Method and device for acquiring published object, electronic equipment and storage medium

Abstract

The invention relates to the technical field of computers, in particular to the technical field of artificial intelligence, and provides a release object acquisition method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring first publishing contents and first content feedback data of a plurality of candidate publishing objects on at least one external platform, and sharing data on an external platform associated with a target platform; for each candidate publishing object, executing the following operations: respectively obtaining a corresponding content label set according to respective basic attribute information of at least one first publishing content associated with one candidate publishing object; according to the obtained at least one content label set, first content feedback data associated with one candidate publishing object and external platform sharing data are combined, and corresponding object potential parameters are obtained; and selecting at least one of the plurality of candidate release objects as a target release object according to the obtained potential parameters of each object. According to the method, the object acquisition accuracy is improved through the multi-dimensional data.

Description

Background Art

[0002] In the information flow recommendation scenario, due to the insufficient content volume of the target platform, it is often necessary to introduce publishing objects from external platforms (such as competitor platforms) to publish content in order to increase the available high-quality content of the recommendation system.

[0003] In related technologies, in order to solve the above problems, the number of followers of accounts on external platforms or the number of content views is often sorted from large to small based on the influence rankings of external platforms, and the publishing objects ranked at the top are introduced.

[0004] Although the number of account followers and content views can, to a certain extent, represent the influence of the publisher on external platforms, this indicator is one-sided because the influence rankings can only represent the overall audience preferences on external platforms, while there may be significant differences between the audience groups on the target platform and the audience groups on external platforms. Therefore, accounts introduced from external platforms often cannot meet the actual needs of the audience groups on the target platform.

[0005] This may easily lead to the final introduced publication object having excellent performance on the external platform but mediocre performance on the target platform, which greatly wastes the traffic of the recommendation system. Therefore, how to accurately introduce more popular publication objects to the target platform is an urgent problem to be solved. Summary of the invention

[0006] The embodiments of the present application provide a method, device, electronic device and storage medium for obtaining a published object, so as to improve the accuracy of obtaining a published object.

[0007] An embodiment of the present application provides a method for obtaining a publication object, including:

[0008] Acquire the first published content and first content feedback data of each of the plurality of candidate publishing objects on at least one external platform, and the external platform sharing data associated with each of the plurality of candidate publishing objects on the target platform; the external platform is a content platform on which the candidate publishing object has registered, and the target platform is a content platform on which the candidate publishing object has not registered;

[0009] For each candidate publishing object, the following operations are performed respectively: according to basic attribute information of at least one first publishing content associated with a candidate publishing object, a corresponding content tag set is obtained respectively; according to the obtained at least one content tag set, in combination with first content feedback data and external platform sharing data associated with the candidate publishing object, a corresponding object potential parameter is obtained, wherein the object potential parameter represents: the estimated popularity of the publishing content uploaded by the candidate publishing object on the target platform;

[0010] At least one is selected from the plurality of candidate release objects as a target release object according to the obtained potential parameters of each object.

[0011] An embodiment of the present application provides a device for obtaining a publication object, including:

[0012] A data acquisition unit, used to acquire first published content and first content feedback data of each of a plurality of candidate publishing objects on at least one external platform, and external platform sharing data associated with each of the plurality of candidate publishing objects on a target platform; the external platform is a content platform on which the candidate publishing object has registered, and the target platform is a content platform on which the candidate publishing object has not registered;

[0013] The potential estimation unit is used to perform the following operations for each candidate publishing object: obtain a corresponding content tag set according to basic attribute information of at least one first published content associated with a candidate publishing object; obtain a corresponding object potential parameter according to the obtained at least one content tag set, in combination with first content feedback data and external platform sharing data associated with the candidate publishing object, wherein the object potential parameter represents: the estimated popularity of the published content uploaded by the candidate publishing object on the target platform;

[0014] The object acquisition unit is used to select at least one from the multiple candidate release objects as a target release object according to the obtained potential parameters of each object.

[0015] Optionally, the potential estimation unit is specifically used for:

[0016] Determining an external platform influence parameter of the candidate publishing object according to first content feedback data associated with the candidate publishing object within a preset time range;

[0017] Predicting a content quality parameter of the candidate publication object based on a content tag set of each of the at least one first publication content associated with the candidate publication object;

[0018] Predicting a target supply and demand parameter of the candidate publishing object on the target platform according to the external platform sharing data associated with the candidate publishing object and the second content feedback data associated with the target platform within the preset time range;

[0019] Determine the object potential parameter of the candidate release object according to the external platform influence parameter, content quality parameter and target supply and demand parameter of the candidate release object.

[0020] Optionally, the potential estimation unit is specifically used for:

[0021] For each foreign platform, do the following:

[0022] Determining an absolute influence parameter of the one candidate publishing object on the one external platform according to first content feedback data associated with each of the one candidate publishing object and other candidate publishing objects on the one external platform within the preset time range;

[0023] Determine, according to first content feedback data associated with the one candidate publishing object on the one external platform within each unit time of the preset time range, an impact quality parameter of the one candidate publishing object on the one external platform;

[0024] An external platform influence parameter of the candidate release object is determined according to the absolute influence parameter and the influence quality parameter of the candidate release object on the at least one external platform.

[0025] Optionally, the first content feedback data associated with each candidate publishing object on an external platform within the preset time range includes: the number of followers currently following the candidate publishing object, the total number of content views and the total number of content interactions within the preset time range;

[0026] The potential estimation unit is specifically used for:

[0027] Obtaining an attention comparison value determined based on the number of attention objects of the one candidate publishing object and the number of attention objects in a specified order among the numbers of attention objects of each of the plurality of candidate publishing objects;

[0028] Obtaining a viewing comparison value determined based on the total number of content views of the one candidate publishing object and the total number of content views of the plurality of candidate publishing objects in a specified order;

[0029] Obtaining an interaction comparison value determined based on the total number of content interactions of the one candidate publishing object and the total number of content interactions of the plurality of candidate publishing objects in a specified order;

[0030] The attention contrast value, the viewing contrast value, and the interaction contrast value are weightedly summed to obtain an absolute influence parameter of the candidate publishing object on the external platform.

[0031] Optionally, within each unit time of the preset time range, the first content feedback data associated with one candidate publishing object on an external platform includes: the number of content views within the unit time, and the number of content interactions within at least one statistical dimension within the unit time;

[0032] The potential estimation unit is specifically used for:

[0033] For each statistical dimension, perform the following operations:

[0034] Obtaining a comparison value of the number of content interactions and the number of content views of the candidate publishing object in each unit time under a statistical dimension;

[0035] Determine the quality parameter under the one statistical dimension based on the weighted square sum of the comparison values ​​of the number of times in each unit time;

[0036] Based on the quality parameters of the candidate release object in each statistical dimension, an impact quality parameter of the candidate release object on the external platform is obtained.

[0037] Optionally, the difference between the current unit time and the latest unit time within the preset time range is negatively correlated with the weight corresponding to the current unit time.

[0038] Optionally, the potential estimation unit is specifically used for:

[0039] Normalizing and averaging the absolute impact parameter and the impact quality parameter of the candidate release object on the same external platform to obtain a corresponding target impact parameter;

[0040] The target influence parameters of the candidate release object on each external platform are weightedly summed to obtain the external platform influence parameter of the candidate release object.

[0041] Optionally, the weight corresponding to each target impact parameter is positively correlated with the similarity between the corresponding external platform and the target platform.

[0042] Optionally, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions;

[0043] The potential estimation unit is specifically used for:

[0044] Based on the content tag set, first tag distribution features of each first published content associated with the candidate published object and second tag distribution features of each target published content are counted, wherein the tag distribution features represent the total number of occurrences of each content tag; the target published content is the first published content whose publishing time is within a preset sequence range among the first published content;

[0045] The first label distribution feature and the second label distribution feature are input into a trained quality assessment model to obtain a content quality parameter of the candidate release object.

[0046] Optionally, the device further comprises:

[0047] The model training unit is used to train the trained quality assessment model in the following way:

[0048] The quality assessment model to be trained is trained iteratively based on the training sample set to obtain a trained quality assessment model; each training sample includes a sample publishing object and a sample label; the sample publishing object has published content on at least one external platform and the target platform, and the sample label is determined according to sample content feedback data associated with the sample publishing object on the target platform; wherein each iterative training performs the following steps:

[0049] Select training samples from the training sample set, and input sample label distribution characteristics of sample release objects in the training samples and associated with the at least one external platform into the quality assessment model to be trained to obtain sample content quality parameters of the sample release objects;

[0050] Based on the difference between the sample content quality parameter and the corresponding sample label, the parameters of the quality assessment model to be trained are adjusted.

[0051] Optionally, the sample label distribution characteristics include: a first sample label distribution characteristic of each sample content published by the sample publishing object on the at least one external platform, and a second sample label distribution characteristic of each target sample content, wherein the target sample content is the sample content whose publishing time is within a preset order range among the sample contents.

[0052] Optionally, the potential estimation unit is specifically used for:

[0053] Determine potential supply and demand parameters of the candidate publishing object according to external platform sharing data associated with the candidate publishing object and other candidate publishing objects respectively within the preset time range;

[0054] Obtaining current supply and demand parameters of the candidate publishing object according to the third content feedback data associated with the target platform within the preset time range and the content tag set of each first published content;

[0055] Based on the potential supply and demand parameters and the current supply and demand parameters, target supply and demand parameters corresponding to the one candidate release object are determined.

[0056] Optionally, the external platform sharing data includes: the total amount of content and the total number of times the first published content is shared to the target platform;

[0057] The potential estimation unit is specifically used for:

[0058] Obtaining a quantity comparison value determined based on the total quantity of content associated with the one candidate publishing object and the total quantity of content in a specified order among the total quantities of content associated with each of the plurality of candidate publishing objects;

[0059] Obtaining a sharing comparison value determined based on the total number of shares associated with the candidate publishing object and the total number of shares associated with each of the plurality of candidate publishing objects in a specified order;

[0060] The sum of the quantity comparison value and the sharing comparison value is used as a potential supply and demand parameter of the candidate release object.

[0061] Optionally, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions; the second content feedback data includes: the number of completed views and the number of uncompleted views of the second published content;

[0062] The potential estimation unit is specifically used for:

[0063] For each content tag, do the following:

[0064] Determining a scarcity parameter corresponding to the content tag according to the number of completed views and the number of uncompleted views of the second published content containing the content tag within the preset time range;

[0065] Determine a frequency parameter corresponding to the one content tag according to each content tag set corresponding to each first published content associated with the one candidate publishing object within the preset time range;

[0066] The target content tags whose frequency parameter ranking is within a specified order range are determined, and the sum of the products of the scarcity parameter and the frequency parameter of each target content tag is used as the current supply and demand parameter of the one candidate release object.

[0067] Optionally, the potential estimation unit is specifically used for:

[0068] For each content tag, do the following:

[0069] Obtaining the number of occurrences of a content tag in the first published content published by the candidate publishing object within the preset time range, and the total number of occurrences of each content tag;

[0070] Obtaining the total number of candidate publishing objects and the number of candidate publishing objects containing the one content tag;

[0071] Based on the number of occurrences, the total number of occurrences, the total data and the quantity, a frequency parameter corresponding to the one content tag is determined.

[0072] Optionally, the basic attribute information has at least one category; and the potential estimation unit is specifically used for:

[0073] For each first publication, perform the following operations:

[0074] Input each category of basic attribute information of a first published content into the trained multi-label classification model and represent them as multiple attribute vectors respectively;

[0075] For each type of basic attribute information, multiple attribute vectors corresponding to the basic attribute information are represented as a target vector through one-dimensional convolution processing; features of the target vector are extracted based on the multimodal attention network in the multi-label classification model to obtain a corresponding target vector;

[0076] The target vectors corresponding to various basic attribute information are spliced ​​to obtain a spliced ​​vector;

[0077] A content tag set of the first published content is obtained based on the splicing vector prediction.

[0078] An electronic device provided by an embodiment of the present application includes a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the above-mentioned methods for obtaining a published object.

[0079] An embodiment of the present application provides a computer-readable storage medium, which includes a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any one of the above-mentioned methods for obtaining a published object.

[0080] An embodiment of the present application provides a computer program product, which includes a computer program, and the computer program is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device performs the steps of any one of the above-mentioned methods for obtaining a published object.

[0081] The beneficial effects of this application are as follows:

[0082] The embodiments of the present application provide a method, device, electronic device and storage medium for obtaining a publishing object. Because in the present application, when introducing a publishing object from an external platform to a target platform, it is not only one-sidedly considering the number of account fans or the number of content views in the external platform, but comprehensively considering the publishing content of the publishing object on the external platform, the related content feedback data, and the external platform sharing data associated with each target platform, to estimate the popularity of the publishing object on the external platform after uploading the publishing content on the target platform, so as to screen out the publishing object that is more suitable for introducing into the target platform based on this parameter. Because in this way, the content tag set and content feedback data of the publishing object's publishing content on the external platform can reflect the influence of the publishing object on the external platform, and further combined with the external platform sharing data of the publishing object, it can reflect the influence of the publishing object on the target platform. This application comprehensively considers the publishing objects suitable for the target platform according to various indicators, and improves the accuracy of obtaining the publishing objects.

[0083] Furthermore, based on the determination of the target publishing object, the target publishing object can be directly introduced into the target platform, or a publishing object with similar characteristics to the target publishing object can be introduced and registered on the target platform, so as to more accurately introduce a more popular publishing object to the target platform.

