Information Push Method and Its Device, Storage Medium, Program Product

Abstract

The present invention discloses an information push method and its device, storage medium, and program product. The method first determines the social objects of the target object, constructs a first heterogeneous network and a second heterogeneous network based on the target object and its social objects, and obtains a first object vector, multiple first push information vectors, a second object vector, and multiple second push information vectors based on the first heterogeneous network and the second heterogeneous network. The first object vector, the second object vector, the multiple first push information vectors, and the multiple second push information vectors are then input into a feature extraction model to obtain an object feature vector and multiple push information feature vectors. The target push information is then determined and pushed to the target object based on the object feature vector and all push information feature vectors. The embodiment of the present invention can improve the recall accuracy of push information and enhance the recall effect of push information. The present invention can be applied to fields such as cloud technology, cloud social networking, smart transportation, and assisted driving.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to an information push method and its device, storage medium, and program product. Background Art

[0002] With the development of internet technology, the internet has become a crucial means for users to obtain information. For example, users can access various push notifications from information providers through internet platforms. To accurately deliver appropriate push notifications to users, a common approach is to recall previously sent push notifications based on the user's historical push notifications. However, compared to the vast amount of push notifications in databases, the amount of previously sent push notifications is very limited. Recalling previously sent push notifications based solely on the user's historical push notifications can lead to low recall accuracy and poor recall effectiveness due to data scarcity. Summary of the Invention

[0003] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0004] The embodiments of the present invention provide an information push method and apparatus, a storage medium, and a program product thereof, which can improve the accuracy of recalling pushed information and enhance the recall effect of pushed information.

[0005] In one aspect, an embodiment of the present invention provides an information push method, comprising the following steps:

[0006] Identify the target audience's social network;

[0007] Constructing a first heterogeneous network and a second heterogeneous network based on the target object and the social object, wherein the first heterogeneous network includes the target object, the social object, and a plurality of first push messages, the first push messages having no interactive relationship with the target object or the social object, and the second heterogeneous network includes the target object, the social object, and a plurality of second push messages, the second push messages having an interactive relationship with the target object or the social object;

[0008] Acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network, wherein the first object vector is a vectorization result of the target object in the first heterogeneous network, and the first push information vector is a vectorization result of the first push information;

[0009] Acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network, wherein the second object vector is a vectorization result of the target object in the second heterogeneous network, and the second push information vector is a vectorization result of the second push information;

[0010] Inputting the first object vector, the second object vector, a plurality of the first push information vectors, and a plurality of the second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and a plurality of push information feature vectors;

[0011] Target push information is determined according to the object feature vector and all the push information feature vectors, and the target push information is pushed to the target object.

[0012] On the other hand, an embodiment of the present invention further provides an information push device, including:

[0013] an object determination unit, configured to determine a social object of a target object;

[0014] a network construction unit, configured to construct a first heterogeneous network and a second heterogeneous network based on the target object and the social object, wherein the first heterogeneous network includes the target object, the social object, and a plurality of first push messages, the first push messages having no interactive relationship with the target object or the social object, and the second heterogeneous network includes the target object, the social object, and a plurality of second push messages, the second push messages having an interactive relationship with the target object or the social object;

[0015] a first vector acquisition unit, configured to acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network, wherein the first object vector is a vectorization result of the target object in the first heterogeneous network, and the first push information vector is a vectorization result of the first push information;

[0016] a second vector acquisition unit, configured to acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network, wherein the second object vector is a vectorization result of the target object in the second heterogeneous network, and the second push information vector is a vectorization result of the second push information;

[0017] a feature extraction unit, configured to input the first object vector, the second object vector, a plurality of the first pushed information vectors, and a plurality of the second pushed information vectors into a feature extraction model for feature extraction, thereby obtaining an object feature vector and a plurality of pushed information feature vectors;

[0018] The information pushing unit is configured to determine target push information according to the object feature vector and all the push information feature vectors, and push the target push information to the target object.

[0019] Optionally, the feature extraction unit is further configured to:

[0020] splicing the first object vector and the second object vector to obtain first splicing information;

[0021] Among the plurality of first push information vectors and the plurality of second push information vectors, splicing the first push information vector and the second push information vector corresponding to the same push information to obtain a plurality of second spliced information;

[0022] Inputting the first splicing information into a feature extraction model to perform feature extraction to obtain an object feature vector;

[0023] The plurality of second splicing information are input into the feature extraction model for feature extraction to obtain a plurality of push information feature vectors.

[0024] Optionally, the network construction unit is further configured to:

[0025] Acquire a plurality of first push information according to the target object and the social object;

[0026] constructing a first heterogeneous network according to the target object, the social object, and the plurality of first push information;

[0027] Acquire a plurality of second push information according to the target object and the social object;

[0028] A second heterogeneous network is constructed according to the target object, the social object, and the plurality of second push information.

[0029] Optionally, the first vector obtaining unit is further configured to:

[0030] Acquire first sequence information in the first heterogeneous network, wherein the first sequence information includes the target object and a plurality of first push information;

[0031] performing vectorization processing on the target object in the first sequence information to obtain the first object vector;

[0032] Vectorization processing is performed on the multiple first push information in the first sequence information to obtain multiple first push information vectors.

[0033] Optionally, the first vector obtaining unit is further configured to:

[0034] Setting a first meta-path according to the first heterogeneous network;

[0035] A random walk is performed in the first heterogeneous network according to the first meta-path to obtain first sequence information.

[0036] Optionally, the second vector obtaining unit is further configured to:

[0037] Acquire second sequence information in the second heterogeneous network, wherein the second sequence information includes the target object and a plurality of second push information;

[0038] performing vectorization processing on the target object in the second sequence information to obtain the second object vector;

[0039] Vectorization processing is performed on the multiple pieces of the second push information in the second sequence information to obtain multiple second push information vectors.

[0040] Optionally, the second vector obtaining unit is further configured to:

[0041] Setting a second meta-path according to the second heterogeneous network;

[0042] A random walk is performed in the second heterogeneous network according to the second meta-path to obtain second sequence information.

[0043] Optionally, the information pushing unit is further configured to:

[0044] Calculating the similarity between the object feature vector and each of the push information feature vectors;

[0045] Target push information is determined according to the similarity.

[0046] Optionally, the information pushing device further includes:

[0047] An information acquisition unit, used to acquire sequence information to be processed;

[0048] an information encoding unit, configured to perform one-hot encoding on the target information in the sequence information to be processed to obtain encoded information;

[0049] A weight acquisition unit, configured to acquire target weight information according to the encoding information;

[0050] A probability prediction unit, configured to perform probability prediction on adjacent information of the target information based on the target weight information to obtain a probability prediction value of the adjacent information;

[0051] a weight correction unit, configured to correct the target weight information when the probability prediction value of the neighboring information does not meet the preset probability value, until the probability prediction value of the neighboring information meets the preset probability value;

[0052] The vector acquisition unit is used to use the corrected target weight information as the vector information of the target information.

[0053] Optionally, the weight acquisition unit is further configured to:

[0054] Initialize the first weight matrix;

[0055] Target weight information is obtained in the first weight matrix according to the encoding information.

[0056] Optionally, the probability prediction unit is further configured to:

[0057] Initialize the second weight matrix;

[0058] Multiplying the target weight information and the second weight matrix to obtain a probability prediction value of each information in the sequence information to be processed;

[0059] The probability prediction value of the adjacent information is determined among the probability prediction values of each of the information.

[0060] On the other hand, an embodiment of the present invention further provides an information push device, including:

[0061] at least one processor;

[0062] at least one memory for storing at least one program;

[0063] When at least one of the programs is executed by at least one of the processors, the information push method described above is implemented.

[0064] On the other hand, an embodiment of the present invention further provides a computer-readable storage medium storing a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to implement the information push method as described above.

[0065] On the other hand, an embodiment of the present invention also provides a computer program product, including a computer program or computer instructions, wherein the computer program or the computer instructions are stored in a computer-readable storage medium, and the processor of a computer device reads the computer program or the computer instructions from the computer-readable storage medium, and the processor executes the computer program or the computer instructions, so that the computer device executes the information push method as described above.

[0066] Embodiments of the present invention include at least the following beneficial effects: by constructing a first heterogeneous network including multiple first push information and a second heterogeneous network including multiple second push information based on the target object and its social object, then obtaining the first object vector of the target object and the first push information vector of the first push information based on the first heterogeneous network, and obtaining the second object vector of the target object and the second push information vector of the second push information based on the second heterogeneous network, then inputting the first object vector, the second object vector, the multiple first push information vectors and the multiple second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors, and then determining the target push information to be pushed to the target object based on the object feature vector and all push information feature vectors. Since a heterogeneous network including push information is constructed by jointly using the target object and its social object, the relationship between different objects and different push information can be obtained through the heterogeneous network, thereby expanding the relationship between the target object and different push information, and further solving the problem of data scarcity, which is conducive to improving the recall accuracy of push information and enhancing the recall effect of push information; in addition, since the first push information has no interactive relationship with the target object or social object, and the second push information has an interactive relationship with the target object or social object, the first heterogeneous network and the second heterogeneous network can fully and accurately express the target object's preference characteristics for push information. Therefore, feature extraction is performed on the first object vector and the first push information vector obtained based on the first heterogeneous network, and the second object vector and the second push information vector obtained based on the second heterogeneous network, so that the object feature vector and the push information feature vector can be obtained more comprehensively and accurately, thereby improving the accuracy of the target push information determined based on the object feature vector and the push information feature vector.

