A method and apparatus for information push
By predicting user behavior categories using a multi-task network model, the problem of low efficiency and accuracy in information push was solved, resulting in a higher conversion rate.
Patent Information
- Application Number
- CN202210235884.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-03-10
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-03-10
AI Technical Summary
Existing technologies have low information push efficiency and low accuracy, especially in e-commerce product push, where it is difficult to effectively predict users' purchase rate, resulting in low conversion rates.
A multi-task network model is adopted, which combines user behavior features and information features. Through a pre-trained regression task model and classification task model, multiple user behavior categories are predicted, and target information is selected and pushed according to the behavior categories.
It improved the accuracy and conversion rate of information push, and enhanced the effectiveness of information push by comprehensively considering multiple push objectives.
Smart Images

Figure CN114610996B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence, and in particular to a method and apparatus for information push. Background Technology
[0002] With the development of information technology and the Internet industry, information overload has become a challenge for users to process information. Therefore, in Internet applications, the method of pushing personalized information to users is usually adopted to reduce the impact of information overload on users, while improving the accuracy of information push.
[0003] Specifically, when the push information is about items sold by e-commerce platforms, it is necessary to consider multiple push objectives for the items sold by e-commerce platforms. For example, when pushing information, in addition to predicting the click-through rate of users on the items sold by e-commerce platforms, it is also necessary to predict the purchase rate of users on the items in order to increase the conversion rate of the push information.
[0004] Currently, machine learning methods are mainly used to build models for a single push target. Subsequent push information is then pushed through this model, which is mainly aimed at a single push target. This results in problems such as low push efficiency and low accuracy. Summary of the Invention
[0005] In view of this, embodiments of the present invention provide a method and apparatus for information push, which can input user behavior features and information features corresponding to the user behavior features into a trained multi-task network model; predict multiple behavior categories of the user on the corresponding information based on the output of the multi-task network model; filter target display information and push the target display information to the terminal based on the behavior categories; by predicting multiple behavior categories of the user through the trained multi-task network model and filtering out the corresponding target display information, the accuracy of information push is improved, thereby improving the conversion rate of information push.
[0006] To achieve the above objectives, according to one aspect of the present invention, an information push method is provided, comprising: acquiring user behavior features of a terminal; inputting the user behavior features and information features of multiple pieces of information in an information pool corresponding to the user behavior features into a trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, the multi-task network model being trained from multiple push targets and samples having the multiple push targets; predicting multiple behavior categories of one or more pieces of information corresponding to the user's operation based on the output of the multi-task network model; filtering target display information from the multiple pieces of information based on the behavior categories, and pushing the target display information to the terminal.
[0007] Optionally, the information push method further includes:
[0008] Acquire historical data for multiple behavioral categories for various training information; generate a first user behavior feature for training based on the historical data; acquire user profile features corresponding to the first user behavior feature for training; add the first user behavior feature for training and the corresponding user profile features to the training sample set; train the multi-task network model using the training sample set.
[0009] Optionally, the information push method further includes:
[0010] The training first user behavior feature and the corresponding profile feature are input into the trained generative adversarial network model, and the generative adversarial network model outputs multiple simulation samples; the simulation samples are added to the training sample set.
[0011] Optionally, the information push method includes a generative adversarial network model comprising a discriminant network and multiple generative networks; using the discriminant network to determine the category of the simulation samples output by the multiple generative networks; the category of the simulation samples includes simulation positive samples or simulation negative samples; training the multi-task network model includes: selecting the simulation positive samples and / or the simulation negative samples according to the service type and adding them to the training sample set corresponding to the service type; and using the training sample set corresponding to the service type to train the multi-task network model.
[0012] Optionally, the method for pushing information includes:
[0013] The multi-task network model includes a multilayer perceptron; training the multi-task network model includes: inputting the first user behavior feature for training and information features of various training information into the multilayer perceptron, using the multilayer perceptron to output the association information between the first user behavior feature for training and the information features; and inputting the association information into the multi-task network model.
[0014] Optionally, the information push method includes: the loss function corresponding to the multi-task network model is constructed based on the regression task function of the regression task model and the classification task function of the classification task model; training the multi-task network model includes: training the regression task model and the classification task model, and controlling the training iteration through the loss function, the regression task function and the classification task function respectively.
