Social Media Promotion Method, Device and Storage Medium Based on Social Action Injection

Abstract

The present invention relates to the fields of machine learning and social media, and discloses a social media promotion method, device, and storage medium based on social action injection. The method includes: registering multiple social media accounts, programming and developing an application program for executing social media actions and binding it to these accounts, and executing the program to generate social robots; using a web crawler or a public API provided by the social media to obtain some user data including the basic information of users and some blog posts, and extracting user-level features x and blog post-level features x from these data t , obtaining a subset of the dataset, using the obtained small amount of data to train a simulator f(·) to learn the interactions in social media; based on the trained simulator f(·), calculating an influence vector, and using the social robots to forward blog posts and modify the basic information to implement promotion. The method is simple to operate, easy to implement, low in cost, and can effectively improve the acceptance degree of target users for specified content.

Description

Technical Field

[0001] The present invention relates to the fields of machine learning and social media. Specifically, it relates to a social media promotion method, device, and storage medium based on social action injection. Background Art

[0002] The rapid development of Internet technology has promoted the rise of social media such as Weibo and Douyin. Data shows that in 2021, the number of daily active users of Weibo was 246 million, and the number of daily active users of Douyin exceeded 600 million. It has become an important platform for people to receive and share information. Due to the characteristics of social media such as extensiveness and openness, people strategically use it for content promotion and have received increasing attention. For example, the health department popularizes the harm of diseases through the platform to arouse people's attention. Given the great practical significance of social media promotion, how to design an efficient promotion strategy is an urgent research problem to be solved.

[0003] Regarding this research problem, marketers have proposed various methods, including advertising placement, dissemination by influential users (such as news media, prestigious celebrities, etc.), and self-media operation. However, these methods still have defects: First, high costs. Advertising expenditures, customized marketing, etc. are all expensive, and the complex social model makes the promotion cost often much greater than the obtained benefits; Second, one-sided audience groups. Promoters formulate accurate user portraits for personalized promotion by analyzing users' behavioral characteristics such as interests and values. However, in many cases, promoting to indifferent users has more direct value; Third, insufficient promotion ability. The main role of promotion methods such as advertising is to expand influence and deepen users' impressions through frequent placements, but how to improve the conversion rate has not been formally analyzed. And this aggressive direct sales is likely to arouse psychological resistance and have the opposite effect.

[0004] Therefore, it is necessary to design an efficient and low-cost promotion strategy to achieve the promotion goal in a subtle way. Summary of the Invention

[0005] The object of the present invention is to provide a social media promotion method based on social action injection, which is simple to operate, easy to implement, low in cost, and can effectively improve the acceptance degree of target users for specified content.

[0006] To achieve the above object, in the first aspect of the present invention, a social media promotion method based on social action injection is provided, and the method includes:

[0007] Let the user set be represented by ; the blog post set be represented by ; the features of the followed, followers, and blog posts be represented by x u , xv and x t denotes; x represents the features of any user; the data set is represented by , and each piece of data in the data set consists of a sample label pair (x u，v，t , y u，v，t ), where x u，v，t = [x u ; x v ; x t represents the concatenation of the features of the followed user u, the follower v, and the blog post t, and y u，v，t represents whether the user u forwards the blog post t of his follower v (y u，v，t = 1) or not (y u，v，t = 0); represents the follow list of the user u, represents the blog post set of the user u;

[0008] Step 1: Register multiple social media accounts, program and develop an application for executing social media actions and bind it to these accounts, and execute the program to generate a social robot;

[0009] Step 2: Use a web crawler or the public API provided by the social media to obtain partial user data including the basic information of the user and some blog posts Extract the user-level features x and the blog post-level features x t from these data to obtain a subset of the data set

[0010] Step 3: Use the small amount of data obtained in Step 2 to train a simulator f(·) to learn the interactions in the social media; the simulator takes the sample x u，v，t as the input, learns whether the user u forwards the blog post t of the user v, and uses to represent the predicted value of the simulator, where are the simulator parameters;

[0011] Step 4: Based on the simulator f(·) trained in Step 3, calculate an influence vector The estimation formula of this influence vector is as follows:

[0012]

[0013] where represents the training loss of the simulator with respect to the sample x fix ; is the training loss of the Hessian matrix; is the promotion loss; Let the set of content to be promoted be represented by and the set of target users of the promoted blog post be represented by Then the promotion loss is defined as follows:

[0014]

[0015] Divide the influence vector calculated in this step into three vectors, namely where have dimensions that are respectively consistent with x u , x v , x t ;

[0016] Step 5: Use the social bots generated in Step 1 to forward blog posts and modify basic profiles to implement promotion.