[0067] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the technical solution of the present invention and do not constitute a limitation to the technical solution of the present invention.

[0069] Figure 1 is a schematic diagram of an implementation environment provided by an embodiment of the present invention;

[0070] Figure 2 is a schematic diagram of another implementation environment provided by an embodiment of the present invention;

[0071] Figure 3 is a schematic diagram of another implementation environment provided by an embodiment of the present invention;

[0072] Figure 4 This is a flow chart of an information push method provided by an embodiment of the present invention;

[0073] Figure 5 is a schematic diagram of a first heterogeneous network provided by an embodiment of the present invention;

[0074] Figure 6 is a schematic diagram of a second heterogeneous network provided by an embodiment of the present invention;

[0075] Figure 7 is a schematic diagram of a heterogeneous network provided by an embodiment of the present invention;

[0076] Figure 8 Schematic diagram of a feature extraction model provided by an embodiment of the present invention;

[0077] Figure 9 This is a schematic diagram of a vectorization process provided by an embodiment of the present invention;

[0078] Figure 10 This is a complete flow chart of an information push method provided by an embodiment of the present invention;

[0079] Figure 11 is a schematic diagram of an information push device provided by an embodiment of the present invention;

[0080] Figure 12 It is a schematic diagram of another information pushing device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The described embodiments should not be considered as limiting the present invention, and all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0082] In the following description, reference is made to “some embodiments”, which describes a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0083] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present invention pertains. The terms used herein are for the purpose of describing embodiments of the present invention only and are not intended to limit the present invention.

[0084] Before further explaining the embodiments of the present invention in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are subject to the following interpretations.

[0085] 1) Heterogeneous Network refers to a network formed by the integration of multiple different types of networks. In a heterogeneous network, there is a specific connection relationship between the nodes belonging to different types of networks.

[0086] 2) Meta-path, which is the concatenation of edge types and node types between two nodes. For example, if we use the letter U to represent users and the letter M to represent movies, then UUM represents a meta-path that indicates that one user followed a movie that another user watched.

[0087] 3) Random walk, similar in concept to Brownian motion, is the mathematical ideal of Brownian motion, which is the perpetual, random movement of particles suspended in a liquid or gas. Brownian motion is a random process. The core concept of a random walk is that any conserved quantity carried by a random walker corresponds to a diffusion transport law.

[0088] 4) One-hot encoding, also known as single-bit encoding, uses an N-bit state register to encode N states. Each state has its own independent register bit, and at any time, only one of the bits is valid.

[0089] 5) Cloud technology refers to a hosting technology that unifies hardware, software, and network resources within a wide area network (WAN) or local area network (LAN) to enable data computing, storage, processing, and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, and application technology based on the cloud computing business model. It can form a resource pool that can be used on demand with flexibility and convenience. Cloud computing technology will become a critical support. Backend services for technical network systems, such as video websites, image websites, e-commerce platforms, and more portals, require extensive computing and storage resources. With the rapid development and application of the internet industry, every item will likely have its own unique identification mark in the future, requiring transmission to backend systems for logical processing. Data of varying levels will be processed separately, and data from all industries will require a strong system backend, which can only be achieved through cloud computing.

[0090] 6) Big data refers to data sets that cannot be captured, managed, and processed within a specific timeframe using conventional software tools. These are massive, rapidly growing, and diverse information assets that require new processing models to enhance decision-making, insight discovery, and process optimization. With the advent of the cloud era, big data has attracted increasing attention. Big data requires specialized technologies to efficiently process large amounts of time-sensitive data. Technologies suitable for big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the internet, and scalable storage systems.

[0091] 7) Cloud Social is a virtual social application model that integrates the Internet of Things, cloud computing, and mobile internet. Its goal is to establish a well-known "resource sharing relationship map" to promote online social interaction. The key feature of Cloud Social is to integrate and evaluate a large amount of social resources, forming an effective resource pool to provide services on demand. The more users who participate in sharing, the greater the value created.

[0092] 8) Blockchain is a new application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks generated using cryptographic methods. Each block contains information about a batch of network transactions, used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform's product and service layer, and the application service layer. Blockchains can include public chains, consortium chains, and private chains. Public chains are blockchains where anyone can access the blockchain network at any time to read data, send data, or compete for recordkeeping. Consortium chains are blockchains jointly managed by several organizations or institutions. Private chains are blockchains with a certain degree of centralized control. Write access to the private chain's ledger is controlled by a specific organization or institution, and data access and use are subject to strict permission management.

[0093] 9) Intelligent Traffic System (ITS), also known as Intelligent Transportation System, is the effective and integrated application of advanced science and technology (information technology, computer technology, data communication technology, sensor technology, electronic control technology, automatic control theory, operations research, artificial intelligence, etc.) to transportation, service control and vehicle manufacturing, strengthening the connection between vehicles, roads and users, thereby forming a comprehensive transportation system that ensures safety, improves efficiency, improves the environment and saves energy.

[0094] 10) Intelligent Vehicle Infrastructure Cooperative Systems (IVICS), also known as VICS, are a development direction of Intelligent Transportation Systems (ITS). IVICS utilizes advanced wireless communications and next-generation internet technologies to implement dynamic, real-time information exchange between vehicles and roads. Based on the collection and integration of dynamic traffic information across time and space, IVICS conducts active vehicle safety control and collaborative road management. This fully realizes effective coordination between people, vehicles, and roads, ensuring traffic safety and improving traffic efficiency, thereby creating a safe, efficient, and environmentally friendly road transportation system.

[0095] To accurately push appropriate push information to users, a common approach is to recall previously pushed information based on the user's historical push information interactions. However, compared to the vast amount of push information in a database, the number of previously interacted push information is very limited. Recalling only based on this historical push information can lead to low recall accuracy and poor recall effectiveness due to data scarcity. Research and analysis have found that a user's preferences may be related to the preferences of their social friends. This relationship can be used to infer the user's preferences from their social friends. Based on this, related art has proposed information push methods that integrate the user's social network. Specifically, the user's age, gender, interests, and social information are used as feature information for the user and mapped into a feature vector. The user's historical push information interactions are mapped into a push information feature vector. The similarity between the user's feature vector and the push information feature vector is then calculated. Based on the calculated similarity, the K most similar push information are then recalled. This method enriches the user's feature information by using the user's social information as its feature information, thereby improving recall accuracy and enhancing recall effectiveness. However, this method directly uses the user's social information as its feature information, ignoring the problem that not all social friends' preferences are consistent with the user's preferences. In fact, most users' preferences are not directly related to their social friends. This results in the obtained user's feature vector failing to accurately express the user's preference for pushed information, resulting in problems such as low recall accuracy and poor recall effect.

[0096] In order to improve the recall accuracy of pushed information and enhance the recall effect of pushed information, an embodiment of the present invention provides an information push method, an information push device, a computer-readable storage medium and a computer program product, which construct a first heterogeneous network including multiple first push information and a second heterogeneous network including multiple second push information according to the target object and its social object, and then obtain the first object vector of the target object and the first push information vector of the first push information based on the first heterogeneous network, and obtain the second object vector of the target object and the second push information vector of the second push information based on the second heterogeneous network, and then input the first object vector, the second object vector, the multiple first push information vectors and the multiple second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors, and then determine the target push information to be pushed to the target object based on the object feature vector and all push information feature vectors. Since a heterogeneous network including push information is constructed by jointly using the target object and its social object, the relationship between different objects and different push information can be obtained through the heterogeneous network, thereby expanding the relationship between the target object and different push information, and further solving the problem of data scarcity, which is conducive to improving the recall accuracy of push information and enhancing the recall effect of push information; in addition, since the first push information has no interactive relationship with the target object or social object, and the second push information has an interactive relationship with the target object or social object, the first heterogeneous network and the second heterogeneous network can fully and accurately express the target object's preference characteristics for push information. Therefore, feature extraction is performed on the first object vector and the first push information vector obtained based on the first heterogeneous network, and the second object vector and the second push information vector obtained based on the second heterogeneous network, so that the object feature vector and the push information feature vector can be obtained more comprehensively and accurately, thereby improving the accuracy of the target push information determined based on the object feature vector and the push information feature vector.

[0097] The solutions provided by the embodiments of the present invention involve cloud technology, cloud social networking, big data analysis, information push and other technologies, which are specifically explained through the following embodiments.

[0098] Figure 1 This is a schematic diagram of an implementation environment provided by an embodiment of the present invention. Figure 1 The implementation environment includes a server 101 and a terminal 102, and the server 101 and the terminal 102 are directly or indirectly connected via wired or wireless communication. The server 101 and the terminal 102 can be nodes in the blockchain, which is not specifically limited in this embodiment.

[0099] Server 101 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0100] The server 101 may be integrated with an information push function, or an information push device for implementing the information push function may be deployed in the server 101.