[0015] Optionally, the method for pushing information includes:
[0016] The training uses the first user behavior features to classify behavior categories; the task type to which the behavior category belongs is determined; the task type is either a regression task or a classification task.
[0017] Training the multi-task network model includes: training a regression task model included in the multi-task network model using training samples belonging to the regression task included in the training sample set; training a classification task model included in the multi-task network model using training samples belonging to the classification task included in the training sample set; and using the output of the regression task model and / or the output of the classification task model as the output of the multi-task network model.
[0018] To achieve the above objectives, according to a second aspect of the present invention, an information push device is provided, comprising: a user information acquisition module, a user behavior prediction module, and a push information acquisition module; wherein,
[0019] The user information acquisition module is used to acquire user behavior features of the terminal; input the user behavior features and information features of multiple pieces of information in the information pool corresponding to the user behavior features into a trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, and the multi-task network model is trained by multiple push targets and samples with the multiple push targets;
[0020] The user behavior prediction module is used to predict multiple behavior categories of the user for one or more pieces of information corresponding to the operation, based on the output of the multi-task network model.
[0021] The module for obtaining push information is used to filter target display information from the multiple pieces of information according to the behavior category, and push the target display information to the terminal.
[0022] Optionally, the information push device further includes:
[0023] Acquire historical data for multiple behavioral categories for various training information; generate a first user behavior feature for training based on the historical data; acquire user profile features corresponding to the first user behavior feature for training; add the first user behavior feature for training and the corresponding user profile features to the training sample set; train the multi-task network model using the training sample set.
[0024] Optionally, the information push device further includes:
[0025] The training first user behavior feature and the corresponding profile feature are input into the trained generative adversarial network model, and the generative adversarial network model outputs multiple simulation samples; the simulation samples are added to the training sample set.
[0026] Optionally, in the information push device, the generative adversarial network model includes a discriminant network and multiple generator networks; the discriminant network is used to determine the category of the simulation samples output by the multiple generator networks; the category of the simulation samples includes simulation positive samples or simulation negative samples; training the multi-task network model includes: selecting the simulation positive samples and / or the simulation negative samples according to the service type and adding them to the training sample set corresponding to the service type; and using the training sample set corresponding to the service type to train the multi-task network model.
[0027] Optionally, the information push device includes:
[0028] The multi-task network model includes a multilayer perceptron; training the multi-task network model includes: inputting the first user behavior feature for training and information features of various training information into the multilayer perceptron, using the multilayer perceptron to output the association information between the first user behavior feature for training and the information features; and inputting the association information into the multi-task network model.
[0029] Optionally, the information push device includes:
[0030] The loss function corresponding to the multi-task network model is constructed based on the regression task function of the regression task model and the classification task function of the classification task model; training the multi-task network model includes: training the regression task model and the classification task model, and controlling the training iteration through the loss function, the regression task function and the classification task function respectively.
[0031] Optionally, the information push device includes: classifying behavior categories for the first user behavior features used for training; determining the task type to which the behavior category belongs; the task type being either a regression task or a classification task; training the multi-task network model, including: training a regression task model included in the multi-task network model using training samples belonging to the regression task included in the training sample set; training a classification task model included in the multi-task network model using training samples belonging to the classification task included in the training sample set; and using the output of the regression task model and / or the output of the classification task model as the output of the multi-task network model.
[0032] To achieve the above objectives, according to a third aspect of the present invention, an electronic device for information push is provided, characterized in that it includes: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement any of the methods described above for information push.
[0033] To achieve the above objectives, according to a fourth aspect of the present invention, a computer-readable medium is provided having a computer program stored thereon, characterized in that, when the program is executed by a processor, it implements any of the methods described above for information push.
[0034] One embodiment of the above invention has the following advantages or beneficial effects: it can input user behavior characteristics and information features corresponding to user behavior characteristics into a trained multi-task network model; predict multiple behavior categories of the user on the corresponding information based on the output of the multi-task network model; filter target display information and push target display information to the terminal based on the behavior categories; by predicting multiple behavior categories of the user through the trained multi-task network model and filtering out the corresponding target display information, the accuracy of information push is improved, thereby improving the conversion rate of information push.