[0084] Other features and advantages of the present application will be described in the following description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0085] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0086] Figure 1 A schematic diagram of an application scenario in an embodiment of the present application;

[0087] Figure 2 A flowchart of a method for obtaining a publication object provided in an embodiment of the present application;

[0088] Figure 3A A schematic diagram of a wide-angle image listed in the embodiments of the present application;

[0089] Figure 3B A schematic diagram of another wide-angle image listed in the embodiments of the present application;

[0090] Figure 4A A schematic diagram of a rotating image listed in an embodiment of the present application;

[0091] Figure 4B This is another schematic diagram of a rotating image listed in the embodiments of the present application;

[0092] Figure 5 This is a flow chart of a method for obtaining a content tag set in an embodiment of the present application;

[0093] Figure 6 A schematic diagram of a method for extracting a content tag set of a video in an embodiment of the present application;

[0094] Figure 7 A flow chart of a method for obtaining object potential parameters in an embodiment of the present application;

[0095] Figure 8 A schematic diagram of the calculation logic of an external platform influencing parameter in an embodiment of the present application;

[0096] Fig. 9 A schematic diagram of a calculation logic of an absolute influencing parameter in an embodiment of the present application;

[0097] Fig.10 A schematic diagram of a calculation logic of a quality parameter affecting an embodiment of the present application;

[0098] Fig.11 A schematic diagram of a decision tree model in an embodiment of the present application;

[0099] Fig.12 A schematic diagram of a calculation logic of a target supply and demand parameter in an embodiment of the present application;

[0100] Fig.13 A logic block diagram of a process for obtaining a published object in an embodiment of the present application;

[0101] Fig.14 A schematic diagram of the composition structure of an acquisition device for publishing an object in an embodiment of the present application;

[0102] Fig.15 A schematic diagram of the structure of an electronic device in an embodiment of the present application;

[0103] Fig.16 It is a schematic diagram of the composition structure of another electronic device in an embodiment of the present application. DETAILED DESCRIPTION

[0104] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical solution of the present application, rather than all of the embodiments. Based on the embodiments recorded in the application documents, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the technical solution of the present application.

[0105] The following is an introduction to some concepts involved in the embodiments of the present application.

[0106] Publishing object: An organization or individual that publishes content on a content platform (such as an information flow platform). The videos or articles it publishes can be browsed by objects and generate other object behaviors. Object behavior refers to the collective term for operations such as browsing, clicking, following, liking, commenting, sharing, and forwarding performed by objects in information flow products, which can be used to construct content feedback data.

[0107] Content feedback data: After the publisher uploads the published content to an external platform, the published content will be recommended to other objects, and the viewing objects will generate some object behaviors in response to the published content. Content feedback data refers to the relevant data obtained based on the statistics of these object behaviors, which can also be called object consumption data, such as the number of followers of the publisher, the number of times the published content is viewed, liked, commented on, and shared, etc.

[0108] External platform and target platform: The external platform is the content platform where the candidate publishing object has been registered, which can also be called the off-site platform. The target platform is the content platform where the candidate publishing object has not yet been registered, which can also be called the on-site platform. Specifically, the target platform in this application refers to the platform where the publishing object needs to be introduced, and the external platform refers to any platform except the target platform.

[0109] External platform sharing data: refers to data related to content shared to the target platform by at least one external platform. It can be obtained by counting the content shared to the site from different external platforms, including but not limited to the total number of first published content shared to the target platform and the total number of shares.

[0110] External platform influence parameter: For a publishing object, this parameter reflects the social influence of the publishing object on the external platform. It can be determined based on the absolute influence parameter and the influence quality parameter. The absolute influence parameter reflects the absolute social influence of an account within a preset time range, while the influence quality parameter reflects the quality of the social influence.

[0111] Content quality parameter: indicates the quality of the content published by the candidate publishing object when it registers on the target platform and uploads the published content.

[0112] Target supply and demand parameters: Indicates the scarcity of the content published when the candidate publisher registers on the target platform and uploads the published content. It can be determined based on the potential supply and demand parameters and the current supply and demand parameters. The potential supply and demand parameters indicate the potential scarcity score when there is no similar content on the target platform and the viewers need to share it from external platforms to the target platform, while the current supply and demand parameters indicate the current scarcity score when there is similar content on the target platform but the content volume is insufficient.

[0113] Object potential parameter: represents the estimated popularity of the content uploaded by a candidate release object on the target platform, that is, the potential of a candidate release object to be introduced into the target platform. Generally, the higher the estimated popularity, the greater the potential of the candidate release object to be introduced, and the larger the value of the corresponding object potential parameter, and vice versa.

[0114] Decision model: It is a mathematical model used for business decision-making. Due to the complexity of the social and economic system and the crisscrossing decision-making factors, it is difficult for any decision maker to make the best decision based on intuition and experience alone. Therefore, in modern scientific decision-making, we often use the methods of natural science and mathematical tools to establish the relationship formula and model between the decision variables to reflect the essence of the decision-making problem and simplify the complex decision-making problem. The general form of the decision model is V = F (Ai, Sj), where: V is the value target; Ai is a controllable decision factor; Sj is an uncontrollable decision factor. For example, a company has two factories that can produce two products, A and B, but their production capacity is different and their profits are also different. In order to achieve the best economic benefits for the enterprise, the optimal production strategy should be sought, that is, the production plan that maximizes the benefits should be made. To this end, the decision model is first listed, and then the optimal one is found through the model.

[0115] Object: The user who uses the terminal, such as a person.

[0116] Information flow products: a product form of terminal program, on which various video and article information can be obtained. Among them, terminal program refers to products running on the terminal, with various functions, such as chat software, news software, etc.

[0117] Content tag: used to indicate the concepts and contents actually contained in the content, including detailed information such as characters, places, scenes, etc. involved in the content. The content tag set in the embodiment of the present application includes at least one content tag.

[0118] Term frequency-inverse document frequency (TF-IDF): is a commonly used weighting technique for information retrieval and data mining. TF stands for term frequency, and IDF stands for inverse document frequency. In the embodiment of the present application, the frequency parameter corresponding to each content tag can be counted based on this technique.

[0119] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0120] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0121] Among them, machine learning (ML) is the core of artificial intelligence, a branch of artificial intelligence, and a large field. It allows computers to have the same learning ability as humans, simulate and realize human learning behavior and ability, and have the ability to identify and judge like humans, which can be regarded as bionics. The core of machine learning is data, algorithm (model), and computing power (computer computing power). It is a multi-field interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in how computers simulate or realize human learning behavior, in order to acquire new knowledge or skills, and reorganize existing knowledge structures to continuously improve their performance. As in the embodiments of the present application, relevant models and networks can be trained by machine learning techniques.

[0122] Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between people and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. As in the embodiments of the present application, content-related text features (such as video subtitles, article text, speech-to-text, etc.) can be processed by natural language processing technology.

[0123] Computer vision (CV) is a simulation of biological vision using computers and related equipment. Its main task is to obtain three-dimensional information of the corresponding scene by processing the collected pictures or videos, just like humans and many other creatures do every day. As in the embodiments of the present application, content-related image features (such as video picture information, picture information in news, etc.) can be processed by natural language processing technology.

[0124] Automatic Speech Recognition (ASR) is a technology that converts human speech into text. Speech recognition is a multidisciplinary field that is closely linked to many disciplines such as acoustics, phonetics, linguistics, digital signal processing theory, information theory, and computer science. Due to the diversity and complexity of speech signals, current speech recognition systems can only achieve satisfactory performance under certain restrictions, or can only be applied to certain specific occasions. The performance of a speech recognition system generally depends on the following four factors: the size of the recognition vocabulary and the complexity of speech; the quality of the speech signal; a single speaker or multiple speakers; and hardware. In an embodiment of the present application, when the content is audio or video, feature extraction of the attributes of the content may involve processing of speech data in audio and video content. Specifically, speech recognition may be performed using ASR technology to obtain corresponding ASR text, and then text features may be further extracted.

[0125] In an embodiment of the present application, quality prediction models, decision models, etc. can be trained based on the above-mentioned machine learning, natural language processing, computer vision, automatic speech recognition and other technologies, so as to extract features related to the published object and content based on the trained models and the network, and realize the prediction of object potential parameters.

[0126] With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

[0127] The following is a brief introduction to the design concept of the embodiment of the present application:

[0128] In the information flow recommendation scenario, due to the lack of content (such as videos), it is often necessary to introduce external publishers to increase the available high-quality content for the recommendation system. There are two mainstream methods. In addition to the above-mentioned methods based on influence rankings, another method is manual judgment. This method first formulates a set of judgment criteria, manually searches for well-known publishers that meet the criteria in multiple platforms outside the information flow, and introduces them.

[0129] However, this method also has certain shortcomings. When making manual judgments, since the judgment criteria can only represent the subjective feelings of a few people such as the standard setters, it cannot represent all the objects on the site. Therefore, although the introduced accounts meet the pre-set standards, they are often not liked by the objects in the recommendation system on the site. In the end, most of the introduced publishing objects are lost, which greatly wastes manpower and material resources.

[0130] In view of this, the embodiments of the present application provide a method, device, electronic device and storage medium for obtaining a publishing object. Because in the present application, when introducing a publishing object from an external platform to a target platform, it is not only one-sidedly considering the number of account fans or the number of content views in the external platform, but comprehensively considering the publishing content of the publishing object on the external platform, the related content feedback data, and the external platform sharing data associated with each target platform, to estimate the popularity of the publishing object on the external platform after uploading the publishing content on the target platform, so as to screen out the publishing object that is more suitable for introducing into the target platform based on this parameter. Because in this way, the content tag set and content feedback data of the publishing object's publishing content on the external platform can reflect the influence of the publishing object on the external platform, and further combined with the external platform sharing data of the publishing object, it can reflect the influence of the publishing object on the target platform. This application comprehensively considers the publishing objects suitable for the target platform according to various indicators, and improves the accuracy of obtaining the publishing objects.

[0131] Furthermore, based on the determination of the target publishing object, the target publishing object can be directly introduced into the target platform, or a publishing object with similar characteristics to the target publishing object can be introduced and registered on the target platform, so as to more accurately introduce a more popular publishing object to the target platform.

[0132] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.

[0133] like Figure 1 As shown, it is a schematic diagram of an application scenario in an embodiment of the present application. The application scenario diagram includes two terminal devices 110 and a server 120.

[0134] In the embodiment of the present application, the terminal device 110 includes but is not limited to mobile phones, tablet computers, laptop computers, desktop computers, e-book readers, intelligent voice interaction devices, smart home appliances, car terminals and other devices; the terminal device may be installed with a client related to content publishing, content recommendation, etc., which may be a software (such as a browser, short video software, content sharing platform, etc.), or a web page, a small program, etc. The server 120 is a background server corresponding to the software or web page, small program, etc., or a server specifically used to obtain the publishing object, which is not specifically limited in the present application. The server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms.

[0135] It should be noted that the method for obtaining the publishing object in each embodiment of the present application can be executed by an electronic device, which can be a terminal device 110 or a server 120, that is, the method can be executed by the terminal device 110 or the server 120 alone, or by the terminal device 110 and the server 120 together. For example, when executed by the server 120 alone, first, the server 120 obtains the first publishing content and first content feedback data of each of the multiple candidate publishing objects on at least one external platform, as well as the external platform sharing data associated with the target platform, and predicts the object potential parameters of each candidate publishing object based on these data. Then, the server 120 selects at least one from the multiple candidate publishing objects as the target publishing object according to the obtained object potential parameters.

[0136] In an optional implementation, the terminal device 110 and the server 120 may communicate with each other via a communication network.

[0137] In an optional implementation, the communication network is a wired network or a wireless network.

[0138] It should be noted that Figure 1 The figure is only an example. In fact, the number of terminal devices and servers is not limited and is not specifically limited in the embodiments of the present application.

[0139] In an embodiment of the present application, when there are multiple servers, the multiple servers can form a blockchain, and the servers are nodes on the blockchain; as disclosed in the embodiment of the present application, the object-related data and content-related data involved in the method for obtaining the published object can be stored on the blockchain, for example, published content, content feedback data, external platform sharing data, basic attribute information, content tag sets, etc.

[0140] In addition, the embodiments of the present application can be applied to various scenarios, including but not limited to cloud technology, artificial intelligence, smart transportation, assisted driving and other scenarios.

[0141] It can be understood that in the specific implementation of this application, data related to object consumption data, accounts, content, etc. is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0142] The following describes the method for obtaining the published object provided by the exemplary embodiment of the present application in combination with the application scenario described above and with reference to the accompanying drawings. It should be noted that the above application scenario is only shown to facilitate understanding of the spirit and principles of the present application, and the implementation of the present application is not limited in this respect.

[0143] See also Figure 2 As shown, it is an implementation flow chart of a method for obtaining a publication object provided in an embodiment of the present application. Taking the server as the execution subject as an example, the specific implementation process of the method is as follows S21-S23:

[0144] S21: The server obtains first published content and first content feedback data of each of the plurality of candidate publishing objects on at least one external platform, and respective external platform sharing data associated with the target platform.

[0145] Among them, the external platform is the content platform where the candidate publishing object has registered, and the target platform is the content platform where the candidate publishing object has not registered. When each publishing object registers on a content platform, it can get an account, so the publishing object in this application can also be called an account or account owner.

[0146] Specifically, the external platform can also be referred to as the off-site platform, and the internal platform can also be referred to as the on-site platform. In the embodiment of the present application, the on-site platform can be set according to actual needs, and can be any content platform that needs to introduce the publishing object (such as information flow product A), and the off-site platform refers to any content platform other than the on-site platform (such as other information flow products other than information flow product A).

[0147] In the embodiment of the present application, content refers to any type of information flow, including but not limited to: videos, articles, information, questions and answers, topics, encyclopedias, posts, products, music, and e-books.

[0148] Candidate publishing objects refer to publishing objects that have published content on external platforms. The method for obtaining publishing objects in the embodiment of the present application can be executed once every preset period (such as one week), or can be triggered to execute irregularly. Each time it is executed, in S21, the first published content, first content feedback data and external platform sharing data within the preset time range can be obtained.

[0149] The preset time range can be set according to actual needs, such as the past 90 days, and this article does not make specific limitations.

[0150] In the embodiment of the present application, after the publishing object uploads the published content to the external platform, these published contents will be recommended to other objects, which can also be called audience groups and viewing objects. The viewing objects will generate some object behaviors for these published contents, such as browsing, clicking, following, liking, commenting, sharing, forwarding, completing playback (abbreviated as completion), fast sliding, etc. in the information flow product. Object behavior can be understood as a general term for these operations. The content feedback data in the embodiment of the present application refers to the relevant data obtained based on the statistics of these object behaviors, such as the number of following publishing objects, the number of related times of watching published content, liking published content, commenting on published content, sharing published content, etc.

[0151] Among them, the viewing of published content can be further divided into two categories: completed viewing and incomplete viewing according to whether the object has completely viewed the published content.

[0152] Specifically, there are two types of object sharing behaviors: sharing content within the site to other objects within the site, and sharing content outside the site to other objects within the site. The external platform sharing data in the embodiment of the present application is obtained by statistics and analysis of the latter. By counting the behaviors of sharing content from different platforms outside the site to the site, the external platform sharing data for the target platform is obtained.