[0101] The server 101 has at least the functions of constructing a first heterogeneous network and a second heterogeneous network based on a target object and its social objects, obtaining an object feature vector and a push information feature vector based on the first heterogeneous network and the second heterogeneous network, and determining target push information based on the object feature vector and the push information feature vector. For example, after determining the target object and its social objects, the server 101 can first construct a first heterogeneous network and a second heterogeneous network based on the target object and its social objects, wherein the first heterogeneous network includes first push information that has no interactive relationship with the target object or the social object, and the second heterogeneous network includes second push information that has an interactive relationship with the target object or the social object. Then, based on the first heterogeneous network, a first object vector and multiple first push information vectors are obtained, and based on the second heterogeneous network, a second object vector and multiple second push information vectors are obtained. Feature extraction is then performed on the first object vector, the second object vector, the first push information vector, and the second push information vector to obtain an object feature vector and multiple push information feature vectors. Then, target push information is determined based on the object feature vectors and the push information feature vectors, and the target push information is pushed to the target object.

[0102] Terminal 102 may include, but is not limited to, a mobile phone, a computer, an intelligent voice interaction device, a smart home appliance, an in-vehicle terminal, an aircraft, etc. Optionally, a user may log in to an internet application platform via terminal 102. The internet application platform may receive push information from server 101 and display or push the push information to the user via terminal 102. The internet application platform may be a social networking platform, an e-commerce platform, a web browsing platform, a payment platform, a content interaction platform, an education platform, or a video sharing platform, etc., and is not specifically limited herein.

[0103] Reference Figure 2As shown, in an application scenario, assuming that a user logs in to a social platform through a terminal 102, in response to detecting that the user has logged in to the social platform, the server 101 first determines the social objects of the user and constructs a first heterogeneous network and a second heterogeneous network based on the user and its social objects, wherein the first heterogeneous network includes first push information that has no interactive relationship with the user or its social objects, and the second heterogeneous network includes second push information that has an interactive relationship with the user or its social objects. Then, the server 101 obtains a first object vector and multiple first push information vectors based on the first heterogeneous network, and obtains a second object vector and multiple second push information vectors based on the second heterogeneous network. Then, the server 101 inputs the first object vector, the second object vector, the multiple first push information vectors, and the multiple second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors. At this time, the server 101 determines the target push information based on the object feature vector and all push information feature vectors, and pushes the target push information to the user through the social platform.

[0104] Reference Figure 3 As shown, in another application scenario, the server 101 first determines each target object and its social object. For each target object, a first heterogeneous network and a second heterogeneous network are constructed based on the current target object and its social object. The first heterogeneous network includes first push information that has no interactive relationship with the current target object or its social object, and the second heterogeneous network includes second push information that has an interactive relationship with the current target object or its social object. Then, the server 101 obtains a first object vector and multiple first push information vectors based on the first heterogeneous network, and obtains a second object vector and multiple second push information vectors based on the second heterogeneous network. Then, the server 101 inputs the first object vector, the second object vector, the multiple first push information vectors, and the multiple second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors. At this time, the server 101 can obtain the object feature vector and multiple push information feature vectors for each target object. In response to the user logging into the social platform through the terminal 102, the server 101 searches for the object feature vector corresponding to the user, determines the target push information based on the object feature vector and all push information feature vectors, and then pushes the target push information to the user through the social platform.

[0105] It should be noted that in various specific embodiments of the present invention, when it comes to the need to perform relevant processing based on data related to the characteristics of the target object, such as the target object attribute information or attribute information set, the permission or consent of the target object will be obtained first, and the collection, use and processing of such data will comply with the relevant laws, regulations and standards of the relevant countries and regions. In addition, when the embodiment of the present invention needs to obtain the attribute information of the target object, it will obtain the separate permission or consent of the target object through a pop-up window or jump to a confirmation page. After clearly obtaining the separate permission or consent of the target object, the necessary target object-related data for the normal operation of the embodiment of the present invention will be obtained.

[0106] Figure 4 This is a flow chart of an information push method provided by an embodiment of the present invention. In this embodiment, the server is used as an example for explanation. Figure 4 The information push method includes but is not limited to steps 110 to 160.

[0107] Step 110: Determine the social objects of the target object.

[0108] In this step, the target object's social objects refer to objects that have social relationships with the target object on internet application platforms such as social platforms, e-commerce platforms, web browsing platforms, payment platforms, content interaction platforms, education platforms, or video sharing platforms. After determining the target object, the corresponding social objects can be determined based on the target object's social relationships.

[0109] It should be noted that there are different ways to determine the social objects of a target object. For example, the server may actively search for each target object and its social objects in a database. In another example, the server may passively trigger the acquisition of the target object's social objects after the target object logs into an Internet application platform.

[0110] In some possible implementations, the server may search for each target object and its social object in the database in advance, so that subsequent steps can perform operations such as building a heterogeneous network, obtaining vector information, and extracting feature vectors based on each target object and its social object in advance. This is beneficial for the server to quickly find the object feature vector corresponding to a target object when the target object logs into the Internet application platform, so that the server can quickly recall the target push information suitable for the target object based on the object feature vector.

[0111] In other possible implementations, the server may determine the social objects of the target object after detecting that the target object has logged into the Internet application platform, so that subsequent steps can directly perform operations such as building a heterogeneous network, obtaining vector information, and extracting feature vectors based on the target object and its social objects, which is beneficial for the server to recall accurate target push information for the target object based on the object feature vector of the target object.

[0112] Step 120: Construct a first heterogeneous network and a second heterogeneous network according to the target object and the social object.

[0113] It should be noted that the first heterogeneous network includes a target object, a social object, and multiple first push messages, and the first push messages have no interactive relationship with the target object or the social object; the second heterogeneous network includes a target object, a social object, and multiple second push messages, and the second push messages have an interactive relationship with the target object or the social object. The interactive relationship can be a click relationship. For example, if the target object or the social object has clicked on a certain push message, it can be considered that the target object or the social object has an interactive relationship with the push message; if the target object or the social object has not clicked on a certain push message, it can be considered that the target object or the social object has no interactive relationship with the push message. When the target object or the social object has an interactive relationship with a certain push message, it means that the target object or the social object is interested in the push message, which is a positive preference; when the target object or the social object has no interactive relationship with a certain push message, it means that the target object or the social object is not interested in the push message, which is a negative preference. Therefore, the first heterogeneous network and the second heterogeneous network constructed based on the target object and the social object can fully and accurately express the target object's preference characteristics for push information, which is conducive to obtaining more comprehensive and accurate object feature vectors and push information feature vectors based on the first heterogeneous network and the second heterogeneous network in subsequent steps, thereby improving the accuracy of the target push information determined based on the object feature vector and the push information feature vector.

[0114] It should be noted that, since different objects correspond to different first push information and second push information, the same push information may exist in the first heterogeneous network and the second heterogeneous network. For example, assume that the first heterogeneous network and the second heterogeneous network include a target object A1 and a social object A2. In the first heterogeneous network, there is no interactive relationship between the target object A1 and the recommended information B1, while in the second heterogeneous network, there is an interactive relationship between the social object A2 and the recommended information B1. Therefore, the first heterogeneous network and the second heterogeneous network have the same recommended information B1.

[0115] In some possible implementations, when constructing a first heterogeneous network based on the target object and the social object, multiple first push information can be obtained based on the target object and the social object first, and then the first heterogeneous network can be constructed based on the target object, the social object and the multiple first push information. When constructing a second heterogeneous network based on the target object and the social object, multiple second push information can be obtained based on the target object and the social object first, and then the second heterogeneous network can be constructed based on the target object, the social object and the multiple second push information. Among them, when obtaining multiple first push information based on the target object and the social object, multiple historical recommendation information that has no interactive relationship with the target object and multiple historical recommendation information that has no interactive relationship with the social object can be first obtained from the database, thereby obtaining multiple first push information; when obtaining multiple second push information based on the target object and the social object, multiple historical recommendation information that has an interactive relationship with the target object and multiple historical recommendation information that has an interactive relationship with the social object can also be first obtained from the database, thereby obtaining multiple second push information. After obtaining the first push information and the second push information, the first heterogeneous network can be constructed based on the first push information, and the second heterogeneous network can be constructed based on the second push information. As Figure 5 and Figure 6 As shown, Figure 5 This is a schematic diagram of the first heterogeneous network in an example. Figure 6 The following is a schematic diagram of the second heterogeneous network in an example. Figure 5 In the first heterogeneous network, there are circular nodes and rectangular nodes. The circular nodes represent objects (including the target object and its social objects), the rectangular nodes represent the first push information, the lines between the circular nodes represent the social relationships between the objects, and the lines between the circular nodes and the rectangular nodes represent the object's preference for the first push information (i.e., negative preference). Therefore, in the first heterogeneous network, not only the social network of the target object can be obtained, but also the recommended information that the target object or its social objects are not interested in can be obtained, which is conducive to the subsequent steps of obtaining more comprehensive and accurate object feature vectors and push information feature vectors based on the first heterogeneous network. Figure 6 In the example, the second heterogeneous network includes circular nodes and hexagonal nodes. The circular nodes represent objects (including the target object and its social objects), the hexagonal nodes represent the second push information, the lines between the circular nodes represent the social relationships between the objects, and the lines between the circular nodes and the hexagonal nodes represent the objects' preference for the second push information (i.e., positive preference). Therefore, in the second heterogeneous network, not only the social network of the target object can be obtained, but also the recommended information of interest to the target object or its social objects can be obtained, which is conducive to the subsequent steps of obtaining more comprehensive and accurate object feature vectors and push information feature vectors based on the second heterogeneous network.

