Information push method, device and related equipment based on application usage records

Through the combination of unsupervised neural networks and transformed neural networks, the mobile application usage records are automatically mined and accurate application embedded vectors are generated, which solves the problem of inefficiency in the existing technology and improves the quality of information push and user experience.

CN111552871BActive Publication Date: 2024-11-26TENCENT TECHNOLOGY (SHENZHEN) CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010274010.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-04-09
Publication Date
2024-11-26
Estimated Expiration
2040-04-09

AI Technical Summary

Technical Problem

The prior art is inefficient in mining information from mobile application usage records, making it difficult to achieve optimal performance, especially in the absence of in-depth understanding of specific tasks, and requires repeated targeted feature engineering.

Method used

The unsupervised neural network model is used to encode the application installation image information to obtain the application installation image low-dimensional dense vector, and the application installation image low-dimensional dense vector, installation sequence information and uninstall sequence information are fused by transforming the neural network model to generate the application Embedded vectors for pushing accurate information.

Benefits of technology

It realizes automatic mining of internal relationships of application usage records, obtains more accurate application embedded vectors, and improves the hit rate and user experience of push information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111552871B_ABST
    Figure CN111552871B_ABST
Patent Text Reader

Abstract

The embodiments of the present disclosure provide an information push method, device, electronic device and computer-readable medium based on application usage records. The method includes: obtaining the application installation image information, application installation sequence information and application uninstallation sequence information of the current device; encoding the application installation image information using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device; fusing the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information using a transformation neural network model to obtain an application embedded vector of the current device; obtaining the push information of the current device based on the application embedded vector of the current device; and pushing the push information to the current device. The technical solution of the embodiments of the present disclosure can deeply fuse the low-dimensional dense vector of the application installation image with the application installation sequence information and the application uninstallation sequence information to obtain a more accurate application embedded vector of the current device.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to an information push method, device, electronic device and computer-readable medium based on application usage records. Background Art

[0002] The user's application usage records for mobile applications contain a lot of information about the user's interests and preferences. For example, if a user has installed multiple mobile photography applications on his mobile phone, then this user is likely to be very interested in mobile photography; if a user has recently installed mobile applications related to travel ticketing and hotel reservations, this user is likely to have plans for a business trip or travel in the near future. Therefore, the user's behavior records for using mobile applications have the potential to play a huge role in advertising and recommendation scenarios.

[0003] However, the traditional approach to mining information from mobile app usage records relies on task-specific manual feature engineering, which is not only inefficient, as different downstream specific tasks require repeated targeted feature engineering, but also difficult to achieve optimal performance with traditional approaches, especially without a deep understanding of specific tasks.

[0004] Therefore, a new information push method, device, electronic device and computer-readable medium based on application usage records are needed.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The embodiments of the present disclosure provide an information push method, device, electronic device and computer-readable medium based on application usage records, which can automatically mine the internal relationships of application usage records, obtain a more accurate application embedded vector of the current device, generate push information with a higher hit rate based on the application embedded vector, and send it to the current device.

[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0008] The disclosed embodiment proposes an information push method based on application usage records, including: obtaining application installation image information, application installation sequence information and application uninstallation sequence information of the current device; encoding the application installation image information by using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device; fusing the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information by using a transformation neural network model to obtain an application embedded vector of the current device; obtaining push information of the current device according to the application embedded vector of the current device; and pushing the push information of the current device to the current device.

[0009] The disclosed embodiment proposes an information push device based on application usage records, including: a device data acquisition module, configured to obtain application installation image information, application installation sequence information and application uninstallation sequence information of the current device; an image information processing module, configured to encode the application installation image information by using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device; an embedded vector generation module, configured to fuse the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information by using a transformation neural network model to obtain an application embedded vector of the current device; a push information generation module, configured to obtain push information of the current device according to the application embedded vector of the current device; and a push information sending module, configured to push the push information of the current device to the current device.

[0010] In an exemplary embodiment, the embedded vector generation module includes an application installation feature sequence generation unit, an application uninstallation feature sequence generation unit, a first feature generation unit, and an embedded vector generation unit. The application installation feature sequence generation unit is configured to generate an application installation feature sequence according to application installation sequence information; the application uninstallation feature sequence generation unit is configured to generate an application uninstallation feature sequence according to application uninstallation sequence information; the first feature generation unit is configured to obtain first feature information according to the application installation image low-dimensional dense vector, the application installation feature sequence, and the application uninstallation feature sequence; the embedded vector generation unit is configured to encode the first feature information through an encoder of a transformation neural network model to obtain an application embedded vector of the current device.

[0011] In an exemplary embodiment, the application installation sequence information includes an application installation object sequence, an application installation time sequence, and an application installation type sequence; wherein the application installation feature sequence generation unit includes an installation type embedded vector subunit, an installation object embedded vector subunit, an installation time embedded vector subunit, and an application installation feature sequence generation subunit. The installation type embedded vector subunit is configured to perform an embedded representation on the application installation type sequence to obtain an application installation type sequence embedded vector; the installation object embedded vector subunit is configured to perform an embedded representation on the application installation object sequence to obtain an application installation object sequence embedded vector; the installation time embedded vector subunit is configured to perform an embedded representation on the application installation time sequence to obtain an application installation time sequence embedded vector; and the application installation feature sequence generation subunit is configured to perform bitwise addition on the application installation type sequence embedded vector, the application installation object sequence embedded vector, and the application installation time sequence embedded vector to obtain an application installation feature sequence.

[0012] In an exemplary embodiment, the application uninstall sequence information includes an application uninstall object sequence, an application uninstall time sequence, and an application uninstall type sequence; wherein the application uninstall feature sequence generation unit includes an uninstall type embedded vector subunit, an uninstall object embedded vector subunit, an uninstall time embedded vector subunit, and an application uninstall feature sequence generation subunit. The uninstall type embedded vector subunit is configured to perform an embedded representation on the application uninstall type sequence to obtain an application uninstall type sequence embedded vector; the uninstall object embedded vector subunit is configured to perform an embedded representation on the application uninstall object sequence to obtain an application uninstall object sequence embedded vector; the uninstall time application embedded vector subunit is configured to perform an embedded representation on the application uninstall time sequence to obtain an application uninstall time sequence embedded vector; the application uninstall feature sequence generation subunit is configured to perform bitwise addition on the application uninstall type sequence embedded vector, the application uninstall time sequence embedded vector, and the application uninstall object sequence embedded vector to obtain an application uninstall feature sequence.

[0013] In an exemplary embodiment, the application installation image information also includes an installation image type sequence; wherein the first feature generation unit includes an installation image type embedded vector subunit, an installation image feature sequence generation subunit, and a first feature generation subunit. The installation image type embedded vector subunit is configured to perform an embedded representation on the installation image type sequence to obtain an installation image type sequence embedded vector; the installation image feature sequence generation subunit is configured to perform bitwise addition of the installation image type embedded vector and the application installation image low-dimensional dense vector to obtain an application installation image feature sequence; the first feature generation subunit is configured to obtain the first feature information according to the application installation image feature sequence, the application installation feature sequence, and the application uninstall feature sequence.

[0014] In an exemplary embodiment, the transformation neural network model further includes a first fully connected neural network; wherein the embedded vector generation unit includes a first initial representation generation subunit, a second initial representation generation subunit and an embedded vector generation subunit. The first initial representation generation subunit is configured to encode the first feature information through the encoder of the transformation neural network model to obtain a first initial representation; the second initial representation generation subunit is configured to intercept the first initial representation according to the position of the low-dimensional dense vector of the application installation image in the first feature information to obtain a second initial representation; the embedded vector generation subunit is configured to process the second initial representation through the first fully connected neural network of the transformation neural network model to obtain an application embedded vector of the current device.