[0153] S22: For each candidate publishing object, the following operations are performed respectively: the server obtains corresponding content tag sets according to basic attribute information of at least one first publishing content associated with a candidate publishing object; and obtains corresponding object potential parameters according to the obtained at least one content tag set in combination with first content feedback data and external platform sharing data associated with a candidate publishing object.

[0154] Among them, the object potential parameter represents: the estimated popularity of the content uploaded by a candidate publishing object on the target platform.

[0155] In an embodiment of the present application, the basic attribute information of the published content is related to the modality of the published content, and specifically refers to information that describes the attributes of the published content itself, such as image information, text information, audio information, etc. For example, when the published content is a video, the basic attribute information includes but is not limited to some or all of the following: video frames, audio, and title. For another example, when the published content is news, the basic attribute information includes but is not limited to some or all of the following: text, title, and illustrations. This article does not make specific limitations.

[0156] By analyzing the basic attribute information of the published content, a content tag set corresponding to the published content can be obtained. The content tag set is used to represent the content quality and includes the category (i.e., content tag) to which the published content belongs under at least one attribute dimension.

[0157] This application takes into account that an account is a collection of many contents, and its representation is a complex problem that cannot be represented by a single label. Therefore, this application introduces multiple labels of content granularity. For example, a food account may be a Western food or a Chinese food account, and its expression method may be street photography or professional photography.

[0158] Optionally, the present application represents an account through a fine-grained content tag set, that is, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions.

[0159] Taking the published content as video as an example, this application establishes a multi-label system for video content quality based on basic attribute information such as video frame extraction, audio, title, etc., which can characterize the quality of video content from multiple angles and in all aspects.

[0160] When building a multi-label system for video content quality, you first need to define labels. Most of the general multi-labels describe the content of the video, such as animation, celebrities, etc. This application proposes that the content labels of video content can be defined from the following five attribute dimensions: geometric structure, content, style, picture quality, and shooting skills.

[0161] The geometric structure includes but is not limited to wide angle, 90-degree rotation, 180-degree rotation, stretching, compression, center composition, horizontal line composition, vertical line composition, symmetrical composition, diagonal composition, and guide line composition; the content includes but is not limited to: magazine covers, news reports, promotional posters, web page screenshots, animation screens, game screens, live broadcast screens, natural scenery, art paintings, posed compositions, city close-ups, building close-ups, pet close-ups, food close-ups, single-person close-ups, group photos, dense crowds, reasonable content distribution but no prominent subject, large areas of text and no subject, small content proportion, title-style text pictures, large areas The main body of the image is text, advertisements, character special effects, agriculture, educational themes, science fiction works, concerts, and cluttered screen targets; the styles include but are not limited to: monochrome, cluttered colors, black and white, dark screen, color cast, excessive cold colors, and excessive warm colors; the picture quality includes but is not limited to: black and white edges, lace, single-sided frosted glass, asymmetrical frosted glass, large-area frosted glass, mosaics, QR codes, spliced ​​pictures, comic strips, text enhancement, picture-in-picture, fresh and pure pictures, ghosting, text overlap, text overlap, flowery screen, excessive special effects, picture mirroring, and PS synthesis; shooting techniques include but are not limited to: blur, out of focus, and overexposure.

[0162] Among them, wide angle refers to a sufficiently wide field of view, about 120 degrees or more, and there may be a certain degree of horizontal line bending and deformation, such as Figure 3A and Figure 3B , which are schematic diagrams of two wide-angle images listed in the embodiments of the present application.

[0163] Rotating 90 degrees specifically refers to rotating the image 90 degrees to the left or right. These images contain reference objects such as people, animals, ground, sky or text, and the image rotation can be easily recognized. Figure 4A and Figure 4B The following are schematic diagrams of two types of rotating images listed in the embodiments of the present application. Figure 4A It means to rotate 90 degrees to the left. Figure 4B Indicates a 90 degree rotation to the right.

[0164] Similarly, rotating 180 degrees or rotating other angles (such as rotating 45 degrees) is similar, and this article will not go into details one by one.

[0165] In the above case, for a video content, its content tag set may include content tags under the above five attribute dimensions. Specifically, each attribute dimension may correspond to at least one content tag, such as: rotate 90 degrees (geometric structure dimension), magazine cover (content dimension), monochrome (style dimension), black and white edges (image quality dimension), blur (shooting technique dimension).

[0166] For example: wide angle + 180 degree rotation (geometric structure dimension), magazine cover + single person close-up (content dimension), messy colors (style dimension), picture-in-picture (picture quality dimension), overexposure (shooting technique dimension).

[0167] It should be noted that the five attribute dimensions listed above and the tag categories under the related attribute dimensions are just simple examples, and any division method of the content tag set is applicable to the embodiments of the present application and is not specifically limited here.

[0168] After defining the labels, you can collect the data set. For example, you can organize about 10,000 labeled data to train a multi-label classification model. Then, you can train the model based on the labeled data.

[0169] For example, taking video as an example, the multi-label classification model corresponding to the video content may include pre-trained Swint-T and tiny-bert, as well as a multi-modal attention network (Multi-model transformer).

[0170] Specifically, the frames, audio text and title of a video (sample content) are input into the pre-trained Swint-T and tiny-bert respectively to obtain multiple vectors, and then one-dimensional convolution (Conv1D) is used to transform the multiple vectors into one vector, and then the vector is input into the Multi-model transformer to obtain the final vector and concatenate them together for classification.

[0171] Afterwards, the model classification results are compared with the content labels annotated with the samples, a loss function (such as the cross entropy loss function) is constructed, and the parameters of the multi-label classification model are adjusted based on the loss function.

[0172] Through the above steps, a multi-label classification model can be trained. Based on the multi-label classification model, a content quality multi-label (i.e., a content label set containing multiple content labels) can be output according to the title, audio and frame extraction of the video.

[0173] Among them, Tiny-BERT is a knowledge distillation method specially designed for transformer-based models. The model size is less than 1 / 7 of BERT, but the speed is increased by 9 times, and there is no obvious decline in performance. In the embodiment of the present application, the use of pre-trained Tiny-BERT to extract ASR text and title features can effectively improve the speed of feature extraction. However, this article is not limited to Tiny-BERT. In addition, other models for feature extraction of text data, such as BERT, can also be used, which is not specifically limited in this article.

[0174] It should be noted that the structure of the multi-label classification model listed above is just a simple example. In addition, other structures can also be used. For example, Swint-T can be replaced by the vit model, etc. This article does not make specific limitations.

[0175] After training multiple label classification models in the above manner, an optional implementation is that the basic attribute information has at least one category, and when extracting the content label set of the published content, for each first published content, the following operations are performed respectively, see Figure 5 As shown, it is a flow chart of a method for obtaining a content tag set in an embodiment of the present application, including the following steps S51 to S54:

[0176] S51: Input each type of basic attribute information of a first published content into the trained multi-label classification model, and represent them as multiple attribute vectors respectively.

[0177] S52: For each type of basic attribute information, multiple attribute vectors corresponding to the basic attribute information are represented as a target vector through one-dimensional convolution processing; features of the target vector are extracted based on the multimodal attention network in the multi-label classification model to obtain the corresponding target vector.

[0178] S53: splicing the target vectors corresponding to the various basic attribute information to obtain a splicing vector.

[0179] S54: Obtain a content tag set of the first published content based on the splicing vector prediction.

[0180] The above S51 to S54 can all be implemented through a trained multi-model classification model.

[0181] like Figure 6 As shown, it is a schematic diagram of a method for extracting a content tag set of a video in an embodiment of the present application. For the frame extraction of the video content, a Swint-t model pre-trained with an image dataset (such as Imagenet) is used to extract the frame extraction features to obtain multiple corresponding attribute vectors, such as Figure 6 The z attribute vectors listed in are m1, m2, …, mz respectively; for the audio of the video content, we can first use the automatic speech recognition technology (Automatic Speech Recognition, ASR) to perform speech recognition to obtain the corresponding ASR text, and use the pre-trained Tiny-bert to extract the ASR text features to obtain the corresponding multiple attribute vectors, such as Figure 6The z attribute vectors listed in are e11, e12, ..., e1z; for the title, the pre-trained Tiny-BERT is used to extract the title features, and the z attribute vectors obtained are e21, e22, ..., e2z. Then, one-dimensional convolution (Conv1D) is used to transform the multiple attribute vectors of each type of basic attribute information into a comprehensive vector. Finally, the target vectors corresponding to the extracted frame, audio version and title are concatenated to obtain the final feature vector, and then the content label set can be predicted based on the final feature vector.

[0182] In the embodiment of the present application, a multi-label system for video content quality is established based on features such as video frame extraction, audio, and title, which can characterize the video content quality from multiple angles and in all aspects.

[0183] It should be noted that the content tag set of the second published content or other content in the embodiment of the present application can be obtained through the above-mentioned method, which will not be described one by one here.

[0184] In order to solve the current problems, this application believes that an account has potential, that is, it is suitable to be introduced into the site and can meet at least one of the following three aspects: outside the site, it should have a wide social influence; within the site, its content should be high-quality; within the site, its content should be scarce.

[0185] Based on the above idea, an optional implementation method is to Figure 7 The process shown in S22 includes the following steps S221 to S224:

[0186] S221: Determine an external platform influence parameter of a candidate publishing object according to first content feedback data associated with a candidate publishing object within a preset time range.

[0187] This application considers that an account has potential, that is, it is suitable to be introduced into the site. The account owner should have a wide social influence outside the site. For example, it has a large number of fans, which proves that it has a good audience base. For example, its content is loved by viewers on multiple information flow platforms, which proves that it has a wide penetration rate. Such characteristics can prove its social influence.

[0188] Based on this, in S221, the first content feedback data associated with a candidate publishing object within a preset time range can be used to analyze and determine the external platform influence parameter of the candidate publishing object, which reflects the social influence of the publishing object on the external platform. Among them, the content feedback data can reflect the number of fans of the account, the audience's preference for the content published by the account, and other information.

[0189] Generally speaking, based on the influence ranking method, the number of fans or object views of the accounts on the external platform are sorted from large to small, and the account owners at the top are introduced. The number of fans of an account and the number of object views can represent the social influence of the account owner to a certain extent, but this indicator is one-sided. For example, the number of fans on platform A can only represent the influence of this account on platform A, and cannot represent that the objects on platform B also like this account. Moreover, the number of fans can only represent the influence in the past to a certain extent, and cannot absolutely represent the social influence of the account owner.

[0190] In the embodiment of the present application, the social influence of a candidate release object on an external platform is represented by two parameters: an absolute influence parameter and an influence quality parameter.

[0191] Among them, the absolute influence parameter reflects the absolute social influence of an account within a preset time frame, while the influence quality parameter reflects the quality of social influence. Even for the same users or fans, the higher the consumption quality, the greater the social influence. For example, if a piece of content is seen by 100 people and these 100 people read it carefully, it will definitely have a greater social impact than if only some people read it carefully.

[0192] See also Figure 8 As shown, it is a schematic diagram of the calculation logic of an external platform influencing parameter in an embodiment of the present application. For an account (candidate release object), taking the preset time range as the last 90 days as an example, first, it is necessary to obtain the object consumption data (i.e., first content feedback data) of the account in multiple external platforms (such as platform A, platform B, platform C, platform D, etc.) in the last 90 days, including but not limited to the number of likes, the number of comments, the number of shares, the number of fans, and the number of views.

[0193] like Figure 8 In , the object consumption data of account 1 on platform A, platform B, platform C and platform D in the last 90 days are obtained respectively. Then, for each external platform, the corresponding absolute impact parameter and impact quality parameter are calculated according to the object consumption data of account 1 on the external platform in the last 90 days, as follows: Figure 8 In the example, the absolute influence parameter of account 1 on platform A is denoted as s a1 , the quality parameter is recorded as s a2 , based on s a1 With s a2 The comprehensive impact parameters of account 1 on platform A can be determined, namely Figure 8 A platform influence parameter s a Similarly, the absolute impact parameter of account 1 on platform B is denoted as s b1 , the quality parameter is recorded as s b2 , based on s b1With s b2 The comprehensive impact parameters of account 1 on platform B can be determined, namely Figure 8 The B platform influences the parameters s b :The absolute impact parameter of account 1 on platform C is denoted as s c1 , the quality parameter is recorded as s c2 , based on s c1 With s c2 The comprehensive impact parameters of account 1 on platform C can be determined, namely Figure 8 The C platform affects the parameters s c ; The absolute impact parameter of account 1 on platform D is denoted as s d1 , the quality parameter is recorded as s d2 , based on s d1 With s d2 The comprehensive impact parameters of account 1 on platform D can be determined, namely Figure 8 The D platform affects the parameters s in d .

[0194] When account 1 only registers and posts on the above four external platforms, the external platform influence parameter of account 1 is the weighted sum of the above four comprehensive influence parameters.

[0195] It should be noted that different candidate publishing objects may have different registered external platforms. Therefore, when calculating the external platform impact parameter of a candidate publishing object, it is determined based on the external platform where the candidate publishing object actually registers and publishes. For example, if a candidate publishing object registers and publishes on platform A, platform D, and platform E, the external platform impact parameter corresponding to the candidate publishing object should be the s corresponding to the candidate publishing object. a 、s d and e Perform weighted summation to obtain .

[0196] An optional implementation is that step S221 may be implemented according to the following steps. Specifically, for each external platform, the following operations are performed respectively:

[0197] S2211: Determine an absolute influence parameter of a candidate publishing object on an external platform according to first content feedback data associated with a candidate publishing object and other candidate publishing objects on an external platform within a preset time range.

[0198] In the embodiment of the present application, when calculating the absolute influence parameter of a candidate publishing object on an external platform, the first content feedback data associated with a candidate publishing object on an external platform may be compared with the first content feedback data associated with other candidate publishing objects on an external platform.

[0199] For example, there are 10 accounts in total (there should be more accounts in actual applications, this is just a simple example). By comparing the first content feedback data of account 1 on platform A in the last 90 days with the first content feedback data of each account from account 2 to account 10 on platform A in the last 90 days, the absolute impact parameter of account 1 on platform A can be analyzed and determined. Similarly, by comparing the first content feedback data of account 2 on platform B in the last 90 days with the first content feedback data of each account from account 1 to account 3 to account 10 on platform B in the last 90 days, the absolute impact parameter of account 2 on platform B can be analyzed and determined. And so on.