[0017] Preferably, the use of social bots to forward blog posts in Step 5 for promotion includes:

[0018] Calculate the influence of each social bot the influence of the users obtained in Step 2 the influence of the blog posts obtained in Step 2 where Greedy select n u , n v , n t of the most influential social bots, observable users, and observable blog posts from among them; Obtain the candidate set of forwarding to be executed according to their dependency relationships where (u, v, t) represents that social bot u forwards the blog post t of its follower v; Finally, according to the influence of sample x u,v,t and Greedy select n of the most influential forwards from to input into the application of the step to let the social bots automatically execute.

[0019] Preferably, the use of social bots to modify basic profiles in Step 5 for promotion includes:

[0020] For social bot According to Generate m cand neighbors for social bot i, where M is a vector with the same dimension as x i and only contains 0s and 1s; And, if the user information corresponding to a certain feature of x i is immutable, then the value at the corresponding position in M is 1; If x iIf a certain feature in it is multi-dimensional, then the corresponding positions in M should all be consistent, that is, all be 0 or all be 1; for each neighbor calculate its influence; let represent the data related to the social robot That is Then, x i is modified to The formula for calculating the influence of promotion is as follows:

[0021]

[0022] Select the one with the greatest influence from these m cand neighbors Modify the basic profile information of social robot i to be the same as x′ i ; for each social robot Repeat this step to modify the basic profile.

[0023] Preferably, the social media actions in step 1 include searching, replying, forwarding, and following.

[0024] Preferably, the basic profile of the user in step 2 includes age, gender, and tags, the features of the user level include age, the number of posts, the number of followers, and the number of followed users, and the features of the post level include the post topic obtained by using the LDA algorithm, the number of post words, the number of likes, and the number of forwards.

[0025] Preferably, the simulator in step 3 can be any model in the field of deep learning.

[0026] The second aspect of the present invention provides a device, which includes a processor and a memory; wherein, the memory is used to store a computer program, and the processor is used to execute the method for promoting social media based on social action injection described in the first aspect of the present invention according to the computer program.

[0027] The third aspect of the present invention provides a computer-readable storage medium, which is used to store a computer program, and the computer program is used to execute the method for promoting social media based on social action injection described in the first aspect of the present invention.

[0028] According to the above technical solution, the present invention trains a simulator through part of the collected data, calculates the influence of users and posts on promotion based on the simulator, and controls a small number of social robots to perform limited but highly influential social actions, including forwarding and modifying basic profiles, so that target users can accept the specified promotion content in a subtle way.

[0029] Other features and advantages of the present invention will be described in detail in the subsequent specific implementation part. Detailed implementation manners

[0030] The following provides a detailed description of the specific implementation manners of the present invention. It should be understood that the specific implementation manners described herein are only for explaining and illustrating the present invention, and are not used to limit the present invention.

[0031] First, a first aspect of the present invention provides a social media promotion method based on social action injection. The method includes:

[0032] Let the user set be represented by ; the blog post set be represented by ; the features of the followed user, the follower, and the blog post be represented by x u , x v and x t ; the feature of any user be represented by x; the data set be represented by . Each piece of data in the data set consists of a sample label pair (x u，v，t , y u，v，t ), where x u，v，t = [x u ; x v ; x t represents the concatenation of the features of the followed user u, the follower v, and the blog post t, and y u，v，t represents whether the user u forwards the blog post t of his follower v (y u，v，t = 1) or not (y u，v，t = 0); represents the follow list of the user u, represents the blog post set of the user u;