[0015] In an exemplary embodiment, the information push device based on application usage records also includes a historical data acquisition module, a historical image information processing module, a historical application embedded vector generation module, a transformation neural network model training module and an unsupervised neural network model training module. Among them, the historical data acquisition module is configured to obtain historical application installation image information, historical application installation sequence information and historical application uninstallation sequence information of historical devices; the historical image information processing module is configured to encode the historical application installation image information through the second fully connected neural network of the unsupervised neural network model to obtain the predicted application installation image low-dimensional dense vector of the historical device; the historical application embedded vector generation module is configured to encode the predicted application installation image low-dimensional dense vector, historical application installation sequence information and historical application uninstallation sequence information through the encoder of the transformation neural network model and the first fully connected neural network to obtain the predicted application embedded vector of the historical device; the transformation neural network model training module is configured to adjust the parameters of the transformation neural network model according to the predicted application embedded vector, historical application installation sequence information and historical application uninstallation sequence information to obtain the trained transformation neural network model; the unsupervised neural network model training module is configured to adjust the parameters of the unsupervised neural network model according to the predicted application embedded vector and the historical application installation image information to obtain the trained unsupervised neural network model.

[0016] In an exemplary embodiment, the unsupervised neural network model further includes a third fully connected neural network; wherein the unsupervised neural network model training module includes a first prediction information generation unit, a first loss function generation unit and an unsupervised neural network model training unit. The first prediction information generation unit is configured to reconstruct the prediction application embedded vector through the third fully connected neural network of the unsupervised neural network model to obtain the first prediction information; the first loss function generation unit is configured to generate a first loss function based on the historical application installation image information and the first prediction information; the unsupervised neural network model training unit is configured to adjust the parameters of the unsupervised neural network model according to the first loss function to obtain a trained unsupervised neural network model.

[0017] In an exemplary embodiment, the transformation neural network model also includes a decoder; the transformation neural network model training module includes an application installation query information generation unit, an application uninstallation query information generation unit, a second feature information generation unit, a second prediction information generation unit, a second loss function generation unit and a transformation neural network model training unit. The application installation query information generation unit is configured to generate application installation query information based on historical application installation sequence information; the application uninstallation query information generation unit is configured to generate application uninstallation query information based on historical application uninstallation sequence information; the second feature information generation unit is configured to obtain second feature information based on the predicted application embedded vector, application installation query information, and application uninstallation query information; the second prediction information generation unit is configured to reconstruct the second feature information through the decoder of the transformation neural network model to obtain second prediction information; the second loss function generation unit is configured to generate a second loss function based on the second prediction information, historical application installation sequence information, and historical application uninstallation sequence information; the transformation neural network model training unit is configured to adjust the parameters of the transformation neural network model according to the second loss function to obtain a trained transformation neural network model.

[0018] In an exemplary embodiment, the historical application installation sequence information includes a historical application installation time sequence and a historical application installation type sequence; wherein the application installation query information generation unit includes a historical installation type embedded vector subunit, a historical installation time embedded vector subunit, and an application installation query information generation subunit. The historical installation type embedded vector subunit is configured to perform an embedded representation on the historical application installation type sequence to obtain a historical application installation type sequence embedded vector; the historical installation time embedded vector subunit is configured to perform an embedded representation on the historical application installation time sequence to obtain a historical application installation time sequence embedded vector; the application installation query information generation subunit is configured to perform bitwise addition of the historical application installation type sequence embedded vector and the historical application installation time sequence embedded vector to obtain application installation query information.

[0019] In an exemplary embodiment, the historical application uninstall sequence information includes a historical application uninstall time sequence and a historical application uninstall type sequence; wherein the application uninstall query information generation unit includes: a historical uninstall type embedded vector subunit, a historical uninstall time embedded vector subunit and an application uninstall query information generation subunit. The historical uninstall type embedded vector subunit is configured to perform an embedded representation on the historical application uninstall type sequence to obtain a historical application uninstall type sequence embedded vector; the historical uninstall time embedded vector subunit is configured to perform an embedded representation on the historical application uninstall time sequence to obtain a historical application uninstall time sequence embedded vector; the application uninstall query information generation subunit is configured to perform bitwise addition on the historical application uninstall type sequence embedded vector and the historical application uninstall time sequence embedded vector to obtain the application uninstall query information.

[0020] An embodiment of the present disclosure proposes an electronic device, including: one or more processors; a 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 information push method based on application usage records as in the above embodiment.

[0021] The embodiment of the present disclosure provides a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the information push method based on application usage records as in the above embodiment is implemented.

[0022] In the technical solutions provided by some embodiments of the present disclosure, the application installation image information is encoded to obtain the low-dimensional dense vector of the application installation image of the current device, and the low-dimensional dense vector of the application installation image is combined with the application installation sequence information and the application uninstallation sequence information for fusion processing, so that the application installation image information, the application installation sequence information, and the application uninstallation sequence information can be compressed and deeply fused to obtain a more accurate application embedded vector of the current device. Accurate push information can be generated according to the application embedded vector of the current device and pushed to the current device, thereby improving the user experience.

[0023] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings herein are incorporated into the specification and constitute a part of the specification, showing embodiments consistent with the present disclosure, and together with the specification, are used to explain the principles of the present disclosure. Obviously, the drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. In the drawings:

[0025] Figure 1 A schematic diagram showing an exemplary system architecture to which the information push method or device based on application usage records according to an embodiment of the present disclosure can be applied;

[0026] Figure 2 A flowchart of an information push method based on application usage records according to an embodiment of the present disclosure is schematically shown;

[0027] Figure 3 A schematic diagram showing the structure of a method or device for pushing information based on application usage records according to an embodiment of the present disclosure;

[0028] Figure 4 A flowchart of an information push method based on application usage records according to an embodiment of the present disclosure is schematically shown;

[0029] Figure 5 A schematic diagram showing the structure of first characteristic information according to an embodiment of the present disclosure;

[0030] Figure 6 A flowchart of an information push method based on application usage records according to an embodiment of the present disclosure is schematically shown;

[0031] Figure 7 A flowchart of an information push method based on application usage records according to an embodiment of the present disclosure is schematically shown;

[0032] Figure 8 A schematic diagram showing the structure of second characteristic information according to an embodiment of the present disclosure;

[0033] Fig. 9 Schematically shows a flow chart of an information push method based on application usage records according to another embodiment of the present disclosure;

[0034] Fig.10 , 11 A diagram schematically shows a display interface of a method for pushing information according to an embodiment of the present disclosure;

[0035] Fig.12 Schematically shows a structural diagram of a transformation neural network model according to an embodiment of the present disclosure;

[0036] Fig.13 The structure diagram of an unsupervised neural network model according to an embodiment of the present disclosure is schematically shown;

[0037] Fig.14 A block diagram schematically shows an object data processing device according to an embodiment of the present disclosure;

[0038] Fig.15A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0039] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be more comprehensive and complete and will fully convey the concept of the example embodiments to those skilled in the art.

[0040] In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the present disclosure.

[0041] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0042] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

[0043] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0044] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. Basic artificial intelligence technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology (NLP), and machine learning / deep learning. Among them, natural language processing research can realize various theories and methods for effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language used by people in daily life, so it is closely related to the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph and other technologies. In the application field of information push, generating the push information of the device based on the device usage record helps to push information to the device user in a targeted manner according to the preferences of the device user. Information push is an important research section in natural language processing. In related technologies, the application usage record of the device includes installation image, installation and uninstallation records. Among them, the application installation image refers to the collection of all device applications on a user's device (such as a mobile phone) at a certain moment. The installation and uninstallation record refers to the record of a user installing and uninstalling device applications. The record includes: the operation type, i.e., installation or uninstallation, the operation object, i.e., the device application identity (ID), and the operation time.

[0045] In the related art, the installation image and the installation and uninstallation records are usually modeled separately to obtain the application embedded vector of the installation image and the application embedded vector of the installation and uninstallation records respectively. The above method cannot deeply mine the associated information between the installation image and the installation and uninstallation records, so that the quality of the obtained application embedded vector is low, which further affects the push quality of the generated push information.

[0046] Therefore, a new information push method, device, electronic device and computer-readable medium based on application usage records are needed.

[0047] Figure 1 A schematic diagram showing an exemplary system architecture to which the method or device for information push based on application usage records according to an embodiment of the present disclosure can be applied is shown.

