Information pushing method and device, electronic equipment and computer readable medium

Abstract

Embodiments of the present disclosure disclose information pushing methods and apparatuses, electronic devices, and computer readable media. A specific implementation of the method includes: obtaining a target user information set; determining target user feature information of each target user information in the target user information set to obtain a target user feature information set; generating a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; generating a second prediction result set based on the first prediction result set, a user hierarchical information set, and a fusion model, wherein the prediction result is used to represent the value degree of a user corresponding to the target user information; and pushing recommendation information to a target terminal corresponding to the target user information in the target user information set according to the second prediction result set. The implementation improves the matching degree of the information pushed to the user and the user, reduces the transmission of unnecessary push information, and reduces the waste of network transmission resources.

Description

TECHNICAL FIELD

[0001] Embodiments of the present disclosure relate to the technical field of computer technology, and in particular, to an information pushing method and device, an electronic device, and a computer readable medium. BACKGROUND

[0002] Customer Lifetime Value (CLV) is an index used to measure the value of a user in a period of time. The value of a user can be measured by setting the consumption level or resource consumption of the user in a period of time. By determining the CLV of a user, the user can be pushed with push information matched with the user. At present, when predicting the CLV of a user, the commonly used method is to use a traditional and single prediction model to predict the CLV.

[0003] However, when the above method is used, the following technical problems often exist:

[0004] The prediction ability of the traditional prediction model for different data sets is not the same, thereby making it difficult to accurately predict the CLV of a user, and further making it impossible to push information matched with the user to the user, and finally resulting in a waste of a large amount of network transmission resources. SUMMARY

[0005] The summary of the present disclosure is used to introduce the concepts in a brief form, which will be described in detail in the specific embodiments part. The summary of the present disclosure is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to be used to limit the scope of the claimed technical solutions.

[0006] Some embodiments of the present disclosure propose an information pushing method, device, electronic device, and computer readable medium to solve one or more of the technical problems mentioned in the background section.

[0007] In a first aspect, some embodiments of the present disclosure provide an information pushing method, which comprises: obtaining a target user information set; determining target user feature information of each target user information in the target user information set to obtain a target user feature information set; generating a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; generating a second prediction result set based on the first prediction result set, a user stratification information set, and a fusion model, wherein the prediction result is used to represent the value of a user corresponding to the target user information; and pushing recommended information to a target terminal corresponding to the target user information in the target user information set according to the second prediction result set.

[0008] Optionally, the pushing of the recommendation information to the target terminal corresponding to each candidate user information in the candidate user information set comprises: in response to determining that the target terminal is in an information notification mode, pushing the recommendation information to the target terminal.

[0009] Optionally, the pushing of the recommendation information to the target terminal corresponding to each candidate user information in the candidate user information set comprises: in response to determining that the target terminal is in an information notification mode, pushing the recommendation information to the target terminal.

[0010] Optionally, the fusion model is trained by the following steps: constructing a training sample data set and a full user data set, wherein the training sample data set comprises a sub-prediction model training sample set and a fusion model training sample set; training a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models; performing hierarchical processing on the full user data set to generate a user hierarchical information set; performing fusion model training on an initial fusion model based on the fusion model training sample set, the user hierarchical information set, and the plurality of trained sub-prediction models to generate a fusion model.

[0011] Optionally, the construction of the training sample data set comprises: obtaining a user behavior information set; performing positive sampling processing and negative sampling processing on the user behavior information set respectively to generate a positive sampling sample set and a negative sampling sample set, wherein the sample quantity in the positive sampling sample set and the negative sampling sample set is in a preset ratio; based on each sample in the positive sampling sample set and the negative sampling sample set, constructing user feature information to generate a training sample data, thereby obtaining the training sample data set.

[0012] Optionally, the training of the plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models comprises: determining the feature information of each sub-prediction model training sample in the sub-prediction model training sample set to obtain at least one feature information set; for the feature information set in the at least one feature information set, training the initial sub-prediction model corresponding to the feature information set in the plurality of initial sub-prediction models to generate a trained sub-prediction model.

[0013] Optionally, the full-amount user data set is hierarchically processed to generate a user hierarchical information set, including: performing clustering processing on the full-amount user data set according to a preset target category number to generate a clustering information group set; and performing clustering processing on the clustering information in the clustering information group set according to the position and category coverage of the category center of each clustering information group in the clustering information group set to generate the user hierarchical information set.

[0014] Optionally, the initial fusion model is trained based on the fusion model training sample set, the user hierarchical information set, and the plurality of trained sub-prediction models to generate a fusion model, including: inputting the fusion model training sample set into each trained sub-prediction model in the plurality of trained sub-prediction models to generate a prediction result, thereby obtaining a prediction result set; and training the initial fusion model based on the prediction result set and the user hierarchical information set to generate the fusion model.

[0015] Optionally, the sub-prediction model training sample set and the fusion model training sample set are generated by: performing random sampling on the training sample data set to generate the sub-prediction model training sample set and the fusion model training sample set.

[0016] In a second aspect, some embodiments of the present disclosure provide an information pushing device, the device comprising: an acquisition unit configured to acquire a target user information set; a determination unit configured to determine target user feature information of each target user information in the target user information set to obtain a target user feature information set; a first generation unit configured to generate a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; a second generation unit configured to generate a second prediction result set based on the first prediction result set, a user hierarchical information set, and a fusion model, wherein the prediction result is used to represent the user value degree of the target user information; and a pushing unit configured to push recommendation information to a target terminal corresponding to the target user information in the target user information set according to the second prediction result set.

[0017] Optionally, the pushing unit is further configured to: filter target user information satisfying a filtering condition from the target user information set as candidate user information according to the second prediction result set to obtain a candidate user information set; and push recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set, wherein the target terminal is a terminal logged in with a user account of a user corresponding to the candidate user information.

[0018] Optionally, the pushing unit is further configured to push the recommendation information to the target terminal in response to determining that the target terminal is in the open information notification mode.

[0019] Optionally, the fusion model is trained by the following steps: constructing a training sample data set and a full-amount user data set, wherein the training sample data set comprises a sub-prediction model training sample set and a fusion model training sample set; training a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models; performing hierarchical processing on the full-amount user data set to generate a user hierarchical information set; and performing fusion model training on an initial fusion model based on the fusion model training sample set, the user hierarchical information set, and the plurality of trained sub-prediction models to generate a fusion model.