[0116] Step 130: Acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network.

[0117] It should be noted that the first object vector is the vectorization result of the target object in the first heterogeneous network, and the first push information vector is the vectorization result of the first push information.

[0118] In this step, since the first heterogeneous network is obtained in step 120, the first object vector and multiple first push information vectors of the target object in the first heterogeneous network can be obtained based on the first heterogeneous network, so that the object feature vector and push information feature vector of the target object can be obtained according to the first object vector and the first push information vector in subsequent steps, which is conducive to determining accurate target push information based on the object feature vector.

[0119] In some possible implementations, when obtaining a first object vector and multiple first push information vectors based on a first heterogeneous network, first sequence information including a target object and multiple first push information can be first obtained in the first heterogeneous network, and then the target object in the first sequence information is vectorized to obtain the first object vector, and the multiple first push information in the first sequence information are vectorized to obtain the multiple first push information vectors. When obtaining the first sequence information in the first heterogeneous network, a first meta-path can be first set according to the first heterogeneous network, and then a random walk can be performed in the first heterogeneous network according to the first meta-path to obtain the first sequence information including the target object and multiple first push information. In addition, when vectorizing the target object and the first push information in the first sequence information, the target object and the first push information in the first sequence information can be vectorized using a network embedding representation method to obtain a first object vector and a first push information vector, wherein the network embedding representation is a technology that maps nodes in a network to a low-dimensional dense space. By mapping the target object and the first push information in the first sequence information to a low-dimensional dense space, the first object vector and the first push information vector are obtained, so that the first object vector and the first push information vector can retain the inherent network structure and semantic association in the first heterogeneous network, so that when the feature extraction model is used in the subsequent steps to extract features from the first object vector and the first push information vector, more comprehensive and accurate feature information can be extracted, which is conducive to improving the recall accuracy of the target push information. It should be noted that the specific content of the vectorization processing of the target object and the first push information in the first sequence information will be given in the subsequent description.

[0120] It should be noted that, in the process of setting the first meta-path according to the first heterogeneous network, the setting strategy of the first meta-path can be determined based on prior knowledge, or the setting strategy of the first meta-path can be randomly selected, and then the first meta-path can be set in the first heterogeneous network according to the setting strategy. For example, assuming that the setting strategy of the first meta-path determined based on prior knowledge is to select objects that have an interactive relationship with the same recommendation information, then the first meta-path set in the first heterogeneous network according to the setting strategy can be Among them, U represents the object, I represents the recommended information, It indicates that there is an interactive relationship between the object and the recommended information. Therefore, the first meta-path can specify the node type of each step in the random walk, and can play a role in extracting the logical relationship of the corresponding meta-path semantics in the first heterogeneous network, so that the first sequence information from the random walk can ensure that the global logical relationship of the first heterogeneous network is not destroyed according to the preset rules. In addition, since the first meta-path can specify the proportion of the number of node types obtained in the process of guiding the random walk, the problem of imbalance of node types in the first sequence information can be avoided. Since the first meta-path specifies the walking rules of the random walk, when a random walk is performed in the first heterogeneous network according to the first meta-path, as the number of random walks increases, the content of the obtained first sequence information will become richer and richer, thereby retaining the network structure information and semantic association information of the first heterogeneous network, which is conducive to obtaining a more accurate first object vector and first push information vector when the first sequence information is vectorized in the subsequent steps.

[0121] Step 140: Acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network.

[0122] It should be noted that the second object vector is the vectorization result of the target object in the second heterogeneous network, and the second push information vector is the vectorization result of the second push information.

[0123] In this step, since the second heterogeneous network is obtained in step 120, the second object vector and multiple second push information vectors of the target object in the second heterogeneous network can be obtained based on the second heterogeneous network, so that the object feature vector and push information feature vector of the target object can be obtained according to the second object vector and the second push information vector in subsequent steps, which is conducive to determining accurate target push information based on the object feature vector.

[0124] In some possible implementations, when obtaining a second object vector and multiple second push information vectors based on a second heterogeneous network, second sequence information including a target object and multiple second push information can be first obtained in the second heterogeneous network, and then the target object in the second sequence information is vectorized to obtain the second object vector, and the multiple second push information in the second sequence information is vectorized to obtain multiple second push information vectors. When obtaining the second sequence information in the second heterogeneous network, a second meta-path can be first set according to the second heterogeneous network, and then a random walk can be performed in the second heterogeneous network according to the second meta-path to obtain the second sequence information including the target object and multiple second push information. In addition, when vectorizing the target object and the second push information in the second sequence information, the target object and the second push information in the second sequence information can be vectorized by using a network embedding representation method to obtain a second object vector and a second push information vector, wherein the network embedding representation is a technology that maps nodes in a network to a low-dimensional dense space. By mapping the target object and the second push information in the second sequence information to a low-dimensional dense space, the second object vector and the second push information vector are obtained, so that the second object vector and the second push information vector can retain the inherent network structure and semantic association in the second heterogeneous network, so that when the feature extraction model is used in the subsequent steps to extract features from the second object vector and the second push information vector, more comprehensive and accurate feature information can be extracted, which is conducive to improving the recall accuracy of the target push information. It should be noted that the specific content of the vectorization processing of the target object and the second push information in the second sequence information will be given in the subsequent description.

[0125] It should be noted that, in the process of setting the second meta-path according to the second heterogeneous network, the setting strategy of the second meta-path can be determined based on prior knowledge, or the setting strategy of the second meta-path can be randomly selected, and then the second meta-path can be set in the second heterogeneous network according to the setting strategy. For example, assuming that the setting strategy of the second meta-path determined based on prior knowledge is to select objects that have an interactive relationship with the same recommendation information, then the second meta-path set in the second heterogeneous network according to the setting strategy can be Among them, U represents the object, I represents the recommended information, It indicates that there is an interactive relationship between the object and the recommended information. Therefore, the second meta-path can specify the node type of each step in the random walk, and can play a role in extracting the logical relationship of the corresponding meta-path semantics in the second heterogeneous network, so that the second sequence information from the random walk can ensure that the global logical relationship of the second heterogeneous network is not destroyed according to the preset rules. In addition, since the second meta-path can specify the proportion of the number of node types obtained in the process of guiding the random walk, the problem of imbalance of node types in the second sequence information can be avoided. Since the second meta-path specifies the walking rules of the random walk, when the random walk is performed in the second heterogeneous network according to the second meta-path, as the number of random walks increases, the content of the obtained second sequence information will become richer and richer, so that the network structure information and semantic association information of the second heterogeneous network can be retained, which is conducive to the subsequent steps of vectorizing the second sequence information to obtain a more accurate second object vector and second push information vector.

[0126] The following is a specific example to illustrate the process of obtaining sequence information based on a heterogeneous network.

[0127] When it is necessary to obtain sequence information based on a heterogeneous network, a meta-path may be first determined based on the heterogeneous network. For example, multiple meta-paths as shown in Table 1 below may be determined.

[0128] Table 1

[0129]

[0130] In Table 1, U represents the object, I represents the recommended information, Represents the relationship between the object and the recommended information, Represents social relationships between objects.

[0131] After determining the meta-path, random walks are performed in the heterogeneous network according to the meta-path to obtain sequence information. Specifically, for a given meta-path Among them, V q is a node in a heterogeneous network, is the relationship between nodes in the heterogeneous network, q is an integer greater than 1, after determining the i-th node, the type of the i+1-th node is determined by the path path[i], and then a node is randomly selected in the heterogeneous network according to the node type determined by path[i]. For example, when V n =V n+1 =U, from V n An object node is randomly selected from the neighbor nodes; when V n =I, V n+1 =U, from V n An object node is randomly selected from the neighbor nodes; when V n=U,V n+1 =I, from V n A recommended information node is randomly selected from the neighbor nodes.

[0132] like Figure 7 As shown in the heterogeneous network, after walking to object node 1 in the heterogeneous network, if the next node specified by the meta-path is of object type, then randomly select object node 4 or object node 6 from the neighboring nodes of object node 1; if the next node specified by the meta-path belongs to the recommended information type, then randomly select recommended information node a, recommended information node b or recommended information node c from the neighboring nodes of object node 1; then, continue to cyclically select the next node until the length of the sequence information obtained by random walking reaches the preset length len. After walking from a node for a preset length len to obtain a sequence information, you can cyclically continue to walk from the node for a preset number of times to ensure that the semantic information carried by the walked sequence information is richer, so that the logical structure relationship in the heterogeneous network can be fully preserved. For the sequence information set that walks out from the same node It can be expressed by the following formula:

[0133]

[0134] In formula (1), walks is the number of cycles starting from the node, S n Represents the sequence information generated by the nth walk. It should be noted that if the sequence information set is obtained based on the first heterogeneous network, it can be saved in the negative corpus for subsequent vectorization and feature extraction. If the sequence information set is obtained based on the second heterogeneous network, it can be saved in the positive corpus for subsequent vectorization and feature extraction.

