WO2019113977A1

WO2019113977A1 - Method, device, and server for processing written articles, and storage medium

Info

Publication number: WO2019113977A1
Application number: PCT/CN2017/116646
Authority: WO
Inventors: 周莜; 徐澜; 谢奕; 阳丹
Original assignee: 腾讯科技（深圳）有限公司
Priority date: 2017-12-15
Filing date: 2017-12-15
Publication date: 2019-06-20
Also published as: CN110325986B; CN110325986A

Abstract

Provided are a method, device, and server for processing written articles, and a server, employing user-generated content platforms, and a storage medium. The method comprises: receiving target articles sent by a client, wherein the client is used to connect to a user-generated content platform, and the target articles are submitted by users of the user-generated content platform by means of the client; determining a promotional article for presenting promotional information from among the target articles, as well as a position for adding the promotional information in the promotional article; determining a target promotional object from among candidate promotional objects stored in the user-generated content media platform and materials matching the target promotional object; generating promotional information according to the determined materials matching the target promotional object; according to the determined promotional position, adding the promotional information to the corresponding position in the promotional article; and sending the promotional article with the added promotional information.

Description

Article processing method, device, server and storage medium

Technical field

The present invention relates to communication technologies, and in particular, to an article processing method, apparatus, server, and storage medium based on a self-media platform.

Background technique

With the development of the Internet's special mobile Internet, micro-blogs, blogs, and public accounts have become popular social channels. In the self-media platform, the promotion of products or services has become a promotion of goods and services. Promote the universal use of technical means.

The articles published in the self-media platform carry the user's appeals for expressing emotions, disseminating information and socializing. The related technologies provide a scheme for adding promotional information to the article. When the article reaches the user and is viewed, the promotional information added in the article is The user presents the article in the process of viewing the article, and realizes the effect of promoting the object.

For the promotion information added in the article, a technical solution adopted by the related technology is to add the promotion information to the article published by the user on the self-media platform, and then push the article with the promotion information to the user, and the technical solution is read for the user. The perception in the process of the article causes a lot of interference, which leads to a decrease in the user acceptance of the published article, which in turn affects the effect of object promotion;

In addition, a technical solution adopted by the related art is that a special account is opened from the media platform, and an article of various promotion information is published through an account, and the access traffic of the dedicated account has a large fluctuation. Sex, especially in the initial stage of account creation, it is difficult to support the timeliness of the promotion target and the needs of a specific user group.

Summary of the invention

In view of this, embodiments of the present invention are expected to provide an article processing party based on a self-media platform. The method, device, server and storage medium can realize the ideal fusion of promotion information in self-media articles, and promote the good timeliness of information reaching users.

To achieve the above objective, the technical solution of the embodiment of the present invention is implemented as follows:

In a first aspect, an embodiment of the present invention provides an article processing method based on a self-media platform, including:

Receiving a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the client;

Determining, in the target article, a promotion article for presenting promotion information, and a promotion location for adding promotion information in the promotion article;

Determining, in a candidate promotion object stored in the self-media platform, a target promotion object and a material matching the target promotion object;

Generating promotion information according to the determined material that matches the target promotion object;

And adding the promotion information to a corresponding promotion location in the promotion article according to the determined promotion location;

The promotion article to which the promotion information is added is sent.

In a second aspect, an embodiment of the present invention provides an article processing method based on a self-media platform, where the method is performed by a server, where the server includes one or more processors and a memory, and one or more programs, where The one or more programs are stored in a memory, the program may include one or more units each corresponding to a set of instructions, the one or more processors being configured to execute instructions; Methods include:

Determining a target promotion object among candidate promotion objects stored in the self-media platform, and a material that matches the target promotion object;

The promotion article to which the promotion information is added is sent.

In a third aspect, an embodiment of the present invention provides an article processing apparatus based on a self-media platform, including:

a receiving unit, configured to receive a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the first client;

a determining unit configured to determine, in the target article, a promotion article for presenting the promotion information, and a promotion location for adding the promotion information in the promotion article;

And configured to determine a target promotion object and a material matching the target promotion object among the candidate promotion objects stored in the self-media platform;

a generating unit, configured to generate promotion information according to the determined material that matches the target promotion object;

Adding a unit, configured to add the promotion information to a corresponding promotion location in the promotion article according to the determined promotion location;

The sending unit is configured to send the promotion article to which the promotion information is added.

In a fourth aspect, an embodiment of the present invention provides a server, including:

a memory configured to store an executable program;

The processor, configured to execute the executable program stored in the memory, implements the above-described self-media platform-based article processing method.

In a fifth aspect, an embodiment of the present invention provides a storage medium, where an executable program is stored, and when the executable program is executed by a processor, the article processing party based on the self-media platform is implemented. law.

The above embodiments of the present invention have the following beneficial effects:

The source of the target article can come from any user terminal in the social network, breaking the limitation of the article collecting the specific topic, and the batching and automatic addition of the promotion information can be realized; the location of the promotion information is automatically selected, and the position is flexible. It can avoid the sudden emergence of promotion information, make the content of the article and the content of the promotion information natural; through the process of publishing the article and reaching the user, the delivery of the promotion information is completed, and the promotion information is promoted by relying on the release/send traffic of the self-media platform itself. The information is covered by the traffic from the media platform and reaches the user in real time.

DRAWINGS

1A is a schematic diagram of an optional application scenario of an article processing method based on a self-media platform according to an embodiment of the present invention;

FIG. 1B is a schematic diagram of an optional application scenario of an article processing method based on a self-media platform according to an embodiment of the present disclosure;

2A is a schematic diagram of an optional presentation manner of a promotion article according to an embodiment of the present invention;

2B is a schematic diagram of an optional presentation manner of a promotion article according to an embodiment of the present invention;

2C is a schematic diagram of an optional presentation manner of a promotion article according to an embodiment of the present invention;

FIG. 3 is a schematic structural diagram of an optional hardware of an article processing apparatus based on a self-media platform according to an embodiment of the present disclosure;

4 is an optional schematic flowchart of an article processing method based on a self-media platform according to an embodiment of the present invention;

FIG. 5A is a schematic diagram of performing topic prediction by using a keyword-topic classifier according to an embodiment of the present invention; FIG.

FIG. 5B is a schematic diagram of performing similarity calculation by using a text-to-text similarity classifier according to an embodiment of the present invention; FIG.

FIG. 5C is a similarity diagram of an image-image similarity classifier according to an embodiment of the present invention; Schematic diagram of the calculation;

5D is a schematic diagram of performing similarity calculation by using a text-image similarity classifier according to an embodiment of the present invention;

6A is an optional schematic diagram of a text material according to an embodiment of the present invention;

FIG. 6B is an optional schematic diagram of a text material according to an embodiment of the present invention; FIG.

FIG. 7A is a schematic diagram of adding promotion information in a promotion article according to an embodiment of the present invention; FIG.

FIG. 7B is a schematic diagram of adding promotion information in a promotion article according to an embodiment of the present invention; FIG.

FIG. 7C is a schematic diagram of adding promotion information in a promotion article according to an embodiment of the present invention; FIG.

FIG. 8 is a schematic diagram of promotion information according to an embodiment of the present invention;

9A is a schematic diagram of a manner of displaying promotion information according to an embodiment of the present invention;

9B is a schematic diagram of a manner of displaying promotion information according to an embodiment of the present invention;

9C is a schematic diagram of a manner of displaying promotion information according to an embodiment of the present invention;

FIG. 10 is an optional schematic flowchart of an article processing method based on a self-media platform according to an embodiment of the present disclosure;

FIG. 11 is an optional schematic flowchart of an article processing method based on a self-media platform according to an embodiment of the present disclosure;

FIG. 12 is a schematic structural diagram of a composition processing apparatus based on a self-media platform according to an embodiment of the present invention.

Detailed ways

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is to be understood that the examples are provided to illustrate the invention and not to limit the invention. In addition, the embodiments provided below are part of the embodiments for implementing the present invention, and do not provide all the embodiments for implementing the present invention. In the case of no conflict, the technical solutions described in the embodiments of the present invention may be combined in any combination. Implementation.

It should be noted that, in the embodiments of the present invention, the terms "including", "including" or any of them are used. The other variations are intended to cover a non-exclusive inclusion, such that a method or apparatus that includes a plurality of elements includes not only the elements that are specifically described, but also other elements that are not explicitly listed, or The inherent elements. In the absence of further limitation, an element defined by the phrase "comprising a ..." does not exclude the presence of additional related elements in the method or device including the element (eg, a step in the method or a unit in the device) The unit here may be part of a circuit, part of a processor, part of a program or software, etc.).

For example, the article processing method based on the self-media platform provided by the embodiment of the present invention includes a series of steps, but the article processing method based on the self-media platform provided by the embodiment of the present invention is not limited to the described steps, and the present invention is similarly The article processing device based on the self-media platform provided by the embodiment includes a series of units, but the device provided by the embodiment of the present invention is not limited to including the unit explicitly described, and may also include when the related information is acquired or processed based on the information. The unit that needs to be set.

Before the embodiments of the present invention are further described in detail, the nouns and terms involved in the embodiments of the present invention are explained. The nouns and terms involved in the embodiments of the present invention are applicable to the following explanations.

1) Self-media platform, also known as self-media, an information platform set up on the Internet for users (including individual users, groups, organizations, etc.) to publish articles, depending on the server and the implementation of the server deployment. Software (supports front-end access and background processing); self-media platforms such as Weibo, blogs, personal websites, forum communities, and public numbers for various social applications.

The accounts from the media platform can be different types of individuals, organizations, groups, and enterprises. After registering the account from the media, the news, dynamics, and other articles submitted by the client are related to the user's own preferences, dynamics, or business. The end is pushed to the appropriate user via the self-media platform.

2) Articles for articles published on the self-media platform, the content of which includes one or a combination of text and images.

3) Promotion of information, for the promotion of the object of promotion, applicable to the spread of the Internet Types of information, the objects advertised in the promotion information are called promotion objects, such as advertisements.