[0048] like Figure 1As shown, the system architecture 100 may include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0049] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

[0050] Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, portable computers, desktop computers, wearable devices, virtual reality devices, smart homes, etc.

[0051] The server 105 may be a server that provides various services. For example, the terminal device 103 (or the terminal device 101 or 102) uploads application installation image information, application installation sequence information, and application uninstallation sequence information to the server 105. The server 105 may obtain the application installation image information, application installation sequence information, and application uninstallation sequence information of the terminal device 103 (current device); encode the application installation image information using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the terminal device 103; fuse the low-dimensional dense vector of the application installation image, the application installation sequence information, and the application uninstallation sequence information using a transformation neural network model to obtain an application embedded vector of the terminal device 103; obtain the push information of the terminal device 103 according to the application embedded vector of the terminal device 103; and push the push information of the terminal device 103 to the terminal device 103. Then, the terminal device 103 may display the push information to the user operating the terminal device 103, or perform subsequent operations on the push information based on the operation behavior of the user operating the terminal device 103.

[0052] It should be emphasized that the method provided in this application requires the use of the user's application usage records during execution. When applying the various embodiments in this application to specific products or technologies, it is necessary to obtain the relevant user's permission or consent before collecting application usage records, and the collection, use and processing of relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0053] Figure 2The flowchart of the information push method based on application usage records according to an embodiment of the present disclosure is schematically shown. The method provided by the embodiment of the present disclosure can be processed by any electronic device with computing processing capabilities, such as the above Figure 1 The server 105 and / or terminal devices 102 and 103 in the embodiment are described by taking the server 105 as the execution subject in the following embodiment, but the present disclosure is not limited to this.

[0054] like Figure 2 As shown, the information push method based on application usage records provided by the embodiment of the present disclosure may include the following steps.

[0055] In step S210, application installation image information, application installation sequence information, and application uninstallation sequence information of the current device are obtained.

[0056] In the embodiment of the present disclosure, the current device is, for example, terminal device 101 (or terminal device 102 or 103). The current device may be one (for example, terminal device 101) or multiple (for example, terminal devices 101, 102, 103). The application installation image information of the current device refers to the collection of all device applications in the current device at a certain moment. The application installation image information may be expressed as follows:

[0057]

[0058] Among them, U is the number of current devices, and M is the number of device applications. The installation image for the current device u can be obtained by vector For example, x um =1 indicates that the current device u is currently installing device application m.

[0059] The application installation sequence information and application uninstallation sequence information of the current device respectively describe the application installation record and application uninstallation record of the current device. Among them, the application installation sequence information may include an application installation object sequence and an application installation time sequence. The application installation object sequence and the application installation time sequence may be respectively expressed as follows:

[0060]

[0061] in, Indicates the device application installed on the current device u, Indicates the time when the current device u is installing the i-th installed device application. I means that the application installation sequence information includes I installation operations.

[0062] In an exemplary embodiment, the application installation sequence information may further include an application installation type sequence. The application installation type sequence describes a type identifier whose operation type is installation. The dimension of the application installation type sequence is the same as the dimension of the application installation object sequence and the application installation time sequence. As mentioned above, when the dimension of the application installation object sequence is 1, the dimension of the application installation type sequence is also 1.

[0063] The application uninstallation sequence information of the current device may include an application uninstallation object sequence and an application uninstallation time sequence. The application uninstallation object sequence and the application uninstallation time sequence may be respectively expressed as follows:

[0064]

[0065] in, Indicates the device application uninstalled by the current device u, Indicates the time when the current device u uninstalls the i-th device application. I means that the application uninstall sequence information includes I uninstall operations.

[0066] In an exemplary embodiment, the application uninstall sequence information may further include an application uninstall type sequence. The application uninstall type sequence describes the type identifier of the operation type being uninstall. The dimension of the application uninstall type sequence is the same as the dimension of the application uninstall object sequence and the application uninstall time sequence. As described above, when the dimension of the application uninstall object sequence is 1, the dimension of the application uninstall type sequence is also 1.

[0067] In step S220, the application installation image information is encoded using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device.

[0068] In the embodiment of the present disclosure, the low-dimensional dense vector of the application installation image is a low-dimensional vector embedded representation that characterizes the characteristics of the application installation image information. The application embedded vector represents an object through a low-dimensional vector, and the object can be a word, a commodity, etc. In the application scenario of the device application usage record of the embodiment of the present disclosure, the embedded vector of the installation image is used to represent the application installation image information through a low-dimensional vector. In the vector of the low-dimensional dense vector of the application installation image, each dimension represents some potential grammatical or semantic feature of the text. The low-dimensional dense vector of the application installation image jointly describes the application installation image information through the information of each dimension.

[0069] In the disclosed embodiment, the unsupervised neural network model may be a stacked autoencoder. An autoencoder is an unsupervised neural network model used to reproduce the input signal as much as possible. It can encode disordered input data, and further reconstruct the original input data with the encoded data. The input vector of the autoencoder is of the same dimension as the output vector. The stacked autoencoder increases the number of intermediate hidden layers in the autoencoder (the number of hidden layers is greater than 1) to obtain stronger encoding and reconstruction capabilities. The output of the lowest dimensional bottleneck layer can be used as a feature after dimensionality reduction. In one embodiment, the application installation image information can be processed by an invisible stacked autoencoder to obtain a low-dimensional dense vector of the application installation image of the current device. In another embodiment, the application installation image information can also be processed by an explicit stacked autoencoder to obtain a low-dimensional dense vector of the application installation image of the current device. This embodiment can improve the efficiency of model training. The structural diagram of the explicit stacked autoencoder can be seen in Fig.13 As shown. Fig.13 As shown, the explicit stacked autoencoder may include an encoding part 1310 and a decoding part 1320. The encoding part is used to encode the disordered input data, and the decoding part is used to further reconstruct the original input data according to the encoded data. The invisible stacked autoencoder only includes the encoding part.

[0070] Figure 3 A schematic diagram showing the structure of a method or device for pushing information based on application usage records according to an embodiment of the present disclosure. Figure 3 As shown, the unsupervised neural network model 2 may include a second fully connected neural network 21. The application installation image information x of the current device u may be obtained by the second fully connected neural network 21. u Processing is performed to obtain the low-dimensional dense vector x of the application installation image of the current device u (1) The processing process of the second fully connected neural network 21 on the application installation image information can be shown as follows:

[0071] x (1) =Leaky_ReLU(x u W (1) +b (1) ) (4)

[0072] Where W (1) As the weight of the second fully connected neural network 21, b (1) is the offset vector. In the embodiment of the present disclosure, W (1) It is also used as the application embedded vector of the globally shared device application, namely W (1) =W a .

[0073] In step S230, the low-dimensional dense vector of the application installation image, the application installation sequence information, and the application uninstallation sequence information are fused using a transformation neural network model to obtain an application embedded vector of the current device.

[0074] In the disclosed embodiment, the application installation sequence information and the application uninstallation sequence information can be embeddedly represented to perform deep fusion processing on the low-dimensional dense vector of the application installation image, the embedded representation of the application installation sequence information and the embedded identifier of the application uninstallation sequence information to obtain the application embedded vector of the current device.

[0075] Among them, the Transformer neural network model (Transformer model) is a deep neural network model based on the attention mechanism. The Transformer model includes an encoder and a decoder structure. The Transformer model is widely used in the field of natural language.

[0076] In an exemplary embodiment, if Figure 3 As shown, the encoder 31 can fuse the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information to obtain the application embedded vector of the current device. The encoder 31 may be composed of a plurality of stacked network blocks, each of which is composed of a multi-head self-attention mechanism and a feed forward network. The multi-head self-attention mechanism refers to a type of artificial neural network structure that considers the association between different elements in a sequence, thereby having the ability to extract multiple aspects of information from a sequence. Specifically, Figure 3 As shown, in the encoder 31, each network block 311 may be composed of a plurality of stacked network blocks 311, and each network block 311 may include a multi-head self-attention mechanism 3111 and a feedforward neural network 3112. Each network block 311 may also include a residual connection and a layer 3113 for normalization (Add & Normalize), which can better optimize the deep network.

