A brand ambassador screening method based on efficient subset sampling

Abstract

This invention presents a brand ambassador screening method based on efficient subset sampling. It consists of two parts: the first part constructs a graph structure G based on the social network of social media, builds a preprocessing structure for subset sampling, sets parameters, and uses greedy selection to generate a user set based on the generated reverse reachable set. Finally, the influence upper bound of the true most influential user set is calculated to provide brand ambassador candidate users based on the social network. The second part processes social network updates; when the social network is updated, the second part is used first, and then the first part is used to obtain brand ambassador candidate users. This ultimately achieves the effect of faster search for the most influential user set in the social network, allowing businesses and social media platforms to quickly identify suitable brand ambassadors.

Description

Technical Field

[0001] The present invention relates to the field of network technology, and in particular to a brand ambassador screening method based on efficient subset sampling. Background Art

[0002] In recent years, with the rapid development of Internet technology, social platforms have become an indispensable part of people's daily lives. According to the "2021 China Social Media Development Trend Report" released by iResearch Consulting, as of the end of 2021, the number of social media users in China has reached 1.093 billion, of which the number of mobile social media users has reached 1.065 billion, accounting for 97.44% of the total number of users. As the number of social media users continues to grow, social media platforms have gradually expanded into the field of social e-commerce, enhancing the commercial value of social media platforms through social shopping methods and social marketing methods. An important part of social e-commerce is product promotion on social media platforms. Brand merchants usually hire influential social media users to serve as brand ambassadors. Brand ambassadors promote products on social media to encourage other users on social media to understand the brand and purchase brand products.

[0003] Choosing the right brand ambassadors is a common challenge for businesses and social media platforms. Businesses often seek to identify a set of one or more users (consisting of the most influential users) on a social media platform as brand ambassadors, thereby increasing their reach. How do you select the most influential user group on a social media platform? Influence maximization is a widely used approach to address this problem. Influence maximization first models social media, forming a directed, weighted graph where nodes represent users and edges represent influence relationships between users. An influence maximization algorithm is then run on this graph to find the most influential user group, one that maximizes the total number of users affected. With the ever-increasing volume of social media and the complex and frequently changing relationships between users, effectively applying influence maximization on such a large and dynamic graph has become a worthy research question. Existing influence maximization methods are either difficult to apply to large social networks or struggle to adapt to the dynamics of social networks. A method that addresses both these needs is urgently needed to help advertisers quickly identify suitable brand ambassadors.

[0004] An existing technology related to the present invention is an online processing algorithm for influence maximization. This method mainly gradually increases the number of generated reverse reachable sets to find the user set with the greatest influence. The specific solution and shortcomings of this method are described below.

[0005] The online processing method for maximizing influence improves the efficiency of solving the influence maximization problem by gradually increasing the number of generated reverse reachable sets. For the modeled social network G, where nodes represent users, the edges between nodes represent the influence relationship between users. For example, if v may receive dynamic information from u (including advertising promotion), there is an edge e = (u, v) pointing from u to v. At the same time, each edge e has a probability p(e), which represents the probability of v receiving dynamic information from u, that is, v receives dynamic information from u with a probability of p(e). First, the online processing method for maximizing influence sets the parameter θ based on the number of graph nodes n, the number of users k included in the required set of users with the greatest influence (when only one brand ambassador is needed, set k = 1), the failure probability δ, and the error parameter ∈ (to ensure that the influence of the obtained user set is not less than (1-1 / e-∈) times the influence of the true set of users with the greatest influence, and the probability of failure is not higher than δ). max Then, the influence maximization online processing method generates two random reverse reachable sets R1 and R2, and the number of reverse reachable sets contained in both sets is θ0. Set the parameters Next, proceed to max Second loop: Each time, greedy selection is used to generate a user set S of size k based on the generated reverse reachable set R1. * , then according to S * To calculate S * The lower bound of influence and the upper bound of the influence of the real most influential user set Then set δ1 and δ2 to be δ / (3i max ).calculate like or i=i max Returns the current S * As the required user set, otherwise continue to generate reverse reachable sets so that the sizes of R1 and R2 are twice as large as before and repeat the above cycle.