[0035] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description
[0036] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:
[0037] Figure 1 This is a flowchart illustrating an information push method according to an embodiment of the present invention;
[0038] Figure 2 This is a schematic diagram of a process for training a multi-task network model according to an embodiment of the present invention;
[0039] Figure 3 This is a schematic diagram of the structure of an information push device according to an embodiment of the present invention;
[0040] Figure 4 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied;
[0041] Figure 5 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation
[0042] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.
[0043] like Figure 1 As shown in the figure, an embodiment of the present invention provides a method for information push, which may include the following steps:
[0044] Step S101: Obtain user behavior features of the terminal; input the user behavior features and information features of multiple pieces of information in the information pool corresponding to the user behavior features into the trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, and the multi-task network model is trained by multiple push targets and samples with the multiple push targets.
[0045] Specifically, user behavior characteristics indicate different user behaviors for different application scenarios. For example, for users using an e-commerce platform, user behavior characteristics include browsing, clicking, purchasing, and rating information within the e-commerce application; for users using video services, user behavior characteristics include browsing, rating, liking, and saving video information. Furthermore, the information pool contains multiple pieces of information to be pushed (e.g., item information, video information, etc.); information characteristics include feature information used to describe the information. For example, when the information is item information, information characteristics include the item's category, model, color, historical sales volume, etc.; when the information is video information, information characteristics include the video's category, content characteristics, click volume, etc.
[0046] Further, the user behavior features and the information features of multiple pieces of information in the information pool corresponding to the user behavior features are input into the trained multi-task network model. The trained multi-task network model includes a regression task model and a classification task model, and is trained from multiple push targets and samples containing those multiple push targets. Specifically, the multi-task network model can be an optimized MMOE model (Modeling TaskRelationships in Multi-task Learning with Multi-gate Mixture-of-Experts model); further, the multi-task network model includes a regression task model and a classification task model, trained from multiple push targets (e.g., click-through rate, browsing duration, purchase rate, etc.) and samples containing those multiple push targets. The description of training the multi-task network model is consistent with the description of steps S201-S204, and will not be repeated here.
[0047] In various information push application scenarios, there is a need to consider multiple push targets. For example, in the application scenario of e-commerce, information push needs to combine user click-through rate and purchase rate to improve the conversion rate of product information; in the application scenario of information (such as video) recommendation, information push needs to combine user browsing rate and evaluation rate to improve the conversion rate of information. Therefore, the multi-task network model trained in this application (including regression task model and classification task model) can take user behavior characteristics as input, and push one or more information to users by integrating multiple user behavior characteristics and information characteristics, thereby improving the accuracy and effect of information push, and thus improving the conversion rate of pushed information.
[0048] Step S102: Based on the output of the multi-task network model, predict multiple behavioral categories of the user for one or more pieces of information corresponding to the operation.
[0049] Specifically, the output of the multi-task network model is generated from the outputs of the regression task model and / or the classification task model. The corresponding task type can be selected based on the category of the push target. For example, if the predicted push target is click-through rate, the output of the classification task model can be used for prediction; if the predicted push target is browsing time, the output of the regression task model can be used to predict the probability corresponding to the browsing time range, etc. Furthermore, the outputs of the classification task model and / or the regression task model are combined as the output of the multi-task network model to predict multiple behavioral categories of one or more pieces of information corresponding to the user's operation. These behavioral categories are set according to different application scenarios. For example, for an e-commerce application scenario, the user's behavioral categories can be clicks, purchases, and shopping cart addition time, etc.; for a video service application scenario, the user's behavioral categories can be clicks, browsing time, and favorites, etc.
[0050] Step S103: Based on the behavior category, filter the target display information from the multiple pieces of information, and push the target display information to the terminal.