[0077] In step S240, push information of the current device is obtained according to the application embedded vector of the current device.

[0078] In the disclosed embodiment, the push information of the current device can be generated offline. The push information of the current device can also be generated in real time when a downstream task (such as a device application with information push function) calls the information push method based on application usage records of the disclosed embodiment.

[0079] In step S250, the push information of the current device is pushed to the current device.

[0080] The information push method based on application usage records provided by the disclosed embodiment encodes the application installation image information to obtain the low-dimensional dense vector of the application installation image of the current device, and fuses the low-dimensional dense vector of the application installation image with the application installation sequence information and the application uninstallation sequence information, so as to compress and deeply fuse the application installation image information, the application installation sequence information and the application uninstallation sequence information, and obtain a more accurate application embedded vector of the current device. Accurate push information can be generated according to the application embedded vector of the current device and pushed to the current device, thereby improving the user experience.

[0081] In some embodiments of the present disclosure, the push information of the current device can be pushed to downstream application software, such as mobile phone management software, so that users can browse the push information when using the mobile phone management software, thereby improving the user's browsing experience. Data mining and user mining can also be performed based on the application embedded vector of the current device to obtain target users in specific scenarios and improve the hit rate of the obtained target users.

[0082] Figure 4 The flowchart of the information push method based on application usage records according to an embodiment of the present disclosure is schematically shown.

[0083] like Figure 4 As shown, the information push method based on application usage records provided by the embodiment of the present disclosure may include the following steps.

[0084] Step S210, obtaining application installation image information, application installation sequence information and application uninstallation sequence information of the current device.

[0085] Step S220: Encode the application installation image information using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device.

[0086] In step S231, an application installation feature sequence is generated according to the application installation sequence information.

[0087] In an exemplary embodiment, the application installation sequence information may include an application installation object sequence, an application installation time sequence, and an application installation type sequence; wherein the application installation type sequence may be embeddedly represented to obtain an application installation type sequence embedded vector; the application installation object sequence may be embeddedly represented to obtain an application installation object sequence embedded vector; the application installation time sequence may be embeddedly represented to obtain an application installation time sequence embedded vector; the application installation type sequence embedded vector, the application installation object sequence embedded vector, and the application installation time sequence embedded vector are bitwise added to obtain an application installation feature sequence.

[0088] The application embedded vector of the application installation type sequence may be the following I-dimensional vector:

[0089] [w n ,…,w n ,…,w n ] (5)

[0090] In the sequence of application installation objects, if the i-th element Then the i-th element The application embedded vector can be expressed as In the application installation time series, if the i-th element Then the i-th element The application embedded vector can be expressed as

[0091] In step S232, an application uninstall feature sequence is generated according to the application uninstall sequence information.

[0092] In an exemplary embodiment, the application uninstall sequence information may include an application uninstall object sequence, an application uninstall time sequence, and an application uninstall type sequence; wherein the application uninstall type sequence may be embeddedly represented to obtain an application uninstall type sequence embedded vector; the application uninstall object sequence may be embeddedly represented to obtain an application uninstall object sequence embedded vector; the application uninstall time sequence may be embeddedly represented to obtain an application uninstall time series embedded vector; the application uninstall type sequence embedded vector, the application uninstall time series embedded vector, and the application uninstall object sequence embedded vector are bitwise added to obtain an application uninstall feature sequence.

[0093] The application embedded vector of the application uninstall type sequence may be the following I-dimensional vector:

[0094] [w l ,...,w l ,...,w l ] (6)

[0095] In the application uninstallation object sequence, if the i-th element Then the i-th element The application embedded vector can be expressed as In the application uninstallation time series, if the i-th element Then the i-th element The application embedded vector can be expressed as

[0096] In step S233, first feature information is obtained according to the low-dimensional dense vector of the application installation image, the application installation feature sequence, and the application uninstallation feature sequence.

[0097] In the disclosed embodiment, the low-dimensional dense vector of the application installation image, the application installation feature sequence, and the application uninstallation feature sequence may be concatenated to obtain the first feature information. Figure 5 FIG. 2 is a schematic diagram showing the structure of the first characteristic information according to an embodiment of the present disclosure. Figure 5 As shown, for the current device u, the installation image sequence embedded vector can be e x .

[0098] In an exemplary embodiment, the application installation image information also includes an installation image type sequence. The installation image type sequence can be embedded to obtain an installation image type sequence embedded vector; the installation image type embedded vector and the application installation image low-dimensional dense vector are bitwise added to obtain an application installation image feature sequence; and the first feature information is obtained according to the application installation image feature sequence, the application installation feature sequence, and the application uninstallation feature sequence.

[0099] In the embodiment of the present disclosure, the installation image type sequence embedded vector can be expressed as follows:

[0100] w n =[w x ,...,w x ,...,w x ] (7)

[0101] The dimension of the embedded vector of the installation image type sequence is the same as the low-dimensional dense vector of the application installation image. n and the low-dimensional dense vector x of the application installation image (1) Perform bitwise addition to obtain the application installation image feature sequence e x .

[0102] Application installation object sequence The embedded vector of the application installation object sequence can be App installation time series The embedded vector of the app installation time series can be The application installation type sequence embedded vector can be [w n ,w n w n ].in, Indicates that the first application installed on the current device u is the application numbered 3. Indicates that the time when the first application is installed on the current device u is 4. The application installation type sequence embedded vector [w n ,w n ,w n ], Application installation object sequence embedded vector App installation time series embedded vector Perform bitwise addition to obtain the application installation feature sequence [e n1 ,e n2 ,e n3 ].

[0103] For the current device u, its application uninstall object sequence The embedded vector of the application uninstall object sequence can be App uninstall time series The embedded vector of the application uninstallation time series can be The application uninstall type sequence embedded vector can be [w l ,w l ,w l ].in, Indicates that the first application uninstalled by the current device is the application numbered 7. Indicates that the time when the current device u uninstalls the first application is 8. The application uninstall type sequence embedded vector [w l , w l , w l ], Application uninstall time series embedded vector Application uninstall object sequence embedded vector Perform bitwise addition to obtain the application uninstall feature sequence [e l1 , e l2 , e l3 ].

[0104] Among them, the application installation image feature sequence e x , application installation feature sequence [e n1 , e n2 , e n3 ]、Application uninstall feature sequence[e l1 , e l2 , e l3 ] are spliced ​​to obtain the first feature information.

[0105] In step S234, the first feature information is encoded by transforming the encoder of the neural network model to obtain an application embedded vector of the current device.

[0106] In the embodiment of the present disclosure, the transformation neural network model may be, for example, Figure 3As shown in the figure 3. The transformation neural network model 3 may include an encoder 31. In an exemplary embodiment, the transformation neural network model also includes a first fully connected neural network 32; wherein, the first feature information can be processed by the encoder 31 of the transformation neural network model 3 to obtain a first initial representation; the first initial representation is intercepted according to the position of the low-dimensional dense vector of the application installation image in the first feature information to obtain a second initial representation; the second initial representation is processed by the first fully connected neural network 32 of the transformation neural network model 3 to obtain the application embedded vector of the current device Through the first fully connected neural network 32, the information can be reduced to the bottleneck layer to obtain the application embedded vector of the current device

[0107] Among them, when the first initial representation is intercepted according to the position of the low-dimensional dense vector of the application installation image in the first feature information, data at the same position as the position of the low-dimensional dense vector of the application installation image in the first feature information in the first initial representation can be intercepted to obtain the second initial representation.

[0108] The first fully connected neural network may have one layer or more layers, and the present disclosure does not specifically limit the specific number of layers of the first fully connected neural network.

[0109] In step S240, push information of the current device is obtained according to the application embedded vector of the current device.

[0110] In step S250, the push information of the current device is pushed to the current device.

[0111] Figure 6 The flowchart of the information push method based on application usage records according to an embodiment of the present disclosure is schematically shown.

[0112] like Figure 6 As shown, the information push method based on application usage records provided by the embodiment of the present disclosure may include the following steps.

[0113] In step S610, historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of the historical device are obtained.