[0020] Optionally, the training sample data set is constructed by: obtaining a user behavior information set; performing positive sampling processing and negative sampling processing on the user behavior information set respectively to generate a positive sampling sample set and a negative sampling sample set, wherein the sample amount in the positive sampling sample set and the negative sampling sample set is in a preset ratio; and constructing user feature information based on each sample in the positive sampling sample set and the negative sampling sample set to generate a training sample data, thereby obtaining the training sample data set.

[0021] Optionally, the plurality of initial sub-prediction models are trained based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models, comprising: determining the feature information of each sub-prediction model training sample in the sub-prediction model training sample set to obtain at least one feature information set; and training the initial sub-prediction model corresponding to the feature information set in the plurality of initial sub-prediction models based on the feature information set in the at least one feature information set to generate a trained sub-prediction model.

[0022] Optionally, the full-amount user data set is hierarchically processed to generate a user hierarchical information set, comprising: performing clustering processing on the full-amount user data set according to a preset target category number to generate a clustering information group set; and performing clustering processing on the clustering information in the clustering information group set according to the position and category coverage range of the class center of each clustering information group in the clustering information group set to generate the user hierarchical information set.

[0023] Optionally, the fusion model training of the initial fusion model based on the fusion model training sample set, the user stratification information set and the plurality of trained sub-prediction models is performed to generate the fusion model, including: inputting the fusion model training sample set into each of the plurality of trained sub-prediction models to generate a prediction result, thereby obtaining a prediction result set; and performing the fusion model training of the initial fusion model based on the prediction result set and the user stratification information set to generate the fusion model.

[0024] Optionally, the sub-prediction model training sample set and the fusion model training sample set are generated by: performing random sampling on the training sample data set to generate the sub-prediction model training sample set and the fusion model training sample set.

[0025] In a third aspect, some embodiments of the present disclosure provide an electronic device, including: one or more processors; a storage device having one or more programs stored thereon, when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any of the implementation manners of the first aspect.

[0026] In a fourth aspect, some embodiments of the present disclosure provide a computer readable medium having a computer program stored thereon, wherein the program is executed by a processor to implement the method described in any of the implementation manners of the first aspect.

[0027] The above various embodiments of the present disclosure have the following beneficial effects: through the information pushing method of some embodiments of the present disclosure, the matching degree of the information pushed by the user and the user is improved. At the same time, the waste of network transmission resources is reduced. Specifically, the reason for the waste of network transmission resources is that the prediction ability of the traditional prediction model for different data sets is different, thereby making it difficult to accurately predict the CLV of the user. Based on this, the information pushing method of some embodiments of the present disclosure first acquires a target user information set. Second, the target user feature information of each target user information in the target user information set is determined to obtain a target user feature information set. In actual situations, the target user information often contains a lot of information corresponding to the target user, such as hobbies, gender, body mass index, height, weight, and income. Most of the feature information in such information has little effect on predicting the CLV of the user. Moreover, too much feature information may increase the training time of the prediction model. Moreover, it may not improve the prediction accuracy of the obtained prediction model. Second, based on the target user feature information set and a plurality of trained sub-prediction models, a first prediction result set is generated. In actual situations, a single, traditional prediction model has different prediction abilities for different data sets, so it is necessary to predict according to the same data set through a plurality of different prediction models, and the prediction ability of the plurality of sub-prediction models is determined according to the obtained first prediction result set. In addition, based on the first prediction result set, a user stratification information set, and a fusion model, a second prediction result set is generated. In actual situations, through the fusion model, the accuracy of the final prediction result can be improved, that is, the accuracy of the CLV prediction of the user is improved. In addition, according to the long tail effect theory and the 80-20 law, users often appear stratification, that is, users at different levels often have different behavior habits. Users at the same level often have the same behavior habits. Since stratification is an important feature, it will affect user classification and the accuracy of CLV. Therefore, through the user stratification information set, the accuracy of the second prediction result generated by the fusion model can be further improved. Finally, according to the second prediction result, recommended information is pushed to the target terminal corresponding to the target user information in the target user information set. In this way, the matching degree of the information pushed to the user and the user can be greatly improved. Thus, unnecessary transmission of information is reduced, and thus the waste of network transmission resources is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0028] The above and other features, advantages, and aspects of embodiments of the present disclosure will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings in which:

[0029] Figure 1 is a schematic diagram of one application scenario of the information pushing method of some embodiments of the present disclosure;

[0030] Figure 2 is a flow chart of some embodiments of the information pushing method according to the present disclosure;

[0031] Figure 3 is a schematic diagram of an authorization prompt box;

[0032] Figure 4 is a flow chart of another embodiments of the information pushing method according to the present disclosure;

[0033] Figure 5 is a schematic diagram of network structures of the first sub-prediction model and the second sub-prediction model;

[0034] Figure 6 is a schematic diagram of the result of the clustering processing;

[0035] Figure 7 is another schematic diagram of the result of the clustering processing;

[0036] Figure 8 is a structural schematic diagram of some embodiments of the information pushing apparatus according to the present disclosure;

[0037] Figure 9 is a structural schematic diagram of an electronic device suitable for being used to implement some embodiments of the present disclosure. DETAILED DESCRIPTION

[0038] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided so as to make the present disclosure more thorough and complete. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0039] It should also be noted that, for the sake of description, only the parts related to the present application are shown in the drawings. The embodiments in the present disclosure and the features in the embodiments can be combined with each other without conflict.

[0040] It should be noted that the concepts of "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0041] It should be noted that the modification of "one", "multiple" mentioned in the present disclosure is illustrative but not restrictive, and those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0042] The names of the messages or information exchanged between the plurality of devices in the embodiments of the present disclosure are only for illustrative purposes, and are not used to limit the scope of the messages or information.

[0043] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0044] Figure 1 is a schematic diagram of one application scenario of the information pushing method of some embodiments of the present disclosure.