[0051] Specifically, based on the numerical values of the behavior categories in the application scenario, for the classification task model, the values can be (0, 1), etc.; for the regression task model, the values can be one or more probability values. The multi-task network model combines the output values of the classification task model and the output values of the regression task model to filter target display information for the user from the multiple pieces of information. Among them, when filtering target display information, a set number of display information can be selected as target display information according to a set order (e.g., sales from high to low), and the target display information is pushed to the terminal to display the corresponding push information on the terminal.
[0052] like Figure 2As shown, this embodiment of the invention provides a method for training a multi-task network model, which may include the following steps:
[0053] Step S201: Obtain historical data for multiple behavioral categories of various training information; generate a first user behavior feature for training based on the historical data; obtain the user profile feature corresponding to the first user behavior feature for training; add the first user behavior feature for training and the corresponding user profile feature to the training sample set.
[0054] Specifically, acquiring the first user behavior feature sample data for training includes acquiring historical data for multiple behavior categories across various training information. Preferably, the historical data can be preprocessed using a training library generation module, such as acquiring user browsing information, click information, and user profile features. The historical data can be combined with the user's device number, and historical data indicating invalid data can be removed, improving the efficiency and effectiveness of training the multi-task network model. Further, a sampling module can be used to sample positive and negative samples based on the original training samples obtained from the training library generation module, determining the ratio of positive and negative samples according to the application scenario. Further, a feature generation module can be used to further extract features from the training samples obtained from the sampling module to generate a training sample set. That is, the first user behavior feature for training is generated based on the historical data; the user profile features corresponding to the first user behavior feature for training are acquired; and the first user behavior feature for training and the corresponding user profile features are added to the training sample set.
[0055] Step S202: Input the first user behavior feature and the corresponding profile feature into the trained generative adversarial network model, and use the generative adversarial network model to output multiple simulation samples; add the simulation samples to the training sample set.
[0056] Furthermore, based on training samples obtained from historical data, a generative adversarial network (GAN) model is used to generate simulation samples; thereby increasing the number and information capacity of training samples, further improving the effect of training the multi-task network model, and improving the prediction accuracy of the multi-task network model. Among them, the generative adversarial network (GAN) is a method of unsupervised learning, which learns by having two neural networks play against each other; the GAN consists of a generator network and a discriminator network. This invention adopts a model with multiple generator networks and a single discriminator network, that is, the GAN model includes a discriminator network and multiple generator networks; in order to improve the simulation degree of the simulation samples, an example of the GAN model is shown in formula (1).
[0057]
[0058] Where G represents the function corresponding to the generator, D represents the function corresponding to the discriminator; x and z represent the input features, and E represents the expected value. That is, the first user behavior feature and the corresponding profile feature are input into the trained generative adversarial network model, and the generative adversarial network model outputs multiple simulation samples; further, the simulation samples are added to the training sample set.
[0059] Preferably, the generative adversarial network model includes a discriminant network and multiple generator networks; the discriminant network is used to determine the category of the simulated samples output by the multiple generator networks; the category of the simulated samples includes simulated positive samples or simulated negative samples; training the multi-task network model includes: selecting the simulated positive samples and / or the simulated negative samples according to the service type and adding them to the training sample set corresponding to the service type; and training the multi-task network model using the training sample set corresponding to the service type. Specifically, according to the application scenarios corresponding to different service types, simulated positive samples and / or simulated negative samples are selected from the simulated samples as simulated samples corresponding to the service type, and the simulated samples are added to the training sample set, which further improves the feature discrimination and sample information capacity of the training samples.
[0060] Before using the generative adversarial network (GAN) model to obtain simulation samples, the GAN model needs to be trained. Specifically, during the training process, the discriminator can be trained with the generator parameters fixed first, and then the generator can be trained with the discriminator parameters fixed, and so on iteratively. Simulation positive samples and / or simulation negative samples are selected according to the business type. Combined with the feature information of real training samples, the ratio of real training samples to simulation samples is adjusted according to the business type, thereby increasing the sample information capacity.
[0061] Step S203: Input the first user behavior feature for training and the information features of various training information into the multilayer perceptron, and use the multilayer perceptron to output the association information between the first user behavior feature for training and the information features; input the association information into the multi-task network model.