4) Word vector, using a word-to-vector mapping model such as word-to-vector (Word2Vector), based on the degree of semantic similarity between different words, the vector obtained by mapping words into vector space, the distance between different word vectors It is negatively related to the degree of semantic similarity of the corresponding words, that is, the smaller the distance of the word vectors of the two words (such as the Euclidean distance), the closer the semantics of the two words are.

5) The topic feature, mapping the keywords representing the theme into corresponding word vectors, and combining them, also called theme feature vectors.

6) Content features, mapping a plurality of feature words extracted from the article into corresponding word vectors, and combining them, also referred to as content feature vectors.

7) Word segmentation, also known as word segmentation, according to a certain word segmentation strategy refers to the division of characters in an article into separate words.

8) Stop words, words that are filtered from the article and do not affect the classification decision of the article; usually the general words do not have a clear meaning (only if they are put into a complete sentence), for example, pronouns , articles such as articles and numerals, modal particles, adverbs, prepositions and conjunctions.

9) Feature words, after the article is segmented, after filtering the stop words from the article, the words that can be extracted from the remaining words can represent the subject of the article.

10) a classifier model, also called a classifier, is a model for classification obtained by means of machine learning, and is used for predicting an article as a target category based on the sample characteristics of the article to indicate that the article is a target category. The probability.

For example, the classifier model in this paper can use the two-classifier model of Support Vector Machines (SVM), the word bag-based classifier model, the classifier model based on prior probability and sparse features, based on neural network and depth. The classifier model of the classifier model such as learning, if not specified, the classifier model described in this paper is used for two classifications, such as judging whether it belongs to a topic, and determining whether the article belongs to the target category.

11) Machine Learning, through the sample of the training set (referred to as The sample is trained on the sample features and whether it belongs to the target category (such as beauty articles), and the classifier model is trained so that the trained classifier model has the performance of determining whether the article sample of the test set belongs to the target category.

12) The training set, including the article that trains the classifier model, the vector representation of the article and the prior classification results are used to construct the training samples to train the classifier model, so that the classifier model has the performance of classifying the target category by the article to be tested. .

13) A test set comprising articles to be tested (classified), the vector representation of the article being used to input a classifier model to predict scores belonging to the target category.

Embodiments of the present invention provide an article processing method based on a self media platform, an article processing device based on a self media platform based on an article processing method based on a self media platform, and an executable program for implementing an article processing method based on a self media platform. Storage medium. For the implementation of the article processing method based on the self-media platform, the embodiment of the present invention provides a solution implemented by the terminal side and the server side. Next, an exemplary implementation scenario of the article processing will be described.

1A and FIG. 1B are schematic diagrams of an optional application scenario of an article processing method based on a self-media platform according to an embodiment of the present invention. As shown in FIG. 1A and FIG. 1B, in the embodiment of the present invention, the user terminal is not limited to a mobile phone. For a tablet, a PC, or the like, the server may be any commercial or dedicated server. In the embodiment of the present invention, based on the functions implemented by the server, the server is divided into two categories, namely, a social network server 21 and an advertisement back server. 22, and in practical applications, each type of server can be set according to the actual situation one or more. The user terminal 11 to the user terminal 15 can exchange information with the social network server 21 and the advertisement background server 22 through a wired network, a wireless network, or a combination of the two, and each user terminal can send and receive information (such as an article) through the server. Delivery, etc. The article processing method of the embodiment of the present invention will be described below with reference to FIG. 1A and FIG. 1B. It should be noted that the network shown in FIG. 1A and FIG. 1B is merely an example for easy understanding and does not constitute the network architecture of the present invention. Any restrictions.

Referring to FIG. 1A, in some embodiments, the advertisement backend server 22 obtains from the advertiser terminal. Taking the promotion article carrying the promotion information (such as advertisement), and then sending the promotion article to the social network server 21 to send the promotion article carrying the promotion information to the social network through the social network server 21, so that the social network user can receive and read the carrier. Promote promotional articles for information.

The promotion article carrying the promotion information may be that the advertiser writes the advertisement for the specific group (such as the public number user) or the specific product (such as a designated shampoo), for each product or a specific group of people for promotion. Articles (also known as soft texts) can be in the form of text and/or images that combine advertising and article content.

For example, FIG. 2A is a schematic diagram of an optional presentation manner of a promotion article provided by an advertiser according to an embodiment of the present invention. The advertiser is an advertiser for the public number user (ie, through social interaction). The web client pays attention to the user of the public number to write the soft text, and then the advertiser sends the soft text to the advertisement background server 22, and the advertisement background server 22 sends the soft text to the terminal of the user who pays attention to the public number through the social network server 21, as shown in FIG. 2A. In the process of accessing the public number, the user jumps to the interface 2 by clicking anywhere in the interface 1, and both sees the content of his own attention and the advertisement of the advertiser.

However, the labor cost of the above implementation method is very high, and it is necessary to write corresponding articles for different advertisers, and it is impossible to achieve scale and batchization, and at the same time, the demand for real-time promotion cannot be met due to the special writing of soft texts.

Referring to FIG. 1A, in some embodiments, the social network server 21 obtains articles for a particular topic collection from a user, and then sends the collected articles to an advertisement backend server 22, which adds promotional information to the collected articles ( For example, an advertisement is sent to the social network through the social network server 21, so that the social network user can receive and read the promotion article with the promotion information.

For example, FIG. 2B is a schematic diagram of an optional presentation manner of a promotion article according to an embodiment of the present invention. The public number initiates an essay activity on a specific theme (such as Valentine's Day). , the social network server 21 that carries the public number function is obtained The article sent by the user participating in the activity sends the obtained article to the advertisement background server 22, and the advertisement background server 22 adds an advertisement at the end of the article, obtains a promotion article for adding the advertisement, and sends the promotion article to the social network server 21 through the public. The number will send the promotion article to the social network, and the social network user jumps to the interface 2 by clicking anywhere in the interface 1 to read the article content and add the advertisement.

However, the source of the above-mentioned implementations has a large limitation due to the fact that the article is collected from the user, and is limited to the user-generated content (UGC) article of the specific topic collected, and the advertisement content is mainly displayed. Users are unable to further manipulate the displayed ads, reducing the user’s desire to purchase ads and understanding desires.

Referring to FIG. 1A, in some embodiments, the social network server 21 obtains an article to be published from a user, and then transmits the obtained article to an advertisement backend server 22, which adds a topic related to the article topic at a specific location of the article. The promotion information (such as an advertisement) is promoted and the promotion article is sent to the social network through the social network server 21.

For example, FIG. 2C is a schematic diagram of an optional presentation manner of the promotion article provided by the social network server 21, and the social network server 21 is operated from the operation user terminal of the news release platform. The side obtains the article to be published, and then sends the obtained article to the advertisement background server 22, and the advertisement background server 22 adds the advertisement related to the article topic to the promotion article at the end of the article, and sends the promotion article to the social network server 21 through the news. The publishing platform sends the promotion article to the social network, and the user jumps to the interface 2 by clicking anywhere in the interface 1, and can see the promotion article with the added advertisement.

However, in the above implementation manner, although the advertisement added in the article has a high degree of fit with the article theme, the combination of the advertisement and the article content is hard due to the fixed location, and the promotion information and the content in the article are not highly correlated or even Unrelated situations reduce the acceptance of promotional information.

Referring to FIG. 1B, in some embodiments, embodiments of the present invention are based on articles from a media platform. The implementation of the method may include: the social network server 21 carrying the self-media platform function receives the to-be-published article submitted by the user of the media platform through the first client connected to the media platform, and sends the to-be-published article to the advertisement background server 22 The advertisement background server 22 determines, in the article to be published, a promotion article for presenting the promotion information, and a promotion location for adding the promotion information in the promotion article; the social network server 21 determines the target promotion of the candidate promotion object that matches the promotion article. The object and the material matching the target promotion object; adding the promotion information including the material to the promotion location of the promotion article according to the determined promotion location; and sending the promotion article with the promotion information to the social network server 21, and the social network server 21 sends The promotion article of the promotion information is added to the second client of the media platform for presentation.

Referring to FIG. 1B, in some embodiments, the implementation of the article processing method based on the self-media platform of the embodiment of the present invention may include: the social network server 21 carrying the self-media platform function receives the user from the media platform by connecting to the media platform. The target article submitted by the first client; the social network server 21 obtains the candidate promotion object from the advertisement background server 22 or obtains the candidate promotion object stored by itself, and stores the material of the candidate promotion object itself, and determines the presentation for the presentation in the target article. a promotion article of the promotion information, and a promotion location for adding the promotion information in the promotion article; the social network server 21 determines the target promotion object matching the promotion article in the candidate promotion object, and the material matching the target promotion object; The location, adding the promotion information including the material to the promotion position of the promotion article; sending the promotion article adding the promotion information to the second client from the media platform for presentation. Among them, the target article submitted by the client may include the article to be published and the original article; the original article here refers to the article that has been released and retracted through the self-media platform.

Next, an exemplary hardware structure of an apparatus corresponding to the self-media platform-based article processing method for implementing an embodiment of the present invention is described with reference to FIG. 3, and the article processing apparatus based on the self-media platform may be implemented in various forms, such as a terminal (eg, Various types of computer equipment, such as desktop computers, laptops or smart phones), servers, etc., are independent of computer equipment such as terminals and servers. The self-media platform-based article processing method of the embodiment of the present invention is implemented in a coordinated manner. The hardware structure of the article processing apparatus based on the self-media platform of the embodiment of the present invention is described in detail below. It can be understood that FIG. 3 only shows an exemplary structure of the article processing apparatus based on the self-media platform, and not all the structures, as needed. Part or all of the structure shown in Fig. 3 is implemented.

Referring to FIG. 3, FIG. 3 is a schematic diagram of an optional hardware structure of an article processing apparatus based on a self-media platform according to an embodiment of the present disclosure, which can be applied to a server in the foregoing application scenario, such as a background server that can be a microblog/WeChat. From the background server of the media website, the article processing apparatus 100 shown in FIG. 3 includes at least one processor 101, a memory 102, and at least one network interface 103. The various components in article processing device 100 are coupled together by bus system 104. It will be appreciated that the bus system 104 is used to implement connection communication between these components. The bus system 104 includes, in addition to the data bus, a power bus, a control bus, and a status signal bus. However, for clarity of description, various buses are labeled as bus system 104 in FIG.