[0045] In Figure 1In the application scenario, first, the computing device 101 can obtain a target user information set 102 (for example, the target user information can be [user number: 1004, user gender: male, user name: Zhu XX, user age: 34 years old, user height: 152 cm, user weight: 200 kg, user body mass index: 43.3, user home address: XX City, XX Street, XX Community, XX Building, XX Layer, XX, user occupation: sound engineer, user education: master, user income: 20,000 yuan / month, purchase times: 20 times, browse times: 32 times, purchase amount: 5,000 yuan, click times: 54 times, collection times: 0 times, add to shopping cart times: 20 times, search times: 23 times]); second, the computing device 101 can determine the target user feature information of each target user information in the above target user information set 102, and obtain a target user feature information set 103 (for example, the target user feature information can be [user number: 1004, user gender: male, user age: 34 years old, user weight: 200 kg, user occupation: sound engineer, user education: master, user income: 20,000 yuan / month, purchase times: 20 times, browse times: 32 times, purchase amount: 5,000 yuan, click times: 54 times, collection times: 0 times, add to shopping cart times: 20 times, search times: 23 times]); in addition, the computing device 101 can generate a first prediction result set 105 based on the above target user feature information set 103 and a plurality of trained sub-prediction models 104 (for example, the first prediction result can be [user number: 1004, user value: 0.4, confidence value: 0.98]); further, the computing device 101 can generate a second prediction result set 108 based on the above first prediction result set 105, a user stratification information set 106 and a fusion model 107 (for example, the second prediction result can be [user number: 1004, user value: 0.396, confidence value: 0.98]), wherein the prediction result is used to represent the user value degree corresponding to the target user information; finally, the computing device 101 can push recommendation information to the target terminal corresponding to the target user information in the above target user information set 102 according to the above second prediction result set 108.

[0046] It should be noted that the above computing device 101 can be hardware or software. When the computing device is hardware, it can be implemented as a distributed cluster composed of multiple servers or terminal devices, or as a single server or single terminal device. When the computing device is software, it can be installed in the above-mentioned hardware devices. It can be implemented as, for example, multiple software or software modules for providing distributed services, or as a single software or software module. No specific limitation is made here.

[0047] It should be understood that Figure 1The number of computing devices in FIG. 1 is merely illustrative. Any number of computing devices can be present according to implementation needs.

[0048] With continued reference to Figure 2 , a flow 200 of some embodiments of the information pushing method according to the present disclosure is shown. The information pushing method comprises the following steps:

[0049] In step 201, a target user information set is acquired.

[0050] In some embodiments, the execution subject of the information pushing method (e.g. Figure 1 the computing device 101 shown) can acquire, for each target user information in the target user information set described above, the target user information described above through wired connection or wireless connection in response to receiving a target authorization signal. The target user information in the target user information set described above can include identity information and behavior information of the target user. The behavior information described above can represent the value operation behavior of the target user in a period of time. The target authorization signal described above can be a signal generated by the target user corresponding to the target user information performing a target operation on a target control. The target control described above can be contained in an authorization prompt box. The authorization prompt box described above can be displayed on a target terminal device. The target terminal device described above can be a terminal device logged in an account corresponding to the user. For example, the terminal device can be a "mobile phone" or a "computer". For example, the target operation described above can be a "click operation" or a "slide operation". The target control described above can be a "confirmation button".

[0051] As an example, the authorization prompt box described above can be as shown in Figure 3 . The authorization prompt box described above can include a prompt information display part 301 and a control 302. The prompt information display part 301 described above can be used to display prompt information. The prompt information described above can be "whether to allow the acquisition of target user information". The control 302 described above can be a "confirmation button" or a "cancel button".

[0052] As another example, the value operation behavior described above can be a collection operation behavior. The value operation behavior described above can also be a purchase operation behavior.

[0053] As yet another example, the identity information of the target user described above can include user number, user gender, user name, user age, user height, user weight, user body mass index, user home address, user occupation, user education, and user income, etc. The behavior information of the target user described above can include purchase times, browsing times, purchase amount, click times, collection times, adding to shopping cart times, and search times, etc.

[0054] Step 202, determine the target user feature information of each target user information in the target user information set, and obtain a target user feature information set.

[0055] In some embodiments, the above execution subject can determine the target user feature information of each target user information in the target user information set, and obtain a target user feature information set. Wherein, the above execution subject can select feature information from the identity information and behavior information included in each target user information in the above target user information set according to the business scenario, to generate target user feature information, and obtain the above target user feature information set.

[0056] As an example, the above business scenario can be "prediction of the purchase amount of the target user in the future period of time". Therefore, the user gender, user age, user occupation, user education, user income, purchase times, browsing times, purchase amount, browsing times, click times, add to cart times and search times can be selected from the target user information to generate user feature information. For example, the user feature information can be: [user gender: male, user age: 24 years old, user occupation: front-end development engineer, user education: master, user income: 30,000 yuan / month, purchase times: 12 times, browsing times: 223 times, purchase amount: 3000 yuan, click times: 200 times, add to cart times: 15 times, search times: 30 times].

[0057] Step 203, based on the target user feature information set and the plurality of trained sub-prediction models, generate a first prediction result set.

[0058] In some embodiments, the above execution subject generates the above first prediction result set based on the above target user feature information set and the above plurality of trained sub-prediction models, which can include the following steps:

[0059] First, encode each target user feature information in the target user feature information set to generate encoded target user feature information, and obtain an encoded target user feature information set.

[0060] Wherein, the above encoding processing can be one-hot encoding processing.

[0061] Second, input the above encoded target user feature information set into each sub-prediction model in the above plurality of trained sub-prediction models to generate a first prediction result, and obtain the above first prediction result set.

[0062] The sub-prediction model can be, but is not limited to, any one of the following: a CNN (Convolutional Neural Networks) model, an RNN (Recurrent Neural Networks) model, an XGBoost model, and an RF (Random Forest) model. The first prediction result can represent a user value degree of the target user. The user value degree can be a CLV (Customer Lifecycle Value).

[0063] For example, the first prediction result can be [user ID: 1002, user value degree: 0.2, confidence value: 0.98]. The "0.2" can represent the value degree of the target user. The "0.98" can represent the confidence value.

