Object Push Method, Device, Electronic Device and Storage Medium

Abstract

The present disclosure relates to an object pushing method, apparatus, electronic device, and storage medium. The method includes: determining a plurality of relationship chain objects from a plurality of objects to be pushed obtained according to the relationship chain data of a first account; determining at least one target object from the plurality of relationship chain objects; at least one target object corresponds to at least one object type, and the object types of each target object in the at least one target object are different; determining preference index data of the first account for each object type in the at least one object type; determining preference index data of the first account for non-relationship chain objects among the plurality of objects to be pushed; and performing object pushing to the first account based on the preference index data of each object type and the preference index data of the non-relationship chain objects. The present disclosure can meet the personalized needs of different users, improve the pushing accuracy, and improve user retention.

Description

Technical Field

[0001] The present disclosure relates to the field of Internet technologies, and in particular, to an object pushing method, apparatus, electronic device, and storage medium. Background Art

[0002] A push system is an information filtering system that filters step by step from a vast amount of materials through processes such as recall and sorting, and selects objects of interest to the user for pushing.

[0003] In related technologies, mainly consumer-oriented positive feedback, such as predicting the click-through rate, like rate, follow rate, long-play rate, etc. of an object, is used to recall and sort the object. This results in the sorting result being biased towards consumption, unable to meet the personalized needs of users, and leading to low push accuracy and poor effects. Summary of the Invention

[0004] The present disclosure provides an object pushing method, apparatus, electronic device, and storage medium. The technical solution of the present disclosure is as follows:

[0005] According to a first aspect of an embodiment of the present disclosure, an object pushing method is provided, including:

[0006] Determine a plurality of relationship-chain objects from a plurality of objects to be pushed obtained according to the relationship-chain data of a first account; the object type of each relationship-chain object in the plurality of relationship-chain objects is determined according to the relationship between the publishing account of the relationship-chain object and the first account;

[0007] Determine at least one target object from the plurality of relationship-chain objects; at least one target object corresponds to at least one object type, and the object type of each target object in the at least one target object is different;

[0008] Determine the preference index data of the first account for each object type in the at least one object type;

[0009] Determine the preference index data of the first account for non-relationship-chain objects in the plurality of objects to be pushed;

[0010] Based on the preference index data of each object type and the preference index data of non-relationship-chain objects, push objects to the first account.

[0011] In some possible embodiments, the relationship-chain data includes associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts that follow the first account, the accounts that mutually follow the first account, and the accounts that have at least one common friend with the first account;

[0012] Determining a plurality of relationship-chain objects from the plurality of objects to be pushed obtained includes:

[0013] Obtain the publishing accounts of each object among multiple objects to be pushed;

[0014] When it is determined that the publishing account is an associated account, determine that the object corresponding to the publishing account is a relationship chain object.

[0015] In some possible embodiments, after determining that the publishing account is an associated account and determining that the object corresponding to the publishing account is a relationship chain object, it further includes:

[0016] When the publishing account is an account followed by the first account, determine that the object corresponding to the publishing account is a relationship chain object of the first object type;

[0017] Or;

[0018] When the publishing account is an account that follows the first account, determine that the object corresponding to the publishing account is a relationship chain object of the second object type;

[0019] Or;

[0020] When the publishing account is an account that mutually follows the first account, determine that the object corresponding to the publishing account is a relationship chain object of the third object type;

[0021] Or;

[0022] When the publishing account is an account that has at least one common friend with the first account, determine that the object corresponding to the publishing account is a relationship chain object of the fourth object type.

[0023] In some possible embodiments, determining at least one target object from multiple relationship chain objects includes:

[0024] Obtain the sorting index of each relationship chain object among multiple relationship chain objects; the sorting index of each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object;

[0025] Classify multiple relationship chain objects to obtain relationship chain object sets corresponding to at least one object type; the object types of each relationship chain object in the relationship chain object set are the same;

[0026] Determine the relationship chain objects with sorting indexes greater than or equal to the preset index in each relationship chain object set corresponding to each object type as target objects, and obtain at least one target object.

[0027] In some possible embodiments, obtaining the sorting index of each relationship chain object among multiple relationship chain objects includes:

[0028] Estimate the operation behavior of each relationship chain object through the operation prediction model to obtain the operation prediction data corresponding to each relationship chain object; the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account;

[0029] Determine the sorting index of each relationship chain object according to the operation prediction data corresponding to each relationship chain object.

[0030] In some possible embodiments, the generation method of the operation prediction model includes:

[0031] Obtain the first training sample set; each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, the feature data of the historical objects published by the associated accounts of the second account, and the actual operation data of the second account on the historical objects; the second account is any user account accessing the push system; the associated accounts of the second account include at least one of the accounts followed by the second account, the accounts following the second account, the accounts mutually following the second account, and the accounts having at least one common friend with the second account;

[0032] Obtain the first preset machine learning model;

[0033] Input the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account into the first preset machine learning model to obtain the predicted operation data corresponding to the second account;

[0034] Train the first preset machine learning model based on the predicted operation data and the actual operation data to obtain the trained operation prediction model.

[0035] In some possible embodiments, the operation prediction data includes multiple predicted data corresponding to multiple operation behaviors; inputting the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account into the first preset machine learning model to obtain the predicted operation data corresponding to the second account includes:

[0036] Determine the first feature vector corresponding to the feature data of the second account according to the first spatial vector representation module in the first preset machine learning model;

[0037] Determine a second feature vector corresponding to the feature data of the associated account of the second account according to the second spatial vector representation module in the first preset machine learning model;

[0038] Determine a third feature vector corresponding to the interaction feature data between the second account and the associated account of the second account according to the third spatial vector representation module in the first preset machine learning model;

[0039] Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the second account according to the fourth spatial vector representation module in the first preset machine learning model;

[0040] Input the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into the corresponding preprocessing neural network respectively, and output the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector;

[0041] Connect the fused feature vector after fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and input them into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the prediction data output by each operation behavior prediction sub-module.

[0042] In some possible embodiments, determining the preference index data of the first account for each object type in at least one object type includes:

[0043] Determine the preference index data corresponding to each target object in at least one target object according to the preference prediction model; the preference prediction model is trained according to the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of the non-relationship-chain objects in the multiple historical push objects corresponding to the third account;

[0044] Based on the object type of each target object, use the preference index data corresponding to each target object as the preference index data of the first account for the object type.

[0045] In some possible embodiments, determining the preference index data corresponding to each target object in at least one target object according to the preference prediction model includes:

[0046] Input the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects in the multiple objects to be pushed into the preference prediction model to obtain the preference index data corresponding to the target object.

[0047] In some possible embodiments, determining preference index data of the first account for non-relationship-chain objects among multiple objects to be pushed includes:

[0048] Inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data of the non-relationship-chain objects.

[0049] In some possible embodiments, the generation method of the preference prediction model includes:

[0050] Obtaining a second training sample set; each second training sample in the second training sample set includes the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account for the historical objects; the third account is any user account accessing the push system; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts following the third account, the accounts mutually following the third account, and the accounts having at least one common friend with the third account;

[0051] Obtaining a second preset machine learning model;

[0052] Inputting the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects; the preference prediction index data of the historical objects represents the preference degree of the third account for the object type of the historical objects;

[0053] Determining the actual preference index data of the historical objects according to the actual behavior data of the third account for the historical objects;

[0054] Training the second preset machine learning model based on the actual preference index data and the preference prediction index data to obtain the trained preference prediction model.

[0055] In some possible embodiments, inputting the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects includes:

[0056] Determine a first feature vector corresponding to the feature data of the third account according to the first spatial vector representation module in the second preset machine learning model;

[0057] Determine a second feature vector corresponding to the feature data of the associated account of the third account according to the second spatial vector representation module in the second preset machine learning model;

[0058] Determine a third feature vector corresponding to the historical behavior data of the third account according to the third spatial vector representation module in the second preset machine learning model;

[0059] Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the third account according to the fourth spatial vector representation module in the second preset machine learning model;

[0060] Determine a fifth feature vector corresponding to the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account according to the fifth spatial vector representation module in the second preset machine learning model;

[0061] After connecting the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector, input them into the behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data;

[0062] Determine the preference prediction index data according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data.

[0063] In some possible embodiments, based on the preference index data of each object type and the preference index data of the non-relationship-chain objects, pushing objects to the first account includes:

[0064] When the preference index data of the non-relationship-chain objects is greater than the preference index data of each object type, use the non-relationship-chain objects among the multiple objects to be pushed as the target push objects; or; when there is a preference index data of an object type greater than the preference index data of the non-relationship-chain objects, use the object type with the preference index data greater than or equal to the preset index data as the target object type, and use the objects among the multiple objects to be pushed whose object type is the target object type as the target push objects;

[0065] Send an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

[0066] According to the second aspect of the embodiments of the present disclosure, there is provided an object push device, including:

[0067] The first determination module is configured to determine a plurality of relationship chain objects from the plurality of objects to be pushed obtained according to the relationship chain data of the first account; the object type of each relationship chain object in the plurality of relationship chain objects is determined according to the relationship between the publishing account of the relationship chain object and the first account;

[0068] The second determination module is configured to determine at least one target object from the plurality of relationship chain objects; the at least one target object corresponds to at least one object type, and the object type of each target object in the at least one target object is different;

[0069] The third determination module is configured to determine the preference index data of the first account for each object type in at least one object type;

[0070] The fourth determination module is configured to determine the preference index data of the first account for the non-relationship chain objects among the plurality of objects to be pushed;

[0071] The push module is configured to perform object pushing to the first account based on the preference index data of each object type and the preference index data of the non-relationship chain objects.