[0200] Optionally, within a preset time range, the first content feedback data associated with each candidate publishing object on an external platform includes: the number of followers currently following the candidate publishing object, the total number of content views and the total number of content interactions within the preset time range.

[0201] Among them, the object following the candidate release object can be called the follow object, and the number of follow objects associated with a candidate release object on an external platform within a preset time range is the current latest number of follow objects. If the current latest number of followers of account 1 on platform A (referring to the number of followers when the method for obtaining the release object is executed, which is essentially irrelevant to the preset time range) is 1 million, then within the last 90 days, the number of follow objects associated with account 1 on platform A is 1 million.

[0202] The total number of views of content associated with a candidate release object on an external platform within a preset time range refers to the total number of views of the first release content released by the candidate release object during that period of time. For example, if the video playback volume of account 1 on platform A increased by 2 million in the last 90 days, it means that the total number of views of content associated with account 1 on platform A is 2 million.

[0203] The total number of content interactions associated with a candidate publishing object on an external platform within a preset time range refers to the total number of interactions between the viewing object and the content published by the candidate publishing object when the viewing object browses the first published content published by the candidate publishing object during this period of time. Among them, interaction refers to viewing by the viewing object, including but not limited to sharing, liking, forwarding, commenting, etc. Take likes as an example. For example, if account 1 has 500,000 new likes on platform A in the past 90 days, it means that the total number of content interactions associated with account 1 on platform A is 500,000.

[0204] In the case where the first content feedback data includes the above data, refer to Fig. 9 The logic block diagram shown, an optional calculation method of the absolute influence parameter is as follows:

[0205] like Fig. 9As shown, it is a schematic diagram of the calculation logic of an absolute influence parameter in an embodiment of the present application. The calculation process is roughly divided into the following two steps:

[0206] Step 1: Calculate the attention contrast value, viewing contrast value, and interaction contrast value respectively.

[0207] 1) Obtaining an attention comparison value determined based on the number of attention objects of a candidate release object and the number of attention objects in a specified order among the numbers of attention objects of each of the plurality of candidate release objects.

[0208] Specifically, the designated order refers to the designated order position after sorting the number of followers of each of the multiple candidate release objects from small to large (or from large to small). For example, when sorting from large to small, the designated order may refer to the first position, which means the largest number of followers. For another example, the designated order may refer to the second position, which means the second largest number of followers, and so on.

[0209] In the embodiments of the present application, the numerical value represented by the specified order is taken as the maximum value for example (the same applies below and will not be repeated).

[0210] Optionally, the attention comparison value in the embodiment of the present application may be specifically determined by the number of attention objects of a candidate publishing object and the maximum number of attention objects corresponding to multiple candidate publishing objects.

[0211] Specifically, the comparison value may be a logarithmic comparison value. The present application further considers that when the minimum value of the number of attention objects is 0, the corresponding logarithmic value is negative infinity. In actual practice, electronic devices cannot calculate negative infinity, and when the minimum value is 1, the logarithmic value is 0. To avoid the situation where the logarithmic value is negative infinity, the present application may first add 1 to the number of attention objects and then calculate the logarithm when calculating the logarithmic comparison value of the number of attention objects. Based on this, an optional calculation method of the attention comparison value is as follows:

[0212]

[0213] Among them, fans represents the number of fans of the account (referring to the current and latest number of fans).

[0214] For example, multiple candidate release objects are account 1 to account 10, one candidate release object is account 1, the other candidate release objects are account 2 to account 10, and one external platform is platform A:

[0215] The numerator of the above formula represents the logarithmic value (base 10) of the number of people followed by account 1 on platform A + 1, and the denominator represents the logarithmic value (base 10) of the maximum number of people followed by accounts 1 to 10 on platform A + 1.

[0216] It should be noted that the minimum value of fans+1 is 1 (the minimum value of fans is 0), that is, log 10 The minimum value of (fans+1) is 0. In this case, the larger the number of fans, the greater the log 10 (fans+1) is the largest, so max(log 10 (fans+1)) Although the formula means the maximum value of the logarithm, its essence also represents the logarithmic value of the maximum number of followers + 1 (the same below, no further explanation).

[0217] 2) Obtaining a viewing comparison value determined based on the total number of content views of one candidate publishing object and the total number of content views of each of the plurality of candidate publishing objects in a specified order.

[0218] Similarly, when calculating the logarithmic comparison value of the total number of content views, the total number of content views can be added by 1, and then the logarithm can be calculated. Based on this, an optional calculation method for the viewing comparison value is as follows:

[0219]

[0220] Take video as an example, views Represents the total number of times the video was viewed.

[0221] Still taking the above 10 accounts as an example, the numerator of the above formula represents the logarithmic value (base 10) of the total number of video views of account 1 on platform A + 1, and the denominator represents the logarithmic value (base 10) of the maximum total number of video views of accounts 1 to 10 on platform A + 1.

[0222] 3) Obtaining a total number of content interactions based on a candidate publishing object and a total number of content interactions in a specified order among the total number of content interactions of a plurality of candidate publishing objects, and determining an interaction comparison value.

[0223] Similarly, when calculating the logarithmic comparison value of the total number of content interactions, you can first add 1 to the total number of content interactions and then calculate the logarithm. Based on this, an optional calculation method for the interaction comparison value is as follows:

[0224]

[0225] Wherein, interaction_count represents the total number of content interactions, such as the total number of content shares, or the total number of content likes, or the total number of content comments.

[0226] Optionally, this application takes into account that there is no way to deduplicate off-site data. Therefore, the total number of content interactions takes the maximum value of likes, comments, and shares. For example, the total number of likes received by Account 1 on Platform A in the past 90 days is 1 million, the total number of comments is 200,000, and the total number of shares is 500,000. For Account 1, the total number of content interactions can be the maximum value of the above-mentioned total number of likes, total number of comments, and total number of shares, that is, the number when the interaction is likes: 1 million.

[0227] Based on the above, the numerator in the above represents the logarithm of the total number of video likes of account 1 on platform A + 1 (base 10), and the denominator represents the logarithm of the maximum total number of video likes on platform A among accounts 1 to 10 + 1 (base 10).

[0228] The calculation order of the attention contrast value, viewing contrast value and interaction contrast value listed above is not specifically limited herein.

[0229] Step 2: Perform weighted summation of the attention comparison value, viewing comparison value, and interaction comparison value to obtain the absolute influence parameter of a candidate release object on an external platform.

[0230]

[0231] In the above formula, s1 calculates the absolute social influence of an account within 90 days, which is the absolute influence parameter in this application. Among them, w1, w2, and w3 represent weights, which can be determined according to actual needs. For example, if the number of followers is required to have a greater impact, w1 can be set to be larger than w2 and w3. For example, if no specific item is required to have a greater impact, w1 = w2 = w3 = 1 / 3 can be set, and so on. This article does not make specific restrictions.

[0232] It should be noted that the above formula 4 takes the consideration of the attention contrast value, the viewing contrast value and the interaction contrast value as an example. When only some of the above three items are considered, the corresponding parameters can be set to 0.

[0233] It should be noted that the above-listed method for obtaining the logarithmic comparison value is only a simple example. In addition to the above method, each item of first content feedback data may be normalized first and then the ratio may be directly obtained, etc., which is not specifically limited in this document. That is, any method for calculating the absolute impact parameter based on the comparison results of the first content feedback data in multiple accounts is applicable to the embodiments of the present application, which is not specifically limited in this document.

[0234] S2212: Determine an impact quality parameter of a candidate publishing object on an external platform according to first content feedback data associated with a candidate publishing object on an external platform within each unit time of a preset time range.

[0235] Optionally, within each unit time of a preset time range, the first content feedback data associated with a candidate publishing object on an external platform includes: the number of content views within the unit time, and the number of content interactions within at least one statistical dimension within the unit time.

[0236] Among them, the unit time can be set, and the preset time range can be divided according to the unit time. For example, when the preset time range is the last 90 days, the unit time can be set to days; for example, when the preset time range is the last 24 hours, the unit time can be set to hours; for example, when the preset time range is the last year, the unit time can be set to months or weeks, etc. This article does not make specific limitations.

[0237] Among them, the statistical dimensions of the total number of content interactions are the sharing, forwarding, likes, comments, etc. listed above.

[0238] The following example takes the preset time range as the last 90 days and the unit time as day:

[0239] The number of content views associated with a candidate publishing object on an external platform within each unit time of the preset time range, that is, the number of new content views added by Account 1 every day in the past 90 days.

[0240] Taking the statistical dimensions including sharing, likes and comments as an example, the number of content interactions associated with a candidate publishing object on an external platform within each unit time of the preset time range, that is, the number of new likes, comments and shares of account 1 per day in the past 90 days.

[0241] In the case where the first content feedback data includes the above data, refer to Fig.10 As shown in the logic block diagram, an optional calculation method affecting the quality parameters is as follows:

[0242] like Fig.10 As shown, it is a schematic diagram of the calculation logic of a quality parameter affecting the embodiment of the present application. The calculation process is roughly divided into the following two steps:

[0243] Step 1: For each statistical dimension, perform the following operations:

[0244] 1) Obtain a comparison value of the number of content interactions and the number of content views of a candidate release object in each unit time under a statistical dimension.

[0245] Specifically, when calculating the comparison value of the number of times, the method of calculating the logarithmic comparison value of the number of times listed above may also be adopted, or the number of content interactions and the number of content views may be normalized first, and then the ratio may be directly calculated, etc. This article does not make any specific limitations.

[0246] The following takes the logarithmic comparison value of the calculation number as an example:

[0247] In the sharing statistical dimension, Fig.10 In the example, the first numerical comparison value can be calculated based on the number of content views and the number of content shares in a unit time. An optional calculation formula is as follows:

[0248]

[0249] Where n represents the nth unit time, which can be counted from the current date and forward 90 days. For example, if today is the first day, then n is 1, yesterday is the second day, then n is 2, the day before yesterday is the third day, then n = 3, ...; share_count n Represents the number of content shares on day n, video_views n Represents the number of content views on day n.

[0250] In the statistical dimension of likes, the second numerical comparison value can be calculated based on the number of content views and the number of content likes in a unit time. An optional calculation formula is as follows:

[0251]

[0252] Among them, n and video_views n The meaning of is the same as that of formula 5 above, and the repeated parts are not repeated here. n Represents the number of likes on the content on day n.

[0253] In the statistical dimension of comments, the third comparison value can be calculated based on the number of content views and the number of content comments in a unit time. An optional calculation formula is as follows:

[0254]

[0255] Among them, n and video_views n The meaning is the same as that of formula 5 above, and the repeated parts are not repeated here. comment_count n Represents the number of content comments on day n.

[0256] 2) Based on the weighted sum of squares of the comparison values ​​of the number of times in each unit time, determine a quality parameter under a statistical dimension.

[0257] Under the statistical dimension of sharing, an optional calculation formula for the quality parameter corresponding to the first numerical comparison value is as follows:

[0258]

[0259] Under the statistical dimension of likes, an optional calculation formula for the quality parameter corresponding to the second numerical comparison value is as follows:

[0260]

[0261] Under the statistical dimension of comments, an optional calculation formula for the quality parameter corresponding to the third comparison value is as follows:

[0262]

[0263] In the above formulas 8, 9 and 10, N is equal to 90. Indicates weight, representing different

[0264] Data for dates have different weights.

[0265] Step 2: Based on the quality parameters of a candidate release object in each statistical dimension, obtain the impact quality parameters of a candidate release object on an external platform.

[0266] Specifically, the above quality parameters may be squared and summed to obtain the final quality influencing parameter. An optional calculation method is as follows:

[0267]

[0268] In the above formula, s2 calculates the quality of social influence, that is, the influence quality parameter in the embodiment of the present application.

[0269] It should be noted that the above formula 11 takes the quality parameters of the candidate release object under three statistical dimensions into consideration as an example. When only some of the three dimensions are considered, the corresponding parameters can be set to 0. Similarly, when other dimensions are considered, the quality parameters can be calculated in a similar manner and participate in the square root sum.

[0270] Optionally, the difference between the current unit time and the latest unit time within the preset time range is negatively correlated with the weight corresponding to the current unit time.

[0271] Still taking the last 90 days as an example, the latest unit time is today, when n=1, the current unit time is also today, the time difference is 0, and the corresponding weight is the largest; when n=2, the current unit time is yesterday, the time difference is 1 day, and the corresponding weight is the second largest...

[0272] That is, the closer the object behavior is to the current date, the greater the corresponding weight. For example, when n=1, it is the current date, and the corresponding weight is 1; when n=2, it is one day away from the current date, and the corresponding weight is 1 / 2; when n=3, it is two days away from the current date, and the corresponding weight is 1 / 3; and so on.

[0273] In the above implementation, combined with the performance data of the account outside the site, the calculation of the social influence of the account outside the site is split into two parts: absolute social influence and social influence quality. Compared with the method of only considering the number of account fans or content views in related technologies, the true social influence of the account outside the site can be determined more accurately.

[0274] S2213: Determine an external platform impact parameter of a candidate release object according to the absolute impact parameter and impact quality parameter of the candidate release object on at least one external platform.

[0275] In the embodiment of the present application, the external platform influence parameter of a candidate publishing object is the comprehensive result of the influence of the candidate publishing object on various external platforms. Therefore, an optional calculation method of the external platform influence parameter is as follows:

[0276] S22131: Normalize and average the absolute impact parameter and impact quality parameter of a candidate release object on the same external platform to obtain a corresponding target impact parameter.

[0277] by Figure 8 As shown in the example, the A platform corresponding to account 1 affects the parameter s a That is: s a =(s a1 +s a2 ) / 2; Similarly, the B platform affects the parameter s b That is: s b =(s b1 +s b2 ) / 2; C platform influence parameter s c That is: s c =(s c1 +s c2 ) / 2; D platform influence parameter s d That is: s d =(s d1 +s d2 ) / 2.

[0278] In summary, after normalizing the above two scores (absolute influence parameter and influence quality parameter) and taking the average, we can get the social influence score (i.e. influence parameter) of an account on an external platform.

[0279] S22132: Perform weighted summation of target influence parameters of a candidate release object on each external platform to obtain an external platform influence parameter of the candidate release object.