[0064] At step 204, a second prediction result set is generated based on the first prediction result set, the user hierarchical information set, and a fusion model.

[0065] In some embodiments, the execution subject generates the second prediction result set based on the first prediction result set, the user hierarchical information set, and the fusion model, which can include the following steps:

[0066] First, each first prediction result in the first prediction result set and corresponding user hierarchical information are spliced to generate candidate information, thereby obtaining a candidate information set.

[0067] The user hierarchical information can represent the level of the target user. The user hierarchical information can include a user ID and a user level value.

[0068] For example, the user hierarchical information can be [user ID: 1002, user level value: 1]. The "1" can represent that the level of the target user with the user ID "1002" is the first level. The user level can also be "2" or "3". The "2" can represent that the level of the target user is the second level. The "3" can represent that the level of the target user is the third level.

[0069] As another example, the candidate information set can be [user ID: 1002, user value degree: 0.2, confidence value: 0.98, user level value: 1].

[0070] Second, each candidate information in the candidate information set is encoded to generate encoded candidate information, thereby obtaining an encoded candidate information set.

[0071] The encoding can be one-hot encoding.

[0072] In the third step, the encoded candidate information in the encoded candidate information set is input into the fusion model to generate a second prediction result, thereby obtaining the second prediction result set.

[0073] The second prediction result set includes second prediction results, and each second prediction result represents a user value degree of a target user. The fusion model can be, but is not limited to, any one of the following: a CNN (Convolutional Neural Networks) model, an RNN (Recurrent Neural Networks) model, an XGBoost model, and an RF (Random Forest) model.

[0074] As an example, the second prediction result can be [user number: 1002, user value degree: 0.3, confidence value: 0.99].

[0075] In the step 205, the recommendation information is pushed to the target terminal corresponding to the target user information in the target user information set according to the second prediction result set.

[0076] In some embodiments, the execution subject can push the recommendation information to the target terminal corresponding to the target user information in the target user information set according to the second prediction result set. The execution subject can push the recommendation information to the target terminal corresponding to the target user information in the target user information set according to the user value degree included in the second prediction result in the second prediction result set in a target order. The target order can be in a descending order or in an ascending order. The recommendation information can be item recommendation information or activity recommendation information. For example, the recommendation information can also be discount recommendation information. The recommendation information can also be detailed introduction information corresponding to an item.

[0077] The above various embodiments of the present disclosure have the following beneficial effects: through the information pushing method of some embodiments of the present disclosure, the matching degree of the information pushed by the user and the user is improved. At the same time, the waste of network transmission resources is reduced. Specifically, the reason for the waste of network transmission resources is that the prediction ability of the traditional prediction model for different data sets is different, thereby making it difficult to accurately predict the CLV of the user. Based on this, the information pushing method of some embodiments of the present disclosure first acquires a target user information set. Second, determine the target user feature information of each target user information in the target user information set to obtain a target user feature information set. In actual situations, the target user information often contains a lot of information corresponding to the target user, such as hobbies, gender, body mass index, height, weight, and income. Most of the feature information in such information has little effect on predicting the CLV of the user. Moreover, too much feature information may increase the training time of the prediction model. Moreover, it may not improve the prediction accuracy of the obtained prediction model. Second, based on the target user feature information set and a plurality of trained sub-prediction models, a first prediction result set is generated. In actual situations, a single, traditional prediction model has different prediction abilities for different data sets, so it is necessary to predict according to the same data set through a plurality of different prediction models, and according to the obtained first prediction result set, the prediction ability of the plurality of sub-prediction models is determined. In addition, based on the first prediction result set, a user stratification information set, and a fusion model, a second prediction result set is generated. In actual situations, through the fusion model, the accuracy of the final prediction result can be improved, that is, the accuracy of the CLV prediction of the user is improved. In addition, according to the long tail effect theory and the 80-20 law, users often appear stratification, that is, users at different levels often have different behavior habits. Users at the same level often have the same behavior habits. Since stratification is an important feature, it will affect user classification and the accuracy of CLV. Therefore, through the user stratification information set, the accuracy of the second prediction result generated by the fusion model can be further improved. Finally, according to the second prediction result, recommended information is pushed to the target terminal corresponding to the target user information in the target user information set. In this way, the matching degree of the information pushed to the user and the user can be greatly improved. Thus, unnecessary transmission of information is reduced, and thus the waste of network transmission resources is reduced.

[0078] Further reference is made to Figure 4 which shows the flow 400 of some other embodiments of the information pushing method. The flow 400 of the information pushing method includes the following steps:

[0079] In step 401, a target user information set is acquired.

[0080] At step 402, target user feature information of each target user information in the target user information set is determined, and a target user feature information set is obtained.

[0081] In some embodiments, the specific implementation of steps 401-402 and the technical effects brought by them can be referred to Figure 2 The steps 201-202 in the corresponding embodiments are not described here again.

[0082] At step 403, a first prediction result set is generated based on the target user feature information set and the plurality of trained sub-prediction models.

[0083] In some embodiments, the execution subject of the information pushing method (for example Figure 1 The computing device 101 shown) generates the first prediction result set based on the target user feature information set and the plurality of trained sub-prediction models, which can include the following steps:

[0084] First, each target user feature information in the target user feature information set is vectorized to generate vectorized target user feature information, and a vectorized target user feature information set is obtained.

[0085] Second, the vectorized target user feature information set is input into each sub-prediction model in the plurality of trained sub-prediction models to generate a first prediction result, and the first prediction result set is obtained.

[0086] Among them, the plurality of trained sub-prediction models can also include a first sub-prediction model and a second sub-prediction model.

[0087] As an example, the network structure of the first sub-prediction model and the second sub-prediction model can be as shown in Figure 5As shown. Among them, the network structure includes: an input layer 501, an embedding layer 502, a feature sharing layer 503, a splicing layer 504, a first full connection layer 505, a second full connection layer 506, a third full connection layer 507, and an output layer 508. The input of the input layer 501 includes continuous features and discrete features corresponding to the target user feature information. The embedding layer 502 is used to convert high-dimensional feature vectors into low-dimensional feature vectors. The feature sharing layer 503 is used to cross-process the low-dimensional feature vectors output by the embedding layer 502, thereby constructing more rich feature combinations. The splicing layer 504 is used to splice the multiple feature vectors output by the feature sharing layer 503 to generate a feature matrix. The number of neurons of the first full connection layer 505 can be 64. The number of neurons of the second full connection layer 506 can be 32. The number of neurons of the third full connection layer 507 can be 16. The output layer 508 is used to output the first prediction result. The first sub-prediction model can use a cross-entropy loss function. The second prediction model can use a mean square error loss function.