The memory 102 may be a volatile memory or a non-volatile memory, and may also include both volatile and non-volatile memory.

The memory 102 in the embodiment of the present invention is used to store various types of data to support the operation of the article processing apparatus 100 based on the self media platform. Examples of such data include: any computer program for operating on the self-media platform-based article processing apparatus 100, such as the executable program 1021, the program implementing the self-media platform-based article processing method of the embodiment of the present invention may be included in It can be executed in the program 1021.

Network interface 103 may include one or more communication modules, such as a mobile communication module and a wireless internet module.

The article processing method based on the self-media platform disclosed in the embodiment of the present invention may be applied to the processor 101 or implemented by the processor 101. Processor 101 may be an integrated circuit chip with signal processing capabilities. In the implementation process, the steps of the method in the embodiment of the present invention may be completed by using an integrated logic circuit of hardware in the processor 101 or an instruction in a software form. abovementioned The processor 101 can be a general purpose processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, or the like. The processor 101 can implement or perform the various methods, steps, and logic blocks disclosed in the embodiments of the present invention. A general purpose processor can be a microprocessor or any conventional processor or the like. The steps of the method disclosed in the embodiment of the present invention may be directly implemented as a hardware decoding processor, or may be performed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, and the storage medium is located in the memory 102. The processor 101 reads the information in the memory 102, and completes the steps of the above-described self-media platform-based article processing method provided by the embodiment of the present invention.

In an exemplary embodiment, the article processing apparatus 100 based on the self-media platform may be configured by one or more Application Specific Integrated Circuits (ASICs), DSPs, Programmable Logic Devices (PLDs), and complexities. A Programmable Logic Device (CPLD) is used to execute the self-media platform-based article processing method of the embodiment of the present invention.

Based on the application scenario of the article processing method based on the self-media platform and the article processing device based on the self-media platform, the implementation process of the article processing method based on the self-media platform according to the embodiment of the present invention is described.

As an optional embodiment of the article processing method based on the self-media platform, the embodiment of the present invention can be implemented as an advertisement background server implemented by the article processing device based on the self-media platform as an information promotion platform function, and the advertisement background server passes the social network. The server receives the to-be-published article from the self-media platform client, determines the promotion article in the article to be published, obtains the candidate promotion object information, determines the target promotion object, and the material corresponding to the target promotion object, and generates promotion information including the material, and determines Promote the promotion location of the promotion information in the article, add the promotion information to the promotion location in the promotion article, and add the promotion article of the promotion information to the client presentation of the self-media platform through the social network server. Of course, based on self-media The article processing device of the platform can also be implemented in other application environments, such as a background server of a news push APP, a background server of a media website, a server of an information promotion platform, etc., and the article processing device based on the self media platform is not excluded from being provided. Any application environment from the media platform function.

As an alternative embodiment of the article processing method based on the self-media platform, FIG. 4 is a schematic flowchart of an article processing method based on the self-media platform provided by the embodiment of the present invention. The embodiment of the present invention will be described as a social network server deployed from a media platform based on a self-media platform. The article processing method based on the self-media platform provided by the embodiment of the present invention includes:

Step 301: The publisher sends the to-be-published article to the social network server by using the first client from the media platform.

Articles to be published may be articles to be published on self-media platforms (such as Weibo, public number, forum community, etc.).

The social network server carries the functions of a self-media platform (such as Weibo, public number, QQ space, forum community, etc.), and the social network may be an Internet social entity based on Weibo, public number, etc., which can be used by users to publish articles, and the publisher of the article. For any user in the social network, after the user completes the writing of the article based on the client of the self-media platform on the terminal, the article to be published is sent to the social network server for sending to the client of the self-media platform through the social network server. Present it. In this way, the self-media platform's spontaneous traffic about the article is used as the carrier of the promotion information, that is, the source of the article can come from any user terminal in the social network, breaking the limitation of the article collecting the specific topic.

Exemplarily, the public number operation user as the publisher of the article may submit the article through the client, for example, may be an article written by the user about the theme of mood, food, makeup, etc., or may be downloaded from the Internet and the above. Topic-related articles, which will be sent by the client to the social network server hosting the public number function to serve through the social network service. Send the article to the user who is following the public number.

For another example, the Weibo user acts as a publisher, and the article to be published is sent to the social network server hosting the Weibo function through the Weibo client, and the social network server sends the article to the user who pays attention to the publisher's Weibo account.

In an embodiment, after receiving the article to be published sent by the publisher terminal, the social network server may directly send the to-be-published article to other users of the social network, or may perform the synthesis process of the promotion information for publishing the article to perform the synthesis. The publication of articles with promotional information.

The following takes the steps 302 to 305 to describe the synthesis processing of the promotion information of the social network server to be published.

Step 302: The social network server determines, in the article to be published, a promotion article for presenting the promotion information, and a location for promoting the promotion information in the promotion article.

When the promotion information is an advertisement, there may be articles that are not suitable for advertisement addition, such as academic articles, so it is necessary to filter the articles to be published, and determine an article capable of presenting (adding) advertisements, that is, promotion articles.

In some embodiments, the social network server may acquire the candidate promotion object through the advertisement background server, and itself stores the material information of the candidate promotion object, and the social network server may determine the promotion article for presenting the promotion information by:

The topic similarity degree is calculated by the topic of the article to be published and the topic represented by the candidate promotion object, and the article to be published that satisfies the topic similarity condition (such as the similarity exceeds the preset topic similarity threshold) is determined as the promotion for presenting the promotion information. article.

In an embodiment, the feature words representing the topic of the article and the candidate promotion object are respectively input into a preset learning model (such as word2vec), and the theme feature vector of the corresponding article theme and the candidate promotion object is obtained through the learning model mapping, and then calculated. The similarity between the topic feature vector of the article and the topic feature vector of the candidate promotion object, and the article whose similarity exceeds the similarity threshold is selected as the promotion article. For example, the topic features and candidate promotion of the article to be published can be achieved as follows: Subject similarity calculation of the subject feature of the object: input the keyword extracted from the article to be published, input the classifier model according to the feature word, and obtain the topic corresponding to the article to be published calculated by the classifier model; The keyword extracted from the material of the object is input into a classifier model for classifying the topic according to the feature word, and the topic corresponding to the candidate promotion object is obtained; and the semantic distance is determined according to the semantic distance of the topic corresponding to the article to be published and the topic corresponding to the candidate promotion object. The topic similarity of negative correlations. For example, calculating the Euclidean distance of the topic vector corresponding to the published article and the topic vector corresponding to the candidate promotion object, and then calculating the reciprocal thereof as the topic similarity.

In some embodiments, the articles and the topics of the candidate promotion objects may be level-set according to a preset classification standard, such as a first-level theme, a second-level theme, and a third-level theme according to a hierarchical level from high to low; The theme of each level may include multiple topics of the next level. For example, the first level theme may be military, sports, entertainment, finance, and the first level theme is entertainment, which may include food, There are a number of second-level themes such as travel, movies, and music. The second-level theme is music. For example, it can include multiple third-level themes such as jazz music and classical music.

Correspondingly, in an embodiment, the social network server divides the topic of the article to be published into the corresponding first-level topic, and calculates the similarity between the first-level topic of the article to be published and the first-level topic of the candidate promotion object. When the similarity satisfies the similarity condition (for example, the topic similarity threshold exceeding the first-level topic), it is determined that the article to be published is a promotion article for presenting the promotion information. That is to say, based on the first-level topic feature, the article is roughly screened to obtain the promotion article, but in other embodiments, the article may be filtered based on the second-level topic or the third-level topic to obtain the promotion article, but Since the first level topic includes a plurality of second level topics and a plurality of third level topics, that is, the feature vector dimension corresponding to the first level theme is lower than the dimension of the feature vector corresponding to the second/third level theme, and the social network The traffic received by the server from the media platform is very large. Therefore, the screening of the article using the first-level theme is obviously faster than the filtering based on the second/third-level theme, that is, the promotion of the promotion information can be quickly determined. Article to reduce network latency.

In an embodiment, the manner in which the topic of the article to be published is divided into the first-level topic is similar to the manner of dividing into the second-level theme and the third-level theme, and can be implemented by using a classifier of the corresponding level as a classifier. For example, by extracting features of the article (at least one of the text feature and the image feature), the extracted feature is mapped to a corresponding feature vector through a preset learning model, and the feature vector of the obtained article is input into a classifier of a different level. The mapping obtains the topic of the corresponding level; for example, input the feature word of the extracted article into the preset word2vec model, obtain the corresponding feature vector of multiple dimensions, and input the obtained feature vector of the plurality of dimensions into the first classifier to obtain the corresponding The first level theme. For example, if the feature vector of the plurality of dimensions of the obtained article is input to the first-level classifier, the probability of the first-level theme is 10% for sports, 80% for entertainment, and 5% for financial. The most probable "entertainment" is selected as the final first-level theme.

In some embodiments, the acquisition of different ranks of classifiers can be obtained in one of the following ways:

1) Supervised learning methods, such as manually annotating text and/or pictures corresponding to several topics, and training a specific level of feature-thematic classifier model with the characteristics of the annotated data (text and/or picture features), obtained through training The feature-theme classifier implements a mapping of the corresponding level of topics. 2) Unsupervised learning method, clustering the text and/or picture features of the article to get the theme of the corresponding article.

The candidate promotion object may be a certain number of promotion objects prior to the prioritization (such as the auction ranking) obtained from the promotion system (such as the advertisement back-end server).

In some embodiments, the social network server may also determine a promotional article for presenting promotional information by:

The social network server matches the subject of the to-be-published article with at least one of the name of the candidate promotion object, the category of the candidate, and the corresponding promotion information keyword, and determines that the matching condition is met (ie, the similarity condition is met, for example, the similarity reaches the threshold). The article is a promotional article for presenting promotional information. The subject of the above-mentioned article to be published may be one of the following two types:

1) keywords of the article to be published;

For example, the keyword may include: a keyword extracted from an article title or article content (such as each paragraph of the article); for example, the title of the article to be published is “Qingdao Food Analysis”, and the extracted keyword is “Gourmet”. .