[0280] The above is a calculation method for the social influence of an external platform. Assuming there are 4 external platforms, after obtaining their social influences, they are given different weights for weighted summation. Then, the comprehensive social influence score of an account on all external platforms (i.e., external platform influence parameter) can be obtained, denoted as s total , an optional calculation formula is as follows:

[0281] s total =w a *s a +w b *s b +w c *s c +w d *s d (Formula 12)

[0282] Optionally, the size of the weight depends on the platform (ie, the target platform) to which the present application is applied, and the weight corresponding to each target impact parameter is positively correlated with the similarity between the corresponding external platform and the target platform.

[0283] For example, if the target platform is close to platform B, then the weight of platform B should be larger.

[0284] Among them, the similarity between the target platform and the external platform can specifically refer to the similarity of the platform audience. For example, the differences in user groups can be seen, which are reflected in the distribution of age and gender. For example, platform B is mostly young people, and the target platform is also mostly young people, while platforms A and C are mostly middle-aged people, and platform D is mostly elderly people. In this case, the similarity between platform B and the target platform is greater than that of other external platforms, and the corresponding weight is also greater.

[0285] It should be noted that the above-mentioned method of analyzing the similarity between platforms based on the age and gender distribution of the user group is just a simple example. In addition, other indicators can also be used for analysis, such as the modality of the published content (video, graphics, etc.), which is not specifically limited in this article.

[0286] Optionally, the calculation process of the above-mentioned external platform influence parameters can also be implemented through an off-site social influence model, such as establishing a model for evaluating the off-site social influence of an account based on its off-site performance data. The specific implementation logic of this model can be referred to above Figures 8 to 10 The repeated parts will not be repeated.

[0287] In the above implementation, the present application calculates the social influence of an account on each external platform and then performs a weighted sum to obtain the social influence score of an account on all platforms outside the site. In this way, the external platform influence parameter can more comprehensively reflect the social influence of an account outside the site.

[0288] S222: Predicting a content quality parameter of a candidate release object based on a content tag set of at least one first release content associated with the candidate release object.

[0289] This application also considers that when an account is suitable for introduction into the site, the content of this account owner should be high-quality. The quality of content does not depend entirely on the content itself, but also on the differences in the viewing target groups. Because different viewing target groups like different content, the potential accounts required by this application must have content that is high-quality for the viewing targets on the site, that is, what the viewing targets like.

[0290] In the embodiment of the present application, the content quality parameter indicates the quality of the content published by the candidate publishing object when the candidate publishing object registers on the target platform and uploads the published content.

[0291] Based on this, in S222, the content quality parameters of a candidate publishing object on the target platform may be evaluated by using a pre-built quality evaluation model.

[0292] Optionally, a trained quality assessment model can be trained as follows:

[0293] Specifically, the quality assessment model to be trained is trained iteratively based on the training sample set to obtain a trained quality assessment model.

[0294] In the embodiment of the present application, each training sample includes a sample publishing object and a sample label.

[0295] It should be noted that the sample publishing object in the embodiment of the present application has published content on at least one external platform and the target platform, that is, an account that has posted both on the site and outside the site can be used as a training sample for training the quality assessment model on the site. For the same account, there may be some differences between its on-site and off-site posts, because the quality assessment model of the present application is mainly used to predict accounts that are not on the site. Therefore, when training the model, the off-site posts of the account are used as input, and the on-site posts are not considered.

[0296] The performance data related to the posting on the site is used to construct the sample label. In the embodiment of the present application, the sample label is determined based on the sample content feedback data associated with the sample publishing object on the target platform.

[0297] Specifically, a sample label is constructed based on the on-site performance data of an account. The sample label is an indicator used to measure the quality of the content of an account on the site.

[0298] For example, the number of video views (vv) of an account within 90 days on the site is used as a training label. Other feedback data can also be used, such as total number of views + total number of likes, total number of views + total number of likes + total number of comments, and completion rate, etc. This article does not make specific restrictions. The following uses the total number of video views of an account within 90 days on the site as a sample label as an example.

[0299] Each iteration of training performs the following steps:

[0300] First, training samples are selected from a training sample set, and sample label distribution characteristics of sample release objects in the training samples associated with at least one external platform are input into a quality assessment model to be trained to obtain sample content quality parameters of the sample release objects.

[0301] Optionally, the sample label distribution characteristics include: a first sample label distribution characteristic of each sample content published by the sample publishing object on at least one external platform, and a second sample label distribution characteristic of each target sample content, where the target sample content is the sample content among all the sample contents whose publishing time is within a preset order range.

[0302] The preset order range refers to a number of preset sorting positions when sorting the contents from small to large or from large to small according to the release time. For example, the target sample content is the 10 most recently released sample content, the 20 most recently released sample content, the 30 earliest released sample content, etc. The following takes the 10 most recently released sample content as an example.

[0303] In this step, for a sample publishing object, first, obtain the sample content published by the sample publishing object within a certain time range, use the multi-label classification model obtained above to infer the posts of the sample publishing object (account) to obtain the content label set information of each sample content.

[0304] Taking video content as an example, a video may have multiple content tags. For example, a video has content tags such as architectural close-up, symmetrical composition, etc. For details, please refer to the above embodiment, and the repeated parts will not be repeated.

[0305] On the basis of the above, statistics are collected on the tag distribution of all posts of an account in the last 90 days and the tag distribution of the last 10 posts.

[0306] For example, in the last 90 days, the account posted 130 articles, the tag "Architecture Close-up" appeared 100 times, and the tag "Symmetrical Composition" appeared 120 times. In the last 10 articles of the account, the tag "Architecture Close-up" appeared 10 times, and the tag "Symmetrical Composition" appeared 9 times.

[0307] For all content tags, a one-hot encoding is constructed, and then, for the content tags of all posts of an account, the one-hot encodings of all corresponding tags are added together to obtain the feature vector of all posts, which is the first sample label distribution feature in the embodiment of the present application. Similarly, the feature vectors of the most recent 10 posts of an account are obtained, which is the second sample label distribution feature.

[0308] For example, there are three content tags A, B, and C. The one-hot of A is [1, 0, 0], the one-hot of B is [0, 1, 0], and the one-hot of C is [0, 0, 1]. For an account, if in all posts in the past 90 days, tag A appears 300 times, tag B appears 200 times, and tag C appears 100 times, then the first sample tag distribution feature corresponding to the account can be expressed as a feature vector [300, 200, 100]. In the most recent 10 posts, tag A appears 10 times and tag B appears 5 times, then the second sample tag distribution feature corresponding to the account can be expressed as a feature vector [10, 5, 0].

[0309] Specifically, after obtaining the first sample label distribution feature and the second sample label distribution feature of an account, the two feature vectors can be input into the quality assessment model to be trained to obtain the sample content quality parameters of the sample release object.

[0310] Furthermore, based on the difference between the sample content quality parameters and the corresponding sample labels, the parameters of the quality assessment model to be trained are adjusted.

[0311] Among them, when the sample label represents the total number of views of the account's videos within 90 days on the site, the sample content quality parameter can represent the estimated total number of views of the content within a certain time range when the sample publishing object posts on the target platform. Correspondingly, the content quality parameters for candidate publishing objects are the same.

[0312] When adjusting parameters based on the difference between the sample content quality parameters and the corresponding sample labels, a cross entropy loss function (or other loss functions) may be constructed, which is not specifically limited in this article.

[0313] Among them, the parameter adjustment of the model involved in this article generally refers to the weights, biases, convolution kernel sizes, etc. contained in the model, and this article does not make specific restrictions.

[0314] Optionally, the quality assessment model in the embodiment of the present application may adopt a decision tree model. For example, based on the video content posted by the account outside the site, the content quality multi-label and the performance data within the site, a decision tree model for assessing the quality of the account's content within the site is established.

[0315] See also Fig.11 As shown, it is a schematic diagram of a decision tree model in an embodiment of the present application, wherein, if there are four features (depending on actual needs, the number of features is not specifically limited in this article), [Feature A, Feature B, Feature C, Feature D], for example, these four features respectively represent some influencing factors that affect the final result in the content quality evaluation process, and each feature will correspond to a corresponding condition, such as Feature A corresponds to Condition a1 and Condition a2. For a content, if Feature A of the content satisfies Condition a1, then Feature B of the content is continued to be analyzed. If Feature A of the content satisfies Condition a2, the evaluation result of this time - Result 2 can be obtained; when Feature B satisfies Condition b1, Feature C of the content is continued to be analyzed. If Feature B of the content satisfies Condition b2, the evaluation result of this time - Result 2 can be obtained. Specifically, based on the above logic, a method such as a [feature] can be automatically learned according to the sample label. Fig.11 The decision tree model shown in Figure 1 is used to obtain the trained quality assessment model.

[0316] It should be noted that in addition to the decision tree models listed above, other model structures can also be used. For example, the decision tree model can be replaced by algorithm models such as eXtreme Gradient Boosting (xgboost) and logistic regression, which is not specifically limited in this article.

[0317] Through the above process, this application obtains a trained quality assessment model that can predict the quality of an account's posts on the site based on the posts.

[0318] Optionally, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions; an optional implementation of S222 is as follows, including steps S2221 to S2222:

[0319] S2221: Based on the content tag set, count the first tag distribution characteristics of each first published content associated with a candidate published object and the second tag distribution characteristics of each target published content.

[0320] In the embodiment of the present application, the tag distribution feature represents the total number of times each content tag appears; the target published content is the first published content whose publishing time is within a preset order range among the first published contents.

[0321] For example, the preset time range is the latest 90 days, and the target release content is the 10 first release contents that are most recently released by the candidate release objects within the latest 90 days.

[0322] If the candidate publishing object is account 1, in S2221, the sum of the number of times each content tag appears in each first published content published by account 1 in the last 90 days can be counted, thereby obtaining the first tag distribution feature. Also, the sum of the number of times each content tag appears in the 10 first published contents most recently published by account 1 can be counted, thereby obtaining the second tag distribution feature.

[0323] For example, if account 1 has published 100 articles in the last 90 days, label A appears 50 times, label B appears 120 times, and label C appears 70 times, then the corresponding first label distribution feature can be represented as the feature vector [50, 120, 70]. If account 1 has published 10 articles in the last 10 articles, label A appears 5 times, label B appears 15 times, and label C appears 6 times, then the corresponding first label distribution feature can be represented as the feature vector [5, 15, 6].

[0324] S2222: Input the first label distribution feature and the second label distribution feature into the trained quality assessment model to obtain a content quality parameter of a candidate release object.

[0325] In the above method, the trained quality assessment model can predict the content quality parameters corresponding to the posting of account 1 on the target platform based on the above two feature vectors [50, 120, 70] and [5, 15, 6].

[0326] In the above implementation, the content quality parameters of the account on the site are estimated based on the content published by the account outside the site, multiple content quality labels and performance data on the site. When training the quality prediction model, accounts that post both outside and on the site are used as training samples, which can effectively improve the estimation accuracy of the content quality parameters.

[0327] S223: Predicting a target supply and demand parameter of a candidate publishing object on the target platform according to the external platform sharing data associated with a candidate publishing object and the second content feedback data associated with the target platform within a preset time range.

[0328] This application also takes into account that when an account is suitable for introduction into the site, the content of this account owner should be scarce within the site. The scarcity of content is divided into two aspects, one is potential scarcity (there are no similar ones within the site, but the viewing objects share from outside the site to the site), and the other is current scarcity (there are similar ones within the site, but the amount of content is not enough). This requires analyzing the potential supply and demand relationship based on the sharing behavior of the viewing object and the current supply and demand relationship based on the content multi-tag system separately. For example, the viewing object shared a video from other information flow platforms to this information flow platform, then this application can mine scarce content and account owners from this behavior. This is a potential demand for viewing objects and is not reflected in any consumption data of the current recommendation system. For example, this application finds from the object behavior of the recommendation system that the content of a certain label is consumed by many viewing objects, but its content supply is insufficient. This kind of content is the current scarcity reflected in the consumption data of the recommendation system, and this application needs to introduce such account owners in a targeted manner.

[0329] Based on this, in S223, the potential supply and demand parameters of the candidate publishing object can be determined through the external platform sharing data, and the current supply and demand parameters of the candidate publishing object can be determined according to the content feedback data of the target platform. The target supply and demand parameters of a candidate publishing object are represented by the two parameters of potential supply and demand parameters and current supply and demand parameters. An optional calculation method is as follows, including the following steps S2231 to S2233:

[0330] S2231: Determine potential supply and demand parameters of a candidate release object according to external platform sharing data of a candidate release object and other candidate release objects respectively associated with each other within a preset time range.

[0331] Among them, the potential supply and demand parameter reflects the potential scarcity score mentioned above. Potential scarcity means that there is no similar content in the target platform, and the viewing object needs to share it from the external platform to the target platform. It is specifically related to the object's sharing behavior, so this parameter can be calculated based on the sharing data of the external platform.

[0332] In an optional implementation, the external platform sharing data includes: the total number of contents and the total number of times the first published content is shared to the target platform; specifically, by collecting the content and object behaviors shared to all other information flow platforms within the site, the total number of contents and the total number of times the content of an account is shared within the site can be obtained. Then, step S2231 can be implemented according to the following steps, including the following processes S22311 to S22313:

[0333] S22311: Obtain a quantity comparison value determined based on a total quantity of content associated with a candidate publishing object and a total quantity of content in a specified order among total quantities of content associated with each of a plurality of candidate publishing objects.

[0334] See also Fig.12 As shown, it is a schematic diagram of the calculation logic of a target supply and demand parameter in an embodiment of the present application. Fig.12 In the calculation of the target supply and demand parameters, it is necessary to first calculate the potential supply and demand parameters and the current supply and demand parameters respectively.

[0335] Among them, the calculation of potential supply and demand parameters based on object sharing behavior involves quantity comparison value and sharing comparison value.

[0336] For example, multiple candidate release objects are account 1 to account 10, one candidate release object is account 1, the other candidate release objects are account 2 to account 10, and one external platform is platform A:

[0337] Assume that the total number of contents in the specified order among the total number of contents associated with each of the multiple candidate release objects refers to the maximum total number of contents. Then the quantity comparison value can be determined based on the maximum value of the total number of first release contents shared to the target platform among the first release contents released by account 1 and the total number of first release contents shared to the target platform among the first release contents released by accounts 1 to 10. Still taking the logarithmic comparison value as an example, an optional calculation method is as follows:

[0338]

[0339] S22312: Obtaining a total number of shares associated with a candidate publishing object, and a total number of shares in a specified order among the total number of shares associated with each of the multiple candidate publishing objects, and determining a sharing comparison value.