[0135] Step 150: Input the first object vector, the second object vector, the plurality of first push information vectors, and the plurality of second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and a plurality of push information feature vectors.

[0136] In this step, since the first object vector and multiple first push information vectors are obtained in step 130, and the second object vector and multiple second push information vectors are obtained in step 140, the first object vector, the second object vector, the multiple first push information vectors and the multiple second push information vectors can be input into the feature extraction model for feature extraction to obtain the object feature vector and multiple push information feature vectors, so that the subsequent steps can determine the target push information based on the object feature vector and these push information feature vectors.

[0137] In some possible embodiments, when the first object vector, the second object vector, multiple first push information vectors, and multiple second push information vectors are input into a feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors, the first object vector and the second object vector can be first spliced to obtain first splicing information, and among the multiple first push information vectors and the multiple second push information vectors, the first push information vector and the second push information vector corresponding to the same push information are spliced to obtain multiple second splicing information, and then the first splicing information is input into the feature extraction model for feature extraction to obtain an object feature vector, and the multiple second splicing information is input into the feature extraction model for feature extraction to obtain multiple push information feature vectors. Among them, before using the feature extraction model to perform feature extraction, the first object vector and the second object vector are first spliced to obtain first splicing information. Since the first object vector is obtained according to the first heterogeneous network and the second object vector is obtained according to the second heterogeneous network, the first splicing information can fully express the target object's preference characteristics for push information. Therefore, when the first splicing information is input into the feature extraction model for feature extraction, a more comprehensive and accurate object feature vector can be obtained, which is conducive to improving the accuracy of the target push information determined according to the object feature vector in subsequent steps. In addition, for the first push information vector and the second push information vector, the first push information vector and the second push information vector corresponding to the same push information are first spliced among multiple first push information vectors and multiple second push information vectors to obtain multiple second spliced information. Since the first push information vector is obtained based on the first heterogeneous network and the second push information vector is obtained based on the second heterogeneous network, the second spliced information can fully express the preference characteristics between the push information and the target object. Therefore, when the second spliced information is input into the feature extraction model for feature extraction, a more comprehensive and accurate push information feature vector can be obtained, which is conducive to improving the accuracy of the target push information determined according to the push information feature vector in subsequent steps.

[0138] In some possible implementations, the feature extraction model may be a double-tower model, such as Figure 8 As shown, Figure 8 This is a structural diagram of the feature extraction model in an example. Figure 8The feature extraction model includes an object model part on the left and a recommendation information model part on the right, wherein the object model part is used to perform feature extraction on the first object vector and the second object vector to output an object feature vector, and the recommendation information model part is used to perform feature extraction on the first push information vector and the second push information vector to output a push information feature vector. The output of the object model part and the output of the recommendation information model part will undergo similarity calculation, and the calculated similarity can be used to update the model parameters of the dual-tower model (including the model parameters of the object model part and the model parameters of the recommendation information model part).

[0139] The following is a specific example to illustrate the process of feature extraction using the feature extraction model.

[0140] like Figure 8 As shown, after obtaining the first object vector, the second object vector, multiple first push information vectors and multiple second push information vectors, the first object vector user1 and the second object vector user2 are first spliced to obtain the first splicing information in user and concatenate a first push information vector ad1 and a second push information vector ad2 corresponding to the same push information to obtain a second concatenated information in ad , where the first splicing information in user The second splicing information inad can be expressed by the following formula:

[0141] in user =concat(user1, user2) (2)

[0142] in ad =concat(ad1, ad2) (3)

[0143] After getting the first splicing information in user and the second splicing information in ad Afterwards, the first splicing information is in user Input the object model part on the left to extract the feature vector of the object, and input the second splicing information into ad Input to the recommendation information model part on the right for feature extraction to obtain the push information feature vector. Among them, the object model part and the recommendation information model part can both include a fully connected layer, which is composed of a linear relationship transformation and an activation function. The linear relationship transformation can be expressed by the following formula (4), and the activation function can be expressed by the following formula (5):

[0144] z=w T x+b (4)

[0145]

[0146] In formula (4) and formula (5), z is the output parameter of the fully connected layer (such as the object feature vector or the push information feature vector), w is the conversion weight matrix, and w T is the transposed matrix of matrix w, x is the input parameter (such as the first splicing information in user Or the second splicing information in ad ), b is a constant, and f(x) is the output of the activation function.

[0147] It should be noted that after the fully connected layer outputs the result, the output of the fully connected layer can be normalized to obtain the object feature vector and the push information feature vector. It is worth noting that the process of feature extraction of the first spliced information by the object model part and the process of feature extraction of the second spliced information by the recommendation information model part do not affect each other. Therefore, when the object model part extracts features from the first spliced information, the second spliced information can be spliced based on the first push information vector and the second push information vector corresponding to any identical push information; and when the recommendation information model part extracts features from the second spliced information, the first spliced information can be spliced based on the first object vector and the second object vector of any object.

[0148] The following describes the training process of the feature extraction model.

[0149] Before using the feature extraction model for feature extraction, the feature extraction model needs to be trained, wherein the cross entropy loss function can be used as the objective function of the feature extraction model. Specifically, first obtain the target object sample and its social object sample, then construct a first heterogeneous network and a second heterogeneous network based on the target object sample and the social object sample, then obtain the first object vector and multiple first push information vectors of the target object sample based on the first heterogeneous network, and obtain the second object vector and multiple second push information vectors of the target object sample based on the second heterogeneous network, then splice the first object vector and the second object vector to obtain the first splicing information, and among the multiple first push information vectors and the multiple second push information vectors, splice the first push information vector and the second push information vector corresponding to the same push information to obtain multiple second splicing information. At this time, the first splicing information can be input into the object model part of the feature extraction model for feature extraction to obtain object features. The second splicing information is input into the recommendation information model part of the feature extraction model for feature extraction to obtain the push information feature vector, and then the similarity between the object feature vector and the push information feature vector is calculated (such as Euclidean distance, Pearson correlation coefficient, cosine similarity, etc.), and then the loss value is calculated according to the similarity and the objective function, and then the model parameters of the object model part and the recommendation information model part are updated according to the loss value, so that the similarity between the target object with an interactive relationship and the push information is greater than the first preset threshold, and the similarity between the target object without an interactive relationship and the push information is less than the second preset threshold, for example, the similarity between the target object with an interactive relationship and the push information is close to 1, and the similarity between the target object without an interactive relationship and the push information is close to 0.

[0150] It should be noted that the cross entropy loss function as the objective function can be expressed by the following formula (6):

[0151]

[0152] In formula (6), L is the loss value; n is the total number of nodes in the heterogeneous network; p i is the similarity; y i Indicates whether there is an interactive relationship between the target object and the pushed information, for example, y i When the value is 1, it means that there is an interactive relationship between the target object and the pushed information. i When the value is 0, it means that there is no interaction between the target object and the pushed information.

[0153] It should be noted that cosine similarity uses the cosine value of the angle between two vectors in the vector space as a measure of the difference between the two vectors. The closer the cosine similarity is to 1, the closer the angle between the two vectors is to 0 degrees, that is, the more similar the two vectors are. Euclidean distance refers to the natural length of a vector in the vector space, such as the distance from the push information feature vector to the object feature vector in the vector space. The Pearson correlation coefficient is obtained by dividing the covariance of two vectors by the standard deviation of the two vectors. When the Pearson correlation coefficient is 1, the two vectors are completely positively correlated; when the Pearson correlation coefficient is -1, the two vectors are completely negatively correlated; the larger the absolute value of the Pearson correlation coefficient, the stronger the correlation between the two vectors; the closer the Pearson correlation coefficient is to 0, the weaker the correlation between the two vectors.

[0154] Step 160: Determine target push information based on the object feature vector and all push information feature vectors, and push the target push information to the target object.

[0155] In this step, since the object feature vector and multiple push information feature vectors are obtained in step 150, the target push information can be determined based on the object feature vector and all push information feature vectors, and then the target push information is pushed to the target object to achieve information push for the target object.

[0156] It should be noted that the target push information can be of many different types. For example, the target push information can be advertising information, application download link information, web page link information, etc., which is not specifically limited here.

[0157] In some possible implementations, when determining the target push information based on the object feature vector and all push information feature vectors, the similarity between the object feature vector and each push information feature vector can be calculated first, and then the target push information can be determined based on the similarity. The number of target push information can be one or more. When the number of target push information is one, when determining the target push information based on the similarity, the push information corresponding to the similarity with the largest value can be determined as the target push information. When the number of target push information is multiple, when determining the target push information based on the similarity, the push information corresponding to the similarity with a value greater than a preset similarity threshold can be determined as the target push information. Alternatively, the calculated similarities can be sorted from large to small, and then the push information corresponding to the top K similarities can be selected as the target push information.

[0158] In other possible implementations, the object feature vectors of each target object and the push information feature vectors of each push information can be obtained and saved offline in advance. When the target push information needs to be pushed to a target object online, it is only necessary to perform a simple similarity calculation between the object feature vector of the target object and the push information feature vectors of each push information. Based on the calculated similarity, accurate target push information can be quickly recalled for the target object, effectively ensuring the user's speed experience during the information push process, thereby effectively improving the user's usage experience.