[0072] In some possible embodiments, the relationship chain data includes the associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts that follow the first account, the accounts that mutually follow the first account, and the accounts that have at least one common friend with the first account;

[0073] The first determination module is further configured to obtain the publishing account of each object among the plurality of objects to be pushed; when it is determined that the publishing account is an associated account, determine the object corresponding to the publishing account as a relationship chain object.

[0074] In some possible embodiments, the first determination module is further configured to, when the publishing account is an account followed by the first account, determine the object corresponding to the publishing account as a relationship chain object of the first object type; or; when the publishing account is an account that follows the first account, determine the object corresponding to the publishing account as a relationship chain object of the second object type; or; when the publishing account is an account that mutually follows the first account, determine the object corresponding to the publishing account as a relationship chain object of the third object type or; when the publishing account is an account that has at least one common friend with the first account, determine the object corresponding to the publishing account as a relationship chain object of the fourth object type.

[0075] In some possible embodiments, the second determination module includes:

[0076] An obtaining sub-module, configured to obtain sorting metrics for each of multiple relationship chain objects; the sorting metric for each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object;

[0077] A classifying sub-module, configured to classify multiple relationship chain objects to obtain a relationship chain object set corresponding to at least one object type; each relationship chain object in the relationship chain object set has the same object type;

[0078] A determining sub-module, configured to determine relationship chain objects with sorting metrics greater than or equal to a preset metric in the relationship chain object set corresponding to each object type as target objects, to obtain at least one target object.

[0079] In some possible embodiments, the obtaining sub-module is further configured to perform prediction of operation behaviors on each relationship chain object through an operation prediction model to obtain operation prediction data corresponding to each relationship chain object; the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account; the sorting metric for each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object.

[0080] In some possible embodiments, it further includes a generation module of the operation prediction model;

[0081] The generation module of the operation prediction model includes:

[0082] A first obtaining sub-module, configured to obtain a first training sample set; each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, the feature data of the historical objects published by the associated accounts of the second account, and the actual operation data of the second account on the historical objects; the second account is any user account accessing the push system; the associated accounts of the second account include at least one of the accounts followed by the second account, the accounts following the second account, the accounts mutually following the second account, and the accounts having at least one common friend with the second account;

[0083] A second obtaining sub-module, configured to obtain a first preset machine learning model;

[0084] An input sub-module, configured to input the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, and the feature data of the historical objects published by the associated account of the second account into a first preset machine learning model to obtain the estimated operation data corresponding to the second account;

[0085] A training sub-module, configured to train the first preset machine learning model based on the estimated operation data and the actual operation data to obtain a trained operation prediction model.

[0086] In some possible embodiments, the input sub-module is configured to perform:

[0087] According to the first spatial vector representation module in the first preset machine learning model, determine the first feature vector corresponding to the feature data of the second account;

[0088] According to the second spatial vector representation module in the first preset machine learning model, determine the second feature vector corresponding to the feature data of the associated account of the second account;

[0089] According to the third spatial vector representation module in the first preset machine learning model, determine the third feature vector corresponding to the interaction feature data between the second account and the associated account of the second account;

[0090] According to the fourth spatial vector representation module in the first preset machine learning model, determine the fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the second account;

[0091] Input the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into the corresponding preprocessing neural networks respectively, and output the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector;

[0092] Connect the fused feature vector obtained by fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and input them into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the estimated data output by each operation behavior prediction sub-module.

[0093] In some possible embodiments, the third determination module includes:

[0094] The first determination sub-module is configured to execute the determination of preference index data corresponding to each target object among at least one target object according to a preference prediction model; the preference prediction model is trained based on the feature data of a third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account.

[0095] The second determination sub-module is configured to execute the operation of taking the preference index data corresponding to each target object as the preference index data of the first account for the object type based on the object type of each target object.

[0096] In some possible embodiments, the first determination sub-module is further configured to execute the operation of inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data corresponding to the target object.

[0097] In some possible embodiments, the fourth determination module is configured to execute the operation of inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data of the non-relationship-chain objects.

[0098] In some possible embodiments, it further includes a generation module for the preference prediction model.

[0099] The generation module for the preference prediction model includes:

[0100] The first acquisition sub-module is configured to execute the acquisition of a second training sample set; each second training sample in the second training sample set includes the feature data of a third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account for the historical objects; the third account is any user account accessing the push system; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts following the third account, the accounts mutually following the third account, and the accounts having at least one common friend with the third account.

[0101] The second acquisition sub-module is configured to execute the acquisition of a second preset machine learning model.

[0102] An input sub-module, configured to input the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of non-relationship-chain objects among multiple historical push objects corresponding to the third account into a second preset machine learning model to obtain the preference prediction index data of the historical objects; the preference prediction index data of the historical objects characterizes the preference degree of the third account for the object type of the historical objects.

[0103] A determination sub-module, configured to determine the actual preference index data of the historical objects according to the actual behavior data of the third account for the historical objects.

[0104] A training sub-module, configured to train the second preset machine learning model based on the actual preference index data and the preference prediction index data to obtain a trained preference prediction model.

[0105] In some possible embodiments, the input sub-module is further configured to perform:

[0106] Determine a first feature vector corresponding to the feature data of the third account according to a first spatial vector representation module in the second preset machine learning model;

[0107] Determine a second feature vector corresponding to the feature data of the associated accounts of the third account according to a second spatial vector representation module in the second preset machine learning model;

[0108] Determine a third feature vector corresponding to the historical behavior data of the third account according to a third spatial vector representation module in the second preset machine learning model;

[0109] Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated accounts of the third account according to a fourth spatial vector representation module in the second preset machine learning model;

[0110] Determine a fifth feature vector corresponding to the feature data of non-relationship-chain objects among multiple historical push objects corresponding to the third account according to a fifth spatial vector representation module in the second preset machine learning model;

[0111] Connect the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector and input them into a behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data;

[0112] Determine the preference prediction index data according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data.

[0113] In some possible embodiments, the push module includes:

[0114] A determining sub-module, configured to execute: when the preference index data of a non-relationship-chain object is greater than the preference index data of each object type, using the non-relationship-chain object among the multiple objects to be pushed as the target push object; or; when the preference index data of a certain object type is greater than the preference index data of the non-relationship-chain object, using the object type whose preference index data is greater than or equal to the preset index data as the target object type, and using the object whose object type is the target object type among the multiple objects to be pushed as the target push object;

[0115] A sending sub-module, configured to execute sending an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

[0116] According to a third aspect of the embodiments of the present disclosure, there is provided an electronic device, including:

[0117] A processor;

[0118] A memory for storing executable instructions of the processor;

[0119] Wherein, the processor is configured to execute instructions to implement the object push method provided in the first aspect of the embodiments of the present disclosure.

[0120] According to a fourth aspect of the embodiments of the present disclosure, there is provided a computer-readable storage medium, when the instructions in the computer-readable storage medium are executed by the processor of the electronic device, enabling the electronic device to execute the object push method provided in the first aspect of the embodiments of the present disclosure.

[0121] According to a fifth aspect of the embodiments of the present disclosure, there is provided a computer program product, the computer program product includes a computer program, the computer program is stored in a readable storage medium, and at least one processor of the computer device reads and executes the computer program from the readable storage medium, enabling the computer device to execute the object push method provided in the first aspect of the embodiments of the present disclosure.

[0122] The technical solutions provided by the embodiments of the present disclosure at least bring the following beneficial effects:

[0123] In the process of pushing objects to the first account, based on the relationship-chain data of the first account, multiple relationship-chain objects are determined from the multiple objects to be pushed obtained, and further the relationship-chain objects that perform better in terms of sociality are screened out, so as to realize pushing objects that the first account is more interested in to the first account, thereby solving the problem of poor push effect caused by the traditional push method easily ignoring relationship-chain objects; and by determining the preferences of different users for different types of relationship-chain objects and non-relationship-chain objects, the personalized needs of different users are met, the push accuracy can be improved, and the user retention can be improved.

[0124] It should be understood that the above general description and the following detailed description are merely exemplary and explanatory, and do not limit the present disclosure. Brief Description of the Drawings

[0125] The drawings herein are incorporated into the specification and form a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure, and do not constitute an undue limitation to the present disclosure.