[0114] In the disclosed embodiment, the historical device may be, for example, terminal device 101 (or terminal device 102 or 103). The historical device may be one (for example, terminal device 101) or multiple (for example, terminal devices 101, 102, 103). Preferably, historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of hundreds of millions of historical devices may be obtained. The representation of the historical application installation image information is similar to that of the application installation image information, and may be, for example, represented as shown in formula (1). The historical application installation sequence information may include a historical application installation object sequence and a historical application installation time sequence. The representation of the historical application installation object sequence and the historical application installation time sequence may be similar to that of the application installation object sequence and the application installation time sequence, and may be, for example, represented as shown in formula (2).

[0115] In step S620, the historical application installation image information is encoded by the second fully connected neural network of the unsupervised neural network model to obtain a low-dimensional dense vector of the predicted application installation image of the historical device.

[0116] In the present disclosure, Figure 3 As shown, the historical application installation image information can be encoded by the second fully connected neural network 21 of the unsupervised neural network model 2 to obtain the low-dimensional dense vector of the predicted application installation image of the historical device. Among them, the second fully connected neural network 22 encodes the historical application installation image information through formula (4) to obtain the low-dimensional dense vector of the predicted application installation image of the historical device.

[0117] In step S630, the predicted application installation image low-dimensional dense vector, historical application installation sequence information and historical application uninstallation sequence information are encoded by transforming the encoder of the neural network model and the first fully connected neural network to obtain the predicted application embedded vector of the historical device.

[0118] In the present disclosure, Figure 3 As shown, the encoder 31 of the transformation neural network model 3 and the first fully connected neural network 32 can be used to encode the low-dimensional dense vector of the predicted application installation image, the historical application installation sequence information and the historical application uninstallation sequence information to obtain the predicted application embedded vector of the historical device.

[0119] In step S640, the parameters of the transformation neural network model are adjusted according to the predicted application embedded vector, the historical application installation sequence information, and the historical application uninstallation sequence information to obtain a trained transformation neural network model.

[0120] In the disclosed embodiment, a first loss function can be generated based on the predicted application embedded vector, historical application installation sequence information, and historical application uninstallation sequence information, so as to adjust the parameters of the transformation neural network model according to the first loss function to obtain a trained transformation neural network model.

[0121] In step S650, the parameters of the unsupervised neural network model are adjusted according to the predicted application embedded vector and the historical application installation image information to obtain a trained unsupervised neural network model.

[0122] In the disclosed embodiment, a second loss function may be generated based on the predicted application embedded vector and the historical application installation image information, so as to adjust the parameters of the unsupervised neural network model according to the second loss function to obtain a trained unsupervised neural network model.

[0123] In an exemplary embodiment, the unsupervised neural network model may further include a third fully connected neural network; wherein the prediction application embedded vector may be processed by the third fully connected neural network of the unsupervised neural network model to obtain first prediction information; a first loss function may be generated according to the historical application installation image information and the first prediction information; and the parameters of the unsupervised neural network model may be adjusted according to the first loss function to obtain a trained unsupervised neural network model. Figure 3 As shown, the unsupervised neural network model 2 also includes a third fully connected neural network 22. The process of processing the prediction application embedded vector by the third fully connected neural network 22 of the unsupervised neural network model 2 to obtain the first prediction information can be expressed as follows:

[0124]

[0125] x (4) =Leaky_ReLU(x (3) W (4) +b (4) ) (8)

[0126] in, Apply the embedded vector for prediction. W (3) , W (4) are the weights of the first fully connected layer and the second fully connected layer of the third fully connected neural network 22, and b (3) , b (4) are respectively the offset vectors of the first fully connected layer and the second fully connected layer of the second fully connected neural network 22. This is only an example, and the present disclosure does not specifically limit the specific number of layers of the third fully connected layer.

[0127] The calculation method of the first loss function can be as follows:

[0128]

[0129] Where D is the number of historical devices. u Install the image information for the history application of the history device u. (4) The first prediction information.

[0130] In step S210, application installation image information, application installation sequence information, and application uninstallation sequence information of the current device are obtained.

[0131] In step S220, the application installation image information is encoded using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device.

[0132] In step S230, the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information are fused using a transformation neural network model to obtain an application embedded vector of the current device.

[0133] In step S240, push information of the current device is obtained according to the application embedded vector of the current device.

[0134] In step S250, the push information of the current device is pushed to the current device.

[0135] Figure 7 The flowchart of the information push method based on application usage records according to an embodiment of the present disclosure is schematically shown.

[0136] like Figure 7 As shown, the transformation neural network model also includes a decoder, and the information push method based on application usage records provided by the embodiment of the present disclosure may include the following steps.

[0137] In step S610, historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of the historical device are obtained.

[0138] In step S620, the historical application installation image information is encoded by a second fully connected neural network of the unsupervised neural network model to obtain a low-dimensional dense vector of the predicted application installation image of the historical device.

[0139] In step S630, the predicted application installation image low-dimensional dense vector, historical application installation sequence information and historical application uninstallation sequence information are encoded by transforming the encoder of the neural network model and the first fully connected neural network to obtain the predicted application embedded vector of the historical device.

[0140] In step S641, application installation query information is generated according to historical application installation sequence information.

[0141] The historical application installation type sequence may be the same as the application installation type sequence, and the dimension of the historical application installation type sequence is the same as the dimension of the historical application installation sequence information. The historical application installation type sequence embedded vector may be as shown in formula (5).

[0142] In step S642, application uninstall query information is generated according to the historical application uninstall sequence information.

[0143] The historical application uninstall type sequence may be the same as the application uninstall type sequence, and the dimension of the historical application uninstall type sequence is the same as the dimension of the historical application uninstall sequence information. The historical application uninstall type sequence embedded vector may be as shown in formula (6).

[0144] In step S643, second feature information is obtained according to the predicted application embedded vector, the application installation query information, and the application uninstallation query information.

[0145] In the disclosed embodiment, the predicted application embedded vector, application installation query information, and application uninstall query information may be concatenated to obtain the second feature information. Fig.12 The structure diagram of the transformation neural network model according to an embodiment of the present disclosure is schematically shown. Fig.12 As shown, the transformation neural network model may be, for example, a transformer model structure. The transformer model structure includes an encoder and a decoder. In the encoder or decoder, the scaled dot-product attention mechanism determines the weight distribution of key values ​​by the degree of similarity between queries and keys. The disclosed embodiment splices the application installation image information on the key and key value of the encoder, and splices the embedded vector of the current device (or historical device) on the key and key value of the decoder, and can deeply fuse the application installation image information, the embedded vector of the current device (or historical device), the application installation sequence information (or historical application installation sequence information) and the application uninstallation sequence information (or historical application uninstallation sequence information), thereby improving the quality of the produced application embedded vectors.

[0146] Figure 8 FIG. 2 is a schematic diagram showing the structure of the second characteristic information according to an embodiment of the present disclosure. Figure 8As shown, in an exemplary embodiment, historical application installation sequence information may include a historical application installation time sequence and a historical application installation type sequence; wherein the historical application installation type sequence may be embeddedly represented to obtain a historical application installation type sequence embedded vector; the historical application installation time sequence may be embeddedly represented to obtain a historical application installation time sequence embedded vector; the historical application installation type sequence embedded representation and the historical application installation time sequence embedded vector are bitwise added to obtain application installation query information.

[0147] like Figure 8 As shown, the historical application installation time series The embedded vector can be The embedded vector of the historical application installation type sequence can be [w n , w n , w n ]. You can install type sequence embedded vectors [w n , w n , w n ], Historical application installation time series embedded vector Perform bitwise addition to obtain application installation query information [e′ n1 , e′ n2 , e′ n3 ].

[0148] In an exemplary embodiment, the historical application uninstall sequence information may include a historical application uninstall time series and a historical application uninstall type series; wherein the historical application uninstall type series may be embeddedly represented to obtain a historical application uninstall type series embedded vector; the historical application uninstall time series may be embeddedly represented to obtain a historical application uninstall time series embedded vector; the historical application uninstall type series embedded representation and the historical application uninstall time series embedded vector are bitwise added to obtain application uninstall query information.