[0033] Step 1: Register multiple social media accounts, program and develop an application for executing social media actions and bind it to these accounts, and execute the program to generate social robots;

[0034] Step 2: Use a web crawler or a public API provided by the social media to obtain partial user data including the basic information of the user and some blog posts Extract the user-level feature x and the blog post-level feature x t from these data to obtain a subset of the data set

[0035] Step 3: Use the small amount of data obtained in Step 2 to train a simulator f(·) to learn the interactions in the social media; the simulator takes the sample x u，v，t as input, learns whether the user u forwards the blog post t of the user v, and uses to represent the predicted value of the simulator, where is the simulator parameter;

[0036] Step 4. Based on the simulator f(·) trained in Step 3, calculate an influence vector The estimation formula of this influence vector is as follows:

[0037]

[0038] where represents the training loss of the simulator with respect to the sample x fix ; is the Hessian matrix of the training loss ; is the generalization loss; let the set of content to be generalized be represented by , and the set of target users of the blog post to be promoted be represented by , then the generalization loss is defined as follows:

[0039]

[0040] Divide the influence vector calculated in this step into three vectors, namely where have dimensions that are respectively the same as those of x u , x v , and x t ;

[0041] Step 5. Use the social bots generated in Step 1 to forward blog posts and modify basic profiles to implement the promotion.

[0042] Specifically, the use of social bots to forward blog posts in Step 5 for promotion includes:

[0043] Calculate the influence of each social bot the influence of the users obtained in Step 2 the influence of the blog posts obtained in Step 2 where Greedy select n u , n v , and n t of the most influential social bots, observable users, and observable blog posts respectively from them; obtain the candidate set of forwarding to be executed where (u, v, t) means that social bot u forwards the blog post t of its follower v; finally, based on the influence u,v,t of the sample x and then greedily select n from the most influential forwarding inputs to the application program in the step to let the social bots Automatically execute.

[0044] The promotion executed by using a social bot to modify the basic profile in step 5 includes:

[0045] For the social bot According to Generate m cand neighbors for the social bot i, where M is a vector with the same dimension as x i and only contains 0 and 1; and, if the user information corresponding to a certain feature of x i is immutable, then the value at the corresponding position in M is 1; if a certain feature in x i is multi-dimensional, then the values at the corresponding positions in M should all be the same, that is, all 0 or all 1; for each neighbor Calculate its influence; let represent the data related to the social bot i.e., Then, x i is modified to The formula for calculating the influence of the promotion is as follows:

[0046]

[0047] Select the one with the greatest influence from these m cand neighbors Modify the basic profile information of the social bot i to be the same as x′ i ; repeat this step for each social bot to modify the basic profile.

[0048] In the actual operation process, the social media actions in step 1 above may include searching, replying, forwarding, and following, etc.

[0049] The basic profile of the user in step 2 includes age, gender, and tags, etc., and the user-level features include age, the number of posts, the number of followers and the number of followed users, etc. that can be obtained, and the post-level features include the post topic obtained by using the LDA algorithm, the number of post words, the number of likes, and the number of forwards, etc.

[0050] In addition, the simulator in step 3 can be any one of the models in the field of deep learning, which can be a multi-layer perceptron, a graph convolutional neural network, or a deep neural network based on the attention mechanism.

[0051] In addition, a second aspect of the present invention provides a device, which includes a processor and a memory; wherein, the memory is used to store a computer program, and the processor is used to execute the above-mentioned social media promotion method based on social action injection according to the computer program.

[0052] In a third aspect of the present invention, there is provided a computer-readable storage medium for storing a computer program for executing the above-mentioned social media promotion method based on social action injection.

[0053] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various simple modifications can be made to the technical solutions of the present invention, and these simple modifications all fall within the protection scope of the present invention.

[0054] In addition, it should be noted that, in the case of no conflict, the various specific technical features described in the above specific embodiments can be combined in any suitable manner. To avoid unnecessary repetition, the present invention will not separately describe various possible combination manners.