2) Using the keyword-topic model to predict the topic of the article;

FIG. 5A is a schematic diagram of a topic prediction using a keyword-topic classifier model according to an embodiment of the present invention. Referring to FIG. 5A, a keyword-topic model may be a classifier model obtained by pre-training, and implements article keywords and articles. The relationship mapping of the topic, two or more keywords through the keyword-topic classifier mapping to get the topic of the article. For example, the keyword extracted from the content of the article is “sugar”, “cookie”, “ instant noodles”, “chocolate”, and the vector corresponding to these keywords is input into the keyword-theme model to predict the theme. The probability of getting the theme "food" is 80%, the probability of getting the theme "entertainment" is 3%, and the result with the highest probability (food) is selected as the subject of the predicted article.

Next, the candidate promotion object in the embodiment of the present invention is described. The candidate promotion object may be a service (such as a movie, a game, etc.) or a product (such as cosmetics, clothes, shoes, etc.); The name, category, and promotion information keyword corresponding to the object may be a service name (such as a movie name), a service category (such as a movie), an advertisement word (such as "Who said the car cannot fly-XXX"); The object is a product. For example, the name, category, and promotion information keyword corresponding to the candidate promotion object may be the product name corresponding to the product (such as women's brand-XX), product category (such as clothes), and advertising words (such as "from France." Romantic, fashion clothing - XX").

Based on the foregoing description of the topic of the article, the candidate promotion object, and the like, in some embodiments, when the topic of the article to be published is matched with at least one of the name of the promotion object, the category of the promotion, and the corresponding promotion information keyword, Promote promotional articles for presenting promotional information in one of the following ways:

1) The content corresponding to at least one of the name of the candidate promotion object, the category to which it belongs, and the corresponding promotion information keyword, including the subject of the article to be published; for example, the promotion information keyword corresponding to the candidate promotion object is: Queen of Fruits - Mangosteen The subject of the article to be published is: fruit, and it is determined that the article to be published satisfies the matching condition;

2) separately calculating the similarity between the name of the candidate promotion object, the category and the corresponding promotion information keyword and the topic of the article to be published, and when at least one of the calculated three similarity values exceeds the preset similarity threshold, Make sure the article to be published meets the matching criteria.

After the promotion article for presenting the promotion information is determined, the promotion position for adding the promotion information in the promotion article is explained next.

In some embodiments, the promotion location for adding the promotion information in the promotion article may be determined by: determining, according to the topic feature included in the promotion article, a paragraph having the included topic feature in the promotion article; when the included topic feature When the topic feature of the promotion information satisfies the topic similarity condition, it is determined that the position of the corresponding paragraph is a promotion position for adding the promotion information.

The promotion article may include one or more topics. Different topics may be distributed in different paragraphs of the promotion article. The positions corresponding to different paragraphs (corresponding to different topics) may be the middle position of the article, the end position of the article, or The position where two adjacent topics (paragraphs) are handed over; for example, when the promotion article contains only one topic, the topic feature and the topic feature of the promotion information satisfy the topic similarity condition, and the position of the paragraph including the topic (the end position of the article) In order to promote the location; when the promotion article contains two or more topics, multiple topics are distributed in different paragraphs, when at least one of the plurality of topic features and the topic feature of the promotion information satisfy the topic similarity condition, The position where the paragraph in which the topic feature satisfying the topic similarity condition is placed and the adjacent paragraph is used as the promotion position. In this way, the selection of the location of the promotion information is automatically realized, the position is flexible, and the occurrence of the promotion information can be avoided, so that the content of the article and the content of the promotion information are naturally connected, and it is easy to be accepted by the user in the process of reading the article.

In some embodiments, the promotion location for adding the promotion information in the promotion article may also be determined by: when the promotion information is added to the position between the adjacent paragraphs in the promotion article, according to the content style of the promotion article Whether the same type of content is segmented by the promotion information, and/or the display ratio occupied by the promotion information in the content style determines a corresponding completeness; when the completeness satisfies the preset completeness condition, determining The location between the adjacent paragraphs is the promotion location where the promotion information is added.

Illustratively, when the promotion information is added at a position where the promotion article can add the promotion information (ie, the candidate location, for example, the middle position of any two paragraphs), according to whether the same type of content in the content style of the article is divided by the promotion information, Determining the completeness of the content style. If the same type of content is segmented by the promotion information, the content style is destroyed, and the corresponding completeness is 0; if the content of the same type in the content style of the article is not segmented by the promotion information, the content style is It is still complete, and the corresponding degree of completeness is 1. The candidate position meets the integrity condition and can be used as a promotion location.

For example, if the content style of the promotion article only contains text, in order to ensure the integrity of the content style in the article, the end position of the article content can be used as a promotion location for adding promotion information (advertisement), so that the content style of the article does not destroyed.

For another example, if the content sample of the promotion article includes a plurality of categories or a plurality of pictures in addition to the text, the end position of each type or each picture may be used as a promotion position for adding the promotion information; When the promotion information is added in the middle of the article, the influence on the content style of the article is minimal, and the ideal integration degree of the promotion information in the article is formed.

Illustratively, when the promotion information is added at a position where the promotion article can add the promotion information (ie, the candidate location, for example, the middle position of any two paragraphs), the display ratio determined by the promotion information in the content style of the article is determined. The degree of completeness, the larger the proportion of the occupied display, the smaller the completeness of the corresponding content, the negative correlation between the two (inverse proportional relationship can be adopted); when the completeness is less than the completeness threshold, the candidate position does not meet the integrity condition .

In actual implementation, the social network server determines the promotion article for presenting the promotion information. And after the promotion location for adding the promotion information in the promotion article, the target promotion object matching the promotion article needs to be determined; the material corresponding to the target promotion object is used to generate the promotion information; that is, step 303 is performed: the social network server determines to match the promotion article. Target promotion object.

According to the above embodiment of the present invention, the target promotion object matching the promotion article may have one or more, and the target promotion object matching the promotion article may be determined by: performing the content feature of the promotion article and the content feature of the candidate promotion object. The content similarity calculation determines the candidate promotion object that satisfies the content similarity condition as the target promotion object.

In an embodiment, the social network server may determine the target promotion object that matches the promotion article by: the social network server may obtain the plurality of promotion objects to be promoted from the advertisement background server, and then perform the obtained plurality of promotion objects. Firstly, the candidate promotion object set matching the promotion article is obtained, and then the obtained candidate promotion object is subjected to secondary screening to obtain the target promotion object matching the promotion article.

For example, the social network server determines a candidate promotion object that satisfies the topic similarity condition of the promotion article, forms a candidate promotion object set, completes a screening, and then determines each candidate promotion object in the candidate promotion object set and the promotion article about at least one type feature. The similarity degree includes the image feature and the text feature, and the candidate promotion object whose similarity satisfies the similarity condition of the corresponding type feature is determined as the target promotion object, and the secondary screening is completed. By determining the target promotion object in the above manner, the first screening is based on the theme, and then the feature is selected for screening, and the purpose is to save all the use characteristics to filter and consume the computing power.

What needs to be explained here is that the similarity of the corresponding type of features mentioned here, that is, the similarity between the image features and the image features, or the similarity between the text features and the text features; that is, the similarity satisfies the similarity of the corresponding type features. The candidate promotion object of the degree condition is determined as the target promotion object, which comprises: determining the similarity between the extracted image feature and the feature image feature of the promotion article, and determining the candidate promotion object as the target promotion when the determined similarity exceeds the similarity threshold of the image feature Object; or, to determine the similarity between the extracted text features and the text features of the promotional article, When the determined similarity exceeds the similarity threshold of the text feature, it is determined that the candidate promotion object is the target promotion object. In actual implementation, the feature extraction operation may be performed before performing feature similarity calculation, such as: extracting image features composed of colors, textures, and shapes; and/or performing word segmentation processing, filtering out stop words for word segmentation results , get the text features composed of feature words.

In some embodiments, the candidate promotion object set matching the promotion article may be obtained by inputting at least one type of feature of the candidate promotion object into a classifier model for class classification according to the feature word, and obtaining a classifier model calculation output. The topic to which the candidate promotion object belongs; when the similarity between the mapped topic and the topic of the promotion article exceeds the topic similarity threshold, the candidate promotion object that satisfies the topic similarity condition with the topic of the promotion article is determined. The process of classifying a topic by a classifier model according to a feature word may include: combining a word vector having a plurality of feature words to form an input vector (a word vector of each feature word is output according to a semantic-vector model), and predicting different according to the input vector The probability of the topic, taking the topic corresponding to the maximum probability as the topic to which the candidate promotion object belongs.

The categorization is divided into three levels for exemplification. In some embodiments, the candidate promotion object set matching the promotion article may be selected from the promotion objects to be promoted based on the second-level theme, for example, by obtaining the following Promoting the matching of candidate objects in the article matching: determining the second-level topic feature of the promotion object and the second-level topic feature of the promotion article, and calculating the similarity between the second-level topic feature of the promotion object and the second-level topic feature of the promotion article, When the preset second-level topic similarity threshold (which can be set according to actual needs, such as 70%) is exceeded, it is determined as a candidate promotion object that matches the promotion article.

Of course, before the second-level topic similarity calculation, it is necessary to obtain the second-level theme of the promotion article and the promotion object, obtain the feature vector of the promotion object/promotion article, and input the acquired feature vector into the second classifier to obtain the corresponding first. Secondary theme. In actual implementation, the feature vector of the multiple dimensions of the promotion object/promotional article is mapped to obtain the corresponding second-level theme. Therefore, the above process of mapping to the corresponding second-level topic is equivalent to performing feature vectors of multiple dimensions. Dimensionality reduction The process, in this way, reduces the difficulty of the algorithm of the article processing.

Next, the extraction of different types of features in the topic mapping process will be described separately.

First, for the extraction of text features, the purpose of the extraction is to obtain a semantic description of the text from the article or the paragraph. In an implementation, the two main operations of preprocessing and text feature extraction may be included, wherein the preprocessing may include the following steps. :

Step 1, invalid character filtering; for example: if the article is derived from a web page, it is usually necessary to filter the HTML tag by means of a regular expression or the like.