[0088] The cross-entropy loss function is as follows:

[0089]

[0090] Where x represents the predicted value. μ represents the mean of log(x). σ represents the variance of log(x). L Lognormal (x; μ; σ) represents the loss value of the cross-entropy loss function.

[0091] The mean square error loss function is as follows:

[0092]

[0093] Where n represents the total number of samples. i represents the serial number. f represents the predicted value of the sample. t represents the true value of the sample. f i represents the predicted value of the i-th sample. t i represents the true value of the i-th sample. MSE represents the loss value of the mean square error loss function.

[0094] Step 404, generating a second prediction result set based on the first prediction result set, the user hierarchical information set, and the fusion model.

[0095] In some embodiments, the execution subject generates a second prediction result set based on the first prediction result set, the user hierarchical information set, and the fusion model can include the following steps:

[0096] In a first step, each first prediction result in the first prediction result set is spliced with user hierarchical information corresponding to the first prediction result to generate candidate information, thereby obtaining a candidate information set.

[0097] In a second step, each candidate information in the candidate information set is subjected to vectorization processing to generate vectorized candidate information, thereby obtaining a vectorized candidate information set.

[0098] In a third step, each vectorized candidate information in the vectorized candidate information set is input into the fusion model to generate a second prediction result, thereby obtaining a second prediction result set.

[0099] The fusion model is trained through the following steps:

[0100] In a first step, a training sample data set and a full-amount user data set are constructed.

[0101] The training sample data set can include a sub-prediction model training sample set and a fusion model training sample set. The sub-prediction model training sample in the sub-prediction model training sample set is used to train a sub-prediction model. The fusion model training sample in the fusion model training sample set is used to train a fusion model.

[0102] Optionally, the sub-prediction model training sample set and the fusion model training sample set can be generated by randomly sampling the training sample data set.

[0103] Optionally, the execution subject constructing the training sample data set can include the following sub-steps:

[0104] In a first sub-step, a user behavior information set is obtained.

[0105] The user behavior information in the user behavior information set can include static features and dynamic features. The static features can represent the identity information of the user. The dynamic features can represent data corresponding to the value operation of the user.

[0106] As an example, the static features can include user gender, user age, user occupation, user address, user education, and user income. The dynamic features can include the number of purchases, purchase amount, browsing times, click times, number of times added to the shopping cart, search times, value operation status, and value operation type within a period of time.

[0107] As an example, the value operation type described above can be "1". Among them, the operation type "1" can represent that the value operation of the user is a purchase operation. The value operation state described above can be "execution success" or "execution failure". Among them, "execution success" can represent that the user purchases successfully. The state "execution failure" can represent that the user fails to purchase.

[0108] The second sub-step is to perform positive sampling processing and negative sampling processing on the user behavior information set respectively to generate a positive sampling sample set and a negative sampling sample set.

[0109] Among them, the sample quantity in the positive sampling sample set and the negative sampling sample set is in a preset ratio. The preset ratio can be manually set. The execution subject can determine the user behavior information with the state "execution success" in the user behavior information set as the positive sampling sample. The user behavior information with the state "execution failure" in the user behavior information set is determined as the negative sampling sample.

[0110] The third sub-step is to construct user feature information based on each sample in the positive sampling sample set and the negative sampling sample set to generate training sample data and obtain the training sample data set.

[0111] Among them, the execution subject can randomly select samples from the positive sampling sample set and the negative sampling sample set as training sample data to obtain the training sample data set.

[0112] The second step is to train a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models.

[0113] Among them, the execution subject can first perform vectorization processing on each sub-prediction model training sample in the sub-prediction model training sample set to generate a vectorization-processed sub-prediction model training sample to obtain a vectorization-processed sub-prediction model training sample set. Secondly, based on the vectorization-processed sub-prediction model training sample set, each initial sub-prediction model in the plurality of initial sub-prediction models is trained.

[0114] Optionally, the execution subject trains a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models can include the following sub-steps:

[0115] The first sub-step is to determine the feature information of each sub-prediction model training sample in the sub-prediction model training sample set to obtain at least one feature information set.

[0116] The execution subject can determine the feature information of each sub-prediction model training sample in the sub-prediction model training sample set according to the business scenario, and obtain the at least one feature information set.

[0117] As an example, the business scenario can be "predicting the number of purchases of a user in a future period of time". Therefore, the dynamic features included in the sub-prediction model training sample can be selected as the feature information of the sub-prediction model training sample, and the feature information set is obtained.

[0118] Second sub-step, for the feature information set in the at least one feature information set, the initial sub-prediction model corresponding to the feature information set in the plurality of initial sub-prediction models is trained to generate a trained sub-prediction model.

[0119] The execution subject can first vectorize each feature information in the feature information set to generate vectorized feature information, and obtain a vectorized feature information set. Secondly, according to the vectorized feature information set, each initial sub-prediction model in the plurality of initial sub-prediction models is trained to generate a trained sub-prediction model, and a plurality of trained sub-prediction models are obtained.

[0120] Third step, performing hierarchical processing on the full user data set to generate a user hierarchical information set.

[0121] The full user data set can represent the user data corresponding to all registered users. The execution subject can perform hierarchical processing on the full user data set by DBSCAN (density-based clustering algorithm) to generate the user hierarchical information set.

[0122] Optionally, the execution subject performing hierarchical processing on the full user data set to generate a user hierarchical information set can include the following sub-steps:

[0123] First sub-step, according to a preset target category number, performing clustering processing on the full user data set to generate a clustering information group set.

[0124] The execution subject can perform clustering processing on the full user data set according to the preset target category number by a clustering algorithm to generate a clustering information group set. The clustering algorithm can be a K-means algorithm. The clustering algorithm can also be a KNN (K-Nearest Neighbor) algorithm.

