatile retention of information. Further, any memory can represent a fixed or removable component of the computer device 1002. In one case, when the processor 1004 executes an associated instruction stored in any memory or a combination of memories, the computer device 1002 can perform any operation of the associated instruction. The computer device 1002 also includes one or more drive mechanisms 1008 for interacting with any storage, such as a hard disk drive mechanism, an optical disk drive mechanism, etc.

[0174] The computer device 1002 may also include an input / output module 1010 (I / O) for receiving various inputs (via input devices 1012) and for providing various outputs (via output devices 1014). A specific output mechanism may include a presentation device 1016 and an associated graphical user interface (GUI) 1018. In other embodiments, the input / output module 1010 (I / O), the input device 1012, and the output device 1014 may not be included, and the computer device 1002 may be used as a computer device in a network. The computer device 1002 may also include one or more network interfaces 1020 for exchanging data with other devices via one or more communication links 1022. One or more communication buses 1024 couple the components described above together.

[0175] The communication link 1022 may be implemented in any manner, for example, through a local area network, a wide area network (e.g., the Internet), a point-to-point connection, etc., or any combination thereof. The communication link 1022 may include any combination of hardwired links, wireless links, routers, gateway functions, name servers, etc. governed by any protocol or combination of protocols.

[0176] Corresponds to Figures 1 to 7 The method in the embodiment of the present specification also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above method are executed.

[0177] The embodiment of the present specification also provides a computer-readable instruction, wherein when the processor executes the instruction, the program therein causes the processor to execute the following Figures 1 to 7 The method shown.

[0178] It should be understood that in the various embodiments of this specification, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this specification.

[0179] It should also be understood that in the embodiments of this specification, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this specification generally indicates that the associated objects before and after are in an "or" relationship.

[0180] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed in this specification can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this specification.

[0181] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0182] In the several embodiments provided in this specification, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0183] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the embodiments of this specification.

[0184] In addition, each functional unit in each embodiment of this specification may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0185] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this specification is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of this specification. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0186] Specific embodiments are used in this specification to illustrate the principles and implementation methods of this specification. The description of the above embodiments is only used to help understand the methods and core ideas of this specification. At the same time, for those skilled in the art, according to the ideas of this specification, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on this specification.

Claims

1. A target channel prediction method based on multiple channels, characterized in that: The method comprises: Inputting sample data of multiple original channels into a channel classification model, obtaining a prediction result of the category attribution of each sample data in each original channel output by the channel classification model to each known channel, wherein the channel classification model is trained by known channel sample data and corresponding channel labels, and the known channel sample data at least includes target channel sample data; Determine the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel; Selecting at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel; The data to be predicted of the target channel is input into a prediction model pre-built based on the similar original channel to obtain a prediction result corresponding to the sample data of the target channel.

2. The method according to claim 1, characterized in that According to the category attribution prediction result of each sample data of each original channel, determining the similarity between each original channel and the target channel includes: Determine the number of category attribution prediction results of sample data of each original channel belonging to each category of known channels; Constructing a similar sample quantity matrix according to the known channels, the original channels and the number of category attribution prediction results of the sample data of each original channel belonging to each category of the known channels; The similarity between each original channel and the target channel is determined according to the similar sample quantity matrix.

3. The method according to claim 2, characterized in that Determining the similarity between each original channel and the target channel according to the similar sample quantity matrix includes: According to the number of prediction results of the category attribution of the sample data of each original channel belonging to each category of known channels and the total amount of sample data of each original channel in the similar sample quantity matrix, determine the proportion of similar samples of the sample data of each original channel belonging to each category of target channels; According to the proportion of similar samples belonging to each target channel in each original channel, calculate the average proportion of similar samples corresponding to the same category attribution prediction result for all original channels; According to the ratio of the similar sample proportion to the average of the similar sample proportions, the approximate multiple of each original channel belonging to the same type of target channel is obtained; The value of the approximate multiple is used as the similarity between the original channel and the target channel.

4. The method according to claim 1, characterized in that: The prediction model is constructed by: If the number of the similar original channels is one, using the sample data of the similar original channels and the channel labels corresponding to the sample data, an initial classification model is trained to construct a first prediction model corresponding to the similar original channels; If there are N similar original channels, at least one original channel is selected according to a preset mode, and an initial classification model is trained using multiple channel labels of sample data of the original channel to construct a second prediction model corresponding to the multiple original channels, where N is a positive integer greater than 1.

5. The method according to claim 4, characterized in that Selecting at least one original channel according to a preset mode, using multiple channel labels of sample data of the original channel to train an initial classification model, and constructing a second prediction model corresponding to the multiple original channels includes: Using sample data of each similar original channel and the channel label of the sample data respectively, an initial classification model is trained to obtain a first number of second prediction models; A preset number of original channels are arbitrarily selected from N similar original channels, and the sample data and channel labels of the arbitrarily selected preset number of original channels are used to train the initial classification model to obtain a second number of second prediction models, and the first number of second prediction models and the second number of second prediction models constitute the second prediction model.

6. The method according to claim 5, characterized in that The channel classification model is trained in the following way: Acquire a sample data set formed by combining sample data of known channels and a channel label corresponding to each sample data, wherein the known channels include a target channel, or a combination of the target channel and at least one original channel; The sample data set is used to train an initial multi-classifier to obtain the channel classification model, and the channel classification model outputs a category attribution prediction result for each sample data.

7. The method according to claim 6, characterized in that If the known channels include a target channel and at least one original channel, the method further includes: Determine the classification prediction accuracy of each sample data in each original channel according to the category attribution prediction result of each sample data in each original channel and the sample channel label of the original channel; Determining the prediction accuracy of the multiple classifiers for each original channel according to the classification prediction accuracy of each sample data in each original channel; According to the prediction accuracy and the preset accuracy threshold, the original channel is divided into a first original channel and a second original channel.

8. The method according to claim 7, characterized in that The method further comprises: If the first similarity between the first original channel and each type of target channel is not higher than the preset threshold, then determining the second similarity between the second original channel and each type of target channel according to the category attribution prediction result of each sample data of the second original channel; A second original channel whose second similarity is greater than a preset threshold is selected as a similar original channel similar to each type of target channel.

9. A target channel prediction device based on multiple channels, characterized in that: The device comprises: A category attribution prediction result determination unit is used to input sample data of multiple original channels into a channel classification model to obtain a category attribution prediction result output by the channel classification model that each sample data in each original channel belongs to each known channel, wherein the channel classification model is trained by known channel sample data and corresponding channel labels, and the known channel sample data at least includes target channel sample data; A similarity determination unit, configured to determine the similarity between each original channel and the target channel according to the category attribution prediction result of each sample data of each original channel; a similar original channel determination unit, configured to select at least one original channel whose similarity is greater than a preset similarity threshold as an original channel similar to the at least one target channel; The prediction result determination unit is used to input the to-be-predicted data of the target channel into a prediction model pre-built based on the similar original channel to obtain the prediction result corresponding to the sample data of the target channel.

10. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images