Step 2, word segmentation;

In actual implementation, it is often necessary to first encode and convert the content obtained in step 1. Then, regular expressions can be used to match the punctuation marks and line breaks to divide the paragraphs of the article into sentences. Finally, the Chinese word segmentation can be used to divide the sentences into one sentence. Separate words.

Step 3. Filter the stop words;

In actual implementation, words of irrelevant semantics such as "", "ground", etc., may be filtered according to a preset dictionary. In some embodiments, after step 3 is performed, feature word extraction may be further performed, so that subsequent text feature extraction is more convenient.

After the execution of the above steps 1 to 3 is completed, the pre-processing of the text feature extraction is completed, and then the text feature can be extracted. In actual implementation, the text feature can be extracted in one of the following ways:

1) Keyword extraction, such as algorithm implementation using word frequency-inverse document frequency (TF-IDF, Term Frequency-Inverse Document Frequency).

2) BOW (Bag of Words) model, ignoring the grammar to represent the text as a set of words, that is, a combination of words.

3) Deep learning models, such as Word Embedding, map words to get word vectors and operate them through word vectors.

Next, the extraction of image features is described. The purpose of the extraction is to obtain the image from the article. The semantic description of the picture, in one implementation, the extraction of image features in the article can be one of the following ways:

1) Using the algebraic features of the picture matrix, for example, by Singular value decomposition (SVD, Singular value decomposition), Programmable Counter Array (PCA), etc., the matrix of the represented picture is reduced in dimension.

2) Adopt global statistical features such as histogram, contrast, geometric invariant moment, and so on.

3) Use local intuitive features, such as texture features (such as Linear Back Projection (LBP), Generalized Search Trees (GIST), corner features (such as Harris corner detection, etc.) , edge features (such as multi-level edge detection algorithm - Canny operator), shape features (such as Hough transform).

4) Feature extraction is performed by using at least one of a Scale-invariant feature transform (SIFT), a Histogram of Oriented Gradient (HOG), and a Haar classifier.

5) Using Convolutional Neural Network (CNN) for feature extraction, CNN network has many specific implementation methods, such as AlexNet, VGG, ResNet, etc. In actual implementation, it can be used for general data training such as ImageNet. The result of the last convolutional layer of the model is characteristic of the CNN model.

Next, the candidate promotion object selected by the second-level topic is subjected to secondary screening to obtain a target promotion object that matches the promotion article. In some embodiments, the candidate promotion object collection may be selected by the following manner. The target promotion object matching the promotion article: extracting the feature of the candidate promotion object, the extracted feature includes at least one type of the image feature and the text feature; calculating the similarity between the extracted feature and the corresponding type feature of the promotion article; When the similarity threshold of the corresponding type feature is exceeded, the target promotion object matching the promotion article is determined. In this way, the adaptation of the target promotion object is automatically realized, so that the promotion information (advertisement) in the promotion article has a high degree of fit with the article content, and does not create a process for the user to read the article. Interference has improved the user's reading experience.

The following describes the target promotion object obtained by pre-training to filter out the target promotion object that matches the promotion article from the candidate promotion object set.

In an example, for the case that the article to be published and the candidate promotion object both have the text material, FIG. 5B is a schematic diagram of the similarity calculation using the text-to-text similarity classifier according to the embodiment of the present invention. Referring to FIG. 5B, the candidate is extracted. Promote the text feature of the object and promote the text feature of the article, and input the corresponding text-text similarity classifier. When the text-text similarity exceeds the text-text similarity threshold, determine that the candidate promotion object matches the promotion article. Target promotion object.

In another example, the case where the article to be published and the candidate promotion object both have the image material, FIG. 5C is a schematic diagram of the similarity calculation using the image-image similarity classifier according to the embodiment of the present invention, and FIG. 5C, the extraction promotion The image feature of the object and the image feature of the article are extended, and the corresponding image-image similarity classifier is input. When the image-image similarity exceeds the image-image similarity threshold, the candidate promotion object is determined to be the target matching the promotion article. Promote the object. In some embodiments, the similarity can be calculated from the feature vector of the image, as obtained in the following manner:

1) Euclidean distance: imagine the vector as the point of the N-dimensional space, and the Euclidean distance measures the distance between the point and the point;

2) Cosine similarity: measure the angle between two vectors;

3) Jaccard similarity: Think of two vectors as a set, measuring the degree of coincidence between sets.

In some embodiments, the target promotion object matching the promotion article may be selected from the candidate promotion object set by calculating a third-level theme of the candidate promotion object in the candidate promotion object set and the third promotion article. The similarity of the level topic, when the third level topic similarity threshold is exceeded, is determined as the target promotion object that matches the promotion article.

In some embodiments, the screening from the set of candidate promotion objects can be implemented in the following manner. Target promotion object matching the promotion article: extracting the image features of the candidate promotion object and promoting the text feature of the article, determining the similarity between the image feature of the candidate promotion object and the text feature of the promotion article; when the text and image similarity threshold is exceeded, Determine the target promotion object that matches the promotion article.

The following is a description of the calculation of the text feature similarity of the image features of the candidate promotion object and the promotion article by using the text-image similarity classifier obtained by the pre-training;

FIG. 5D is a schematic diagram of performing similarity calculation by using a text-image similarity classifier according to an embodiment of the present invention. Referring to FIG. 5D, extracting image features of the promotion object and text features of the promotion article, and inputting corresponding text-image similarity classification. When the text-image similarity exceeds the text-image similarity threshold, it is determined that the promotion object is a promotion object that matches the promotion article.

In some embodiments, the target promotion object matching the promotion article may be determined by: calculating a similarity between the obtained promotion object to be promoted and at least one of the image feature and the text feature of the promotion article; determining that the similarity satisfies the corresponding The promotion object of the similarity condition of the type feature is the target promotion object matching the promotion article. As an implementation manner, for example, performing feature extraction operations on at least one of the following types of material of the candidate promotion object and the promotion article: extracting image features composed of colors, textures, and shapes; performing word segmentation processing, filtering and filtering the word segmentation results Using the word, the text feature composed of the feature word is obtained; determining the similarity between the candidate promotion object and the promotion article regarding the at least one type feature: the candidate promotion object satisfying the similarity condition of the corresponding type feature is determined as the target promotion object. For example, calculating the similarity between the image features extracted by the picture material of the candidate promotion object (pre-stored in the media platform) and the image features of the image in the promotion article, if the image similarity condition is satisfied (greater than the image feature similarity threshold) , the candidate promotion object is the target promotion object; for example, calculating the text feature of the candidate promotion object text material (pre-stored in the media platform, such as classification information, advertisement words), and promoting the text feature of the text in the article Similarity, if the picture similarity condition is satisfied (greater than the text feature similarity threshold), the candidate promotion object is the target promotion object. This implementation eliminates the need to The process of determining the set of candidate promotion objects directly determines the target promotion object based on the image features and/or text features of the promotion article.

In some embodiments, the target promotion object matching the promotion article may also be determined by determining the similarity between the image feature of the candidate promotion object and the text feature of the promotion article; when the determined similarity exceeds the text and image similarity threshold When the candidate promotion object is determined as the target promotion object. It should be noted that, because the feature vector refers to the feature vector in the embodiment of the present invention, the similarity between the candidate promotion object and the different types of features of the promotion article can be calculated, and then the threshold comparison is performed to determine the target promotion object. However, the image features of the candidate promotion object are used here, and the similarity calculation is performed using the text features of the article because: for all candidate promotion objects, there will be corresponding image material in the self-media platform, and all the articles are Including text, it can guarantee that the similarity of the two can always be calculated; avoiding the problem that the similarity cannot be calculated by using the same type of features because the text material of the candidate promotion object is missing from the media platform and the image material is missing in the text.

It should be noted that, in the embodiment of the present invention, there is no dependency relationship between step 302 and step 303, and the execution order is interchangeable.

Next, step 304 is performed: the social network server determines the material that matches the target promotion object, and forms promotion information including the material.

In some embodiments, the material that matches the target promotion object can be determined as follows:

Extracting the character keyword from the promotion article; combining at least one of the character keyword and the tag keyword of the target promotion object with the template content of the promotion object to form a first text material that matches the promotion object.

In actual implementation, the character keyword can be the title of the author or other person appearing in the article, such as: American friends, star balls, etc.; and the way to extract the character keywords can be extracted based on the semantic analysis method.

The tag keyword is a keyword used to identify the features, functions, etc. of the promotion object, and each promotion The object has a corresponding label keyword for identifying the feature, function, and the like of the promotion object. For example, if the promotion object is a certain mask, the label keyword may be: moisturizing and hydrating.

In an embodiment of the present invention, a template for generating a text material (which may be a unified template or a template for classification of promotion objects for different topics) is preset for the promotion object, and a fixed text description is set in the template. And the blank text position to be supplemented, when the character keyword and/or the tag key of the promotion object are substituted into the template, the text material corresponding to the promotion object is formed.

An example of the implementation of the text material corresponding to the promotion object after the label keyword of the promotion object is substituted into the template is as follows: FIG. 6A is an optional schematic diagram of the text material provided by the embodiment of the present invention; The tag keyword is popular, and after entering it into the template, the text material generated by the text template + dynamic text (ie, the tag keyword) is: This is also the popular recommendation.

An example of the implementation of the text material corresponding to the promotion object after the character keyword and the tag keyword of the promotion object are substituted into the template: FIG. 6B is an optional schematic diagram of the text material provided by the embodiment of the present invention; The character of the promotion object is the big cake, the label keyword is popular, and the text material generated by the text template + dynamic text (ie, the label keyword and the character keyword) is substituted into the template: this is also a big cake. Popular items recommended by the main focus.

For example, when the promotion object is a mask, the extracted character keyword is a star ball, and the label keyword of the promotion object is moisturizing or hydrating, the template is combined with the mask template to form a corresponding text material: star ball A highly recommended mask that moisturizes and replenishes water.

In some embodiments, the social network server can determine the material that matches the target promotion object by:

Image recognition is performed on the target promotion object, and the image recognition result representing the attribute of the target promotion object is obtained; the image recognition result is combined with the description information of the target promotion object to form a second text material that matches the target promotion object. In some embodiments, the material matching the target promotion object may include at least one of the first text material and the second text material.