[0126] Figure 1 is a schematic diagram of an application environment shown according to an exemplary embodiment;

[0127] Figure 2 is a flowchart of an object pushing method shown according to an exemplary embodiment;

[0128] Figure 3 is a flowchart of determining a plurality of relationship chain objects from a plurality of objects to be pushed obtained according to an exemplary embodiment;

[0129] Figure 4 is a flowchart of determining at least one target object from a plurality of relationship chain objects according to an exemplary embodiment;

[0130] Figure 5 is a flowchart of obtaining a sorting index for each relationship chain object among a plurality of relationship chain objects according to an exemplary embodiment;

[0131] Figure 6 is a flowchart of a generation method of an operation prediction model shown according to an exemplary embodiment;

[0132] Figure 7 is a structural diagram of a first preset machine learning model shown according to an exemplary embodiment;

[0133] Figure 8 is a flowchart of determining preference index data of a first account for each object type among at least one object type according to an exemplary embodiment;

[0134] Figure 9 is a flowchart of a generation method of a preference prediction model shown according to an exemplary embodiment;

[0135] Figure 10 is a structural diagram of a second preset machine learning model shown according to an exemplary embodiment;

[0136] Figure 11 is a block diagram of an object pushing device shown according to an exemplary embodiment;

[0137] Figure 12It is a block diagram of an electronic device for object pushing shown according to an exemplary embodiment. Detailed implementation manners

[0138] In order to enable those of ordinary skill in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below with reference to the accompanying drawings.

[0139] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar first objects, and do not necessarily need to describe a specific order or sequence. It should be understood that such used data can be interchanged under appropriate circumstances so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present disclosure. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0140] It should be noted that the user information involved in the present disclosure (including but not limited to user device information, user personal information, etc.) is all information authorized by the user or fully authorized by all parties.

[0141] Please refer to Figure 1 , which shows a schematic diagram of an application environment of an object pushing method shown according to an exemplary embodiment. The application environment may include a terminal 110 and a server 120, and the terminal 110 and the server 120 may be connected through a wired network or a wireless network.

[0142] The terminal 110 may be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto. The terminal 110 may be installed with client software providing a human-computer interaction function, such as an application program (abbreviated as App). The application program may be an independent application program or a subroutine in an application program. Exemplarily, the application program may be a news application program, a live broadcast application program, or a video application program, etc. The user of the terminal 110 may log in to the application program through pre-registered user information, and the user information may include an account and a password.

[0143] The server 120 can be a server that provides background services for the applications in the terminal 110. Specifically, the service provided by the server 120 can be an object push service, and the object can be determined according to specific application scenarios, for example, it can include but is not limited to short videos, news, advertisements, and so on. The server 120 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms.

[0144] In some possible embodiments, the user information pre-registered by the user of the terminal 110 includes a first account. The user can log in to the application through the first account to use the object push service provided by the server 120. The terminal 110 can generate an object push request based on the user operation and send the object push request to the server 120.

[0145] In response to the object push request of the first account, the server 120 determines multiple relationship chain objects from the multiple objects to be pushed according to the relationship chain data of the first account. The object type of each relationship chain object in the multiple relationship chain objects is determined according to the relationship between the publishing account of the relationship chain object and the first account. Then, the server 120 determines at least one target object from the multiple relationship chain objects. The at least one target object corresponds to at least one object type, and the object type of each target object in the at least one target object is different. Then, the server 120 determines the preference index data of the first account for each object type in the at least one object type, and determines the preference index data of the first account for the non-relationship chain objects in the multiple objects to be pushed. Then, the server 120 performs object push to the first account based on the preference index data of each object type and the preference index data of the non-relationship chain objects.

[0146] In the embodiments of the present disclosure, by using the relationship chain data of the user, deeper implicit connections between the objects to be pushed and the user are explored. In the process of object push, considering the user's preference for the relationship chain objects can solve the problem of poor push effect caused by the traditional push method easily ignoring the relationship chain objects, improve the push accuracy, enhance the user's relationship chain network, and improve user retention.

[0147] It should be understood that Figure 1 The application environment shown is only an example. In actual applications, the object push method of the embodiments of the present disclosure can be independently executed by the terminal or the server, or can be executed by the terminal and the server in cooperation. The embodiments of the present disclosure do not limit the specific application environment.

[0148] Figure 2 is a flowchart of an object pushing method shown according to an exemplary embodiment. As Figure 2 shown, taking the object pushing method used in the Figure 1 server as an example, it includes the following steps:

[0149] In step S201, according to the relationship chain data of the first account, multiple relationship chain objects are determined from the multiple objects to be pushed obtained; the object type of each relationship chain object in the multiple relationship chain objects is determined according to the relationship between the publishing account of the relationship chain object and the first account.

[0150] In the embodiments of the present disclosure, the object is content matching the actual application scenario; in some possible application scenarios, the object may include music, video, goods, news, advertisements, etc. The server responds to the object pushing request of the first account and obtains the relationship chain data of the first account from the preset storage area; wherein, the preset storage area is protected by security measures. The server obtains the relationship chain data of the first account from the preset storage area through a secure acquisition channel.

[0151] Among them, the definition of the relationship chain is that if each person is regarded as a node in the entire social group, then the connection line between every two points is called a relationship chain; the relationship chain data includes the own attribute information of each node and the information flow between every two points.

[0152] In a specific application scenario, such as in an application program with social functions, each user account can establish a connection with other user accounts, so that each user account can be an associated account of other user accounts; the server can be the background server of the application program to provide corresponding computing services;

[0153] Correspondingly, in some possible embodiments, the above relationship chain data includes the associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts following the first account, the accounts mutually following the first account, and the accounts having at least one common friend with the first account;

[0154] Then, the above determining multiple relationship chain objects from the multiple objects to be pushed obtained may include the following steps as Figure 3 shown:

[0155] In step S301, obtain the publishing account of each object in the multiple objects to be pushed.

[0156] In step S303, when it is determined that the publishing account is an associated account, determine the object corresponding to the publishing account as a relationship chain object.

[0157] In the embodiments of the present disclosure, the multiple objects to be pushed are objects that the server recalls from a massive object resource pool and that the first account may consume; in the related art, after sorting the multiple objects to be pushed according to a certain rule, the server pushes the top N objects to the user. For example, in the video push scenario, the server estimates behaviors such as click-through rate, like rate, and follow rate for all recalled videos, and sorts all the videos to be pushed according to the estimation results; however, this method of recalling and sorting videos using consumer-oriented positive feedback, such as indicators like click-through rate, like rate, and follow rate, will cause the final video sorting result to be biased towards consumption, and does not consider the relationship and behavior between the user and the video publisher, nor does it consider the user's preference behavior for content in the relationship chain, resulting in a lack of personalization in the sorting result and causing a poor push effect.

[0158] Therefore, in the embodiments of the present disclosure, after obtaining the multiple objects to be pushed, the server obtains the publishing account of each object among the multiple objects, where the publishing account is the account of the author who publishes the object; then, the server determines for each object's publishing account, and when it is determined that the publishing account of the object is an associated account of the first account, the corresponding object is regarded as a relationship-chain object; that is, a relationship-chain object refers to an object published by an associated account of the first account. In this way, through the relationship-chain data of the first account, the relationship-chain objects among the multiple objects to be pushed are determined. There is a deeper implicit connection between the relationship-chain objects and the first account, which can play a positive role in realizing personalized push and can improve the push effect.

[0159] In some possible embodiments, the object type of each relationship-chain object is determined according to the relationship between the publishing account of the relationship-chain object and the first account; if the relationship between the first account and the publishing account is different, the object type of the object corresponding to the publishing account is different; for example, as mentioned in the above embodiments, the relationship between the first account and the publishing account can be a two-way follow relationship, or a one-way follow relationship, including the first account following the publishing account one-way, or the publishing account following the first account one-way; there is no follow relationship between the first account and the publishing account, but there is at least one common friend between the first account and the publishing account.

[0160] Thus, after determining that the publishing account is an associated account and determining that the object corresponding to the publishing account is a relationship-chain object as described above, the following steps may be included:

[0161] When the publishing account is an account followed by the first account, determine that the object corresponding to the publishing account is a relationship-chain object of the first object type;

[0162] Or; when the publishing account is an account that follows the first account, determine that the object corresponding to the publishing account is a relationship-chain object of the second object type;

[0163] Alternatively, when the publishing account is an account that follows the first account mutually, determine that the object corresponding to the publishing account is a relationship chain object of the third object type;

[0164] Alternatively, when the publishing account is an account that has at least one mutual friend with the first account, determine that the object corresponding to the publishing account is a relationship chain object of the fourth object type.