[0159] In this embodiment, an information push method including the previous steps 110 to 160 is used to construct a first heterogeneous network including multiple first push information and a second heterogeneous network including multiple second push information based on the target object and its social object, and then obtain the first object vector of the target object and the first push information vector of the first push information based on the first heterogeneous network, and obtain the second object vector of the target object and the second push information vector of the second push information based on the second heterogeneous network, and then input the first object vector, the second object vector, the multiple first push information vectors and the multiple second push information vectors into the feature extraction model for feature extraction to obtain an object feature vector and multiple push information feature vectors, and then determine the target push information to be pushed to the target object based on the object feature vector and all push information feature vectors. Since a heterogeneous network including push information is constructed by jointly using the target object and its social object, the relationship between different objects and different push information can be obtained through the heterogeneous network, thereby expanding the relationship between the target object and different push information, and further solving the problem of data scarcity, which is conducive to improving the recall accuracy of push information and enhancing the recall effect of push information; in addition, since the first push information has no interactive relationship with the target object or social object, and the second push information has an interactive relationship with the target object or social object, the first heterogeneous network and the second heterogeneous network can fully and accurately express the target object's preference characteristics for push information. Therefore, feature extraction is performed on the first object vector and the first push information vector obtained based on the first heterogeneous network, and the second object vector and the second push information vector obtained based on the second heterogeneous network, so that the object feature vector and the push information feature vector can be obtained more comprehensively and accurately, thereby improving the accuracy of the target push information determined based on the object feature vector and the push information feature vector.

[0160] In some possible implementations, the vectorization processing of the target object and the first push information in the first sequence information, or the vectorization processing of the target object and the second push information in the second sequence information, may include: first, obtaining the sequence information to be processed, performing one-hot encoding on the target information in the sequence information to be processed to obtain the encoding information, then obtaining the target weight information based on the encoding information, and performing probability prediction on the adjacent information of the target information based on the target weight information to obtain the probability prediction value of the adjacent information, when the probability prediction value of the adjacent information does not meet the preset probability value (for example, the probability prediction value of the adjacent information is not the maximum probability prediction value), the target weight information is corrected until the probability prediction value of the adjacent information meets the preset probability value (for example, the probability prediction value of the adjacent information is the maximum probability prediction value), and then the corrected target weight information is used as the vector information of the target information. Wherein, when obtaining the target weight information according to the encoding information, the first weight matrix can be initialized first, and then the target weight information can be obtained in the first weight matrix according to the encoding information. For example, the encoding information can be multiplied by the first weight matrix to obtain the target weight information. When the probability prediction of the adjacent information of the target information is performed based on the target weight information to obtain the probability prediction value of the adjacent information, the second weight matrix can be initialized first, and then the target weight information and the second weight matrix are multiplied to obtain the probability prediction value of each information in the sequence information to be processed, and then the probability prediction value of the adjacent information is determined from the probability prediction value of each information. For example, first the target information v in the sequence information to be processed i Perform one-hot encoding to obtain the encoded information x i , and then encode the information x i With the first weight matrix W x Multiply and encode the information x i Mapped to low-dimensional space, the target weight information h is obtained i =W x ·x i , then, the Skip-gram model can be used to train the target information v i The representation vector, specifically, the target weight information h i and the second weight matrix K i Multiply and get the target information u i Neighboring information (such as v i-1 or v i+1 ) probability prediction value, when the probability prediction value of the adjacent information does not meet the preset probability value, the target weight information h i Correction is performed until the probability prediction value of the adjacent information meets the preset probability value. At this time, the corrected target weight information h i As target information v i For example, for a heterogeneous network G = (V, E, T), |TV |>1, the training goal is to give a target node v (such as target information v i ), we hope to maximize the probability of the context node of the target node v, where the probability of the context node can be expressed as:

[0161]

[0162] Among them, T v Indicates the type of the target node v, N t (v) represents the t-th type of the target node v in the sequence information to be processed, and the probability function p(c t |v; θ) is an activation function, θ is the model parameter of the Skip-gram model (i.e. the first weight matrix W x ).

[0163] It should be noted that when the target object in the first sequence information and the first push information are vectorized, the sequence information to be processed obtained is the first sequence information. At this time, the target information to be one-hot encoded can be the target object in the first sequence information or the first push information; when the target object in the second sequence information and the second push information are vectorized, the sequence information to be processed obtained is the second sequence information. At this time, the target information to be one-hot encoded can be the target object in the second sequence information or the second push information.

[0164] The following is an example to illustrate the vectorization process.

[0165] like Figure 9 As shown, Figure 9 This is a flow chart of vectorization processing in an example. When the sequence information to be processed is obtained by random walk in a heterogeneous network (such as Figure 9 After the sequence information U1→U3→I1 and sequence information U3→U4→I2 in the sequence information are processed, the information in the sequence information to be processed can be deduplicated first, and then the sequence information to be processed that has completed the deduplication process can be sorted. After the sorting is completed, the target information in the sequence information to be processed (such as Figure 9 U2) in the single hot encoding to obtain the encoding information |V i |(such as Figure 9 00000010000 in), then, the coded information |V i |With the first weight matrix W x Multiply to get the target weight information h i , then, the target weight information h i and the second weight matrix K iMultiply them to get the probability prediction value of each information in the sequence information to be processed. At this time, the adjacent information (such as Figure 9 If the probability prediction value of the adjacent information is not the largest one among the probability prediction values of all information, then the first weight matrix W x Make corrections (actually the target weight information h i Correction) until the probability prediction value of the adjacent information is the largest one among the probability prediction values of all information, then the corrected target weight information h i As target information (such as Figure 9 Vector information of U2) in .

[0166] The principle of the information push method provided by the embodiment of the present invention is fully explained below with specific examples.

[0167] Reference Figure 10 As shown, Figure 10 This is a complete flow chart of the information push method provided by an embodiment of the present invention. The information push method specifically includes the following contents:

[0168] In the first step, based on the historical interaction information of sample users on historical push information and the social network information of sample users, a first sample heterogeneous network is constructed to characterize users with a negative preference (i.e., no interest) for push information and a second sample heterogeneous network is constructed to characterize users with a positive preference (i.e., interest) for push information.

[0169] In step 2, a first sample meta-path is set according to the first sample heterogeneous network, and a second sample meta-path is constructed according to the second sample heterogeneous network.

[0170] In step 3, a first sample sequence information is obtained by performing a random walk in the first sample heterogeneous network according to the first sample meta-path, and the first sample sequence information is stored in the negative sample library; and a second sample sequence information is obtained by performing a random walk in the second sample heterogeneous network according to the second sample meta-path, and the second sample sequence information is stored in the positive sample library.

[0171] In step 4, the first sample sequence information in the negative sample library is represented by a network embedding, and the first sample vector of the sample user in the first sample heterogeneous network and multiple first pushed sample vectors are obtained. The second sample sequence information in the positive sample library is represented by a network embedding, and the second sample vector of the sample user in the second sample heterogeneous network and multiple second pushed sample vectors are obtained.

[0172] Step 5: Build a dual-tower model and train the dual-tower model using the first sample vector, the second sample vector, multiple first push sample vectors, and multiple second push sample vectors in step 4 to obtain a trained dual-tower model.

[0173] In step 6, when it is necessary to push information to the target user, a test dataset is constructed based on the target user's historical interaction information on the historical pushed information and the target user's social network information. Specifically, first, based on the target user's historical interaction information on historical push information and the target user's social network information, a first heterogeneous network is constructed to characterize the user's negative preference (i.e., lack of interest) for push information, and a second heterogeneous network is constructed to characterize the user's positive preference (i.e., interest) for push information; then, a first meta-path is set based on the first heterogeneous network, and a second meta-path is constructed based on the second heterogeneous network; then, a random walk is performed in the first heterogeneous network according to the first meta-path to obtain first sequence information, and the first sequence information is stored in a negative corpus; and a random walk is performed in the second heterogeneous network according to the second meta-path to obtain second sequence information, and the second sequence information is stored in a positive corpus; then, a network embedding representation is performed on the first sequence information in the negative corpus to obtain a first object vector and multiple first push information vectors of the target user in the first heterogeneous network, and a network embedding representation is performed on the second sequence information in the positive corpus to obtain a second object vector and multiple second push information vectors of the target user in the second heterogeneous network; then, a test dataset is constructed based on the first object vector, the second object vector, the multiple first push information vectors, and the multiple second push information vectors.

[0174] In step 7, the first object vector, the second object vector, the first push information vector, and the second push information vector in the test data set are input into the trained dual-tower model for feature extraction to obtain the object feature vector and the push information feature vector.

[0175] Step 8: Calculate the similarity between the object feature vector and the feature vectors of each push message, determine the target push message suitable for the target user based on the calculated similarity, and then push the target push message to the target user.