[0062] Specifically, the multi-task network model includes a multilayer perceptron (MLP). Preferably, the training first user behavior feature and various information features of the training information are input into the multilayer perceptron, and the multilayer perceptron outputs the association information between the training first user behavior feature and the information features. By strengthening the understanding of the association information of the sample input features, the information capacity of the training samples is further expanded.
[0063] Furthermore, the associated information is input into the multi-task network model to train the multi-task network model.
[0064] Step S204: Train the multi-task network model using the training sample set.
[0065] Specifically, a training sample set is obtained for training, and the multi-task network model is trained.
[0066] Furthermore, the loss function corresponding to the multi-task network model is constructed based on the regression task function of the regression task model and the classification task function of the classification task model; training the multi-task network model includes: training the regression task model and the classification task model, and controlling the training iteration through the loss function, the regression task function and the classification task function respectively.
[0067] Specifically, the loss function corresponding to the multi-task network model is shown in Equation (2), where ω represents the model parameters required for the multi-task network model, σ represents the variance, and ∑ represents the sum of k tasks (classification tasks and / or regression tasks).
[0068]
[0069] For example, the classification task function of the classification task model is shown in formula (3).
[0070] -logSoftmax(y, f) W (x)) (3)
[0071] The variance of the regression task function associated with the regression task model is shown in Equation (4).
[0072] ||yf W (x)|| 2 (4)
[0073] That is, the loss function corresponding to the multi-task network model is constructed based on the regression task function of the regression task model and the classification task function of the classification task model; training the multi-task network model includes: training the regression task model and the classification task model, controlling the training iterations through the loss function, the regression task function, and the classification task function, respectively. By iteratively training the loss function, the regression task function, and the classification task function corresponding to the multi-task network model, an optimized multi-task network model is generated through model training.
[0074] Further, the training uses first user behavior features to classify behavior categories; determines the task type to which the behavior category belongs; the task type is either a regression task or a classification task; training the multi-task network model includes: training a regression task model included in the multi-task network model using training samples belonging to the regression task included in the training sample set; training a classification task model included in the multi-task network model using training samples belonging to the classification task included in the training sample set; and using the output of the regression task model and / or the output of the classification task model as the output of the multi-task network model. Specifically, behavior categories are divided based on the first user behavior features used for training (e.g., clicks, purchases, browsing, etc.). Push targets can be generated based on user behavior (e.g., click-through rate, purchase rate, browsing duration range probability, etc.). Further, task types are determined for behavior categories; for example, if the behavior category is "click," the corresponding task type is a classification task type; if the behavior category is "browsing duration," it is determined as a regression task type, etc. Here, the behavior category is associated with the push target. For example, if the behavior category is "click," the associated push target can be "click-through rate." When training the multi-task network model, it includes training the corresponding task model using training samples of different types, that is, using the training samples in the training sample set that belong to the regression task to train the regression task model included in the multi-task network model, using the training samples in the training sample set that belong to the classification task to train the classification task model included in the multi-task network model, and further using the output of the regression task model and / or the output of the classification task model as the output of the multi-task network model to improve the training effect of the multi-task network model. That is, the multi-task network model, and the regression task model or classification task model included in the multi-task network model, are trained by multiple push targets and samples with said multiple push targets.
[0075] like Figure 3 As shown, this embodiment of the invention provides an information push device 300, including: a user information acquisition module 301, a user behavior prediction module 302, and a push information acquisition module 303; wherein,
[0076] The user information acquisition module 301 is used to acquire user behavior features of the terminal; input the user behavior features and information features of multiple pieces of information in the information pool corresponding to the user behavior features into a trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, and the multi-task network model is trained by multiple push targets and samples with the multiple push targets;
[0077] The user behavior prediction module 302 is used to predict multiple behavior categories of the user for one or more pieces of information corresponding to the operation based on the output of the multi-task network model.
[0078] The push information acquisition module 303 is used to filter target display information from the multiple pieces of information according to the behavior category, and push the target display information to the terminal.
[0079] This invention also provides an electronic device for information push, comprising: one or more processors; and a storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method provided in any of the above embodiments.
[0080] This invention also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the method provided in any of the above embodiments.
[0081] Figure 4 An exemplary system architecture 400 is shown, in which the information push method or information push apparatus of the present invention can be applied.