[0165] In a specific application scenario, the server provides a video push service to user account A. First, the set of videos to be pushed recalled by the server includes Video to be Pushed 1, Video to be Pushed 2, Video to be Pushed 3, and Video to be Pushed 4. The publishing accounts obtained by the server for the above 4 videos to be pushed are user account B, user account C, user account D, and user account E respectively. Then, based on the relationship chain data of user account A, the server determines that the above user accounts B, C, and D are associated accounts of user account A, where user account B is an account followed by user account A, user account C is an account that follows user account A, user account D is an account that follows user account A mutually, and there is no connection between user account E and user account A. Then, the server can determine that the above Videos to be Pushed 1, 2, and 3 are relationship chain videos.

[0166] In the above embodiments, considering that different users have different preferences for different relationship chain objects, the object types of the relationship chain objects are divided according to the relationship between the publishing account and the first account. Then, through subsequent steps, the preferences of the first account for different types of relationship chain objects can be determined. Thus, when pushing to the first account, by combining the preferences of the first account for specific types of relationship chain objects, the push result can better meet the personalized needs of the first account, thereby improving user retention.

[0167] In step S203, determine at least one target object from multiple relationship chain objects; the at least one target object corresponds to at least one object type, and the object types of each target object in the at least one target object are different.

[0168] In the embodiments of the present disclosure, after determining multiple relationship chain objects from multiple objects to be pushed and determining the object types of each relationship chain object, there may be multiple relationship chain objects of the same object type. At this time, the server selects one relationship chain object from the multiple relationship chain objects of the same object type as the target object. Thus, in subsequent steps, by determining the preference index data of the first account for the target object, the preference index data of the first account for the object type of the target object can be determined.

[0169] In some possible embodiments, determining at least one target object from multiple relationship chain objects may specifically include the following steps as Figure 4 shown:

[0170] In step S401, obtain the sorting index of each relationship chain object among the multiple relationship chain objects.

[0171] Among them, the sorting index of each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object.

[0172] In the embodiments of the present disclosure, by predicting multiple operation behaviors for each relationship chain object, the operation prediction data of each relationship chain object is obtained. The operation prediction data includes the probability of the first account executing each operation behavior. For example, in the video push scenario, the operation prediction data includes click-through rate, like rate, follow rate, comment rate, and long-play rate, etc. The sorting index of each relationship chain object can be obtained by calculating the operation prediction data.

[0173] In some possible embodiments, obtaining the sorting index of each relationship chain object among the multiple relationship chain objects may specifically include the following steps as Figure 5 shown:

[0174] In step S501, use the operation prediction model to predict the operation behaviors of each relationship chain object, and obtain the operation prediction data corresponding to each relationship chain object.

[0175] Among them, the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and its associated accounts, and the feature data of the historical objects published by the associated accounts of the second account.

[0176] In the embodiments of the present disclosure, use the trained operation prediction model to predict the operation behaviors of each relationship chain object, and obtain the operation prediction data corresponding to each relationship chain object.

[0177] In some possible embodiments, the generation method of the operation prediction model may specifically include the following steps as Figure 6 shown:

[0178] In step S601, obtain the first training sample set.

[0179] Among them, each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, the feature data of the historical objects published by the associated account of the second account, and the actual operation data of the second account on the historical objects; the associated accounts of the second account include at least one of the accounts followed by the second account, the accounts that follow the second account, the accounts that mutually follow the second account, and the accounts that have at least one common friend with the second account.

[0180] In the embodiments of the present disclosure, the second account is any user account accessing the push system; the interaction feature data between the second account and the associated account of the second account may include the operation behavior data executed by the second account on the historical objects published by its associated account; in the video push scenario, the interaction feature data may include the viewing data of the historical videos published by the associated account of the second account, such as the viewing duration, as well as the comment behavior, like behavior, forwarding behavior, etc. on the video; correspondingly, the actual operation data of the second account on the historical objects may include the actual viewing data, actual comment behavior, actual like behavior, etc.

[0181] In step S603, obtain a first preset machine learning model.

[0182] In step S605, input the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, and the feature data of the historical objects published by the associated account of the second account into the first preset machine learning model to obtain the estimated operation data corresponding to the second account.

[0183] In step S607, based on the estimated operation data and the actual operation data, train the first preset machine learning model to obtain a trained operation prediction model.

[0184] In the above embodiments, through the first preset machine learning model, learn the features in multiple different dimensions to obtain the estimated operation data of the second account for the historical objects, then compare the estimated operation data with the actual operation data of the second account to calculate the loss value, and after multiple iterations, the loss value converges to obtain a trained operation prediction model; in actual applications, input the corresponding features of each relationship chain object into this operation prediction model to obtain the estimated operation data corresponding to the first account for each relationship chain object.

[0185] In some possible embodiments, the operation prediction data includes multiple pieces of prediction data corresponding to multiple operation behaviors; among them, the multiple operation behaviors may include click behavior, like behavior, comment behavior, follow behavior, forward behavior, long video play behavior, etc.; the prediction data corresponding to each operation behavior can be predicted by a corresponding task sub-model, that is, the first preset machine learning model includes multiple task sub-models, and each task sub-model is used to predict an operation behavior; the prediction data output by each task sub-model is a value between zero and one, representing the probability that the user will perform the corresponding operation behavior. Correspondingly, the loss function used in a single task sub-model can be a cross-entropy loss function, and the cross-entropy loss function is as shown in the following formula (1):

[0186] Loss i =-(y * log(y')+(1 - y) * log(1 - y')) (1)

[0187] Where Loss i represents the error of predicting operation behavior i; y' represents the probability of the occurrence of predicting operation behavior i, that is, the prediction data corresponding to operation behavior i; y is the sample label, taking the value of 1 for positive samples and 0 for negative samples.

[0188] Correspondingly, the loss value of the first preset machine learning model is the sum of the errors of each task sub-model.

[0189] In some possible embodiments, as Figure 7 shown, the first preset machine learning model includes a first spatial vector representation module (embedding1), a second spatial vector representation module (embedding2), a third spatial vector representation module (embedding3), a fourth spatial vector representation module (embedding4), each preprocessing neural network connected to each spatial vector representation module, a feature fusion module, a feature connection module, and multiple operation behavior prediction sub-modules;

[0190] Correspondingly, the above-mentioned step of inputting the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, and the feature data of the historical objects published by the associated account of the second account into the first preset machine learning model to obtain the predicted operation data corresponding to the second account may specifically include the following steps:

[0191] First, according to the first spatial vector representation module in the first preset machine learning model, determine the first feature vector corresponding to the feature data of the second account; according to the second spatial vector representation module in the first preset machine learning model, determine the second feature vector corresponding to the feature data of the associated account of the second account; according to the third spatial vector representation module in the first preset machine learning model, determine the third feature vector corresponding to the interaction feature data between the second account and the associated account of the second account; according to the fourth spatial vector representation module in the first preset machine learning model, determine the fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the second account; input the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into the corresponding preprocessing neural networks respectively, and output the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector.

[0192] Secondly, connect the fused feature vector after fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and input them into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the prediction data output by each operation behavior prediction sub-module.

[0193] In the above embodiment, in addition to considering the object and the user's own features, the interaction features between users are also considered. Through the learning of multi-dimensional features, the prediction data output by the model is more accurate and comprehensive; and the above first preset machine learning model adopts a multi-task model, sharing underlying features. In this way, the model structure can be simplified, the parameter storage can be reduced, and thus the online prediction performance can be improved and the overall push efficiency can be improved.

[0194] In step S503, according to the operation prediction data corresponding to each relationship chain object, determine the sorting index of each relationship chain object.

[0195] In the embodiment of the present disclosure, when the operation prediction data includes multiple prediction data corresponding to multiple operation behaviors, different weights can be assigned to each operation behavior, and then based on the weights corresponding to each operation behavior, the multiple prediction data are weighted and summed to obtain the weighted sum result corresponding to each relationship chain object, and this weighted sum result is the sorting index of this relationship chain object.

[0196] In step S403, classify multiple relationship chain objects to obtain a set of relationship chain objects corresponding to at least one object type; the object types of each relationship chain object in the set of relationship chain objects are the same.

[0197] In this step, according to the object type of each relationship chain object, a plurality of relationship chain objects are classified to obtain at least one relationship chain object set, and the relationship chain objects in the relationship chain object set have the same object type.

[0198] In step S405, the relationship chain objects in the relationship chain object set corresponding to each object type with a sorting index greater than or equal to a preset index are determined as target objects, and at least one target object is obtained.

[0199] In the embodiments of the present disclosure, one target object can be selected corresponding to each object type. Therefore, the server only needs to determine one target object from each relationship chain object set. Thus, the server can use the relationship chain object with the largest sorting index in each relationship chain object set as the target object; that is, the preset index corresponding to each relationship chain object set is numerically equal to the sorting index of the target object.

[0200] In the above embodiments, by determining the sorting index of each relationship chain object, the sorting index to a certain extent reflects the degree of interest of the first account in the relationship chain object. By classifying a plurality of relationship chain objects, for each object type, a more representative relationship chain object, that is, a relationship chain object with good performance in the sorting index, is selected as the target object. In subsequent steps, by determining the preference index data of the first account for the target object, the preference index data of the first account for a specific object type is determined.