[0176] Through the information push method including steps 1 to 8, since a heterogeneous network including push information is constructed by jointly using the target user and its social network, the relationship between different users and different push information can be obtained through the heterogeneous network, thereby expanding the relationship between the target user and different push information, thereby solving the problem of data scarcity, and facilitating improving the recall accuracy of push information and enhancing the recall effect of push information. In addition, since the first heterogeneous network can represent the user's negative preference (i.e., lack of interest) for push information, and the second heterogeneous network can represent the user's positive preference (i.e., interest) for push information, the first heterogeneous network and the second heterogeneous network can fully and accurately express the target user's preference characteristics for push information. Therefore, feature extraction of the first object vector and the first push information vector obtained based on the first heterogeneous network and the second object vector and the second push information vector obtained based on the second heterogeneous network can more comprehensively and accurately obtain the object feature vector and the push information feature vector, thereby improving the accuracy of the target push information determined based on the object feature vector and the push information feature vector.

[0177] The following uses some practical examples to illustrate the application scenarios of the embodiments of the present invention.

[0178] It should be noted that the information push method provided by the embodiment of the present invention can be applied to a variety of different application scenarios such as product push scenarios, movie push scenarios, music push scenarios, book push scenarios, video push scenarios, tourist attraction push scenarios, and food push scenarios. The following is an explanation using product push scenarios, movie push scenarios, music push scenarios, and book push scenarios as examples.

[0179] Scene 1

[0180] The information push method provided in an embodiment of the present invention can be applied to a product push scenario. Specifically, in response to detecting that a consumer has logged into an e-commerce platform via a device such as a smartphone or computer, the server determines the consumer's social objects and, based on the consumer and its social objects, constructs a first heterogeneous network including multiple first product information and a second heterogeneous network including multiple second product information. The first product information is product information that the consumer or its social objects have not purchased or browsed, and the second product information is product information that the consumer or its social objects have purchased or browsed. The server then obtains the consumer's first object vector and the first product information vector of the first product information based on the first heterogeneous network, and obtains the consumer's second object vector and the second product information vector of the second product information based on the second heterogeneous network. The server then inputs the first object vector, the second object vector, the multiple first product information vectors, and the multiple second product information vectors into a feature extraction model for feature extraction, obtaining an object feature vector and multiple product information feature vectors for the consumer. At this point, the server determines target product information to be pushed to the consumer based on the consumer's object feature vector and all product information feature vectors, and then pushes the target product information to the consumer.

[0181] Scene 2

[0182] The information push method provided by the embodiment of the present invention can also be applied to movie push scenarios. Specifically, in response to detecting that a viewer logs into a movie playback platform through a device such as a smartphone or a computer, the server determines the viewer's social objects and constructs a first heterogeneous network including multiple first movie information and a second heterogeneous network including multiple second movie information based on the viewer and its social objects, wherein the first movie information is movie information that the viewer or its social objects have not watched, and the second movie information is movie information that the viewer or its social objects have watched. Then, the server obtains the viewer's first object vector and the first movie information vector of the first movie information based on the first heterogeneous network, and obtains the viewer's second object vector and the second movie information vector of the second movie information based on the second heterogeneous network. Then, the first object vector, the second object vector, the multiple first movie information vectors, and the multiple second movie information vectors are input into a feature extraction model for feature extraction to obtain the viewer's object feature vector and multiple movie information feature vectors. At this time, the server determines the target movie information to be pushed to the viewer based on the viewer's object feature vector and all movie information feature vectors, and then pushes the target movie information to the viewer.

[0183] Scene 3

[0184] The information push method provided by the embodiment of the present invention can also be applied to music push scenarios. Specifically, in response to detecting that a user logs in to a music playback platform through a device such as a smart phone or a car terminal, the server determines the user's social objects, and constructs a first heterogeneous network including multiple first music information and a second heterogeneous network including multiple second music information based on the user and its social objects, wherein the first music information is music information that the user or its social objects have not heard, and the second music information is music information that the user or its social objects have heard. Then, the server obtains the user's first object vector and the first music information vector of the first music information based on the first heterogeneous network, and obtains the user's second object vector and the second music information vector of the second music information based on the second heterogeneous network. Then, the first object vector, the second object vector, the multiple first music information vectors, and the multiple second music information vectors are input into the feature extraction model for feature extraction to obtain the user's object feature vector and multiple music information feature vectors. At this time, the server determines the target music information to be pushed to the user based on the user's object feature vector and all music information feature vectors, and then pushes the target music information to the user.

[0185] Scene 4

[0186] The information push method provided by an embodiment of the present invention can also be applied to a book push scenario. Specifically, in response to detecting that a reader has logged into a literary work sharing platform via a device such as a smartphone or computer, the server determines the reader's social objects and, based on the reader and its social objects, constructs a first heterogeneous network including multiple first book information and a second heterogeneous network including multiple second book information. The first book information is book information that the reader or its social objects have not read, and the second book information is book information that the reader or its social objects have read. The server then obtains the reader's first object vector and the first book information vector of the first book information based on the first heterogeneous network, and obtains the reader's second object vector and the second book information vector of the second book information based on the second heterogeneous network. The server then inputs the first object vector, the second object vector, the multiple first book information vectors, and the multiple second book information vectors into a feature extraction model for feature extraction, obtaining the reader's object feature vector and multiple book information feature vectors. At this point, the server determines the target book information to be pushed to the reader based on the reader's object feature vector and all book information feature vectors, and then pushes the target book information to the reader.

[0187] It will be appreciated that, although the various steps in the above-mentioned various flow charts are shown in sequence according to the indication of the arrows, these steps are not necessarily performed in sequence according to the order indicated by the arrows. Unless clearly stated in the present embodiment, the execution of these steps does not have strict order restrictions, and these steps can be performed in other orders. Moreover, at least a portion of the steps in the above-mentioned flow charts can include multiple steps or multiple stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a portion of the steps or stages in other steps or other steps.

[0188] Reference Figure 11 The embodiment of the present invention further discloses an information push device. The information push device 1100 can implement the information push method of the previous embodiment. The information push device 1100 includes:

[0189] An object determination unit 1110 is configured to determine a social object of a target object;

[0190] A network construction unit 1120 is configured to construct a first heterogeneous network and a second heterogeneous network based on the target object and the social object, wherein the first heterogeneous network includes the target object, the social object, and a plurality of first push messages, and the first push messages have no interactive relationship with the target object or the social object; and the second heterogeneous network includes the target object, the social object, and a plurality of second push messages, and the second push messages have an interactive relationship with the target object or the social object.

[0191] A first vector acquisition unit 1130 is configured to acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network, wherein the first object vector is a vectorization result of a target object in the first heterogeneous network, and the first push information vector is a vectorization result of the first push information;

[0192] A second vector acquisition unit 1140 is configured to acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network, wherein the second object vector is a vectorization result of the target object in the second heterogeneous network, and the second push information vector is a vectorization result of the second push information;

[0193] A feature extraction unit 1150 is configured to input the first object vector, the second object vector, the plurality of first push information vectors, and the plurality of second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and a plurality of push information feature vectors;

[0194] The information pushing unit 1160 is configured to determine target push information according to the object feature vector and all push information feature vectors, and push the target push information to the target object.

[0195] In one embodiment, the feature extraction unit 1150 is further configured to:

[0196] splicing the first object vector and the second object vector to obtain first splicing information;

[0197] Among the plurality of first push information vectors and the plurality of second push information vectors, splicing the first push information vectors and the second push information vectors corresponding to the same push information to obtain a plurality of second spliced information;

[0198] Inputting the first splicing information into a feature extraction model to perform feature extraction to obtain an object feature vector;

[0199] The plurality of second splicing information are input into a feature extraction model for feature extraction to obtain a plurality of push information feature vectors.

[0200] In one embodiment, the network construction unit 1120 is further configured to:

[0201] Acquire multiple first push messages according to the target object and the social object;

[0202] Building a first heterogeneous network according to the target object, the social object, and the plurality of first push information;

[0203] Acquire multiple second push messages according to the target object and the social object;

[0204] A second heterogeneous network is constructed according to the target object, the social object, and the plurality of second push information.

[0205] In one embodiment, the first vector obtaining unit 1130 is further configured to:

[0206] Acquire first sequence information in the first heterogeneous network, wherein the first sequence information includes a target object and a plurality of first push information;

[0207] Performing vectorization processing on the target object in the first sequence information to obtain a first object vector;

[0208] Vectorization processing is performed on the multiple first push information in the first sequence information to obtain multiple first push information vectors.

[0209] In one embodiment, the first vector obtaining unit 1130 is further configured to:

[0210] Setting a first meta-path according to the first heterogeneous network;

[0211] A random walk is performed in the first heterogeneous network according to the first meta-path to obtain first sequence information.

[0212] In one embodiment, the second vector obtaining unit 1140 is further configured to:

[0213] Acquire second sequence information in the second heterogeneous network, wherein the second sequence information includes a target object and a plurality of second push information;

[0214] performing vectorization processing on the target object in the second sequence information to obtain a second object vector;

[0215] Vectorization processing is performed on the multiple second push information in the second sequence information to obtain multiple second push information vectors.

[0216] In one embodiment, the second vector obtaining unit 1140 is further configured to:

[0217] Setting a second meta-path according to the second heterogeneous network;

[0218] A random walk is performed in the second heterogeneous network according to the second meta-path to obtain second sequence information.

[0219] In one embodiment, the information push unit 1160 is further configured to:

[0220] Calculate the similarity between the object feature vector and the feature vectors of each push information;

[0221] Determine the target push information based on similarity.