[0082] like Figure 4 As shown, system architecture 400 may include terminal devices 401, 402, and 403, a network 404, and a server 405. Network 404 serves as the medium for providing communication links between terminal devices 401, 402, and 403 and server 405. Network 404 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.
[0083] Users can use terminal devices 401, 402, and 403 to interact with server 405 via network 404 to receive or send messages, etc. Various client applications can be installed on terminal devices 401, 402, and 403, such as e-commerce client applications, web browser applications, and information service applications.
[0084] Terminal devices 401, 402, and 403 can be various electronic devices with displays and supporting various client applications, including but not limited to smartphones, tablets, laptops, and desktop computers.
[0085] Server 405 can be a server that provides various services, such as a backend management server that supports client applications used by users through terminal devices 401, 402, and 403. The backend management server can process received requests to retrieve push notifications and return the selected target information to the terminal device.
[0086] It should be noted that the information push method provided in the embodiments of the present invention is generally executed by server 405, and correspondingly, the information push device is generally set in server 405.
[0087] It should be understood that Figure 4 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.
[0088] The following is for reference. Figure 5 It shows a schematic diagram of the structure of a computer system 500 suitable for implementing a terminal device of the present invention. Figure 5 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of the present invention.
[0089] like Figure 5 As shown, the computer system 500 includes a central processing unit (CPU) 501, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 502 or programs loaded from storage section 508 into random access memory (RAM) 503. The RAM 503 also stores various programs and data required for the operation of the system 500. The CPU 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.
[0090] The following components are connected to I / O interface 505: an input section 506 including a keyboard, mouse, etc.; an output section 507 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 508 including a hard disk, etc.; and a communication section 509 including a network interface card such as a LAN card, modem, etc. The communication section 509 performs communication processing via a network such as the Internet. A drive 510 is also connected to I / O interface 505 as needed. A removable medium 511, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 510 as needed so that computer programs read from it can be installed into storage section 508 as needed.
[0091] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 509, and / or installed from removable medium 511. When the computer program is executed by central processing unit (CPU) 501, it performs the functions defined above in the system of this invention.
[0092] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.
[0093] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.
[0094] The modules and / or units described in the embodiments of the present invention can be implemented in software or hardware. The described modules and / or units can also be housed in a processor; for example, a processor can be described as including a user information acquisition module, a user behavior prediction module, and a push notification acquisition module. The names of these modules do not necessarily limit the module itself; for example, the push notification acquisition module can also be described as "a module that filters target display information from multiple pieces of information based on the behavior category."
[0095] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs, which, when executed by the device, cause the device to include: acquiring user behavior features of a terminal; inputting the user behavior features and information features of multiple pieces of information in an information pool corresponding to the user behavior features into a trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, the multi-task network model being trained from multiple push targets and samples having the multiple push targets; predicting multiple behavior categories of one or more pieces of information corresponding to the user's operation based on the output of the multi-task network model; filtering target display information from the multiple pieces of information based on the behavior categories, and pushing the target display information to the terminal.
[0096] In embodiments of the present invention, user behavior features and corresponding information features are input into a trained multi-task network model; based on the output of the multi-task network model, multiple behavior categories of the user on the corresponding information are predicted; based on the behavior categories, target display information is filtered and pushed to the terminal; by predicting multiple behavior categories of the user through the trained multi-task network model and filtering out the corresponding target display information, the accuracy of information push is improved, thereby improving the conversion rate of information push.
[0097] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.