[0125] As an example, the preset target category number can be 6. For example, the target category number can be 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, or 100. Figure 6As shown. Among them, Figure 6 The class center 601 of the cluster information group in the cluster information group set, the coordinate points 602 corresponding to each cluster information in the cluster information group set, and the category coverage range 603 of the cluster information group in the cluster information group set.

[0126] Second sub-step, according to the position of the class center of each cluster information group in the cluster information group set and the category coverage range, the cluster information in the cluster information group set is clustered to generate the user hierarchical information set.

[0127] Among them, the execution subject can cluster the cluster information in the cluster information group set through the clustering algorithm according to the position of the class center of each cluster information group in the cluster information group set and the category coverage range, to generate the user hierarchical information set. The category coverage range can represent the area covering the coordinate points corresponding to each cluster information in the cluster information group.

[0128] As an example, as shown Figure 7 Among them, the cluster information corresponding to each coordinate point 602 in "region A" is of the same category. The cluster information corresponding to each coordinate point 602 in "region B" is of the same category. The cluster information corresponding to each coordinate point 602 in "region C" is of the same category.

[0129] Fourth, based on the fusion model training sample set, the user hierarchical information set and the plurality of trained sub-prediction models, the initial fusion model is trained to generate a fusion model.

[0130] Among them, the execution subject can first splice each fusion model training sample in the fusion model training sample set and the user hierarchical information set to generate a spliced training sample, and obtain a spliced training sample set. Secondly, input the spliced training sample set into each sub-prediction model in the plurality of trained sub-prediction models to generate prediction information, and obtain a prediction information set. Finally, according to the prediction information set, the initial fusion model is trained to generate the fusion model. The initial fusion model can be an RF model.

[0131] Optionally, the execution subject trains the initial fusion model based on the fusion model training sample set, the user hierarchical information set and the plurality of trained sub-prediction models to generate a fusion model, which can include the following sub-steps:

[0132] A first sub-step, inputting the fusion model training sample set into each of the plurality of trained sub-prediction models to generate a prediction result, thereby obtaining a prediction result set.

[0133] The execution subject can input the vectorized fusion model training sample set into the trained sub-prediction model to generate a prediction result.

[0134] A second sub-step, based on the prediction result set and the user hierarchical information set, performing fusion model training on the initial fusion model to generate the fusion model.

[0135] The execution subject can first splice each prediction result in the prediction result set with the user hierarchical information set to generate a spliced feature vector, thereby obtaining a spliced feature vector set. Then, based on the spliced feature vector set, the execution subject performs fusion model training on the initial fusion model to generate the fusion model.

[0136] Step 405, according to the second prediction result set, filtering target user information that meets a filtering condition from the target user information set as candidate user information, thereby obtaining a candidate user information set.

[0137] In some embodiments, the execution subject can filter target user information that meets a filtering condition from the target user information set as candidate user information according to the second prediction result set, thereby obtaining a candidate user information set. The filtering condition can be that the second prediction result corresponding to the target user information includes a user value degree greater than a preset threshold.

[0138] As an example, the preset threshold can be 0.7. The second prediction result can be [user number: 1004, user value degree: 0.8, confidence value: 0.99]. The corresponding target user information can be [user number: 1004, user gender: male, user name: Zhu XX, user age: 34 years old, user height: 152 cm, user weight: 200 kg, user body mass index: 43.3, user home address: XX City, XX Street, XX Community, XX Building, XX Layer, XX House, user occupation: sound engineer, user education: master, user income: 20,000 yuan / month, purchase times: 20 times, browsing times: 32 times, purchase amount: 5,000 yuan, click times: 54 times, collection times: 0 times, add to shopping cart times: 20 times, search times: 23 times].

[0139] Step 406, pushing the recommendation information to the target terminal corresponding to each candidate user information in the candidate user information set.

[0140] In some embodiments, the execution subject can push the recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set. The target terminal is a terminal logged in with a user account of a user corresponding to the candidate user information. For example, the target terminal can be a "mobile phone". The target terminal can also be a "computer".

[0141] Optionally, the execution subject can push the recommendation information to the target terminal in response to determining that the target terminal has turned on the information notification mode.

[0142] The execution subject can first obtain terminal state information of the target terminal, then determine an information notification mode state value from the terminal state information, and finally determine whether the target terminal has turned on the information notification mode according to the information notification mode state value.

[0143] For example, the terminal state information can be [terminal number: AXX102, terminal state: 30 hours of booting, information notification mode state value: 1]. The information notification mode state value of "1" can indicate that the terminal with the terminal number "AXX102" has turned on the information notification mode.

[0144] From Figure 4 It can be seen that, compared with the description of some embodiments corresponding to Figure 2 The first sub-prediction model and the second sub-prediction model are first introduced in some embodiments of the present disclosure. Since the existing prediction model has a fixed model structure, and the existing prediction model is difficult to meet the prediction requirements in various scenarios. Therefore, the first sub-prediction model and the second sub-prediction model are introduced. The first sub-prediction model and the second sub-prediction model have good performance and scalability. At the same time, for different scenarios, such as "large amount of prediction data and high prediction accuracy" and "small amount of prediction data and high prediction accuracy", the prediction accuracy can be maintained. In addition, in order to avoid infringing the privacy of users, the execution subject only pushes information to the target terminal when the target terminal has turned on the information notification mode.

[0145] Further reference Figure 8 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of an information pushing device. The device embodiments correspond to the method embodiments shown in Figure 2 The device can be applied to various electronic devices.

[0146] As Figure 8As shown, the information pushing apparatus 800 of some embodiments comprises an obtaining unit 801, a determining unit 802, a first generating unit 803, a second generating unit 804 and a pushing unit 805. The receiving unit 801 is configured to receive a webpage browsing request of a user. The obtaining unit 801 is configured to obtain a target user information set. The determining unit 802 is configured to determine target user feature information of each target user information in the target user information set to obtain a target user feature information set. The first generating unit 803 is configured to generate a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models. The second generating unit 804 is configured to generate a second prediction result set based on the first prediction result set, a user stratification information set and a fusion model, wherein the prediction result is used to represent the user value degree of the target user information. The pushing unit 805 is configured to push recommendation information to a target terminal corresponding to the target user information in the target user information set according to the second prediction result set.