[0149] like Figure 8 As shown, the historical application uninstall time series The embedded vector can be The embedded vector of the historical application uninstallation type sequence can be [w l , w l , w l ]. The embedded vector of type sequence [w l , w l , w l ], Historical application uninstall time series embedded vector Perform bitwise addition to obtain application uninstall query information [e′ l1 , e′ l2 , e′ l3 ].

[0150] Among them, for the historical device u, the embedded vector can be applied to the prediction Application installation query information[e′ n1 , e′ n2 , e′ n3 ]、Application uninstall query information[e′ l1 , e′ l2 , e′ l3 ] are spliced ​​to obtain the second feature information.

[0151] In step S644, the second feature information is processed by the decoder of the transformation neural network model to obtain second prediction information.

[0152] like Figure 3 As shown, the decoder 33 of the transformation neural network model 3 may be composed of a plurality of stacked network blocks 331, each of which is composed of a multi-head self-attention mechanism (Multi-Head Self-Attention) and a feedforward neural network (FeedForward Network). Each network block 331 may include a multi-head self-attention mechanism 3311 and a feedforward neural network 3312. Among them, each network block 331 may also include a residual connection and a layer 3313 for normalization (Add&Normalize), which can better optimize the deep network.

[0153] In step S645, a second loss function is generated according to the second prediction information, the historical application installation sequence information, and the historical application uninstallation sequence information.

[0154] In the disclosed embodiment, the second prediction information may be a reconstruction of the historical application installation sequence information and the historical application uninstallation sequence information by the decoder, wherein the second prediction information may be a prediction value, and the historical application installation sequence information and the historical application uninstallation sequence information may be an expected value to construct a second loss function. The second loss function may be, for example, but not limited to, a cross entropy loss function.

[0155] In step S646, the parameters of the transformation neural network model are adjusted according to the second loss function to obtain a trained transformation neural network model.

[0156] In step S650, the parameters of the unsupervised neural network model are adjusted according to the predicted application embedded vector and the historical application installation image information to obtain a trained unsupervised neural network model.

[0157] In step S210, application installation image information, application installation sequence information, and application uninstallation sequence information of the current device are obtained.

[0158] In step S220, the application installation image information is encoded using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device.

[0159] In step S230, the low-dimensional dense vector of the application installation image, the application installation sequence information and the application uninstallation sequence information are fused using a transformation neural network model to obtain an application embedded vector of the current device.

[0160] In step S240, push information of the current device is obtained according to the application embedded vector of the current device.

[0161] In step S250, the push information of the current device is pushed to the current device.

[0162] According to the information push method based on application usage records of the embodiment of the present disclosure, the application installation image information is encoded through an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device, and the low-dimensional dense vector of the application installation image and the application installation sequence information and the application uninstall sequence information are jointly processed through a transformation neural network model, so that the application installation image information and the application installation sequence information and the application uninstall sequence information can be compressed and deeply integrated to obtain a more accurate application embedded vector of the current device. It is possible to compress as much object information as possible in the output layer and generate an application embedded vector of the current device with higher quality. It is possible to generate accurate push information based on the application embedded vector of the current device and push it to the current device, thereby improving the user experience.

[0163] Fig. 9 The flowchart of the information push method based on application usage records according to another embodiment of the present disclosure is schematically shown. Fig. 9 As shown, the information push method based on application usage records in the embodiment of the present disclosure may include the following steps.

[0164] In step S910, historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of the historical device are obtained.

[0165] In step S920, a transformation neural network model and an unsupervised neural network model are trained according to the historical application installation image information, the historical application installation sequence information, and the historical application uninstallation sequence information.

[0166] In the disclosed embodiment, the training process of the transformation neural network model and the unsupervised neural network model may adopt steps similar to steps S610 to S650 and steps S641 to S646.

[0167] In step S930, the application installation image information, application installation sequence information, and application uninstallation sequence information of the current device are pre-processed.

[0168] In step S940, the transformation neural network model and the unsupervised neural network model are called to process the application installation image information, the application installation sequence information and the application uninstallation sequence information of the current device to obtain the application embedded vector of the current device.

[0169] In step S950, a calling instruction of a downstream task is received through an interface, and an application embedded vector in response to the calling instruction is obtained by transforming the application installation image information, application installation sequence information and application uninstallation sequence information in the calling instruction through a transformation neural network model and an unsupervised neural network model.

[0170] The downstream task may be generated by built-in software in the device end, for example. The built-in software may generate push information according to the application embedded vector generated in step S950, and display the push information in the device end. Fig.10 , 11 The following schematically shows a display interface diagram of push information according to an embodiment of the present disclosure.

[0171] The following describes an apparatus embodiment of the present disclosure, which can be used to execute the information push method based on application usage records of the present disclosure. For details not disclosed in the apparatus embodiment of the present disclosure, please refer to the embodiment of the information push method based on application usage records of the present disclosure.

[0172] Fig.14 The block diagram schematically shows an information push device based on application usage records according to an embodiment of the present disclosure.

[0173] Reference Fig.14 As shown, according to an embodiment of the present disclosure, an information push device 1400 based on application usage records may include: a device data acquisition module 1410, a mirror information processing module 1420, an embedded vector generation module 1430, a push information generation module 1440 and a push information sending module 1450.

[0174] The device data acquisition module 1410 may be configured to acquire application installation image information, application installation sequence information, and application uninstallation sequence information of the current device.

[0175] The image information processing module 1420 may be configured to encode the application installation image information using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device.

[0176] The embedded vector generation module 1430 can be configured to use a transformation neural network model to fuse the low-dimensional dense vector of the application installation image, the application installation sequence information, and the application uninstallation sequence information to obtain the application embedded vector of the current device.

[0177] The push information generation module 1440 may be configured to generate push information for the current device according to the application embedded vector of the current device.

[0178] The push information sending module 1450 may be configured to push the push information of the current device to the current device.

[0179] In an exemplary embodiment, the embedded vector generation module 1430 may include an application installation feature sequence generation unit, an application uninstallation feature sequence generation unit, a first feature generation unit, and an embedded vector generation unit. Among them, the application installation feature sequence generation unit may be configured to generate an application installation feature sequence according to the application installation sequence information; the application uninstallation feature sequence generation unit may be configured to generate an application uninstallation feature sequence according to the application uninstallation sequence information; the first feature generation unit may be configured to obtain the first feature information according to the application installation image low-dimensional dense vector, the application installation feature sequence, and the application uninstallation feature sequence; the embedded vector generation unit may be configured to encode the first feature information through the encoder of the transformation neural network model to obtain the application embedded vector of the current device.

[0180] In an exemplary embodiment, the application installation sequence information may include an application installation object sequence, an application installation time sequence, and an application installation type sequence; wherein the application installation feature sequence generation unit may include an installation type embedded vector subunit, an installation object embedded vector subunit, an installation time embedded vector subunit, and an application installation feature sequence generation subunit. The installation type embedded vector subunit may be configured to perform an embedded representation on the application installation type sequence to obtain an application installation type sequence embedded vector; the installation object embedded vector subunit may be configured to perform an embedded representation on the application installation object sequence to obtain an application installation object sequence embedded vector; the installation time embedded vector subunit may be configured to perform an embedded representation on the application installation time sequence to obtain an application installation time sequence embedded vector; and the application installation feature sequence generation subunit may be configured to perform bitwise addition on the application installation type sequence embedded vector, the application installation object sequence embedded vector, and the application installation time sequence embedded vector to obtain an application installation feature sequence.

[0181] In an exemplary embodiment, the application uninstall sequence information may include an application uninstall object sequence, an application uninstall time sequence, and an application uninstall type sequence; wherein the application uninstall feature sequence generation unit may include an uninstall type embedded vector subunit, an uninstall object embedded vector subunit, an uninstall time embedded vector subunit, and an application uninstall feature sequence generation subunit. The uninstall type embedded vector subunit may be configured to perform an embedded representation on the application uninstall type sequence to obtain an application uninstall type sequence embedded vector; the uninstall object embedded vector subunit may be configured to perform an embedded representation on the application uninstall object sequence to obtain an application uninstall object sequence embedded vector; the uninstall time embedded vector subunit may be configured to perform an embedded representation on the application uninstall time sequence to obtain an application uninstall time sequence embedded vector; the application uninstall feature sequence generation subunit may be configured to perform bitwise addition on the application uninstall type sequence embedded vector, the application uninstall time sequence embedded vector, and the application uninstall object sequence embedded vector to obtain an application uninstall feature sequence.