Claims
1. A method of information push, characterized by, include: Acquire user behavior characteristics of the terminal; wherein, for the application scenario of the user using an e-commerce platform, the user behavior characteristics include one or more of the following: browsing, clicking, purchasing, and rating information contained in the e-commerce platform application; for the application scenario of the user using video services, the user behavior characteristics include one or more of the following: browsing, rating, liking, and collecting video information. The user behavior features and the information features of multiple pieces of information in the information pool corresponding to the user behavior features are input into the trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, and the multi-task network model is trained from multiple push targets and samples with the multiple push targets; wherein, the push targets are generated based on user behavior; Based on the output of the multi-task network model, predict multiple behavioral categories of the user's interaction with one or more messages; wherein, the behavioral categories are associated with the push target; Based on the behavior category, target display information is filtered from the multiple pieces of information, and the target display information is pushed to the terminal; The method further includes: The training first user behavior features are used to classify behavior categories; wherein, the training first user behavior features are generated based on historical data; Determine the task type to which the behavior category belongs; the task type is either a regression task or a classification task. Training the multi-task network model includes: The regression task model contained in the multi-task network model is trained using the training samples belonging to the regression task contained in the training sample set. The classification task model contained in the multi-task network model is trained using the training samples belonging to the classification task contained in the training sample set. The output of the regression task model and / or the output of the classification task model are used as the output of the multi-task network model.
2. The method of claim 1, wherein, Also includes: Acquire historical data for multiple behavioral categories targeting various training information; generate first user behavior features for training based on the historical data; Obtain the user profile features corresponding to the first user behavior features used for training; The first user behavior feature used for training and the corresponding user profile feature are added to the training sample set. The multi-task network model is trained using the training sample set.
3. The method of claim 2, wherein, Also includes: The training first user behavior feature and the corresponding profile feature are input into the trained generative adversarial network model, and the generative adversarial network model is used to output multiple simulation samples. The simulation sample is added to the training sample set.
4. The method according to claim 3, characterized in that, The generative adversarial network model includes a discriminant network and multiple generative networks; The discriminant network is used to determine the category of the simulation samples output by the plurality of generator networks; The categories of simulated samples include simulated positive samples or simulated negative samples; Training the multi-task network model includes: Based on the business type, select the simulated positive samples and / or the simulated negative samples and add them to the training sample set corresponding to the business type; The multi-task network model is trained using the training sample set corresponding to the business type.
5. The method according to claim 2, characterized in that, The multi-task network model includes a multilayer perceptron; Training the multi-task network model includes: The training first user behavior feature and various training information features are input into the multilayer perceptron, and the multilayer perceptron outputs the association information between the training first user behavior feature and the information features. The associated information is input into the multi-task network model.
6. The method according to any one of claims 2 to 5, characterized in that, The loss function corresponding to the multi-task network model is constructed based on the regression task function of the regression task model and the classification task function of the classification task model; Training the multi-task network model includes: Train the regression task model and the classification task model. The training iterations are controlled by the loss function, the regression task function, and the classification task function, respectively.
7. An apparatus for information push, characterized by include: The system includes a user information acquisition module, a user behavior prediction module, and a push notification acquisition module; among them, The user information acquisition module is used to acquire user behavior characteristics of the terminal; input the user behavior characteristics and information characteristics of multiple pieces of information in the information pool corresponding to the user behavior characteristics into a trained multi-task network model; the multi-task network model includes a regression task model and a classification task model, and the multi-task network model is trained by multiple push targets and samples with the multiple push targets; wherein, the push targets are generated based on user behavior; wherein, for the application scenario of users using e-commerce, the user behavior characteristics include one or more of the following: browsing, clicking, purchasing, and rating information contained in the e-commerce application; for the application scenario of users using video services, the user behavior characteristics include one or more of the following: browsing, rating, liking, and collecting video information. The user behavior prediction module is used to predict multiple behavior categories of the user interacting with one or more messages based on the output of the multi-task network model; wherein the behavior categories are associated with the push target. The module for obtaining push information is used to filter target display information from the multiple pieces of information according to the behavior category, and push the target display information to the terminal; The device is further configured to classify the first user behavior features for training into behavior categories; wherein the first user behavior features for training are generated based on historical data; Determine the task type to which the behavior category belongs; the task type is either a regression task or a classification task. Training the multi-task network model includes: The regression task model contained in the multi-task network model is trained using the training samples belonging to the regression task contained in the training sample set. The classification task model contained in the multi-task network model is trained using the training samples belonging to the classification task contained in the training sample set. The output of the regression task model and / or the output of the classification task model are used as the output of the multi-task network model.
8. An electronic device, comprising: include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-6.
9. A computer readable medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-6.
Citation Information
Patent Citations
Recommendation method and device, storage medium and electronic equipment
CN113742590A