The image recognition result represents the attributes of the target promotion object: such as the name (what is the promotion object, such as clothes and shoes), color, style, etc.;

The description information of the target promotion object may be information that is presented in a keyword form, and identifies related content of the target promotion object from different dimensions, such as a price description and a source of the target promotion object; and the description information of the target promotion object often includes A hyperlink that enables the user to interact with the target promotion object, so that when the user clicks on the description information, the page jumps to the corresponding page.

In actual implementation, the target promotion information includes image material in addition to the formed text material, and the image material can be obtained by: image feature of the image material of the candidate promotion object and image features of the promotion article When the matching condition of the image feature is satisfied (for example, the similarity exceeds the preset threshold), the image material that matches the target promotion object is determined.

In some embodiments, the image material can also be obtained by:

The image is directly extracted from the original promotion information of the target promotion object, and is used as an image material matching the target promotion object, and then the extracted image is synthesized as an image material together with the formed text material and the promotion article. It should be noted that the image material included in the promotion information may be one or more images, and the image may be a picture corresponding to the target promotion object, or may be another picture associated with the picture corresponding to the target promotion object. As shown in FIG. 7A and FIG. 7B, FIG. 7A and FIG. 7B are schematic diagrams showing the addition of promotion information to the end position of a picture in a promotion article according to an embodiment of the present invention. In FIG. 7A and FIG. 7B, block 72 is a picture in the article. The promotion information added at the end position, wherein the block 71 is a text module for carrying the text material generated by the promotion object-based template included in the promotion information, and the block 73 is a picture module for carrying the image material included in the promotion information.

In some embodiments, the text material portion included in the promotion information includes the above description information in addition to the text material generated based on the template of the promotion object. As shown in FIG. 7C, FIG. 7C is a schematic diagram of adding promotion information at the end position of the promotion article according to an embodiment of the present invention. In FIG. 7C, the block 70 corresponds to the promotion information, and the block 77 is a text module, and is used to carry the promotion information. of The text material generated based on the template of the promotion object, the block 78 is a picture module, and is used to carry the image material included in the promotion information, and the block 79 is a description information module, and is used to carry the description information of the target promotion object included in the promotion information (such as the promotion object). Details and sources).

In some embodiments, the material matched by the target promotion object may be composed of image material and description information of the target promotion object. As shown in FIG. 8, FIG. 8 is a schematic diagram of the promotion information provided by the embodiment of the present invention. In FIG. 8, the block 81 is a picture module, and is used to carry the image material included in the promotion information, and the block 82 is a description information module, and is used for Carrying description information of the target promotion object included in the promotion information.

In some embodiments, generating promotional information based on the obtained material may be accomplished by obtaining fixed content for first (time or location first) presentation in the promotional information, the fixed content being used to guide viewing additions After the promotion information; the fixed content obtained, and the obtained material are filled into the promotion information template to obtain the promotion information.

After the social network server forms the promotion information based on the determined text material, step 305 is performed: adding the promotion information to the promotion location according to the determined promotion location. In this way, the synthetic promotion article and the promotion information including the material are synthesized, and the synthesized article is obtained.

In some embodiments, the social network server may also set the display manner of the promotion information. Referring to FIG. 9A to FIG. 9C, FIG. 9A to FIG. 9C are schematic diagrams showing the display manner of the promotion information provided by the embodiment of the present invention, for example, setting the text module correspondingly. The content is hidden, that is, the display manner of the text material generated by the template based on the promotion object in the promotion information is hidden, as shown in FIG. 9A, when the user clicks on the text module part in the picture, the hidden text material can be displayed; or, for example, As shown in FIG. 9B, the text module of the promotion information is displayed to display the fixed content (the promotion information after the viewing is added); or, as shown in FIG. 9C, the text module of the promotion information is dynamically displayed, and the text generated based on the template of the promotion object is dynamically displayed. The content of the material, such as scrolling through the content of the text material.

At this point, the social network server performs the process of adding the promotion information to the promotion article, and then performs step 306: the social network server sends the added promotion article to the self-media level. The second client in Taichung.

In some embodiments, the social network server may obtain the user's article preference based on the self-media account that the user logs in (eg, the article category that is preferred by the user based on the user's article reading record, and the largest number of users that can be read by the user. As a user-favored article category, the article actively pushes a promotion article with promotion information to the client of the user's self-media platform based on the user's preference to the client from the media platform for presentation.

In another embodiment, the social network server may send the promotion article with the promotion information to the client of the self-media platform for presentation based on the read request sent by the user terminal (ie, terminal pull).

After the second client in the media platform receives the added promotion article, step 307 is performed to display the promotion article. In this way, the user can understand the added promotion information while seeing the content of the article that he is concerned about, and the user's reading experience is enhanced due to the natural transition of the promotion information and the article.

The first client and the second client in the self-media platform are described, and the publisher of the article may also be the reader of the article (ie, the first client and the second client are the same client), and the publisher of the article The same user is the reader of the article. In this scenario, after the first client submits the article to be published to the social network server, if the submitted article to be published is determined to be the promotion article and there is a target promotion object matching the same When the first client obtains the article submitted by itself, it also receives the promotion information added with the material matching the target promotion object, and then presents the original content of the article, and displays the promotion location according to the promotion information in the article. Promotional information is presented when the original content of the article is promoted to the appropriate location.

In another scenario, the publisher of the article is not the same user as the reader of the article (ie, the first client and the second client are different clients), and at this time, the second client can pull according to the user's access request. A promotion article published in a social network, or a promotion article pushed by a social network server according to a social (concern/subscription) relationship with a publishing user, and then presenting the original article Start content, and according to the promotion location of the promotion information in the article, when the original content of the promotion article is displayed to the corresponding location, the promotion information is presented.

In some embodiments, based on the setting of the display manner of the promotion information, when the original content of the promotion article is displayed to the corresponding location, the promotion information may be presented in one of the following ways:

1) presenting the image material in the promotion information, and presenting the text material when the text material in the promotion information is triggered (such as receiving the user's click operation); for example, starting to display only the image material, the text material is hidden and invisible. Rendering text material when the user clicks on the location of the text material;

2) in response to the operation of fixing the fixed content presented by the promotion location of the promotion information (such as a click operation), presenting the text material and the image material in the promotion information; the fixed content is used to guide the viewing of the promotion information; that is, the presentation is started. Fixed content for guiding the user to view the promotion information, and displaying the image material and the text material when the user triggers;

It can be seen that the above two methods only display the text material and the image material in the promotion information only when the user triggers, which reduces the interference to the user to read the article to a certain extent;

3) When the original content of the promotion article is displayed to the corresponding location, the text material and the image material in the promotion information are directly presented.

Applying the above embodiments, the following beneficial effects are obtained:

1) Using the self-media platform to spontaneously report the traffic of the article as the carrier of the promotion information, that is, the source of the article can come from any user terminal in the social network, breaking the limitation of the article collecting the specific topic, and realizing the batch of the promotion information. Adding and automating the addition, with the characteristics of frequent traffic with the self-media platform, the promotion information can be reached in time to reach the user;

2) Automatically realize the promotion object and the adaptation of the corresponding text material, so that the promotion information (advertisement) in the promotion article has a high degree of fit with the article content, and does not interfere with the process of the user reading the article, thereby improving the user's Reading experience

3) Automatically realize the location of the promotion information, the location is flexible, and can avoid the promotion of information. The abruptness makes the content of the article and the content of the promotion information natural, and is easy to accept in the process of reading the article.

As an alternative embodiment of the above-described self-media platform-based article processing method, FIG. 10 is a schematic flowchart of an optional article processing method based on the self-media platform provided by the server side in the embodiment of the present invention. For example, the promotion information is used as an advertisement, and the promotion object is an advertisement object (advertisement product). For example, referring to FIG. 10, the article processing method based on the self-media platform provided by the embodiment of the present invention includes:

Step 401: The server performs semantic analysis on the article.

The article mentioned here is the self-media article to be published by the server or the article that has been published on the self-media platform but has been withdrawn. The semantics of the article is used to understand the title of the article, from the media name (such as the public name). ), the author's name, understand the text of the entire article. Select the topic that belongs to this article as the basis for matching the advertising object. In turn, you can filter out articles that match the ad object.

The feature information that the advertisement object can match includes: a category of the advertisement object, an advertisement word, a name of the advertisement object, and the like. The feature information that the article can match includes keywords and the like.

Semantic analysis, that is, semantic understanding, refers to the transformation of unstructured or semi-structured natural language text into structured information that can be deeply processed by computers, and classified, analyzed, and so on.

Step 402: Identify the subject of the article.

In actual implementation, it can be obtained by training the keyword-topic model, extracting the keywords of the article through semantic analysis, and then inputting the trained keyword-topic model to get the topic of the article.

Step 403: Determine whether the article matches the advertisement object based on the article theme. If the matching performs step 404, if there is no match, the article does not advertise.

In actual implementation, the server may match the topic of the article with at least one of the name, category, and advertisement word corresponding to the advertisement object, and determine that the article satisfying the matching condition can be matched widely. The article of the object.

If it is determined by the above matching that the article is not suitable for adding an advertisement, that is, there is no matching advertisement, no advertisement is added to the article, and the article is directly posted to the social network.

Step 404: Perform semantic analysis of the segmentation of the article.

A semantic analysis of the segmentation of the article yields whether the article has multiple (two or more) topics.

Step 405: Identify whether the article has multiple topics.

Step 406: Determine that there are multiple topics in the article, and perform step 408.

Step 407: Determine that the article has a single topic, and perform step 409.

Step 408: Mark multiple locations where the advertisement is added in the article.

In actual implementation, when it is determined that there are multiple topics in the article, it is optional to add an advertisement at the location of the corresponding topic, such as at the junction of two adjacent topics.

Step 409: Add the location of the advertisement at the end of the article.

Step 410: Select a set of alternative advertisement object sets from the advertisement object library according to the article theme.

The set of alternative advertising objects (eg, determining that the similarity reaches a preset threshold) can be determined by calculating the similarity of the subject of the advertising object to the topic of the article.