[0006] The key technique of the online influence maximization method is to gradually increase the number of reverse reachable sets to find the set of users with the greatest influence. If the error requirement is met, the search stops and the method returns. This method has two drawbacks. The first is that the generation of reverse reachable sets is slow. As shown in the execution process of this method, how to efficiently generate reverse reachable sets is the key to improving efficiency. The process of generating reverse reachable sets repeatedly solves the subset sampling problem. The online influence maximization method uses the simplest approach to this problem, resulting in low efficiency. The second drawback is that it can only solve static social media problems and cannot handle dynamic updates such as the establishment of new links between users (such as following and liking), the deletion of new links (such as unfollowing), the registration of new users, and the deregistration of users. In summary, the efficiency and dynamic adaptability of the online influence maximization method need to be improved.

[0007] The problems to be solved by this invention mainly include two aspects: 1) improving the efficiency of existing technologies to quickly return the most influential user set as reference candidates for brand ambassadors; 2) implementing the update of social networks so that even after social media updates, the technology of this invention can still be used to return reference candidates for brand ambassadors. Summary of the Invention

[0008] To this end, the present invention first proposes a brand ambassador screening method based on efficient subset sampling, which includes two parts: the first part provides brand ambassador candidate users based on social networks, and the second part processes the updates of social networks;

[0009] Input a social network graph G = (V, E), where V is the vertex set, i.e., the users in the social network, and E is the edge set, representing the influence relationship between users. If there is an edge e = (u, v) ∈ E, then user u can influence user v. Each edge e = (u, v) has a probability p(e) ∈ [0, 1], representing the probability that user u's brand promotion will influence user v. That is, a larger p(e) indicates that user u's brand promotion is more likely to influence user v. In addition to the graph G, you also need to input the number of brand ambassadors k to be recruited, an error parameter ∈, and a failure probability δ (this ensures that the total number of users that the candidate set of brand ambassadors (consisting of k users) can influence is no less than (1 - 1 / e-∈) times the total number of users that the actual set of the most influential brand ambassadors can influence, and the failure probability is no higher than δ).

[0010] When there is an update on the social network, the second part is used to update first, and then the first part is used to obtain the candidate set of brand ambassadors;

[0011] The specific process of the first part is:

[0012] S1, for each user node in the input graph G, build a preprocessing structure for subset sampling;

[0013] S2, according to the number of graph nodes n, the number of users k included in the required most influential user set, the failure probability δ, and the error parameter ∈, sets the upper limit θ of the number of random reverse reachable sets max and the number of random reverse reachable sets generated for the first time θ0, θ=θ0 represents the current number of random reverse reachable sets;

[0014] S3, set parameters Indicates the upper limit of the number of next iterations;

[0015] S4, using the subset sampling preprocessing structure generated in S1, generates random reverse reachable sets and puts them into the two sets R1 and R2 until the number of reverse reachable sets contained in R1 and R2 is θ;

[0016] S5, each time using greedy selection to generate a user set S of size k based on the generated reverse reachable set R1 * ;

[0017] S6, set δ1 and δ2 to be δ / (3i max ), S obtained according to S5 * To calculate S * The lower bound of influence (in represents the lower bound of influence) and the real set of users with the greatest influence S o The upper bound of influence σ u (S o )(where σ u (·) indicates the upper bound of influence);

[0018] S7, calculation like or i=i max Returns the current S * As the desired user set, otherwise double θ and repeat S4-S7;

[0019] The specific steps of S1 are as follows:

[0020] S11, for each user node u, suppose it has m incoming neighbors, namely {v1,v2,…v i …,v m}, each incoming neighbor v i The probability of connecting to node u is p(v i ), first according to p(v i )Build the first layer grouping structure and get the group Each group The corresponding probability is

[0021] S12, according to and Construct the second layer grouping structure and obtain m1 groups Each group The corresponding probability is

[0022] S13, according to and Construct a lookup table structure and get OK, Table T of columns;