[0055] Furthermore, any combination can be made between various different embodiments of the present invention, as long as it does not violate the idea of the present invention, and it should also be regarded as the content disclosed by the present invention.

Claims

1. A social media promotion method based on social action injection, characterized in that, The method includes: Let the user set be represented by ; the blog post set be represented by T; the features of the followed, follower, and blog post be represented by x u , x v and x t ; let x represent the feature of any user; the data set be represented by ; each piece of data in the data set consists of a sample label pair (x u,v,t , y u,v,t ), where x u，v，t = [x u ; x v ; x t represents the concatenation of the features of the followed user u, the follower v, and the blog post t, and y u,v,t represents whether user u forwards the blog post t of their follower v (y u,v,t = 1) or not (y u,v,t = 0); represents the follow list of user u, represents the blog post set of user u; Step 1: Register multiple social media accounts, program and develop an application for performing social media actions and bind it to these accounts, and execute the program to generate social robots; Step 2: Obtain partial user data by using web crawlers or public APIs provided by social media including basic user information and some blog posts Extract user-level features x and blog post-level features x from this data t to obtain a dataset subset Step 3: Use the small amount of data obtained in Step 2 to train a simulator f(·) to learn the interactions in social media; the simulator takes a sample x u,v,t as input and learns whether user u forwards the blog post t of user v, denoted by the predicted value of the simulator, where are the simulator parameters; Step 4. Based on the simulator f(·) trained in Step 3, calculate an influence vector The estimation formula of this influence vector is as follows: Among them, denote the training loss of the simulator with respect to the sample x fix ; is the Hessian matrix of the training loss ; is the generalization loss; let the set of content to be generalized be denoted by , and the set of target users of the blog posts to be generalized be denoted by , then the generalization loss is defined as follows: The influence vector calculated in this step is divided into three vectors, namely where have dimensions that are respectively the same as those of x u , x v , x t ; Step 5: Use the social robots generated in Step 1 to forward blog posts and modify basic profiles for promotion; wherein, The promotion by using the social robots to forward blog posts in Step 5 includes: Calculate the influence of each social robot separately The influence of the users obtained in step 2 The influence of the blog posts obtained in step 2 Among them Greedy select n u ,n v ,n t of the most influential social robots, observable users, and observable blog posts respectively; obtain the candidate set of forwarding to be executed according to their dependency relationships Among them, (u, v, t) represents that social robot u forwards blog post t of its follower v; finally, according to the influence of sample x u,v,t Then select n of the most influential forwards from and input them into the application in step 1 to let the social robots automatically execute;​ The promotion by using the social robots to modify basic profiles in Step 5 includes: For a social robot According to Generate m neighbors for social robot i, where M is a vector with the same dimension as x cand and only contains 0 and 1; and, if the user information corresponding to a certain feature of x i is immutable, then the value at the corresponding position in M is 1; if a certain feature in x i is multi-dimensional, then the values at the corresponding positions in M should all be the same, that is, all 0 or all 1; for each neighbor i calculate its influence; let denote the data related to the social robot i.e., Then, x is modified to i The formula for calculating the influence of promotion is as follows: ​ Select the one with the greatest influence from these m cand neighbors. Modify the basic profile information of social robot i to be the same as that of x i ′ ; Repeat this step for each social robot to modify the basic profile.

2. The method according to claim 1, wherein The social media actions in Step 1 include searching, replying, forwarding, and following.

3. The method according to claim 1, wherein The basic profiles of the users in Step 2 include age, gender, and tags, the characteristics of the user level include age, the number of blog posts, the number of followers and the number of followed users, and the characteristics of the blog post level include the blog post theme obtained by using the LDA algorithm, the number of words in the blog post, the number of likes, and the number of forwards.

4. The method according to claim 1, wherein The simulator in Step 3 can be any one of the models in the field of deep learning.

5. A device, characterized in that, The device includes a processor and a memory; wherein, the memory is used for storing a computer program, and the processor is used for executing the social media promotion method based on social action injection according to any one of claims 1-4.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used for storing a computer program, and the computer program is used for executing the social media promotion method based on social action injection according to any one of claims 1-4.

Images