Step 411: Match the advertisement object according to the text content of the article.

In actual implementation, the pre-trained text-image similarity classifier can be used to input the text feature of the article and the image feature of the advertisement object to obtain the similarity between the two, when only the position where the advertisement is added is marked at the end of the article. You only need to match the most similar ad object. When you mark multiple ads in the article, you can select the corresponding number of ad objects according to the similarity.

Step 412: Perform image recognition on the matched advertising object, and obtain description information of the advertising object.

The advertising object may only have image information, and the image recognition of the advertising object can be obtained. Corresponding to the material information of the advertisement object, such as the specific content of the advertisement object, such as clothes, shoes, and the like. The description information of the advertisement object includes the source, price, description details, and the like of the advertisement object.

Step 413: Add text according to the image recognition result and the description information of the advertisement object.

In actual implementation, the above-mentioned added text is the text material of the advertisement for adding the article.

Step 414: The description information of the extracted advertisement object is displayed on the advertisement image.

In actual implementation, the description information of the corresponding advertisement object includes a corresponding hyperlink, and when the user clicks, jumps to the corresponding page, such as jumping to the purchase page of the advertisement object.

Step 415: Add the added text and the interactive advertising object as an advertisement to the article.

The interactive means refers to a hyperlink included in the advertisement, and the user clicks to perform a page jump.

In the foregoing embodiment of the article processing method based on the self-media platform, the server carrying the self-media platform first determines the promotion article for presenting the promotion information, and determines the target promotion object based on the determined promotion article, but in actual application, The target promotion object can be determined first, and then the promotion article for presenting the promotion information can be determined. Next, the article processing method based on the self-media platform is described in detail.

As an alternative embodiment of the article processing method based on the self-media platform, FIG. 11 is a schematic flowchart of an optional article processing method based on the self-media platform provided by the server side in the embodiment of the present invention. The self-media platform can be carried on a social network server with social functions, and the promotion information is an advertisement, and the promotion object is an advertisement object (advertisement product). For example, referring to FIG. 11 , the self-media provided by the embodiment of the present invention is provided. The article's article processing methods include:

Step 501: The first client sends the target article to the self-media platform.

Here, the client connects to the self-media platform, and the target article is submitted by the user of the self-media platform through the client. In practical applications, the target article includes the article to be published and the original article.

Step 502: Determine a target promotion object and a target promotion object in the candidate promotion object. Matching material.

In an embodiment, the self-media platform may determine the target promotion object among the candidate promotion objects stored in the self-media platform by:

Performing content similarity calculation on the content feature of the historical target article and the content feature of the candidate promotion object, and determining the candidate promotion object satisfying the content similarity condition as the target promotion object; wherein the historical target article precedes the target The article is received and sent on the self-media platform.

For example, determining a candidate promotion object that satisfies a topic similarity condition with a topic of the historical target article, forming a candidate promotion object set, and determining a similarity between each candidate promotion object in the candidate promotion object set and the historical target article regarding at least one type feature, The feature includes an image feature and a text feature; the candidate promotion object whose similarity satisfies the similarity condition of the corresponding type feature is determined as the target promotion object.

Step 503: Determine, according to the determined target promotion object, a promotion article for presenting the promotion information in the received target article.

In an embodiment, based on the determined target promotion object, the theme feature of the target promotion object and the theme feature of the target article are subjected to topic similarity calculation, and the target article satisfying the topic similarity condition is determined as Describe the promotion article. For example, the keyword extracted from the target article is input into a classifier model that classifies the topic according to the feature word, and the topic corresponding to the target article calculated by the classifier model is obtained; the keyword extracted from the material of the candidate promotion object is input according to the keyword. The classifier model of the feature word is used to obtain the topic corresponding to the candidate promotion object; according to the semantic distance of the topic corresponding to the target article and the topic corresponding to the candidate promotion object, the topic similarity with the negative relationship of the semantic distance is determined, and the received feature is received. The target article in the target article that satisfies the topic similarity condition is determined as a promotion article.

Step 504: Determine a promotion location for adding promotion information in the promotion article.

In an embodiment, the promotion location for adding promotional information in the promotion article may be determined as follows:

Determining a paragraph having the included topic feature in the promotion article according to the topic feature included in the promotion article; determining the location of the paragraph for adding when the included topic feature and the topic feature of the promotion information satisfy the topic similarity condition The promotion location of the promotion information.

Step 505: Generate promotion information according to the determined material that matches the target promotion object.

In an embodiment, the promotion information can be generated by:

The fixed content for first presentation in the promotion information is obtained, and the fixed content is used to guide the viewing of the added promotion information; and the obtained fixed content and the obtained material are filled into the promotion information template to obtain the promotion information.

Step 506: Add the promotion information to the corresponding promotion location in the promotion article according to the determined promotion location.

Step 507: Send a promotion article to which the promotion information is added to the second client.

Step 508: The second client displays the promotion article.

The embodiment of the present invention further provides an article processing apparatus 300 based on a self-media platform. Referring to FIG. 12, FIG. 12 is a schematic structural diagram of an article processing apparatus based on a self-media platform according to an embodiment of the present invention, including:

The receiving unit 31 is configured to receive a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the first client;

The determining unit 32 is configured to determine, in the target article, a promotion article for presenting the promotion information, and a promotion location for adding the promotion information in the promotion article;

The generating unit 33 is configured to generate promotion information according to the determined material that matches the target promotion object;

The adding unit 34 is configured to add the promotion information according to the determined promotion location Add to the corresponding promotion location in the promotion article;

The sending unit 35 is configured to send the promotion article to which the promotion information is added.

In some embodiments, the determining unit 32 is further configured to perform topic similarity calculation on the topic feature of the target article and the topic feature of the candidate promotion object, and determine the target article that satisfies the topic similarity condition as the promotion. article;

And configuring, according to the determined promotion article, performing content similarity calculation on the content feature of the promotion article and the content feature of the candidate promotion object, and determining the candidate promotion object satisfying the content similarity condition as the target promotion object.

In some embodiments, the determining unit 32 is further configured to perform content similarity calculation on the content feature of the historical target article and the content feature of the candidate promotion object, and determine the candidate promotion object that satisfies the content similarity condition as the target. Promotion target;

The historical target article is received and sent by the self-media platform prior to the target article;

And configuring, according to the determined target promotion object, performing theme similarity calculation on the theme feature of the target promotion object and the theme feature of the target article, and determining a target article satisfying the topic similarity condition as the promotion article .

In some embodiments, the determining unit 32 is further configured to determine, in the promotion article, a paragraph having the included topic feature according to the topic feature included in the promotion article;

When the included topic feature and the topic feature of the promotion information satisfy the topic similarity condition, determining the location of the paragraph is a promotion location for adding the promotion information.

In some embodiments, the determining unit 32 is further configured to: when the promotion information is added at a position between adjacent paragraphs in the promotion article, according to whether the same type of content in the content style of the promotion article is Segmented by the promotion information, and/or a display ratio occupied by the promotion information in the content style, determining a corresponding completeness;

When the completeness satisfies the integrity condition, determining the position between the adjacent paragraphs is Add the promotion location of the promotion information.

In some embodiments, the determining unit 32 is further configured to input a keyword extracted from the to-be-published article into a classifier model that classifies topics according to feature words, and obtain the calculated output of the classifier model. The topic corresponding to the article to be published;

Inputting a keyword extracted from the material of the candidate promotion object into a classifier model that performs topic classification according to the feature word, and obtaining a topic corresponding to the candidate promotion object;

And determining a topic similarity that is negatively related to the semantic distance according to a semantic distance of a topic corresponding to the to-be-published article and a topic corresponding to the candidate promotion object.

In some embodiments, the determining unit 32 is further configured to perform a feature extraction operation on at least one of the following types of material of the candidate promotion object and the promotion article: extracting image features composed of colors, textures, and shapes Perform word segmentation processing, filter out the stop words on the word segmentation results, and obtain text features composed of feature words;

Determining the similarity between the candidate promotion object and the promotion article regarding at least one type of feature:

The candidate promotion object that satisfies the similarity condition of the corresponding type feature is determined as the target promotion object.

In some embodiments, the determining unit 32 is further configured to: determine a candidate promotion object that satisfies a topic similarity condition with a topic of the promotion article, and form a candidate promotion object set;

Determining a similarity between each candidate promotion object in the candidate promotion object set and the promotion article with respect to at least one type of feature, the image includes an image feature and a text feature;

The candidate promotion object whose similarity satisfies the similarity condition of the corresponding type feature is determined as the target promotion object.

In some embodiments, the determining unit 32 is further configured to input at least one type of feature of the candidate promotion object into a classifier model that performs topic classification according to the feature word, and obtain the candidate of the classifier model calculation output. Promote the subject to which the object belongs;

When the similarity between the mapped topic and the topic of the promotion article exceeds the topic similarity threshold, the candidate promotion object that satisfies the topic similarity condition with the topic of the promotion article is determined.

In some embodiments, the determining unit 32 is further configured to determine a similarity between an image feature of the candidate promotion object and a text feature of the promotion article;

When the determined similarity exceeds the text and image similarity threshold, it is determined that the candidate promotion object is the target promotion object.

In some embodiments, the determining unit 32 is further configured to extract a character keyword from the promotion article;

And combining at least one of the character keyword and the tag keyword of the target promotion object with the template content of the target promotion object to form a character material corresponding to the target promotion object.

In some embodiments, the determining unit 32 is further configured to perform image recognition on the target promotion object, and obtain an image recognition result that is used to represent the promotion object attribute;

Combining the image recognition result with the description information of the target promotion object to form a text material corresponding to the target promotion object.

In some embodiments, the determining unit 32 is further configured to: when the image feature of the image material of the candidate promotion object and the image feature of the promotion article satisfy the matching condition of the image feature, the image that satisfies the matching condition is The material is an image material corresponding to the target promotion object.

In some embodiments, the generating unit 33 is further configured to obtain fixed content for first presentation in the promotion information, where the fixed content is used to guide viewing the added promotion information;

The fixed content and the obtained material are filled into a promotion information template to obtain the promotion information.