[0147] In some optional implementations of some embodiments, the pushing unit is further configured to: filter out target user information satisfying a filtering condition from the target user information set as a candidate user information set according to the second prediction result set; and push recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set, wherein the target terminal is a terminal logged in with a user account of a user corresponding to the candidate user information.

[0148] In some optional implementations of some embodiments, the pushing unit is further configured to: push recommendation information to the target terminal in response to determining that the target terminal opens an information notification mode.

[0149] In some optional implementations of some embodiments, the fusion model is trained by the following steps:

[0150] constructing a training sample data set and a full user data set, wherein the training sample data set comprises a sub-prediction model training sample set and a fusion model training sample set; training a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models; performing stratification processing on the full user data set to generate a user stratification information set; and performing fusion model training on an initial fusion model based on the fusion model training sample set, the user stratification information set and the plurality of trained sub-prediction models to generate a fusion model.

[0151] In some optional implementations of some embodiments, the constructing the training sample data set includes: obtaining a user behavior information set; performing positive sampling processing and negative sampling processing on the user behavior information set respectively to generate a positive sampling sample set and a negative sampling sample set, wherein the sample quantity in the positive sampling sample set and the negative sampling sample set is in a preset ratio; and based on each sample in the positive sampling sample set and the negative sampling sample set, constructing user feature information to generate a training sample data, thereby obtaining the training sample data set.

[0152] In some optional implementations of some embodiments, the training, based on the sub-prediction model training sample set, a plurality of initial sub-prediction models to generate a plurality of trained sub-prediction models includes: determining feature information of each sub-prediction model training sample in the sub-prediction model training sample set to obtain at least one feature information set; and for a feature information set in the at least one feature information set, training an initial sub-prediction model corresponding to the feature information set in the plurality of initial sub-prediction models to generate a trained sub-prediction model.

[0153] In some optional implementations of some embodiments, the hierarchical processing of the full-amount user data set to generate a user hierarchical information set includes: performing clustering processing on the full-amount user data set according to a preset target category number to generate a clustering information group set; and performing clustering processing on clustering information in the clustering information group set according to the position and category coverage range of the class center of each clustering information group in the clustering information group set to generate the user hierarchical information set.

[0154] In some optional implementations of some embodiments, the fusion model training, based on the fusion model training sample set, the user hierarchical information set, and the plurality of trained sub-prediction models, of an initial fusion model to generate a fusion model includes: inputting the fusion model training sample set into each trained sub-prediction model in the plurality of trained sub-prediction models to generate a prediction result, thereby obtaining a prediction result set; and performing fusion model training on the initial fusion model based on the prediction result set and the user hierarchical information set to generate the fusion model.

[0155] In some optional implementations of some embodiments, the sub-prediction model training sample set and the fusion model training sample set are generated by: performing random sampling on the training sample data set to generate the sub-prediction model training sample set and the fusion model training sample set.

[0156] It can be understood that the units described in the apparatus 800 are described with reference to the accompanying drawings Figure 2The individual steps in the described methods correspond. Thus, the operations, features, and advantages described above for the methods apply equally to the apparatus 800 and the units contained therein, which will not be described again here.

[0157] Reference is made below to Figure 9 which shows an electronic device, such as a Figure 1 computing device 101) 900 suitable for implementing some embodiments of the present disclosure. Figure 9 The electronic device shown is merely an example and should not bring any limitation to the function and scope of use of embodiments of the present disclosure.

[0158] As shown in Figure 9 The electronic device 900 can include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 901 that can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 902 or loaded into a random access memory (RAM) 903 from a storage device 908. In the RAM 903, various programs and data required for the operation of the electronic device 900 are also stored. The processing device 901, the ROM 902, and the RAM 903 are connected to each other through a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0159] In general, the following devices can be connected to the I / O interface 905: input devices 906 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 907 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 908 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 909. The communication devices 909 can allow the electronic device 900 to communicate wirelessly or wired with other devices to exchange data. Although Figure 9 The electronic device 900 is shown with various devices, but it should be understood that all of the shown devices are not required to be implemented or present. More or fewer devices can alternatively be implemented or present. Figure 9 Each block shown in the flowchart in

[0160] In particular, the processes described above with reference to the flowcharts can be implemented as a computer software program according to some embodiments of the present disclosure. For example, some embodiments of the present disclosure include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for performing the methods illustrated by the flowcharts. In some such embodiments, the computer program can be downloaded and installed from a network via the communication device 909, or installed from the storage device 908, or installed from the ROM 902. When the computer program is executed by the processing device 901, the above-mentioned functions defined in the methods of some embodiments of the present disclosure are performed.

[0161] It should be noted that the computer readable medium in some embodiments of the present disclosure can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium may, for example, be, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the above. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer readable storage medium can be any tangible medium that contains or stores a program used by or in connection with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, the computer readable signal medium can include a data signal carried in a baseband or as part of a carrier wave, in which the computer readable program code is carried. Such a propagated data signal can take any of a variety of forms, including but not limited to electro-magnetic, optical, or any suitable combination thereof. The computer readable signal medium can also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained in the computer readable medium can be transmitted by any suitable medium, including but not limited to wire, cable, RF, etc., or any suitable combination of the above.

[0162] In some embodiments, the client, server, or both can communicate using any known or future developed network protocols, such as the HyperText Transfer Protocol (HTTP), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), the Internet, and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current or future developed network.

[0163] The computer-readable medium described above can be included in the electronic device described above; alternatively, the computer-readable medium can exist on a separate entity that is connected to the electronic device, and is accessed by the electronic device through a network, for example. The computer-readable medium described above can carry one or more programs which, when executed by the electronic device, cause the electronic device to perform the following operations: obtaining a target user information set; determining target user feature information of each target user information in the target user information set, to obtain a target user feature information set; generating a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; generating a second prediction result set based on the first prediction result set, a user hierarchical information set, and a fusion model, wherein the prediction result is used to represent the user value degree corresponding to the target user information; and pushing recommendation information to a target terminal corresponding to the target user information in the target user information set according to the second prediction result set.