[0340] Similarly, the frequency comparison value can be determined based on the total number of times the first published content shared to the target platform in the first published content published by account 1 is shared (regardless of whether it is the same first published content, the number of times is +1 each time it is shared), and the maximum value of the total number of times the first published content shared to the target platform in the first published content published by accounts 1 to 10 is shared. Still taking the logarithmic comparison value as an example, an optional calculation method is as follows:

[0341]

[0342] S22313: The sum of the quantity comparison value and the sharing comparison value is used as a potential supply and demand parameter of a candidate release object.

[0343]

[0344] Among them, Score1 represents the potential supply and demand parameters (i.e. potential scarcity) of an account, such as Fig.12As shown, in the embodiment of the present application, the larger the Score1 is, the more scarce the Account 1 is.

[0345] It should be noted that the above formula 15 takes the consideration of both the frequency comparison value and the quantity comparison value as an example. When only the two are considered, the corresponding parameters can be set to 0. Similarly, when other dimensions are considered, they can be summed up in a similar manner.

[0346] It should be noted that, in addition to the logarithmic method listed above, other methods may be used to calculate any comparison value in the embodiments of the present application, which will not be described in detail herein.

[0347] In the above implementation, the potential supply and demand relationship based on the object-based sharing behavior can effectively reflect the potential scarcity of content when an account publishes content on the target platform.

[0348] S2232: Obtain current supply and demand parameters of a candidate publishing object according to third content feedback data associated with the target platform within a preset time range and respective content tag sets of each first published content.

[0349] Among them, the current supply and demand parameter reflects the above-mentioned current scarcity score. The current scarcity means that there is similar content in the target platform, but the content volume is not enough. It is specifically related to the content tag of the published content. Therefore, this parameter can be calculated based on the third content feedback data associated with the target platform and the content tag set of each first published content.

[0350] In an optional implementation, the content tag set of each content includes: content tags of the content under multiple attribute dimensions; the second content feedback data includes: the number of completed views and the number of uncompleted views of the second published content; step S2232 can be implemented according to the following steps, including S22321 to S22323:

[0351] S22321: For each content tag, determine a scarcity parameter corresponding to a content tag according to the number of completed views and the number of uncompleted views of the second published content containing the content tag within a preset time range.

[0352] Taking video content as an example, first, the multi-label classification model obtained above can be used to infer the posts on the target platform of the registered account to obtain the content label set of each video content (i.e., the second published content).

[0353] Secondly, according to the object consumption of each video in the station (i.e., the third content feedback data), the completion of the video is calculated. The so-called completion means that the viewing object not only saw the video, but also watched the video from the beginning to the end. For example, object a watched 10 videos today. These 10 videos have the "Architectural Close-up" label and the "Symmetrical Composition" label. Among them, 7 videos have been completed and 3 videos have not been completed. The "Architectural Close-up" label has been completed 7 times and incompleted 3 times. For all second publishing objects in the target platform, this application can divide all the playback times into two distributions according to the label: completed playback and incomplete playback.

[0354] like Fig.12 As shown, the calculation of the current supply and demand parameters based on the content multi-tag system involves scarcity parameters and frequency parameters. Among them, the scarcity parameter is used to indicate the scarcity of specific content tags of the content on the site. On the basis of the above, an optional calculation method of the scarcity parameter is as follows:

[0355]

[0356] Among them, Score i Indicates the scarcity parameter corresponding to content label i (can be referred to as label i for short).

[0357] In the embodiment of the present application, Formula 16 is taken as an example of video content. For video content, completed playback means completed viewing, and incomplete playback means incomplete viewing. For each content tag i, the corresponding scarcity parameter can be calculated based on the above Formula 16 to indicate the scarcity of the tag.

[0358] When calculating the scarcity parameter, the ratio of completed views to uncompleted views is not directly calculated, but a certain value is added to each before comparison. It should be noted that the above 1000 and 500 are just simple distances, and can be other values, which are not specifically limited in this article. In addition, the ratio of completed views to uncompleted views can also be directly used as a scarcity parameter, which is not specifically limited in this article.

[0359] In addition, it should be noted that the scarcity parameter in the embodiment of the present application can be updated every 24 hours to indicate the scarcity of specific tags of the content on the site in real time.

[0360] S22322: For each content tag, determine a frequency parameter corresponding to a content tag according to content tag sets corresponding to respective first published contents associated with a candidate published object within a preset time range.

[0361] This step also needs to use the multi-label classification model obtained above to infer the off-site posts of the candidate publishing objects to obtain the content label set corresponding to each video content (the first published content).

[0362] The frequency parameter is used to indicate the frequency of occurrence of specific content tags of off-site content, which can be obtained by comparing the content tag sets of the content published by each candidate publishing object. For each content tag, an optional calculation method is as follows:

[0363] First, the number of occurrences of a content tag and the total number of occurrences of each content tag in the first published content published by a candidate publishing object within a preset time range are obtained; and the total number of each candidate publishing object and the number of candidate publishing objects containing a content tag are obtained.

[0364] Furthermore, based on the number of occurrences, the total number of occurrences, the total data and the quantity, a frequency parameter corresponding to a content tag is determined.

[0365] like Fig.12 As shown, Fig.12 The label distribution in specifically includes the number of occurrences, the total number of occurrences, the total data and the quantity listed. When calculating the frequency parameter based on the label distribution, an optional calculation formula is as follows:

[0366]

[0367] Among them, TF-IDf i It represents the frequency parameter corresponding to tag i. The number of times tag i appears in the current account means the number of times tag i appears in the content published within the preset time range. The total number of tags in the current account refers to the total number of times all content tags appear in the current account, that is, the total number of times each content tag appears in this article. The total number of accounts is the total number of candidate release objects. The number of accounts containing tag i is the number of candidate release objects whose published content contains tag i.

[0368] For each content tag i, the corresponding frequency parameter can be calculated based on the above formula 17.

[0369] It should be noted that the calculation method of the frequency parameters listed in the above formula 17 is just a simple example. Any method of determining the frequency parameters based on the number of occurrences, the total number of occurrences, the total data and the quantity is applicable to the embodiments of the present application and will not be described one by one here.

[0370] S22323: Determine each target content tag whose frequency parameter ranking is within a specified order range, and use the sum of the products of the scarcity parameter and the frequency parameter of each target content tag as the current supply and demand parameter of a candidate release object.

[0371] The specified order range refers to a number of specified sorting positions when sorting the frequency parameters from small to large or from large to small. For example, if the target content tag is determined based on the tf-idf score, the top 10 content tags of an account, then when calculating the current supply and demand parameters, the TF-IDf of each target content tag is used. i Score i Multiply, add and normalize to get the final current supply and demand parameters. An optional calculation method is as follows:

[0372]

[0373] In Formula 18, N is equal to 10, and the value range of i is 1 to 10. Specifically, i here represents the i-th target content tag. The larger the Score2 is, the more scarce the account 1 is.

[0374] In the above implementation, based on the potential supply and demand relationship based on the content multi-tag system, it can effectively reflect the current scarcity of content when an account publishes content on the target platform.

[0375] S2233: Based on the potential supply and demand parameters and the current supply and demand parameters, determine a target supply and demand parameter corresponding to a candidate release object.

[0376] Specifically, by merging Score1 and Score2, the final supply and demand relationship score of an account (i.e., target supply and demand parameters) can be obtained.

[0377] In the embodiment of the present application, the merging method can be weighted summation or averaging, etc., which is not specifically limited in this article.

[0378] Optionally, the calculation process of the above-mentioned target supply and demand parameters can also be implemented through a supply and demand relationship model. For example, by analyzing the supply and demand relationship within the account site, mainly including the potential supply and demand relationship based on object sharing behavior and the current supply and demand relationship based on the content multi-tag system, a model for evaluating the supply and demand relationship of the account within the site is established. The specific implementation logic of the model can be referred to the above 12 parts, and the repeated parts will not be repeated.

[0379] In the above implementation, the target supply and demand parameters are determined by combining the potential demand parameters and the current demand parameters. The data referenced in the calculation process is more comprehensive, and the calculated target supply and demand parameters are more accurate, which can better reflect the real scarcity of the account in the site.

[0380] S224: Determine an object potential parameter of a candidate release object according to an external platform influence parameter, content quality parameter, and target supply and demand parameter of a candidate release object.

[0381] In an embodiment of the present application, when calculating the potential score of an account (i.e., object potential parameters), the external platform influence parameters, content quality parameters, and target supply and demand parameters of the above-mentioned account may be weighted summed or averaged, or other calculation methods may be used, which are not specifically limited in this article.

[0382] An optional calculation method is as follows:

[0383] Score final =(external platform influence parameter+η1) α *(Content quality parameter + η2) β

[0384] *(target supply and demand parameter + η3) γ (Formula 19)

[0385] Among them, Score final Indicates the object potential parameter that an account ultimately corresponds to. η1, η2, and η3 are all preset adjustment parameters. η1 is generally one-tenth of the maximum value of the external platform influence parameter, η2 is generally one-tenth of the maximum value of the content quality parameter, and η3 is generally one-tenth of the maximum value of the target supply and demand parameter. α, β, and γ are hyperparameters that are adjusted based on actual application experience. They are generally set to values ​​such as 1, 1.5, and 0.5, and this article does not make specific restrictions.

[0386] It should be noted that the maximum value of the external platform influencing parameter η1 is fixed for any account, and the same applies to η2 and η3.

[0387] In addition, it should be noted that the above formula 19 takes into account the social influence of the account outside the site, the quality of the content within the site, and the supply and demand relationship within the site. If only some of the above three items are considered, the corresponding parameters can be 0.

[0388] Optionally, the calculation process of the above-mentioned object potential parameters can also be implemented through a decision-making model, such as establishing a decision-making model (which can be recorded as a potential discrimination model) based on the results of the above-mentioned supply and demand relationship model, quality assessment model and off-site social influence model, and outputting the final object potential parameters.

[0389] It should be noted that the decision tree model can also be replaced by algorithms such as xgboost and logistic regression, which is not specifically limited in this article.

[0390] In the above implementation, the potential score of the account is calculated based on the three aspects of the account's social influence outside the site, the quality of the content on the site, and the supply and demand relationship on the site. The account is introduced based on the score. Compared with the method of introducing accounts only based on the number of account fans or the number of content views in the related art, it can ensure that the introduced accounts meet the needs of the site and are loved by the audience group on the site, making the introduced accounts more accurate and the introduction effect better.

[0391] S23: The server selects at least one from a plurality of candidate publishing objects as a target publishing object according to the obtained potential parameters of each object.

[0392] In an embodiment of the present application, on the basis of determining the target publishing object, the target publishing object can also be directly introduced into the target platform, or a publishing object with similar characteristics to the target publishing object can be introduced and registered on the target platform, so as to more accurately introduce more popular publishing objects to the target platform.

[0393] The publishing object having similar characteristics to the target publishing object may specifically have similar published content, similar object attributes, etc., which is not specifically limited in this article.

[0394] The following example takes the introduction of at least one published object into the target platform for registration:

[0395] In the embodiment of the present application, with the above model, the potential score of any account on an off-site platform can be scored to determine whether the account is suitable for introduction into the site.

[0396] For example, a prediction may be made for off-site accounts once a week (i.e., the method for obtaining the publishing object in the embodiment of the present application is executed once), and accounts ranked in the top 1000 in terms of potential scores may be introduced as target publishing objects.

[0397] See also Fig.13 As shown, it is a logic block diagram of a process for obtaining a published object in an embodiment of the present application.

[0398] like Fig.13As shown, after the account owner posts outside the site, the content of a single account owner within a certain period of time is collected and summarized, and the content will be extracted by multiple algorithm models to obtain video content features such as frame extraction, title, audio, etc.; after regularly crawling and counting the object behaviors of these contents of the account owner outside the site, the corresponding off-site object consumption data (i.e., the first content feedback data in this article) can be obtained, such as the number of likes, the number of fans, etc.; when some account owners post both on the site and outside the site, after their content is distributed on the consumer end, the object can see multiple contents on his or her terminal. At this time, he or she will click, play and comment on the content he or she is interested in. This application performs simple statistics on this behavior to obtain the on-site object consumption data corresponding to the account (i.e., the second content feedback data), such as completion rate, vv, etc.; in addition, this application obtains the on-site sharing data of the object (i.e., the external platform sharing data in this article) by counting the behavior of content from different platforms outside the site being shared to the site. Modeling is performed based on the data obtained above. The model is used to predict the potential score of an account (i.e., the object potential parameter in this article). During the account introduction process, the account is introduced based on the potential score. After the account is introduced, its posts will be distributed at the content consumption end.

[0399] Similarly, it is also possible to analyze whether there are publishing objects with similar object characteristics to the target publishing object based on the publishing content, object attributes, etc. If so, such objects can also be introduced into the target platform for registration.

[0400] In the above implementation mode, the present application proposes a method of modeling and scoring the social influence of an account outside the site, the quality of the content within the site, and the supply and demand relationship within the site, respectively, to obtain an account potential discrimination model, and apply it in the actual account owner introduction process. Manual input is no longer required, and the model can automatically predict the object potential parameters, thereby improving the introduction effect and saving manpower. In addition, it can be applied to any platform, especially for introduction based on the scarcity of the platform and the interest of the object.

[0401] Based on the same inventive concept, the embodiment of the present application also provides a device for obtaining a published object. Fig.14 As shown, it is a schematic diagram of the structure of the acquisition device 1400 of the published object, which may include:

[0402] The data acquisition unit 1401 is used to acquire the first published content and first content feedback data of each of the multiple candidate publishing objects on at least one external platform, and the external platform sharing data associated with the target platform; the external platform is the content platform where the candidate publishing object has registered, and the target platform is the content platform where the candidate publishing object has not registered;

[0403] The potential estimation unit 1402 is used to perform the following operations for each candidate publishing object: obtain a corresponding content tag set according to basic attribute information of at least one first published content associated with a candidate publishing object; obtain a corresponding object potential parameter according to the obtained at least one content tag set, combined with first content feedback data and external platform sharing data associated with a candidate publishing object, wherein the object potential parameter represents: the estimated popularity of the published content uploaded by a candidate publishing object on the target platform;

[0404] The object acquisition unit 1403 is used to select at least one from a plurality of candidate release objects as a target release object according to the obtained potential parameters of each object.