The embodiment of the invention further provides a server, including:

a memory configured to store an executable program;

The embodiment of the present invention further provides a readable storage medium, which may include: a mobile storage device, a random access memory (RAM), a read-only memory (ROM), a magnetic disk, or A variety of media such as optical discs that can store program code. The readable storage medium stores an executable program;

The executable program is configured to implement the above-described self-media platform-based article processing method when executed by a processor.

The above is only a specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any person skilled in the art can easily think of changes or substitutions within the technical scope of the present invention. It should be covered by the scope of the present invention. Therefore, the scope of the invention should be determined by the scope of the appended claims.

Industrial applicability

The embodiment of the present invention receives a target article sent by a client, where the client is used to connect to the self-media platform, and the target article is submitted by a user of the self-media platform through the client; Determining, in the article, a promotion article for presenting the promotion information, and a promotion location for adding the promotion information in the promotion article; determining a target promotion object and the target in the candidate promotion object stored in the self-media platform Promoting the matching material of the object; generating the promotion information according to the determined material matching the target promotion object; adding the promotion information to the corresponding promotion location in the promotion article according to the determined promotion location; sending The promotion article with the promotion information is added. In this way, the source of the target article can come from any user terminal in the social network, breaking the limitation of the article collecting the specific topic, and realizing the batching and automatic addition of the promotion information; automatically selecting the location of the promotion information, the location Flexible, able to avoid the abrupt appearance of promotional information, making the content of the article and the promotion information The content is connected naturally; through the release of the article and the process of reaching the user, the delivery of the promotion information is completed, and the promotion information is realized by relying on the release/send traffic of the self-media platform itself, and the promotion information can cover the access traffic from the media platform and reach the user in real time.

Claims

An article processing method based on a self-media platform, comprising:

Receiving a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the client;

Determining, in the target article, a promotion article for presenting promotion information, and a promotion location for adding promotion information in the promotion article;

Determining, in a candidate promotion object stored in the self-media platform, a target promotion object and a material matching the target promotion object;

Generating promotion information according to the determined material that matches the target promotion object;

And adding the promotion information to a corresponding promotion location in the promotion article according to the determined promotion location;

The promotion article to which the promotion information is added is sent.
The method of claim 1 wherein said determining a promotional article for presenting promotional information in said target article comprises:

Performing topic similarity calculation on the topic feature of the target article and the topic feature of the candidate promotion object, and determining the target article satisfying the topic similarity condition as the promotion article;

Determining the target promotion object in the candidate promotion objects stored in the self-media platform, including:

Based on the determined promotion article, the content feature of the promotion article and the content feature of the candidate promotion object are subjected to content similarity calculation, and the candidate promotion object satisfying the content similarity condition is determined as the target promotion object.
The method of claim 1, wherein the determining the target promotion object among the candidate promotion objects stored in the self-media platform comprises:

Performing content similarity calculation on the content feature of the historical target article and the content feature of the candidate promotion object, and determining the candidate promotion object satisfying the content similarity condition as the target promotion object;

The historical target article is received and sent by the self-media platform prior to the target article;

Determining, in the target article, a promotion article for presenting promotion information, including:

Based on the determined target promotion object, the theme feature of the target promotion object and the theme feature of the target article are subjected to topic similarity calculation, and the target article satisfying the topic similarity condition is determined as the promotion article.
The method according to claim 1 or 2, wherein said determining a promotion location for adding promotion information in said promotion article comprises:

Determining a paragraph having the included topic feature in the promotion article according to the topic feature included in the promotion article;

When the included topic feature and the topic feature of the promotion information satisfy the topic similarity condition, determining the location of the paragraph is a promotion location for adding the promotion information.
The method according to claim 1 or 2, wherein said determining a promotion location for adding promotion information in said promotion article comprises:

When the promotion information is added at a position between adjacent paragraphs in the promotion article,

Whether the content of the same type in the content style of the promotion article is segmented by the promotion information, and/or the display ratio occupied by the promotion information in the content style, determining a corresponding completeness;

When the degree of completeness satisfies the integrity condition, determining a position between the adjacent paragraphs is a promotion position to which the promotion information is added.
The method according to claim 2, wherein the content feature of the determined promotion article and the content feature of the candidate promotion object are subjected to content similarity calculation, and the candidate promotion object satisfying the content similarity condition is determined as Target promotion targets, including:

Determining a candidate promotion object that satisfies a topic similarity condition with a topic of the promotion article, and forming a candidate promotion object set;

Determining a similarity between each candidate promotion object in the candidate promotion object set and the promotion article with respect to at least one type of feature, the image includes an image feature and a text feature;

The candidate promotion object whose similarity satisfies the similarity condition of the corresponding type feature is determined as the target promotion object.
The method of claim 6, wherein the determining the set of candidate promotion objects that satisfy the topic similarity condition with the topic of the promotion article comprises:

Entering at least one type of feature of the candidate promotion object into a classifier model for classifying the topic according to the feature word, and obtaining a topic to which the candidate promotion object is calculated and output by the classifier model;

When the similarity between the mapped topic and the topic of the promotion article exceeds the topic similarity threshold, the candidate promotion object that satisfies the topic similarity condition with the topic of the promotion article is determined.
The method of claim 1, wherein the determining the material that matches the target promotion object comprises:

Performing image recognition on the target promotion object, and obtaining an image recognition result characterizing the attribute of the promotion object;

Combining the image recognition result with the description information of the target promotion object to form a text material corresponding to the target promotion object.
The method of claim 1, wherein the determining the material that matches the target promotion object comprises:

When the image feature of the image material of the candidate promotion object and the image feature of the promotion article satisfy the matching condition of the image feature, the image material that satisfies the matching condition is used as the image material corresponding to the target promotion object.
The method of claim 1, wherein the generating the promotion information according to the determined material that matches the target promotion object comprises:

Obtaining fixed content for first presentation in the promotion information, the fixed content being used for Guiding to view the added promotional information;

The fixed content and the obtained material are filled into a promotion information template to obtain the promotion information.
An article processing method based on a self-media platform, the method being performed by a server, the server comprising one or more processors and a memory, and one or more programs, wherein the one or more programs Stored in a memory, the program can include one or more units each corresponding to a set of instructions, the one or more processors being configured to execute instructions; the method comprising:

Receiving a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the client;

Determining, in the target article, a promotion article for presenting promotion information, and a promotion location for adding promotion information in the promotion article;

Determining, in a candidate promotion object stored in the self-media platform, a target promotion object and a material matching the target promotion object;

Generating promotion information according to the determined material that matches the target promotion object;

And adding the promotion information to a corresponding promotion location in the promotion article according to the determined promotion location;

The promotion article to which the promotion information is added is sent.
The method of claim 11 wherein said determining a promotional article for presenting promotional information in said target article comprises:

A topic similarity calculation is performed on the topic feature of the target article and the topic feature of the candidate promotion object, and the target article satisfying the topic similarity condition is determined as a promotion article.
The method of claim 12, wherein the subject feature of the target article and the topic feature of the candidate promotion object are subjected to topic similarity calculation, including:

Subdividing the keywords extracted from the target article into sub-categories based on feature words a classifier model, obtaining a theme corresponding to the target article output by the classifier model;

Inputting a keyword extracted from the material of the candidate promotion object into a classifier model that performs topic classification according to the feature word, and obtaining a topic corresponding to the candidate promotion object;

And determining a topic similarity that is negatively related to the semantic distance according to a semantic distance of a topic corresponding to the target article and a topic corresponding to the candidate promotion object.
The method of claim 11, wherein the determining the target promotion object among the candidate promotion objects stored in the self-media platform comprises:

And determining the content similarity of the determined content feature of the promotion article and the content feature of the candidate promotion object, and determining the candidate promotion object that satisfies the content similarity condition as the target promotion object.
The method according to claim 14, wherein the content feature of the determined promotion article and the content feature of the candidate promotion object are subjected to content similarity calculation, and the promotion object satisfying the content similarity condition is determined as a target. Promotion targets, including:

Performing a feature extraction operation on at least one of the following types of material of the candidate promotion object and the promotion article: extracting image features composed of colors, textures, and shapes; performing word segmentation processing, filtering out stop words for word segmentation results, and obtaining a text feature consisting of feature words;

Determining the similarity between the candidate promotion object and the promotion article regarding at least one type of feature:

The candidate promotion object that satisfies the similarity condition of the corresponding type feature is determined as the target promotion object.
The method according to claim 14, wherein the content feature of the determined promotion article and the content feature of the candidate promotion object are subjected to content similarity calculation, and the candidate promotion object satisfying the content similarity condition is determined as Target promotion targets, including:

Determining a similarity between an image feature of the candidate promotion object and a text feature of the promotion article;

Determining the candidate push when the determined similarity exceeds a text and image similarity threshold A wide object is a target promotion object.
The method of claim 11 wherein said determining a material that matches said target promotion object comprises:

Extracting a character keyword from the promotion article;

And combining at least one of the character keyword and the tag keyword of the target promotion object with the template content of the target promotion object to form a character material corresponding to the target promotion object.
An article processing device based on a self-media platform, comprising:

a receiving unit, configured to receive a target article sent by the client, where the client is used to connect to the self-media platform, and the target article is submitted by the user of the self-media platform through the client;

a determining unit configured to determine, in the target article, a promotion article for presenting the promotion information, and a promotion location for adding the promotion information in the promotion article;

And configured to determine a target promotion object and a material matching the target promotion object among the candidate promotion objects stored in the self-media platform;

a generating unit, configured to generate promotion information according to the determined material that matches the target promotion object;

Adding a unit, configured to add the promotion information to a corresponding promotion location in the promotion article according to the determined promotion location;

The sending unit is configured to send the promotion article to which the promotion information is added.
A server that includes:

a memory configured to store an executable program;

The self-media platform-based article processing method according to any one of claims 1 to 10, when the processor is configured to execute the executable program stored in the memory.
A storage medium storing an executable program, the executable program being executed by a processor In the line, the self-media platform-based article processing method according to any one of claims 1 to 10 is implemented.
A storage medium storing an executable program, the executable program being executed by a processor, implementing the self-media platform-based article processing method according to any one of claims 11 to 17.