[0201] In step S205, the preference index data of the first account for each object type in at least one object type is determined.

[0202] In step S207, the preference index data of the first account for the non-relationship chain objects among the multiple objects to be pushed is determined.

[0203] In the embodiments of the present disclosure, considering that different users may have different preferences for different types of relationship chain objects at different times, therefore, by determining the preference index data of the first account for each object type and the preference index data for non-relationship chain objects, the target object to be finally pushed to it is determined. Among them, the non-relationship chain object refers to an object published by an account that has no connection such as attention and mutual friends with the first account.

[0204] In some possible embodiments, the above determination of the preference index data of the first account for each object type in at least one object type may include the following steps as Figure 8 shown:

[0205] In step S801, according to the preference prediction model, the preference index data corresponding to each target object in at least one target object is determined.

[0206] Among them, the preference prediction model is trained based on the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account.

[0207] In a specific embodiment, according to the preference prediction model, determining the preference index data corresponding to each target object among at least one target object may include: inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data corresponding to the target object.

[0208] Among them, the historical operation data of the first account may include the data of the objects operated by the first account within a recent preset time period, including the feature data of the operated objects and the specific operation behaviors performed on these objects; the non-relationship-chain objects among the multiple objects to be pushed refer to the objects published by accounts that have no association with the first account during the current push process, such as the video 4 to be pushed in the above example.

[0209] In step S803, based on the object type of each target object, the preference index data corresponding to each target object is used as the preference index data of the first account for the object type.

[0210] Correspondingly, in some possible embodiments, the above-mentioned determination of the preference index data of the first account for the non-relationship-chain objects among the multiple objects to be pushed may include: inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data of the non-relationship-chain objects.

[0211] In the above embodiments, the preference prediction model learns multiple different-dimensional features related to the target object and predicts the preference index data of the first account for the target object; the preference index data of the target object can represent the degree of interest of the first account in the target object, or can also represent the degree of interest of the first account in the object type of the target object; in addition, if no feature data of relationship-chain objects is input into the preference prediction model, the preference index data of the first account for non-relationship-chain objects can be obtained. In this way, through the preference prediction model to perform preference prediction on multiple different types of target objects and non-relationship-chain objects, the degree of interest of the first account in different object types and non-relationship-chain objects can be obtained, and the user's needs can be accurately identified to achieve personalized push.

[0212] In some possible embodiments, the manner of generating the preference prediction model may include the following steps as shown in Figure 9 :

[0213] In step S901, obtain a second training sample set.

[0214] Among them, each second training sample in the second training sample set includes the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account towards the historical objects; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts that follow the third account, the accounts that mutually follow the third account, and the accounts that have at least one common friend with the third account.

[0215] In the embodiments of the present disclosure, the third account is any user account accessing the push system; the above-mentioned historical behavior data of the third account may include the operation behaviors performed by the third account on the operated objects within a preset past time period, such as viewing behaviors, liking behaviors, commenting behaviors, etc.; the non-relationship-chain objects among the multiple historical push objects corresponding to the third account refer to the objects published by the accounts that have no connection with the third account among the multiple historical push objects finally sent by the server during the historical push to the third account.

[0216] In step S903, obtain a second preset machine learning model.

[0217] In step S905, input the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects.

[0218] Among them, the preference prediction index data of the historical objects characterizes the preference degree of the third account for the object type of the historical objects.

[0219] In the embodiments of the present disclosure, through the second preset machine learning model, learn the features in multiple different dimensions to obtain the predicted behavior data of the third account for the historical objects, and based on the predicted behavior data, determine the preference prediction index data of the third account for the historical objects, that is, the preference degree of the third account for the object type of the historical objects.

[0220] In a specific embodiment, the predicted behavior data may include positive behaviors such as clicks, follows, comments, etc., may also include negative behaviors such as short broadcasts, disapprovals, etc., and may also include data reflecting the long-term behaviors of users, such as viewing duration, tendency to continue browsing, etc.; when the predicted behavior data includes any of the above-mentioned multiple behaviors, different weights can be assigned to different predicted behaviors according to actual needs to adapt to different application scenarios and improve the object push effect.

[0221] In step S907, based on the actual behavior data of the third account for historical objects, the actual preference index data of the historical objects is determined.

[0222] Correspondingly, the actual behavior data corresponds to the above-mentioned predicted behavior data, and the corresponding actual preference index data is determined according to the positive behavior, negative behavior or long-term behavior actually performed by the third account.

[0223] In step S909, based on the actual preference index data and the preference prediction index data, the second preset machine learning model is trained to obtain a trained preference prediction model.

[0224] In the above embodiment, by constructing the second preset machine learning model, learning the features in multiple different dimensions, obtaining the preference prediction index data of the third account for historical objects, comparing the preference prediction index data with the actual preference index data, calculating the loss value, and after multiple iterations, the loss value converges to obtain a trained preference prediction model; in actual application, inputting the corresponding features of each target object into the preference prediction model, the preference index data of the first account for each target object can be obtained.

[0225] In some possible embodiments, as Figure 10 shown, the second preset machine learning model includes a first spatial vector representation module (embedding1), a second spatial vector representation module (embedding2), a third spatial vector representation module (embedding3), a fourth spatial vector representation module (embedding4), and a fifth spatial vector representation module (embedding5), a feature connection module, and a behavior prediction module;

[0226] Correspondingly, the above-mentioned input of the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects may include the following steps:

[0227] First, according to the first spatial vector representation module in the second preset machine learning model, determine the first feature vector corresponding to the feature data of the third account; according to the second spatial vector representation module in the second preset machine learning model, determine the second feature vector corresponding to the feature data of the associated accounts of the third account; according to the third spatial vector representation module in the second preset machine learning model, determine the third feature vector corresponding to the historical behavior data of the third account; according to the fourth spatial vector representation module in the second preset machine learning model, determine the fourth feature vector corresponding to the feature data of the historical objects published by the associated accounts of the third account; according to the fifth spatial vector representation module in the second preset machine learning model, determine the fifth feature vector corresponding to the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account.

[0228] Secondly, connect the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector and input them into the behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data.

[0229] Among them, the multiple behavior prediction data may include the prediction data corresponding to any one of the positive behaviors, negative behaviors, or long-term behaviors mentioned in the above embodiments.

[0230] Secondly, according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data, determine the preference prediction index data.

[0231] In a specific embodiment, the following formula (2) can be used to determine the preference prediction index data:

[0232] R = Σaction * weight action (2)

[0233] Among them, R represents the preference prediction index data; action represents the behavior prediction data; weight action represents the weight information corresponding to the behavior prediction data.

[0234] In the above embodiments, through a specific model structure, different spatial vector representation modules are used to learn the features of each dimension of historical objects, and trained with a large number of samples of different types of historical objects, so that the preference index data output by the trained preference prediction model is more in line with the actual preferences of users, in order to achieve personalized control of the exposure frequency of relationship-chain objects in the push, which can improve the distribution of relationship-chain objects, enhance the stickiness between users and the platform, and improve user retention.

[0235] In step S209, based on the preference index data of each object type and the preference index data of non-relationship-chain objects, push objects to the first account.

[0236] In the embodiments of the present disclosure, relatively better preference index data is selected from the preference index data of each object type and the preference index data of non-relational chain objects. For example, object types or non-relational chain objects whose preference index data meets certain numerical conditions are all pushed to the first account. It can be understood that there may be multiple relational chain objects under each object type. When actually pushing to the first account, all relational chain objects under the object type whose preference index data meets the conditions can be pushed to the first account, or, according to the sorting index of each relational chain object calculated in the above embodiments, the top N relational chain objects are pushed to the first account; where the size of N can be determined according to actual needs.

[0237] In some possible embodiments, the object pushing to the first account based on the preference index data of each object type and the preference index data of non-relational chain objects may include the following steps:

[0238] First, when the preference index data of the non-relational chain object is greater than the preference index data of each object type, the non-relational chain object among the multiple objects to be pushed is used as the target push object; or; when the preference index data of one object type is greater than the preference index data of the non-relational chain object, the object type whose preference index data is greater than or equal to the preset index data is used as the target object type, and the objects among the multiple objects to be pushed whose object type is the target object type are used as the target push objects;

[0239] Among them, the preset index data can be a fixed value determined according to experience, or it can also be dynamically adjusted during each push process; that is, the preset index data can change dynamically according to the preference index data of each object type and the preference index data of non-relational chain objects in actual applications; for example, the preset index data can be equal to the largest preference index data among the preference index data of each object type and the preference index data of non-relational chain objects.