[0182] In an exemplary embodiment, the application installation image information may further include an installation image type sequence; wherein the first feature generation unit may include an installation image type embedded vector subunit, an installation image feature sequence generation subunit, and a first feature generation subunit. The installation image type embedded vector subunit may be configured to perform an embedded representation on the installation image type sequence to obtain an installation image type sequence embedded vector; the installation image feature sequence generation subunit may be configured to perform bitwise addition of the installation image type embedded vector and the application installation image low-dimensional dense vector to obtain an application installation image feature sequence; the first feature generation subunit may be configured to obtain the first feature information according to the application installation image feature sequence, the application installation feature sequence, and the application uninstall feature sequence.

[0183] In an exemplary embodiment, the transformation neural network model may further include a first fully connected neural network; wherein the embedded vector generation unit includes a first initial representation generation subunit, a second initial representation generation subunit and an embedded vector generation subunit. The first initial representation generation subunit may be configured to process the first feature information through the encoder of the transformation neural network model to obtain a first initial representation; the second initial representation generation subunit may be configured to intercept the first initial representation according to the position of the low-dimensional dense vector of the application installation image in the first feature information to obtain a second initial representation; the embedded vector generation subunit may be configured to process the second initial representation through the first fully connected neural network of the transformation neural network model to obtain an application embedded vector of the current device.

[0184] In an exemplary embodiment, the object data processing device 1400 may also include a historical data acquisition module, a historical image information processing module, a historical application embedded vector generation module, a transformation neural network model training module, and an unsupervised neural network model training module. Among them, the historical data acquisition module can be configured to obtain historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of historical devices; the historical image information processing module can be configured to process the historical application installation image information through the second fully connected neural network of the unsupervised neural network model to obtain the predicted application installation image low-dimensional dense vector of the historical device; the historical application embedded vector generation module can be configured to process the predicted application installation image low-dimensional dense vector, historical application installation sequence information, and historical application uninstallation sequence information through the encoder of the transformation neural network model and the first fully connected neural network to obtain the predicted application embedded vector of the historical device; the transformation neural network model training module can be configured to adjust the parameters of the transformation neural network model according to the predicted application embedded vector, the historical application installation sequence information, and the historical application uninstallation sequence information to obtain the trained transformation neural network model; the unsupervised neural network model training module can be configured to adjust the parameters of the unsupervised neural network model according to the predicted application embedded vector and the historical application installation image information to obtain the trained unsupervised neural network model.

[0185] In an exemplary embodiment, the unsupervised neural network model may further include a third fully connected neural network; wherein the unsupervised neural network model training module may include a first prediction information generation unit, a first loss function generation unit, and an unsupervised neural network model training unit. The first prediction information generation unit may be configured to process the prediction application embedded vector through the third fully connected neural network of the unsupervised neural network model to obtain the first prediction information; the first loss function generation unit may be configured to generate a first loss function based on the historical application installation image information and the first prediction information; the unsupervised neural network model training unit may be configured to adjust the parameters of the unsupervised neural network model according to the first loss function to obtain a trained unsupervised neural network model.

[0186] In an exemplary embodiment, the transformation neural network model may further include a decoder; the transformation neural network model training module may include an application installation query information generation unit, an application uninstallation query information generation unit, a second feature information generation unit, a second prediction information generation unit, a second loss function generation unit and a transformation neural network model training unit. The application installation query information generation unit may be configured to generate application installation query information based on historical application installation sequence information; the application uninstallation query information generation unit may be configured to generate application uninstallation query information based on historical application uninstallation sequence information; the second feature information generation unit may be configured to obtain second feature information based on the predicted application embedded vector, application installation query information, and application uninstallation query information; the second prediction information generation unit may be configured to process the second feature information through the decoder of the transformation neural network model to obtain second prediction information; the second loss function generation unit may be configured to generate a second loss function based on the second prediction information, historical application installation sequence information, and historical application uninstallation sequence information; the transformation neural network model training unit may be configured to adjust the parameters of the transformation neural network model based on the second loss function to obtain a trained transformation neural network model.

[0187] In an exemplary embodiment, the historical application installation sequence information may include a historical application installation time sequence and a historical application installation type sequence; wherein the application installation query information generation unit may include a historical installation type embedded vector subunit, a historical installation time embedded vector subunit, and an application installation query information generation subunit. The historical installation type embedded vector subunit may be configured to perform an embedded representation on the historical application installation type sequence to obtain a historical application installation type sequence embedded vector; the historical installation time embedded vector subunit may be configured to perform an embedded representation on the historical application installation time sequence to obtain a historical application installation time sequence embedded vector; the application installation query information generation subunit may be configured to perform bitwise addition of the historical application installation type sequence embedded vector and the historical application installation time sequence embedded vector to obtain application installation query information.

[0188] In an exemplary embodiment, the historical application uninstall sequence information may include a historical application uninstall time sequence and a historical application uninstall type sequence; wherein the application uninstall query information generation unit may include: a historical uninstall type embedded vector subunit, a historical uninstall time embedded vector subunit, and an application uninstall query information generation subunit. The historical uninstall type embedded vector subunit may be configured to perform an embedded representation on the historical application uninstall type sequence to obtain a historical application uninstall type sequence embedded vector; the historical uninstall time embedded vector subunit may be configured to perform an embedded representation on the historical application uninstall time sequence to obtain a historical application uninstall time sequence embedded vector; the application uninstall query information generation subunit may be configured to perform bitwise addition on the historical application uninstall type sequence embedded vector and the historical application uninstall time sequence embedded vector to obtain the application uninstall query information.

[0189] In some embodiments of the present disclosure, the object data processing device provided encodes the application installation image information to obtain the low-dimensional dense vector of the application installation image of the current device, and fuses the low-dimensional dense vector of the application installation image with the application installation sequence information and the application uninstallation sequence information, so that the application installation image information, the application installation sequence information and the application uninstallation sequence information can be compressed and deeply fused to obtain a more accurate application embedded vector of the current device. Accurate push information can be generated according to the application embedded vector of the current device and pushed to the current device, thereby improving the user experience.

[0190] Fig.15 The structure diagram of the computer system suitable for implementing the electronic device of the embodiment of the present disclosure is shown. It should be noted that: Fig.15 The computer system 1500 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0191] like Fig.15 As shown, the computer system 1500 includes a central processing unit (CPU) 1501, which can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 1502 or the program loaded from the storage part 1508 to the random access memory (RAM) 1503. In the RAM 1503, various programs and data required for system operation are also stored. The CPU 1501, the ROM 1502, and the RAM 1503 are connected to each other through a bus 1504. An input / output (I / O) interface 1505 is also connected to the bus 1504.

[0192] The following components are connected to the I / O interface 1505: an input section 1506 including a keyboard, a mouse, etc.; an output section 1507 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 1508 including a hard disk, etc.; and a communication section 1509 including a network interface card such as a LAN card, a modem, etc. The communication section 1509 performs communication processing via a network such as the Internet. A drive 1510 is also connected to the I / O interface 1505 as needed. A removable medium 1511, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 1510 as needed, so that a computer program read therefrom is installed into the storage section 1508 as needed.

[0193] In particular, according to an embodiment of the present disclosure, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part 1509, and / or installed from a removable medium 1511. When the computer program is executed by a central processing unit (CPU) 1501, various functions defined in the system of the present application are executed.