[0164] Computer program code for carrying out operations of some embodiments of the present disclosure can be written in any combination of one or more programming languages, including an object oriented programming language such as Java, Smalltalk, C++ or the like, and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network ("LAN") or a wide area network ("WAN"), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0165] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing devices to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0166] The units described in some embodiments of the present disclosure can be implemented by means of software, or by means of hardware. The units described can also be provided in a processor, for example, a processor can be described as comprising an obtaining unit, a determining unit, a first generating unit, a second generating unit and a pushing unit. In some cases, the names of the units do not constitute a limitation on the units themselves, for example, the obtaining unit can also be described as a "unit for obtaining a target user information set".

[0167] The functions described above in the detailed description can be performed by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system on a chip (SOCs), complex programmable logic devices (CPLDs), etc.

[0168] The above description is merely illustrative of the exemplary embodiments of the present disclosure and the technical principles of the present disclosure. It should be understood by those skilled in the art that the inventive scope of the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by the combinations of the above technical features or their equivalent features without departing from the above inventive concept. For example, the above technical features can be replaced with the technical features disclosed in the embodiments of the present disclosure (but not limited to) having similar functions to form the technical solutions.

Claims

1. An information pushing method, comprising: obtaining a target user information set; determining target user feature information of each target user information in the target user information set to obtain a target user feature information set; generating a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; generating a second prediction result set based on the first prediction result set, a user stratification information set and a fusion model, wherein the prediction result is used to represent the value degree of a user corresponding to the target user information, including splicing each first prediction result in the first prediction result set and the corresponding user stratification information to generate candidate information to obtain a candidate information set; and generating a second prediction result set based on the candidate information set and the fusion model; pushing recommendation information to a target terminal corresponding to target user information in the target user information set according to the second prediction result set, including: selecting target user information satisfying a screening condition from the target user information set as candidate user information according to the second prediction result set to obtain a candidate user information set; pushing recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set; wherein the fusion model is trained by the following steps: constructing a training sample data set and a full-amount user data set, wherein the training sample data set includes a sub-prediction model training sample set and a fusion model training sample set; training a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models; stratifying the full-amount user data set to generate a user stratification information set; training an initial fusion model based on the fusion model training sample set, the user stratification information set and the plurality of trained sub-prediction models to generate a fusion model.

2. The method of claim 1, wherein, The target terminal is a terminal that logs in a user account of a user corresponding to the candidate user information.

3. The method of claim 2, wherein, The pushing of recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set includes: in response to determining that the target terminal opens an information notification mode, pushing recommendation information to the target terminal.

4. The method of claim 1, wherein, The construction of the training sample data set includes: obtaining a user behavior information set; respectively performing positive sampling processing and negative sampling processing on the user behavior information set to generate a positive sampling sample set and a negative sampling sample set, wherein the sample amount in the positive sampling sample set and the negative sampling sample set is in a preset ratio; constructing user feature information based on each sample in the positive sampling sample set and the negative sampling sample set to generate a training sample data, thereby obtaining the training sample data set.

5. The method of claim 1, wherein, The training of a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models includes: determining feature information of each sub-prediction model training sample in the sub-prediction model training sample set to obtain at least one feature information set; For a feature information set in the at least one feature information set, an initial sub-prediction model corresponding to the feature information set in the plurality of initial sub-prediction models is trained to generate a trained sub-prediction model.

6. The method of claim 1, wherein, The hierarchical processing of the full-amount user data set to generate a user hierarchical information set comprises: According to a preset target category number, the full-amount user data set is clustered to generate a cluster information group set; According to the position and category coverage range of the class center of each cluster information group in the cluster information group set, the cluster information in the cluster information group set is clustered to generate the user hierarchical information set.

7. The method of claim 1, wherein, The fusion model training of the initial fusion model based on the fusion model training sample set, the user hierarchical information set and the plurality of trained sub-prediction models to generate a fusion model comprises: The fusion model training sample set is input into each trained sub-prediction model in the plurality of trained sub-prediction models to generate a prediction result, thereby obtaining a prediction result set; The fusion model training of the initial fusion model based on the prediction result set and the user hierarchical information set to generate the fusion model.

8. The method of claim 1, wherein, The sub-prediction model training sample set and the fusion model training sample set are generated by the following steps: The training sample data set is randomly sampled to generate the sub-prediction model training sample set and the fusion model training sample set.

9. An information pushing device, comprising: an acquisition unit configured to acquire a target user information set; a determination unit configured to determine target user feature information of each target user information in the target user information set to obtain a target user feature information set; a first generation unit configured to generate a first prediction result set based on the target user feature information set and a plurality of trained sub-prediction models; a second generation unit configured to generate a second prediction result set based on the first prediction result set, a user hierarchical information set and a fusion model, wherein the prediction result is used to represent the value degree of a user corresponding to the target user information, comprising: splicing each first prediction result in the first prediction result set and corresponding user hierarchical information to generate candidate information, thereby obtaining a candidate information set; and generating a second prediction result set based on the candidate information set and the fusion model; a pushing unit configured to push recommendation information to a target terminal corresponding to target user information in the target user information set according to the second prediction result set, comprising: filtering target user information satisfying a filtering condition from the target user information set as candidate user information according to the second prediction result set, thereby obtaining a candidate user information set; pushing recommendation information to a target terminal corresponding to each candidate user information in the candidate user information set; wherein the fusion model is trained by the following steps: constructing a training sample data set and a full user data set, wherein the training sample data set comprises a sub-prediction model training sample set and a fusion model training sample set; training a plurality of initial sub-prediction models based on the sub-prediction model training sample set to generate a plurality of trained sub-prediction models; performing hierarchical processing on the full user data set to generate a user hierarchical information set; performing fusion model training on an initial fusion model based on the fusion model training sample set, the user hierarchical information set, and the plurality of trained sub-prediction models to generate a fusion model.

10. An electronic device, comprising: one or more processors; storage having one or more programs stored thereon; when the one or more programs are executed by the one or more processors, the one or more programs cause the one or more processors to implement the method of any of claims 1-8.

11. A computer readable medium having stored thereon a computer program, wherein, The program is executed by the processor to implement the method of any of claims 1-8.

Images