[0240] In a specific embodiment, as mentioned in the above embodiment, if the preference index data of the non-relationship-chain object - the video 4 to be pushed is greater than the preference index data of the other three relationship-chain objects, then the video 4 to be pushed is taken as the target push object. During the subsequent push process, the server pushes the video 4 to the user account A, and the videos 1 - 3 to be pushed are not pushed to the first account; or, when the preference index data of the video 1 to be pushed is the largest, the video type of the video 1 to be pushed is taken as the target video type. Since the publishing account B of the video 1 is the user account followed by the user A, therefore, if there is also a video 5 to be pushed in the set of videos to be pushed, and the publishing account of the video 5 to be pushed is the user account F, and the user account F is also the account followed by the user account A, then at this time, the server pushes the video 1 and the video 5 to the user account A.

[0241] Secondly, send an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

[0242] In a specific embodiment, the push system of the server sends an object push feedback to the terminal corresponding to the first account. The object push feedback includes the identification information of the target push object. The terminal corresponding to the first account can obtain the target push object from the object resource pool of the server according to the identification information of the target push object to display it on the terminal interface.

[0243] In summary, in the embodiments of the present disclosure, during the process of pushing an object to the first account, based on the relationship-chain data of the first account, relationship-chain objects that perform better in terms of sociality are selected from multiple objects to be pushed, so as to push objects that the first account is more interested in to the first account, which can solve the problem of poor push effect caused by the traditional push method easily ignoring relationship-chain objects; and by determining the preferences of different users for different types of relationship-chain objects or non-relationship-chain objects, the personalized needs of different users can be met, the push accuracy can be improved, and the user retention can be increased.

[0244] Figure 11 is a block diagram of an object push device shown according to an exemplary embodiment, the object push device. Refer to Figure 11 The device includes a first determination module 1101, a second determination module 1102, a third determination module 1103, a fourth determination module 1104, and a push module 1105;

[0245] The first determination module 1101 is configured to determine multiple relationship-chain objects from multiple objects to be pushed obtained according to the relationship-chain data of the first account; the object type of each relationship-chain object in the multiple relationship-chain objects is determined according to the relationship between the publishing account of the relationship-chain object and the first account;

[0246] The second determination module 1102 is configured to determine at least one target object from multiple relationship chain objects; the at least one target object corresponds to at least one object type, and the object type of each target object in the at least one target object is different;

[0247] The third determination module 1103 is configured to determine the preference index data of the first account for each object type in the at least one object type;

[0248] The fourth determination module 1104 is configured to determine the preference index data of the first account for the non-relationship chain objects among the multiple objects to be pushed;

[0249] The push module 1105 is configured to perform object pushing to the first account based on the preference index data of each object type and the preference index data of the non-relationship chain objects.

[0250] In some possible embodiments, the relationship chain data includes the associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts that follow the first account, the accounts that mutually follow the first account, and the accounts that have at least one common friend with the first account;

[0251] The first determination module 1101 is further configured to obtain the publishing account of each object among the multiple objects to be pushed; when it is determined that the publishing account is an associated account, determine the object corresponding to the publishing account as a relationship chain object.

[0252] In some possible embodiments, the first determination module 1101 is further configured to, when the publishing account is the account followed by the first account, determine the object corresponding to the publishing account as a relationship chain object of the first object type; or; when the publishing account is the account that follows the first account, determine the object corresponding to the publishing account as a relationship chain object of the second object type; or; when the publishing account is the account that mutually follows the first account, determine the object corresponding to the publishing account as a relationship chain object of the third object type; or; when the publishing account is the account that has at least one common friend with the first account, determine the object corresponding to the publishing account as a relationship chain object of the fourth object type.

[0253] In some possible embodiments, the second determination module 1102 includes:

[0254] An acquisition sub-module is configured to obtain the sorting index of each relationship chain object among the multiple relationship chain objects; the sorting index of each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object;

[0255] A classification sub-module, configured to perform classification on multiple relationship chain objects to obtain a set of relationship chain objects corresponding to at least one object type; each relationship chain object in the set of relationship chain objects has the same object type;

[0256] A determination sub-module, configured to perform determining relationship chain objects with a sorting index greater than or equal to a preset index in the set of relationship chain objects corresponding to each object type as target objects, to obtain at least one target object.

[0257] In some possible embodiments, the acquisition sub-module is further configured to perform predicting the operation behavior of each relationship chain object through an operation prediction model to obtain operation prediction data corresponding to each relationship chain object; the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account; and determine the sorting index of each relationship chain object according to the operation prediction data corresponding to each relationship chain object.

[0258] In some possible embodiments, it further includes a generation module of the operation prediction model;

[0259] The generation module of the operation prediction model includes:

[0260] A first acquisition sub-module, configured to perform acquiring a first training sample set; each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, the feature data of the historical objects published by the associated accounts of the second account, and the actual operation data of the second account on the historical objects; the second account is any user account accessing the push system; the associated accounts of the second account include at least one of the accounts followed by the second account, the accounts following the second account, the accounts mutually following the second account, and the accounts having at least one common friend with the second account;

[0261] A second acquisition sub-module, configured to perform acquiring a first preset machine learning model;

[0262] An input sub-module, configured to perform inputting the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account into the first preset machine learning model to obtain predicted operation data corresponding to the second account;

[0263] A training sub-module, configured to perform training on a first preset machine learning model based on predicted operation data and actual operation data to obtain a trained operation prediction model.

[0264] In some possible embodiments, an input sub-module is configured to perform:

[0265] Determine a first feature vector corresponding to the feature data of a second account according to a first spatial vector representation module in the first preset machine learning model;

[0266] Determine a second feature vector corresponding to the feature data of an associated account of the second account according to a second spatial vector representation module in the first preset machine learning model;

[0267] Determine a third feature vector corresponding to the interaction feature data between the second account and the associated account of the second account according to a third spatial vector representation module in the first preset machine learning model;

[0268] Determine a fourth feature vector corresponding to the feature data of a historical object published by an associated account of the second account according to a fourth spatial vector representation module in the first preset machine learning model;

[0269] Input the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into corresponding preprocessing neural networks respectively, and output the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector;

[0270] Connect the fused feature vector after fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and input them into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the predicted data output by each operation behavior prediction sub-module.

[0271] In some possible embodiments, the third determination module 1103 includes:

[0272] A first determination sub-module, configured to determine preference index data corresponding to each target object in at least one target object according to a preference prediction model; the preference prediction model is trained according to the feature data of a third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of non-relationship-chain objects among multiple historical push objects corresponding to the third account;

[0273] A second determination sub-module, configured to use the preference index data corresponding to each target object as the preference index data of the first account for the object type based on the object type of each target object.

[0274] In some possible embodiments, the first determination sub-module is further configured to input the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of the non-relational chain objects among the multiple objects to be pushed into the preference prediction model, so as to obtain the preference index data corresponding to the target object.

[0275] In some possible embodiments, the fourth determination module 1104 is configured to input the obtained feature data of the first account, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of the non-relational chain objects among the multiple objects to be pushed into the preference prediction model, so as to obtain the preference index data of the non-relational chain objects.

[0276] In some possible embodiments, it further includes a generation module of the preference prediction model;

[0277] The generation module of the preference prediction model includes:

[0278] The first acquisition sub-module is configured to acquire a second training sample set; each second training sample in the second training sample set includes the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, the feature data of the non-relational chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account for the historical objects; the third account is any user account accessing the push system; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts following the third account, the accounts mutually following the third account, and the accounts having at least one common friend with the third account;

[0279] The second acquisition sub-module is configured to acquire a second preset machine learning model;

[0280] The input sub-module is configured to input the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of the non-relational chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model, so as to obtain the preference prediction index data of the historical objects; the preference prediction index data of the historical objects represents the preference degree of the third account for the object type of the historical objects.

[0281] The determination sub-module is configured to determine the actual preference index data of the historical objects according to the actual behavior data of the third account for the historical objects;

[0282] A training sub-module, configured to perform training on a second preset machine learning model based on actual preference metric data and preference prediction metric data to obtain a trained preference prediction model.

[0283] In some possible embodiments, the input sub-module is further configured to perform:

[0284] Determine a first feature vector corresponding to the feature data of the third account according to the first spatial vector representation module in the second preset machine learning model;

[0285] Determine a second feature vector corresponding to the feature data of the associated account of the third account according to the second spatial vector representation module in the second preset machine learning model;

[0286] Determine a third feature vector corresponding to the historical behavior data of the third account according to the third spatial vector representation module in the second preset machine learning model;

[0287] Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the third account according to the fourth spatial vector representation module in the second preset machine learning model;

[0288] Determine a fifth feature vector corresponding to the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account according to the fifth spatial vector representation module in the second preset machine learning model;

[0289] After connecting the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector, input them into the behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data;

[0290] Determine the preference prediction metric data according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data.

[0291] In some possible embodiments, the push module 1105 includes:

[0292] A determination sub-module, configured to perform when the preference metric data of the non-relationship-chain object is greater than the preference metric data of each object type, use the non-relationship-chain object among the multiple objects to be pushed as the target push object; or; when there is a preference metric data of an object type greater than the preference metric data of the non-relationship-chain object, use the object type with the preference metric data greater than or equal to the preset metric data as the target object type, and use the object with the object type of the target object type among the multiple objects to be pushed as the target push object;

[0293] A sending sub-module, configured to perform sending an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

[0294] Regarding the device in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0295] Figure 12 It is a block diagram of an electronic device for object push shown according to an exemplary embodiment.