[0194] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0195] The flow chart and block diagram in the accompanying drawings illustrate the possible implementation of the system, method and computer program product according to various embodiments of the present disclosure, namely, architecture, function and operation. In this regard, each box in the flow chart or block diagram can represent a module, program segment, or a part of code, and the above-mentioned module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the box can also occur in a sequence different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0196] The modules and / or units and / or sub-units involved in the embodiments described in the present disclosure may be implemented by software or hardware, and the modules and / or units and / or sub-units described may also be set in a processor. The names of these modules and / or units and / or sub-units do not, in some cases, constitute limitations on the modules and / or units and / or sub-units themselves.

[0197] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being installed in the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the following embodiment. For example, the electronic device may implement the following method. Figure 2 or Figure 4 or Figure 6 or Figure 7 or Fig. 9 The steps shown.

[0198] It should be noted that, although several modules or units or subunits of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units or subunits described above can be embodied in one module or unit or subunit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units or subunits to be embodied.

[0199] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0200] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0201] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for pushing information based on application usage records, characterized in that: include: Get the application installation image information, application installation sequence information and application uninstallation sequence information of the current device; The application installation image information includes an installation image type sequence; Encoding the application installation image information by using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device; Generate an application installation feature sequence according to the application installation sequence information; Generate an application uninstall feature sequence according to the application uninstall sequence information; Performing an embedded representation on the installation image type sequence to obtain an installation image type sequence embedded vector; Add the installation image type sequence embedded vector and the application installation image low-dimensional dense vector bit by bit to obtain an application installation image feature sequence; Obtaining first feature information according to the application installation image feature sequence, the application installation feature sequence, and the application uninstallation feature sequence; Encoding the first feature information by transforming an encoder of a neural network model to obtain an application embedded vector of the current device; Obtaining push information of the current device according to the application embedded vector of the current device; The push information of the current device is pushed to the current device.

2. The method according to claim 1, characterized in that The application installation sequence information includes an application installation object sequence, an application installation time sequence, and an application installation type sequence; wherein generating an application installation feature sequence according to the application installation sequence information includes: Performing embedded representation on the application installation type sequence to obtain an application installation type sequence embedded vector; Performing an embedded representation on the application installation object sequence to obtain an embedded vector of the application installation object sequence; Performing an embedded representation on the application installation time series to obtain an embedded vector of the application installation time series; The application installation type sequence embedded vector, the application installation object sequence embedded vector, and the application installation time sequence embedded vector are bitwise added to obtain the application installation feature sequence.

3. The method according to claim 1, characterized in that The application uninstall sequence information includes an application uninstall object sequence, an application uninstall time sequence, and an application uninstall type sequence; wherein generating an application uninstall feature sequence according to the application uninstall sequence information includes: Performing an embedded representation on the application uninstallation type sequence to obtain an application uninstallation type sequence embedded vector; Performing embedded representation on the application uninstallation object sequence to obtain an embedded vector of the application uninstallation object sequence; Performing an embedded representation on the application uninstallation time series to obtain an embedded vector of the application uninstallation time series; The application uninstallation type sequence embedded vector, the application uninstallation time sequence embedded vector, and the application uninstallation object sequence embedded vector are bitwise added to obtain the application uninstallation feature sequence.

4. The method according to claim 1, characterized in that The transformation neural network model also includes a first fully connected neural network; wherein encoding the first feature information through an encoder of the transformation neural network model to obtain the application embedded vector of the current device includes: Encoding the first feature information through the encoder of the transformation neural network model to obtain a first initial representation; intercepting the first initial representation according to the position of the low-dimensional dense vector of the application installation image in the first feature information to obtain a second initial representation; The second initial representation is subjected to dimensionality reduction processing by using the first fully connected neural network to obtain the application embedded vector of the current device.

5. The method according to claim 1, characterized in that Before obtaining the application installation image information, application installation sequence information, and application uninstallation sequence information of the current device, the method further includes: Obtain historical application installation image information, historical application installation sequence information, and historical application uninstallation sequence information of historical devices; Encoding the historical application installation image information through a second fully connected neural network of the unsupervised neural network model to obtain a low-dimensional dense vector of the predicted application installation image of the historical device; The predicted application installation image low-dimensional dense vector, the historical application installation sequence information, and the historical application uninstallation sequence information are encoded by the encoder of the transformation neural network model and the first fully connected neural network to obtain the predicted application embedded vector of the historical device; Adjusting the parameters of the transformation neural network model according to the predicted application embedded vector, the historical application installation sequence information, and the historical application uninstallation sequence information to obtain the trained transformation neural network model; The parameters of the unsupervised neural network model are adjusted according to the predicted application embedded vector and the historical application installation image information to obtain the trained unsupervised neural network model.

6. The method according to claim 5, characterized in that The unsupervised neural network model further includes a third fully connected neural network; wherein, adjusting the parameters of the unsupervised neural network model according to the predicted application embedded vector and the historical application installation image information to obtain the trained unsupervised neural network model includes: Reconstructing the prediction application embedded vector through the third fully connected neural network of the unsupervised neural network model to obtain first prediction information; generating a first loss function according to the historical application installation image information and the first prediction information; The parameters of the unsupervised neural network model are adjusted according to the first loss function to obtain the trained unsupervised neural network model.

7. The method according to claim 5, characterized in that The transformation neural network model also includes a decoder; wherein, adjusting the parameters of the transformation neural network model according to the predicted application embedded vector, the historical application installation sequence information and the historical application uninstallation sequence information to obtain the trained transformation neural network model includes: Generate application installation query information according to the historical application installation sequence information; Generate application uninstall query information according to the historical application uninstall sequence information; Obtaining second feature information according to the predicted application embedded vector, the application installation query information, and the application uninstallation query information; Reconstructing the second feature information through the decoder of the transformation neural network model to obtain second prediction information; generating a second loss function according to the second prediction information, the historical application installation sequence information, and the historical application uninstallation sequence information; The parameters of the transformation neural network model are adjusted according to the second loss function to obtain the trained transformation neural network model.

8. The method according to claim 7, characterized in that The historical application installation sequence information includes a historical application installation time sequence and a historical application installation type sequence; wherein generating application installation query information according to the historical application installation sequence information includes: Performing an embedded representation on the historical application installation type sequence to obtain an embedded vector of the historical application installation type sequence; Performing an embedded representation on the historical application installation time series to obtain an embedded vector of the historical application installation time series; The historical application installation type sequence embedded vector and the historical application installation time sequence embedded vector are bitwise added to obtain the application installation query information.

9. The method according to claim 7, characterized in that The historical application uninstallation sequence information includes a historical application uninstallation time sequence and a historical application uninstallation type sequence; wherein generating application uninstallation query information according to the historical application uninstallation sequence information includes: Performing embedded representation on the historical application uninstallation type sequence to obtain an embedded vector of the historical application uninstallation type sequence; Performing an embedded representation on the historical application uninstallation time series to obtain an embedded vector of the historical application uninstallation time series; The historical application uninstallation type sequence embedded vector and the historical application uninstallation time sequence embedded vector are bitwise added to obtain the application uninstallation query information.

10. An information push device based on application usage records, characterized in that: include: A device data acquisition module is configured to acquire application installation image information, application installation sequence information, and application uninstallation sequence information of the current device; The application installation image information includes an installation image type sequence; An image information processing module, configured to encode the application installation image information using an unsupervised neural network model to obtain a low-dimensional dense vector of the application installation image of the current device; an embedded vector generation module, configured to generate an application installation feature sequence according to the application installation sequence information; Generate an application uninstall feature sequence according to the application uninstall sequence information; Performing an embedded representation on the installation image type sequence to obtain an installation image type sequence embedded vector; Add the installation image type sequence embedded vector and the application installation image low-dimensional dense vector bit by bit to obtain an application installation image feature sequence; Obtaining first feature information according to the application installation image feature sequence, the application installation feature sequence, and the application uninstallation feature sequence; Encoding the first feature information by transforming an encoder of a neural network model to obtain an application embedded vector of the current device; A push information generation module, configured to obtain push information of the current device according to the application embedded vector of the current device; The push information sending module is configured to push the push information of the current device to the current device.

11. An electronic device, characterized in that: include: one or more processors; A storage device for storing one or more programs; When one or more programs are executed by one or more processors, the one or more processors implement the method according to any one of claims 1 to 9.

12. A computer readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 9 is implemented.