[0405] Optionally, the potential estimation unit 1402 is specifically used for:

[0406] Determine an external platform influence parameter of a candidate publishing object according to first content feedback data associated with a candidate publishing object within a preset time range;

[0407] Predicting a content quality parameter of a candidate release object based on a content tag set of at least one first release content associated with the candidate release object;

[0408] Predicting target supply and demand parameters of a candidate release object on a target platform based on external platform sharing data associated with a candidate release object and second content feedback data associated with a target platform within a preset time range;

[0409] Determine the object potential parameters of a candidate release object based on the external platform influence parameters, content quality parameters and target supply and demand parameters of a candidate release object.

[0410] Optionally, the potential estimation unit 1402 is specifically used for:

[0411] For each foreign platform, do the following:

[0412] Determine an absolute influence parameter of a candidate publishing object on an external platform according to first content feedback data associated with a candidate publishing object and other candidate publishing objects on an external platform within a preset time range;

[0413] Determine an impact quality parameter of a candidate publishing object on an external platform according to first content feedback data associated with a candidate publishing object on an external platform within each unit time of a preset time range;

[0414] An external platform impact parameter of a candidate release object is determined according to the respective absolute impact parameter and impact quality parameter of a candidate release object on at least one external platform.

[0415] Optionally, within a preset time range, the first content feedback data associated with each candidate publishing object on an external platform includes: the number of followers currently following the candidate publishing object, the total number of content views and the total number of content interactions within the preset time range;

[0416] The potential estimation unit 1402 is specifically used for:

[0417] Obtaining an attention comparison value determined based on the number of attention objects of a candidate publishing object and the number of attention objects in a specified order among the numbers of attention objects of each of the plurality of candidate publishing objects;

[0418] Obtaining a viewing comparison value determined based on a total number of content views of a candidate publishing object and a total number of content views of a plurality of candidate publishing objects in a specified order;

[0419] Obtaining a total number of content interactions based on a candidate publishing object and a total number of content interactions in a specified order among the total number of content interactions of each of a plurality of candidate publishing objects, and determining an interaction comparison value;

[0420] The attention comparison value, the viewing comparison value, and the interaction comparison value are weightedly summed to obtain the absolute influence parameter of a candidate release object on an external platform.

[0421] Optionally, within each unit time of the preset time range, the first content feedback data associated with a candidate publishing object on an external platform includes: the number of content views within the unit time, and the number of content interactions in at least one statistical dimension within the unit time;

[0422] The potential estimation unit 1402 is specifically used for:

[0423] For each statistical dimension, perform the following operations:

[0424] Obtain a comparison value of the number of content interactions and the number of content views of a candidate release object per unit time under a statistical dimension;

[0425] Based on the weighted square sum of the comparison values ​​of the number of times in each unit time, a quality parameter under a statistical dimension is determined;

[0426] Based on the quality parameters of a candidate release object in each statistical dimension, an impact quality parameter of the candidate release object on an external platform is obtained.

[0427] Optionally, the difference between the current unit time and the latest unit time within the preset time range is negatively correlated with the weight corresponding to the current unit time.

[0428] Optionally, the potential estimation unit 1402 is specifically used for:

[0429] Normalize and average the absolute impact parameter and impact quality parameter of a candidate release object on the same external platform to obtain the corresponding target impact parameter;

[0430] The target influence parameters of a candidate release object on each external platform are weighted summed to obtain an external platform influence parameter of the candidate release object.

[0431] Optionally, the weight corresponding to each target impact parameter is positively correlated with the similarity between the corresponding external platform and the target platform.

[0432] Optionally, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions;

[0433] The potential estimation unit 1402 is specifically used for:

[0434] Based on the content tag set, the first tag distribution characteristics of each first published content associated with a candidate published object and the second tag distribution characteristics of each target published content are counted, where the tag distribution characteristics represent the total number of occurrences of each content tag; the target published content is the first published content in each first published content whose publishing time is within a preset sequence range;

[0435] The first label distribution feature and the second label distribution feature are input into the trained quality assessment model to obtain a content quality parameter of a candidate release object.

[0436] Optionally, the device further comprises:

[0437] The model training unit 1404 is used to obtain a trained quality assessment model by training in the following manner:

[0438] The quality assessment model to be trained is trained iteratively based on the training sample set to obtain a trained quality assessment model; each training sample includes a sample publishing object and a sample label; the sample publishing object has published content on at least one external platform and a target platform, and the sample label is determined according to sample content feedback data associated with the sample publishing object on the target platform; wherein each iterative training performs the following steps:

[0439] Select training samples from the training sample set, and input sample label distribution characteristics of sample release objects in the training samples and associated with at least one external platform into the quality assessment model to be trained to obtain sample content quality parameters of the sample release objects;

[0440] Based on the difference between the sample content quality parameters and the corresponding sample labels, the parameters of the quality assessment model to be trained are adjusted.

[0441] Optionally, the sample label distribution characteristics include: a first sample label distribution characteristic of each sample content published by the sample publishing object on at least one external platform, and a second sample label distribution characteristic of each target sample content, where the target sample content is the sample content among all the sample contents whose publishing time is within a preset order range.

[0442] Optionally, the potential estimation unit 1402 is specifically used for:

[0443] Determine potential supply and demand parameters of a candidate release object according to external platform sharing data of a candidate release object and other candidate release objects respectively associated with each other within a preset time range;

[0444] Obtaining current supply and demand parameters of a candidate publishing object according to third content feedback data associated with the target platform within a preset time range and respective content tag sets of each first published content;

[0445] Based on the potential supply and demand parameters and the current supply and demand parameters, a target supply and demand parameter corresponding to a candidate release object is determined.

[0446] Optionally, the external platform sharing data includes: the total number of contents and the total number of shares of the first published contents shared to the target platform;

[0447] The potential estimation unit 1402 is specifically used for:

[0448] Obtaining a quantity comparison value determined based on a total quantity of content associated with a candidate publishing object and a total quantity of content in a specified order among total quantities of content associated with each of a plurality of candidate publishing objects;

[0449] Obtaining the total number of shares associated with a candidate release object, the total number of shares associated with each of multiple candidate release objects in a specified order, and the determined sharing comparison value;

[0450] The sum of the quantity comparison value and the sharing comparison value is used as a potential supply and demand parameter of a candidate release object.

[0451] Optionally, the content tag set of each content includes: respective content tags of the content under multiple attribute dimensions; the second content feedback data includes: the number of completed views and the number of uncompleted views of the second published content;

[0452] The potential estimation unit 1402 is specifically used for:

[0453] For each content tag, do the following:

[0454] Determine a scarcity parameter corresponding to a content tag according to the number of completed views and the number of uncompleted views of the second published content containing the content tag within a preset time range;

[0455] Determine a frequency parameter corresponding to a content tag according to content tag sets corresponding to first published contents associated with a candidate publishing object within a preset time range;

[0456] The target content tags whose frequency parameter ranking is within the specified order range are determined, and the sum of the products of the scarcity parameter and the frequency parameter of each target content tag is used as the current supply and demand parameter of a candidate release object.

[0457] Optionally, the potential estimation unit 1402 is specifically used for:

[0458] For each content tag, do the following:

[0459] Obtaining the number of occurrences of a content tag and the total number of occurrences of each content tag in the first published content published by a candidate publishing object within a preset time range;

[0460] Get the total number of candidate release objects and the number of candidate release objects containing a content tag;

[0461] Based on the number of occurrences, the total number of occurrences, the total data and the quantity, a frequency parameter corresponding to a content tag is determined.

[0462] Optionally, the basic attribute information has at least one category; the potential estimation unit 1402 is specifically used to:

[0463] For each first publication, perform the following operations:

[0464] Input each category of basic attribute information of a first published content into the trained multi-label classification model and represent them as multiple attribute vectors respectively;

[0465] For each type of basic attribute information, multiple attribute vectors corresponding to the basic attribute information are represented as a target vector through one-dimensional convolution processing; and features of the target vector are extracted based on the multimodal attention network in the multi-label classification model to obtain a corresponding target vector;

[0466] The target vectors corresponding to various basic attribute information are spliced ​​to obtain a spliced ​​vector;

[0467] A content label set of the first published content is obtained based on the splicing vector prediction.

[0468] In view of this, when introducing a publishing object from an external platform to a target platform, this application does not only unilaterally consider the number of account followers or content viewing times in the external platform, but comprehensively considers the publishing content of the publishing object on the external platform, the relevant content feedback data, and the external platform sharing data associated with each target platform, to estimate the popularity of the publishing object on the external platform after uploading the published content on the target platform, so as to screen out the publishing object that is more suitable for introduction into the target platform based on this parameter. Because in this way, the content tag set and content feedback data of the publishing object's published content on the external platform can reflect the influence of the publishing object on the external platform, and further combined with the external platform sharing data of the publishing object, it can reflect the influence of the publishing object on the target platform. This application comprehensively considers the publishing objects suitable for the target platform based on various indicators to improve the accuracy of obtaining the publishing objects.

[0469] Furthermore, based on the determination of the target publishing object, the target publishing object can be directly introduced into the target platform, or a publishing object with similar characteristics to the target publishing object can be introduced and registered on the target platform, so as to more accurately introduce a more popular publishing object to the target platform.

[0470] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.

[0471] After introducing the method and device for acquiring a publication object according to an exemplary embodiment of the present application, next, an electronic device according to another exemplary embodiment of the present application is introduced.

[0472] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.

[0473] Based on the same inventive concept as the above method embodiment, an electronic device is also provided in the embodiment of the present application. In one embodiment, the electronic device may be a server, such as Figure 1 In this embodiment, the structure of the electronic device can be as follows: Fig.15 As shown, it includes a memory 1501 , a communication module 1503 and one or more processors 1502 .

[0474] The memory 1501 is used to store computer programs executed by the processor 1502. The memory 1501 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and programs required for running the instant messaging function, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0475] The memory 1501 may be a volatile memory, such as a random-access memory (RAM); the memory 1501 may also be a non-volatile memory, such as a read-only memory, a flash memory, a hard disk drive (HDD) or a solid-state drive (SSD); or the memory 1501 may be any other medium that can be used to carry or store a desired computer program in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto. The memory 1501 may be a combination of the above memories.

[0476] The processor 1502 may include one or more central processing units (CPU) or a digital processing unit, etc. The processor 1502 is used to implement the above-mentioned method for obtaining the published object when calling the computer program stored in the memory 1501 .

[0477] The communication module 1503 is used to communicate with terminal devices and other servers.

[0478] The specific connection medium between the memory 1501, the communication module 1503 and the processor 1502 is not limited in the embodiment of the present application. Fig.15 In the embodiment, the memory 1501 and the processor 1502 are connected via a bus 1504. The bus 1504 is connected to the processor 1502 via a bus 1504. Fig.15 The connections between the other components are only for illustration and are not intended to be limiting. The bus 1504 can be divided into an address bus, a data bus, a control bus, etc. For ease of description, Fig.15 The diagram shows that only one thick line is used, but this does not mean that there is only one bus or one type of bus.

[0479] The memory 1501 stores a computer storage medium, and the computer storage medium stores computer executable instructions, and the computer executable instructions are used to implement the method for obtaining the published object in the embodiment of the present application. The processor 1502 is used to execute the above-mentioned method for obtaining the published object, such as Figure 2 shown.

[0480] In another embodiment, the electronic device may also be other electronic devices, such as Figure 1 The terminal device 110 shown in FIG. 1 is a terminal device 110 shown in FIG. 1 . In this embodiment, the structure of the electronic device can be as follows: Fig.16 As shown, it includes: a communication component 1610, a memory 1620, a display unit 1630, a camera 1640, a sensor 1650, an audio circuit 1660, a Bluetooth module 1670, a processor 1680 and other components.

[0481] The communication component 1610 is used to communicate with the server. In some embodiments, a circuit wireless fidelity (Wireless Fidelity, WiFi) module may be included. The WiFi module belongs to a short-range wireless transmission technology. The electronic device can help the user to send and receive information through the WiFi module.

[0482] The memory 1620 can be used to store software programs and data. The processor 1680 executes various functions and data processing of the terminal device 110 by running the software programs or data stored in the memory 1620. The memory 1620 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage device. The memory 1620 stores an operating system that enables the terminal device 110 to run. In the present application, the memory 1620 can store an operating system and various application programs, and can also store a computer program for executing the acquisition method of the object published in the embodiment of the present application.

[0483] The display unit 1630 may also be used to display information input by the user or information provided to the user and a graphical user interface (GUI) of various menus of the terminal device 110. Specifically, the display unit 1630 may include a display screen 1632 disposed on the front of the terminal device 110. The display screen 1632 may be configured in the form of a liquid crystal display, a light emitting diode, etc. The display unit 1630 may be used to display the XX user interface in the embodiment of the present application, etc.

[0484] The display unit 1630 can also be used to receive input digital or character information and generate signal input related to user settings and function control of the terminal device 110. Specifically, the display unit 1630 may include a touch screen 1631 arranged on the front of the terminal device 110, which can collect user touch operations on or near it, such as clicking a button, dragging a scroll box, etc.

[0485] The touch screen 1631 can be covered on the display screen 1632, or the touch screen 1631 and the display screen 1632 can be integrated to realize the input and output functions of the terminal device 110. The integrated touch screen can be referred to as a touch display screen. In this application, the display unit 1630 can display the application and the corresponding operation steps.

[0486] The camera 1640 can be used to capture still images, and the user can publish the images taken by the camera 1640 through the application. The camera 1640 can be one or more. The object generates an optical image through the lens and projects it onto the photosensitive element. The photosensitive element can be a charge coupled device (CCD) or a complementary metal-oxide-semiconductor (CMOS) phototransistor. The photosensitive element converts the light signal into an electrical signal, and then passes the electrical signal to the processor 1680 to convert it into a digital image signal.

[0487] The terminal device may further include at least one sensor 1650, such as an acceleration sensor 1651, a distance sensor 1652, a fingerprint sensor 1653, and a temperature sensor 1654. The terminal device may also be configured with other sensors such as a gyroscope, a barometer, a hygrometer, a thermometer, an infrared sensor, a light sensor, and a motion sensor.

[0488] The audio circuit 1660, the speaker 1661, and the microphone 1662 can provide an audio interface between the user and the terminal device 110. The audio circuit 1660 can transmit the electrical signal converted from the received audio data to the speaker 1661, which is converted into a sound signal for output. The terminal device 110 can also be configured with a volume button for adjusting the volume of the sound signal. On the other hand, the microphone 1662 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1660 and converted into audio data, and then the audio data is output to the communication component 1610 to be sent to, for example, another terminal device 110, or the audio data is output to the memory 1620 for further processing.