[0222] In one embodiment, the information push device 1100 further includes:

[0223] An information acquisition unit, used to acquire sequence information to be processed;

[0224] An information encoding unit, configured to perform one-hot encoding on target information in the sequence information to be processed to obtain encoded information;

[0225] A weight acquisition unit, configured to acquire target weight information according to the encoding information;

[0226] A probability prediction unit, configured to perform probability prediction on adjacent information of the target information based on the target weight information to obtain a probability prediction value of the adjacent information;

[0227] A weight correction unit is used to correct the target weight information when the probability prediction value of the adjacent information does not meet the preset probability value until the probability prediction value of the adjacent information meets the preset probability value;

[0228] The vector acquisition unit is used to use the corrected target weight information as vector information of the target information.

[0229] In one embodiment, the weight acquisition unit is further configured to:

[0230] Initialize the first weight matrix;

[0231] Target weight information is obtained in the first weight matrix according to the encoding information.

[0232] In one embodiment, the probability prediction unit is further configured to:

[0233] Initialize the second weight matrix;

[0234] Multiply the target weight information and the second weight matrix to obtain the probability prediction value of each information in the sequence information to be processed;

[0235] The probability prediction values of adjacent information are determined from the probability prediction values of each information.

[0236] It should be noted that since the information push device 1100 of this embodiment can implement the information push method described in the previous embodiment, the information push device 1100 of this embodiment and the information push method described in the previous embodiment have the same technical principles and the same beneficial effects. In order to avoid repetition, they will not be repeated here.

[0237] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0238] Reference Figure 12 The embodiment of the present invention further discloses an information push device, the information push device 1200 comprising:

[0239] at least one processor 1201;

[0240] At least one memory 1202, configured to store at least one program;

[0241] When at least one program is executed by at least one processor 1201 , the information push method described above is implemented.

[0242] An embodiment of the present invention further discloses a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to implement the above information push method.

[0243] An embodiment of the present invention also discloses a computer program product, including a computer program or computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer program or computer instructions from the computer-readable storage medium, and the processor executes the computer program or computer instructions, so that the computer device executes the information push method as described above.

[0244] The terms "first," "second," "third," "fourth," and the like (if any) in the description of the present invention and the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or precedence. It should be understood that the terms used in this manner are interchangeable where appropriate, so that the embodiments of the present invention described herein can, for example, be implemented in orders other than those illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0245] It should be understood that in the present invention, "at least one (item)" refers to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can represent: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0246] In the several embodiments provided by the present invention, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0247] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0248] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0249] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk, and other media that can store program code.

[0250] The step numbers in the above method embodiment are only provided for the convenience of explanation and do not limit the order of the steps. The execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

Claims

1. An information push method, characterized in that: The following steps are involved: Identify the target audience's social network; Constructing a first heterogeneous network and a second heterogeneous network based on the target object and the social object, wherein the first heterogeneous network includes the target object, the social object, and a plurality of first push messages, the first push messages having no interactive relationship with the target object or the social object, and the second heterogeneous network includes the target object, the social object, and a plurality of second push messages, the second push messages having an interactive relationship with the target object or the social object; Acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network, wherein the first object vector is a vectorization result of the target object in the first heterogeneous network, and the first push information vector is a vectorization result of the first push information; Acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network, wherein the second object vector is a vectorization result of the target object in the second heterogeneous network, and the second push information vector is a vectorization result of the second push information; Inputting the first object vector, the second object vector, a plurality of the first push information vectors, and a plurality of the second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and a plurality of push information feature vectors; Target push information is determined according to the object feature vector and all the push information feature vectors, and the target push information is pushed to the target object.

2. The information push method according to claim 1, characterized in that: The step of inputting the first object vector, the second object vector, the plurality of first push information vectors, and the plurality of second push information vectors into a feature extraction model for feature extraction to obtain an object feature vector and a plurality of push information feature vectors includes: splicing the first object vector and the second object vector to obtain first splicing information; Among the plurality of first push information vectors and the plurality of second push information vectors, splicing the first push information vector and the second push information vector corresponding to the same push information to obtain a plurality of second spliced information; Inputting the first splicing information into a feature extraction model to perform feature extraction to obtain an object feature vector; The plurality of second splicing information are input into the feature extraction model for feature extraction to obtain a plurality of push information feature vectors.

3. The information push method according to claim 1, wherein: The constructing of the first heterogeneous network and the second heterogeneous network according to the target object and the social object includes: Acquire a plurality of first push information according to the target object and the social object; constructing a first heterogeneous network according to the target object, the social object, and the plurality of first push information; Acquire a plurality of second push information according to the target object and the social object; A second heterogeneous network is constructed according to the target object, the social object, and the plurality of second push information.

4. The information push method according to claim 1, wherein: The acquiring a first object vector and a plurality of first push information vectors based on the first heterogeneous network includes: Acquire first sequence information in the first heterogeneous network, wherein the first sequence information includes the target object and a plurality of first push information; performing vectorization processing on the target object in the first sequence information to obtain the first object vector; Vectorization processing is performed on the multiple first push information in the first sequence information to obtain multiple first push information vectors.

5. The information push method according to claim 4, characterized in that: The acquiring first sequence information in the first heterogeneous network includes: Setting a first meta-path according to the first heterogeneous network; A random walk is performed in the first heterogeneous network according to the first meta-path to obtain first sequence information.

6. The information push method according to claim 1, characterized in that: The acquiring a second object vector and a plurality of second push information vectors based on the second heterogeneous network includes: Acquire second sequence information in the second heterogeneous network, wherein the second sequence information includes the target object and a plurality of second push information; performing vectorization processing on the target object in the second sequence information to obtain the second object vector; Vectorization processing is performed on the multiple pieces of the second push information in the second sequence information to obtain multiple second push information vectors.

7. The information push method according to claim 6, characterized in that: The acquiring the second sequence information in the second heterogeneous network includes: Setting a second meta-path according to the second heterogeneous network; A random walk is performed in the second heterogeneous network according to the second meta-path to obtain second sequence information.

8. The information push method according to claim 1, characterized in that: The determining the target push information according to the object feature vector and all the push information feature vectors includes: Calculating the similarity between the object feature vector and each of the push information feature vectors; Target push information is determined according to the similarity.

9. The information push method according to claim 4 or 6, characterized in that: The vectorization process includes the following steps: Get the sequence information to be processed; Performing one-hot encoding on the target information in the sequence information to be processed to obtain encoded information; Obtaining target weight information according to the encoding information; Performing probability prediction on the adjacent information of the target information based on the target weight information to obtain a probability prediction value of the adjacent information; When the probability prediction value of the neighboring information does not meet the preset probability value, the target weight information is modified until the probability prediction value of the neighboring information meets the preset probability value; The corrected target weight information is used as the vector information of the target information.

10. The information push method according to claim 9, characterized in that: The acquiring target weight information according to the encoding information includes: Initialize the first weight matrix; Target weight information is obtained in the first weight matrix according to the encoding information.

11. The information push method according to claim 9, characterized in that: The performing probability prediction on the neighboring information of the target information based on the target weight information to obtain a probability prediction value of the neighboring information includes: Initialize the second weight matrix; Multiplying the target weight information and the second weight matrix to obtain a probability prediction value of each information in the sequence information to be processed; The probability prediction value of the adjacent information is determined among the probability prediction values of each of the information.

12. An information push device, characterized in that: include: an object determination unit, configured to determine a social object of a target object; a network construction unit, configured to construct a first heterogeneous network and a second heterogeneous network based on the target object and the social object, wherein the first heterogeneous network includes the target object, the social object, and a plurality of first push messages, the first push messages having no interactive relationship with the target object or the social object, and the second heterogeneous network includes the target object, the social object, and a plurality of second push messages, the second push messages having an interactive relationship with the target object or the social object; a first vector acquisition unit, configured to acquire a first object vector and a plurality of first push information vectors based on the first heterogeneous network, wherein the first object vector is a vectorization result of the target object in the first heterogeneous network, and the first push information vector is a vectorization result of the first push information; a second vector acquisition unit, configured to acquire a second object vector and a plurality of second push information vectors based on the second heterogeneous network, wherein the second object vector is a vectorization result of the target object in the second heterogeneous network, and the second push information vector is a vectorization result of the second push information; a feature extraction unit, configured to input the first object vector, the second object vector, a plurality of the first pushed information vectors, and a plurality of the second pushed information vectors into a feature extraction model for feature extraction, thereby obtaining an object feature vector and a plurality of pushed information feature vectors; The information pushing unit is configured to determine target push information according to the object feature vector and all the push information feature vectors, and push the target push information to the target object.

13. An information push device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When at least one of the programs is executed by at least one of the processors, the information push method according to any one of claims 1 to 11 is implemented.

14. A computer-readable storage medium, characterized in that A processor-executable program is stored therein, and when the processor-executable program is executed by the processor, it is used to implement the information push method according to any one of claims 1 to 11.

15. A computer program product comprising a computer program or computer instructions, characterized in that The computer program or the computer instruction is stored in a computer-readable storage medium, the processor of the computer device reads the computer program or the computer instruction from the computer-readable storage medium, and the processor executes the computer program or the computer instruction, so that the computer device executes the information push method according to any one of claims 1 to 11.

Images