[0296] The electronic device can be a server or a terminal device, and its internal structure diagram can be as Figure 12 shown. The electronic device includes a processor, a memory, and a network interface connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it realizes an object push method.

[0297] Those skilled in the art can understand that Figure 12 the structure shown in

[0298] is only a block diagram of a part of the structure related to the solution of the present disclosure, and does not constitute a limitation on the electronic device to which the solution of the present disclosure is applied. The specific electronic device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0299] In an exemplary embodiment, an electronic device is further provided, including: a processor; a memory for storing executable instructions of the processor; wherein, the processor is configured to execute the instructions to implement the object push method as in the embodiments of the present disclosure.

[0300] In an exemplary embodiment, a computer-readable storage medium is further provided. When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device can execute the object push method in the embodiments of the present disclosure.

[0301] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above various methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0302] After considering the specification and practicing the invention disclosed herein, those skilled in the art will readily conceive of other embodiments of the present disclosure. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and examples are only to be considered as exemplary, and the true scope and spirit of the present disclosure are pointed out by the following claims.

[0303] It should be understood that the present disclosure is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of the present disclosure is only limited by the appended claims.

Claims

1. An object pushing method, characterized in that, Including: Determine a plurality of relationship chain objects from the plurality of objects to be pushed obtained according to the relationship chain data of the first account; The object type of each relationship chain object among the plurality of relationship chain objects is determined according to the relationship between the publishing account of the relationship chain object and the first account; the relationship chain object is an object published by an associated account of the first account; Determine at least one target object from the plurality of relationship chain objects; the at least one target object corresponds to at least one object type, and the object type of each target object among the at least one target object is different; Determine the preference index data of the first account for each object type among the at least one object type; Determine the preference index data of the first account for non-relationship chain objects among the plurality of objects to be pushed; the non-relationship chain object refers to an object published by an account that has no connection with the first account; Based on the preference index data of each object type and the preference index data of the non-relationship chain objects, push objects to the first account; The pushing objects to the first account based on the preference index data of each object type and the preference index data of the non-relationship chain objects includes: When the preference index data of the non-relationship chain object is greater than the preference index data of each object type, use the non-relationship chain object among the plurality of objects to be pushed as the target push object; or; when there is a preference index data of an object type greater than the preference index data of the non-relationship chain object, use the object type whose preference index data is greater than or equal to the preset index data as the target object type, and use the object whose object type among the plurality of objects to be pushed is the target object type as the target push object; Send an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

2. The object pushing method according to claim 1, wherein The relationship chain data includes the associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts that follow the first account, the accounts that mutually follow the first account, and the accounts that have at least one common friend with the first account; The determining a plurality of relationship chain objects from the plurality of objects to be pushed obtained includes: Obtain the publishing account of each object among the plurality of objects to be pushed; When it is determined that the publishing account is the associated account, determine the object corresponding to the publishing account as the relationship chain object.

3. The object pushing method according to claim 2, wherein After the determining the object corresponding to the publishing account as the relationship chain object when it is determined that the publishing account is the associated account, it further includes: When the publishing account is the account followed by the first account, determine the object corresponding to the publishing account as a relationship chain object of the first object type; Or; When the publishing account is the account that follows the first account, determine the object corresponding to the publishing account as a relationship chain object of the second object type; Or; When the publishing account is the account that is mutually followed with the first account, determine that the object corresponding to the publishing account is a relationship chain object of the third object type; Or; When the publishing account is the account that has at least one common friend with the first account, determine that the object corresponding to the publishing account is a relationship chain object of the fourth object type.

4. The object pushing method according to claim 1, characterized in that The determining of at least one target object from the multiple relationship chain objects includes: Obtaining a sorting index for each relationship chain object among the multiple relationship chain objects; the sorting index of each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object; Classifying the multiple relationship chain objects to obtain a set of relationship chain objects corresponding to at least one object type; each relationship chain object in the set of relationship chain objects has the same object type; Determine the relationship chain objects with sorting indexes greater than or equal to a preset index in the set of relationship chain objects corresponding to each object type as the target objects, to obtain the at least one target object.

5. The object pushing method according to claim 4, wherein The obtaining of the sorting index for each relationship chain object among the multiple relationship chain objects includes: Estimating the operation behavior of each relationship chain object through an operation prediction model to obtain the operation prediction data corresponding to each relationship chain object; the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account; Determine the sorting index of each relationship chain object according to the operation prediction data corresponding to each relationship chain object.

6. The object pushing method according to claim 5, characterized in that The generation method of the operation prediction model includes: Obtaining a first training sample set; each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, the feature data of the historical objects published by the associated accounts of the second account, and the actual operation data of the second account on the historical objects; the second account is any user account accessing the push system; the associated accounts of the second account include at least one of the accounts followed by the second account, the accounts following the second account, the accounts mutually followed with the second account, and the accounts that have at least one common friend with the second account; Obtaining a first preset machine learning model; Inputting the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account into the first preset machine learning model to obtain the predicted operation data corresponding to the second account; Training the first preset machine learning model based on the predicted operation data and the actual operation data to obtain the trained operation prediction model.

7. The object pushing method according to claim 6, wherein The operation prediction data includes multiple prediction data corresponding to multiple operation behaviors; the step of inputting the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account into the first preset machine learning model to obtain the predicted operation data corresponding to the second account includes: Determining a first feature vector corresponding to the feature data of the second account according to the first spatial vector representation module in the first preset machine learning model; Determining a second feature vector corresponding to the feature data of the associated accounts of the second account according to the second spatial vector representation module in the first preset machine learning model; Determining a third feature vector corresponding to the interaction feature data between the second account and the associated accounts of the second account according to the third spatial vector representation module in the first preset machine learning model; Determining a fourth feature vector corresponding to the feature data of the historical objects published by the associated accounts of the second account according to the fourth spatial vector representation module in the first preset machine learning model; Respectively inputting the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into corresponding preprocessing neural networks, and outputting the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector; Connecting the fused feature vector obtained by fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and inputting the result into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the prediction data output by each operation behavior prediction sub-module.

8. The object pushing method according to claim 1, wherein The step of determining the preference index data of the first account for each object type in the at least one object type includes: Determining the preference index data corresponding to each target object in the at least one target object according to a preference prediction model; the preference prediction model is trained according to the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of non-relationship-chain objects among the multiple historical push objects corresponding to the third account; Based on the object type of each target object, using the preference index data corresponding to each target object as the preference index data of the first account for the object type.

9. The object pushing method according to claim 8, wherein The step of determining the preference index data corresponding to each target object in the at least one target object according to the preference prediction model includes: Inputting the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data corresponding to the target object.

10. The object pushing method according to claim 8 or 9, characterized in that Determining the preference index data of the non-relationship-chain objects among the multiple objects to be pushed by the first account includes: Inputting the feature data of the first account obtained, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data of the non-relationship-chain objects.

11. The object pushing method according to claim 8, wherein The generation method of the preference prediction model includes: Obtaining a second training sample set; each second training sample in the second training sample set includes the feature data of a third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account for the historical objects; the third account is any user account accessing the push system; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts following the third account, the accounts mutually following the third account, and the accounts having at least one common friend with the third account; Obtaining a second preset machine learning model; Inputting the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects; the preference prediction index data of the historical objects characterizes the preference degree of the third account for the object type of the historical objects; Determining the actual preference index data of the historical objects according to the actual behavior data of the third account for the historical objects; Training the second preset machine learning model based on the actual preference index data and the preference prediction index data to obtain a trained preference prediction model.

12. The object pushing method according to claim 11, wherein The step of inputting the feature data of the third account, the feature data of the associated account of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated account of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical objects includes: Determining a first feature vector corresponding to the feature data of the third account according to the first spatial vector representation module in the second preset machine learning model; Determining a second feature vector corresponding to the feature data of the associated account of the third account according to the second spatial vector representation module in the second preset machine learning model; Determining a third feature vector corresponding to the historical behavior data of the third account according to the third spatial vector representation module in the second preset machine learning model; Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated accounts of the third account according to the fourth spatial vector representation module in the second preset machine learning model; Determine a fifth feature vector corresponding to the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account according to the fifth spatial vector representation module in the second preset machine learning model; Connect the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector and input them into the behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data; Determine the preference prediction index data according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data.