[0489] The Bluetooth module 1670 is used to exchange information with other Bluetooth devices having Bluetooth modules through the Bluetooth protocol. For example, the terminal device can establish a Bluetooth connection with a wearable electronic device (such as a smart watch) that also has a Bluetooth module through the Bluetooth module 1670 to exchange data.

[0490] The processor 1680 is the control center of the terminal device. It uses various interfaces and lines to connect various parts of the entire terminal. It executes various functions of the terminal device and processes data by running or executing software programs stored in the memory 1620 and calling data stored in the memory 1620. In some embodiments, the processor 1680 may include one or more processing units; the processor 1680 may also integrate an application processor and a baseband processor, wherein the application processor mainly processes the operating system, the user interface, and the application program, and the baseband processor mainly processes wireless communication. It is understandable that the above-mentioned baseband processor may not be integrated into the processor 1680. In the present application, the processor 1680 can run the operating system, the application program, the user interface display and the touch response, as well as the acquisition method of the published object in the embodiment of the present application. In addition, the processor 1680 is coupled to the display unit 1630.

[0491] In some possible implementations, various aspects of the method for obtaining a published object provided in the present application may also be implemented in the form of a program product, which includes a computer program. When the program product is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of the method for obtaining a published object according to various exemplary implementations of the present application described above in this specification. For example, the electronic device may execute the following steps: Figure 2 Follow the steps shown in .

[0492] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0493] The program product of the embodiment of the present application may adopt a portable compact disk read-only memory (CD-ROM) and include a computer program, and can be run on an electronic device. However, the program product of the present application is not limited thereto, and in this document, a readable storage medium may be any tangible medium containing or storing a program, which can be used by or in combination with a command execution system, apparatus, or device.

[0494] A readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable computer program is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with a command execution system, apparatus, or device.

[0495] The computer program embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0496] The computer program for performing the operation of the present application can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and also conventional procedural programming languages ​​such as "C" language or similar programming languages. The computer program can be executed entirely on the user electronic device, partially on the user electronic device, as a separate software package, partially on the user electronic device and partially on a remote electronic device, or entirely on a remote electronic device or server. In the case of a remote electronic device, the remote electronic device can be connected to the user electronic device through any type of network including a local area network (LAN) or a wide area network (WAN), or can be connected to an external electronic device (for example, using an Internet service provider to connect through the Internet).

[0497] It should be noted that, although several units or subunits of the device are mentioned in the above detailed description, this division is merely exemplary and not mandatory. In fact, according to the embodiments of the present application, the features and functions of two or more units described above can be embodied in one unit. Conversely, the features and functions of one unit described above can be further divided into multiple units to be embodied.

[0498] In addition, although the operations of the method of the present application are described in a specific order in the drawings, this does not require or imply that the operations must be performed in this specific order, or that all the operations shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps.

[0499] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain a computer-usable computer program.

[0500] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0501] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for obtaining a publication object, characterized in that: The method comprises: Acquire the first published content and first content feedback data of each of the plurality of candidate publishing objects on at least one external platform, and the external platform sharing data associated with each of the plurality of candidate publishing objects on the target platform; the external platform is a content platform on which the candidate publishing object has registered, and the target platform is a content platform on which the candidate publishing object has not registered; For each candidate publishing object, the following operations are performed respectively: according to basic attribute information of at least one first publishing content associated with a candidate publishing object, a corresponding content tag set is obtained respectively; according to the obtained at least one content tag set, in combination with first content feedback data and external platform sharing data associated with the candidate publishing object, a corresponding object potential parameter is obtained, wherein the object potential parameter represents: the estimated popularity of the publishing content uploaded by the candidate publishing object on the target platform; At least one is selected from the plurality of candidate release objects as a target release object according to the obtained potential parameters of each object.

2. The method according to claim 1, characterized in that The obtaining of corresponding object potential parameters according to the obtained at least one content tag set and combining the first content feedback data and external platform sharing data associated with the candidate publishing object includes: Determining an external platform influence parameter of the candidate publishing object according to first content feedback data associated with the candidate publishing object within a preset time range; Predicting a content quality parameter of the candidate publication object based on a content tag set of each of the at least one first publication content associated with the candidate publication object; Predicting a target supply and demand parameter of the candidate publishing object on the target platform according to the external platform sharing data associated with the candidate publishing object and the second content feedback data associated with the target platform within the preset time range; Determine the object potential parameter of the candidate release object according to the external platform influence parameter, content quality parameter and target supply and demand parameter of the candidate release object.

3. The method according to claim 2, characterized in that The determining, according to the first content feedback data associated with the candidate publishing object within a preset time range, an external platform influence parameter of the candidate publishing object includes: For each foreign platform, do the following: Determine an absolute influence parameter of the one candidate publishing object on the one external platform according to first content feedback data associated with each of the one candidate publishing object and other candidate publishing objects on the one external platform within the preset time range; Determine, according to first content feedback data associated with the one candidate publishing object on the one external platform within each unit time of the preset time range, an impact quality parameter of the one candidate publishing object on the one external platform; An external platform influence parameter of the candidate release object is determined according to the absolute influence parameter and the influence quality parameter of the candidate release object on the at least one external platform.

4. The method according to claim 3, characterized in that The first content feedback data associated with each candidate publishing object on an external platform within the preset time range includes: the number of followers currently following the candidate publishing object, the total number of content views and the total number of interactions within the preset time range; The determining, according to the first content feedback data associated with each of the one candidate publishing object and other candidate publishing objects on the one external platform, an absolute influence parameter of the one candidate publishing object on the one external platform comprises: Obtaining an attention comparison value determined based on the number of attention objects of the one candidate publishing object and the number of attention objects in a specified order among the numbers of attention objects of each of the plurality of candidate publishing objects; Obtaining a viewing comparison value determined based on the total number of content views of the one candidate publishing object and the total number of content views of the plurality of candidate publishing objects in a specified order; Obtaining an interaction comparison value determined based on the total number of content interactions of the one candidate publishing object and the total number of content interactions of the plurality of candidate publishing objects in a specified order; The attention contrast value, the viewing contrast value, and the interaction contrast value are weightedly summed to obtain an absolute influence parameter of the candidate publishing object on the external platform.

5. The method according to claim 3, characterized in that In each unit time of the preset time range, the first content feedback data associated with one candidate publishing object on an external platform includes: the number of content views in the unit time, and the number of content interactions in at least one statistical dimension in the unit time; Determining the influence quality parameter of the candidate publishing object on the external platform according to the first content feedback data associated with the candidate publishing object on the external platform within each unit time of the preset time range includes: For each statistical dimension, perform the following operations: Obtaining a comparison value of the number of content interactions and the number of content views of the candidate publishing object in each unit time under a statistical dimension; Determine the quality parameter under the one statistical dimension based on the weighted square sum of the comparison values ​​of the number of times in each unit time; Based on the quality parameters of the candidate release object in each statistical dimension, an impact quality parameter of the candidate release object on the external platform is obtained.

6. The method according to claim 5, characterized in that The difference between the current unit time and the latest unit time within the preset time range is negatively correlated with the weight corresponding to the current unit time.

7. The method according to claim 3, characterized in that The determining the external platform influence parameter of the candidate release object according to the absolute influence parameter and the influence quality parameter of the candidate release object on the at least one external platform includes: Normalizing and averaging the absolute impact parameter and the impact quality parameter of the candidate release object on the same external platform to obtain a corresponding target impact parameter; The target influence parameters of the candidate release object on each external platform are weightedly summed to obtain the external platform influence parameter of the candidate release object.

8. The method according to claim 7, characterized in that The weight corresponding to each target impact parameter is positively correlated with the similarity between the corresponding external platform and the target platform.

9. The method according to any one of claims 2 to 8, characterized in that: The content tag set of each content includes: the content tags of the content under multiple attribute dimensions; The predicting the content quality parameter of the candidate publication object based on the content tag set of the at least one first publication content associated with the candidate publication object comprises: Based on the content tag set, first tag distribution features of each first published content associated with the candidate published object and second tag distribution features of each target published content are counted, wherein the tag distribution features represent the total number of occurrences of each content tag; the target published content is the first published content whose publishing time is within a preset sequence range among the first published content; The first label distribution feature and the second label distribution feature are input into a trained quality assessment model to obtain a content quality parameter of the candidate release object.

10. The method according to claim 9, characterized in that The trained quality assessment model is trained in the following way: Performing cyclic iterative training on the quality assessment model to be trained based on the training sample set to obtain a trained quality assessment model; each training sample includes a sample release object and a sample label; The sample publishing object has published content on at least one external platform and the target platform, and the sample label is determined according to sample content feedback data associated with the sample publishing object on the target platform; wherein each iterative training performs the following steps: Select training samples from the training sample set, and input sample label distribution characteristics of sample release objects in the training samples and associated with the at least one external platform into the quality assessment model to be trained to obtain sample content quality parameters of the sample release objects; Based on the difference between the sample content quality parameter and the corresponding sample label, the parameters of the quality assessment model to be trained are adjusted.

11. The method according to claim 10, characterized in that The sample label distribution characteristics include: the first sample label distribution characteristics of each sample content published by the sample publishing object on the at least one external platform, and the second sample label distribution characteristics of each target sample content, wherein the target sample content is the sample content whose publishing time is within a preset order range among the sample contents.

12. The method according to any one of claims 2 to 8, characterized in that: The predicting, based on the external platform sharing data associated with the candidate publishing object and the second content feedback data associated with the target platform within the preset time range, the target supply and demand parameters of the candidate publishing object on the target platform includes: Determine potential supply and demand parameters of the candidate publishing object according to external platform sharing data associated with the candidate publishing object and other candidate publishing objects respectively within the preset time range; Obtaining current supply and demand parameters of the candidate publishing object according to the third content feedback data associated with the target platform within the preset time range and the content tag set of each first published content; Based on the potential supply and demand parameters and the current supply and demand parameters, target supply and demand parameters corresponding to the one candidate release object are determined.

13. The method according to claim 12, characterized in that The external platform sharing data includes: the total number of first published contents shared to the target platform and the total number of sharing times; Determining potential supply and demand parameters of the candidate publishing object according to the external platform sharing data associated with the candidate publishing object and other candidate publishing objects respectively within the preset time range includes: Obtaining a quantity comparison value determined based on the total quantity of content associated with the one candidate publishing object and the total quantity of content in a specified order among the total quantities of content associated with each of the plurality of candidate publishing objects; Obtaining a sharing comparison value determined based on the total number of shares associated with the candidate publishing object and the total number of shares associated with each of the plurality of candidate publishing objects in a specified order; The sum of the quantity comparison value and the sharing comparison value is used as a potential supply and demand parameter of the candidate release object.

14. The method according to claim 12, characterized in that The content tag set of each content includes: respective content tags of the content under multiple attribute dimensions; the second content feedback data includes: the number of completed views and the number of uncompleted views of the second published content; The obtaining of the current supply and demand parameters of the candidate publishing object according to the second content feedback data associated with the target platform within the preset time range and the content tag set of each first published content includes: For each content tag, do the following: Determining a scarcity parameter corresponding to the content tag according to the number of completed views and the number of uncompleted views of the second published content containing the content tag within the preset time range; According to the content tag sets corresponding to the first published contents associated with the candidate publication object within the preset time range, a frequency parameter corresponding to the content tag is determined; target content tags whose frequency parameter ranking is within a specified order range are determined, and the sum of the products of the scarcity parameters of the target content tags and the frequency parameters is used as the current supply and demand parameter of the candidate publication object.

15. The method according to claim 14, characterized in that The determining, according to each set of content tags corresponding to each first published content associated with the candidate published object within the preset time range, a frequency parameter corresponding to the content tag includes: For each content tag, do the following: Obtaining the number of occurrences of a content tag in the first published content published by the candidate publishing object within the preset time range, and the total number of occurrences of each content tag; Obtaining the total number of candidate publishing objects and the number of candidate publishing objects containing the one content tag; Based on the number of occurrences, the total number of occurrences, the total data and the quantity, a frequency parameter corresponding to the one content tag is determined.

16. The method according to any one of claims 1 to 8, characterized in that: The basic attribute information has at least one category; the corresponding content tag sets are obtained according to the basic attribute information of at least one first published content associated with a candidate published object, including: For each first publication, perform the following operations: Input each category of basic attribute information of a first published content into the trained multi-label classification model and represent them as multiple attribute vectors respectively; For each type of basic attribute information, multiple attribute vectors corresponding to the basic attribute information are represented as a target vector through one-dimensional convolution processing; features of the target vector are extracted based on the multimodal attention network in the multi-label classification model to obtain a corresponding target vector; The target vectors corresponding to various basic attribute information are spliced ​​to obtain a spliced ​​vector; A content tag set of the first published content is obtained based on the splicing vector prediction.

17. A device for obtaining a publication object, characterized in that: include: A data acquisition unit, used to acquire first published content and first content feedback data of each of the plurality of candidate publishing objects on at least one external platform, and external platform sharing data associated with the target platform; The external platform is a content platform where the candidate publishing object has been registered, and the target platform is a content platform where the candidate publishing object has not been registered; The potential estimation unit is used to perform the following operations for each candidate publishing object: obtain a corresponding content tag set according to basic attribute information of at least one first published content associated with a candidate publishing object; obtain a corresponding object potential parameter according to the obtained at least one content tag set, in combination with first content feedback data and external platform sharing data associated with the candidate publishing object, wherein the object potential parameter represents: the estimated popularity of the published content uploaded by the candidate publishing object on the target platform; The object acquisition unit is used to select at least one from the multiple candidate release objects as a target release object according to the obtained potential parameters of each object.

18. An electronic device, characterized in that: It comprises a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of any one of the methods of claims 1 to 16.

19. A computer-readable storage medium, characterized in that: It comprises a computer program. When the computer program is run on an electronic device, the computer program is used to enable the electronic device to execute the steps of any method described in claims 1 to 16.

20. A computer program product, characterized in that It includes a computer program, which is stored in a computer-readable storage medium; when a processor of an electronic device reads the computer program from the computer-readable storage medium, the processor executes the computer program, so that the electronic device executes the steps of any method described in claims 1 to 16.

Images