13. An object pushing device, characterized in that, Includes: A first determination module configured to execute determining multiple relationship-chain objects from the multiple objects to be pushed obtained according to the relationship-chain data of the first account; The object type of each relationship-chain object in the multiple relationship-chain objects is determined according to the relationship between the publishing account of the relationship-chain object and the first account; The relationship-chain object is an object published by an associated account of the first account; A second determination module configured to execute determining at least one target object from the multiple relationship-chain objects; the at least one target object corresponds to at least one object type, and the object type of each target object in the at least one target object is different; A third determination module configured to execute determining the preference index data of the first account for each object type in the at least one object type; A fourth determination module configured to execute determining the preference index data of the first account for the non-relationship-chain objects among the multiple objects to be pushed; the non-relationship-chain object refers to an object published by an account that has no connection with the first account; A push module configured to execute pushing objects to the first account based on the preference index data of each object type and the preference index data of the non-relationship-chain objects; The push module includes: A determination sub-module configured to execute when the preference index data of the non-relationship-chain object is greater than the preference index data of each object type, using the non-relationship-chain object among the multiple objects to be pushed as the target push object; or; when there is an object type whose preference index data is greater than the preference index data of the non-relationship-chain object, using the object type whose preference index data is greater than or equal to the preset index data as the target object type, and using the object whose object type in the multiple objects to be pushed is the target object type as the target push object; A sending sub-module configured to execute sending an object push feedback to the terminal corresponding to the first account; the object push feedback includes the identification information of the target push object.

14. The object pushing device according to claim 13, wherein The relationship chain data includes the associated accounts of the first account; the associated accounts of the first account include at least one of the accounts followed by the first account, the accounts that follow the first account, the accounts that mutually follow the first account, and the accounts that have at least one common friend with the first account; The first determination module is further configured to execute obtaining the publishing account of each object to be pushed; when determining that the publishing account is the associated account, determining the object corresponding to the publishing account as the relationship chain object.

15. The object push device according to claim 14, wherein The first determination module is further configured to execute when the publishing account is the account followed by the first account, determining the object corresponding to the publishing account as a relationship chain object of the first object type; or when the publishing account is the account that follows the first account, determining the object corresponding to the publishing account as a relationship chain object of the second object type; or when the publishing account is the account that mutually follows the first account, determining the object corresponding to the publishing account as a relationship chain object of the third object type or when the publishing account is the account that has at least one common friend with the first account, determining the object corresponding to the publishing account as a relationship chain object of the fourth object type.

16. The object pushing device according to claim 13, wherein The second determination module includes: An acquisition sub-module, configured to execute obtaining the sorting index of each relationship chain object in the multiple relationship chain objects; the sorting index of each relationship chain object is determined according to the operation prediction data corresponding to each relationship chain object; A classification sub-module, configured to execute classifying the multiple relationship chain objects to obtain a relationship chain object set corresponding to at least one object type; the object types of each relationship chain object in the relationship chain object set are the same; A determination sub-module, configured to execute determining the relationship chain objects with sorting indexes greater than or equal to a preset index in the relationship chain object set corresponding to each object type as the target objects, to obtain the at least one target object.

17. The object push device according to claim 16, wherein The acquisition sub-module is further configured to execute predicting the operation behavior of each relationship chain object through an operation prediction model to obtain the operation prediction data corresponding to each relationship chain object; the operation prediction model is trained according to the feature data of the second account, the feature data of the associated accounts of the second account, the interaction feature data between the second account and the associated accounts of the second account, and the feature data of the historical objects published by the associated accounts of the second account; and determining the sorting index of each relationship chain object according to the operation prediction data corresponding to each relationship chain object.

18. The object pushing device according to claim 17, characterized in that, It further includes a generation module of the operation prediction model; The generation module of the operation prediction model includes: The first acquisition sub-module is configured to execute the acquisition of the first training sample set; each first training sample in the first training sample set includes the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, the feature data of the historical objects published by the associated account of the second account, and the actual operation data of the second account on the historical objects; the second account is any user account accessing the push system; the associated account of the second account includes at least one of the accounts followed by the second account, the accounts following the second account, the accounts mutually following the second account, and the accounts having at least one common friend with the second account; The second acquisition sub-module is configured to execute the acquisition of the first preset machine learning model; The input sub-module is configured to execute the input of the feature data of the second account, the feature data of the associated account of the second account, the interaction feature data between the second account and the associated account of the second account, and the feature data of the historical objects published by the associated account of the second account into the first preset machine learning model to obtain the estimated operation data corresponding to the second account; The training sub-module is configured to execute the training of the first preset machine learning model based on the estimated operation data and the actual operation data to obtain the trained operation prediction model.

19. The object pushing device according to claim 18, wherein, The input sub-module is configured to execute: Determine the first feature vector corresponding to the feature data of the second account according to the first spatial vector representation module in the first preset machine learning model; Determine the second feature vector corresponding to the feature data of the associated account of the second account according to the second spatial vector representation module in the first preset machine learning model; Determine the third feature vector corresponding to the interaction feature data between the second account and the associated account of the second account according to the third spatial vector representation module in the first preset machine learning model; Determine the fourth feature vector corresponding to the feature data of the historical objects published by the associated account of the second account according to the fourth spatial vector representation module in the first preset machine learning model; Respectively input the first feature vector, the second feature vector, the third feature vector, and the fourth feature vector into the corresponding preprocessing neural network, and output the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector; Connect the fused feature vector after fusing the first feature vector and the second feature vector with the preprocessed first feature vector, second feature vector, third feature vector, and fourth feature vector, and input them into multiple operation behavior prediction sub-modules in the first preset machine learning model to obtain the estimated data output by each operation behavior prediction sub-module.

20. The object pushing device according to claim 13, characterized in that, The third determination module includes: The first determination sub-module is configured to determine, according to the preference prediction model, the preference index data corresponding to each target object among the at least one target object; the preference prediction model is trained according to the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account. The second determination sub-module is configured to, based on the object type of each target object, use the preference index data corresponding to each target object as the preference index data of the first account for the object type.

21. The object push device according to claim 20, wherein The first determination sub-module is further configured to input the obtained feature data of the first account, the feature data of the publishing account of the target object, the feature data of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data corresponding to the target object.

22. The object push device according to claim 20 or 21, wherein The fourth determination module is configured to input the obtained feature data of the first account, the feature data of the publishing account of the target object, the historical operation data of the first account, and the feature data of the non-relationship-chain objects among the multiple objects to be pushed into the preference prediction model to obtain the preference index data of the non-relationship-chain objects.

23. The object pushing device according to claim 20, characterized in that, It further includes a generation module of the preference prediction model; The generation module of the preference prediction model includes: The first acquisition sub-module is configured to acquire a second training sample set; each second training sample in the second training sample set includes the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account, and the actual behavior data of the third account for the historical objects; the third account is any user account accessing the push system; the associated accounts of the third account include at least one of the accounts followed by the third account, the accounts following the third account, the accounts mutually following the third account, and the accounts having at least one common friend with the third account. The second acquisition sub-module is configured to acquire a second preset machine learning model; An input sub-module, configured to input the feature data of the third account, the feature data of the associated accounts of the third account, the historical behavior data of the third account, the feature data of the historical objects published by the associated accounts of the third account, and the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account into the second preset machine learning model to obtain the preference prediction index data of the historical object; the preference prediction index data of the historical object represents the preference degree of the third account for the object type of the historical object. A determination sub-module, configured to determine the actual preference index data of the historical object according to the actual behavior data of the third account for the historical object. A training sub-module, configured to train the second preset machine learning model based on the actual preference index data and the preference prediction index data to obtain a trained preference prediction model.

24. The object pushing device according to claim 23, characterized in that, The input sub-module is further configured to perform: Determine a first feature vector corresponding to the feature data of the third account according to the first spatial vector representation module in the second preset machine learning model; Determine a second feature vector corresponding to the feature data of the associated accounts of the third account according to the second spatial vector representation module in the second preset machine learning model; Determine a third feature vector corresponding to the historical behavior data of the third account according to the third spatial vector representation module in the second preset machine learning model; Determine a fourth feature vector corresponding to the feature data of the historical objects published by the associated accounts of the third account according to the fourth spatial vector representation module in the second preset machine learning model; Determine a fifth feature vector corresponding to the feature data of the non-relationship-chain objects among the multiple historical push objects corresponding to the third account according to the fifth spatial vector representation module in the second preset machine learning model; Connect the first feature vector, the second feature vector, the third feature vector, the fourth feature vector, and the fifth feature vector and input them into the behavior prediction module in the second preset machine learning model to obtain multiple behavior prediction data; Determine the preference prediction index data according to the multiple behavior prediction data and the weight information of each behavior prediction data in the multiple behavior prediction data.

25. An electronic device, characterized in that, Including: A processor; A memory for storing instructions executable by the processor; Wherein, the processor is configured to execute the instructions to implement the object push method according to any one of claims 1 to 12.

26. A computer-readable storage medium, characterized in that, When the instructions in the computer-readable storage medium are executed by the processor of the electronic device, the electronic device is enabled to execute the object push method according to any one of claims 1 to 12.

27. A computer program product, characterized in that, The computer program product includes a computer program, the computer program is stored in a readable storage medium, and at least one processor of the computer device reads and executes the computer program from the readable storage medium, so that the computer device executes the object push method according to any one of claims 1 to 12.

Images