[0023] The specific steps for generating a reverse reachable set in S4 are as follows. To generate multiple reverse reachable sets, repeat the following process:

[0024] S41, for all nodes v in the social network, set A[v] = 0, randomly select a node u in the social network, set A[u] = 1, set the queue Q to be empty, and add node u to the queue;

[0025] S42, take out the first node u in the queue, and use the preprocessing structure to sample the neighbors of u in turn. For each successfully sampled neighbor v, first check whether A[v] is equal to 1. If so, do nothing. Otherwise, put v into queue Q and set A[v] = 1. Repeat S42 until Q is empty.

[0026] S43, all nodes v with A[v]=1 are grouped into a set S, which is a random reverse reachable set.

[0027] The process of the second part is as follows:

[0028] T1. Identify new user links in the social network. Assume that a new edge e = (u, v) is added to the graph structure G. First, according to p(e), u is added to the first layer of the grouping structure of v. Assume that u is added to

[0029] T2. Due to Change, recalculate because To change, you need to change Take it out from the original group and put it into the new group, assuming from Transfer-in

[0030] T3. Due to and Change, recalculate and And recalculate the corresponding row number in the table.

[0031] The technical effects to be achieved by the present invention are:

[0032] The method provided by the present invention utilizes an efficient subset sampling method to more quickly find the most influential user groups in social networks, allowing businesses and social media platforms to quickly locate suitable brand ambassadors. Secondly, the present invention has a rationally designed storage structure that can process social network updates in real time, allowing the subset sampling-based brand ambassador screening method to be quickly run based on the new social network even after the social network is updated. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 A flowchart of a brand ambassador screening method based on subset sampling provided by an embodiment of the present invention;

[0034] Figure 2 This is a specific flow chart of step S2 in an embodiment of the present invention;

[0035] Figure 3 This is a specific flow chart of step S5 in an embodiment of the present invention;

[0036] Figure 4 This is a flowchart of adapting to dynamic updates of social networks provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following is a preferred embodiment of the present invention and combines the accompanying drawings to further describe the technical solution of the present invention, but the present invention is not limited to this embodiment.

[0038] The present invention proposes a brand ambassador screening method based on efficient subset sampling.

[0039] Figure 1 A flowchart of a brand ambassador screening method based on subset sampling provided in an embodiment of the present invention, the method comprising:

[0040] S1, for each user node, build a preprocessing structure for subset sampling;

[0041] like Figure 2 As shown, step S2 of the brand ambassador screening method based on subset sampling in an embodiment of the present invention specifically includes:

[0042] S11. For each user node u, let its incoming neighbors be {v1,v2,…,v m}, each incoming neighbor v i The probability of connecting to node u is p(v i ). First, according to p(v i )Build the first layer grouping structure and get the group Each group The corresponding probability is

[0043] This step will construct the first layer of grouping structure. The core content is to divide each incoming neighbor into groups according to probability. For m incoming neighbors, first construct groups, namely For grouping The included incoming neighbors satisfy 2 -i <p(v i )≤2 -i+1 ; For the last group The included incoming neighbors satisfy For each group calculate

[0044] S12, according to and Construct the second layer grouping structure and get the grouping Each group The corresponding probability is

[0045] This step will construct the second layer of grouping structure. The core content is to get the Divide into groups according to probability. For m0 groups, first construct groups, namely For grouping The grouping of the previous layer it contains satisfies For the last group The grouping of the previous layer it contains satisfies For each group calculate

[0046] S13, according to and Construct a lookup table structure and get OK, Table T of columns.

[0047] First calculate For the For any subset B, fill in p(B) Bs, where

[0048] Record the current line number as

[0049] For other rows, first calculate based on the row number Then fill in the entries of the entire row in the same way.

[0050] S2, set the parameter θ according to the number of graph nodes n, the number of users k included in the required most influential user set, the failure probability δ, and the error parameter ∈ max and θ0, where θ0=θ max ·ε 2 k / n, θ=θ0;

[0051] The setting here is to ensure that the influence of the user set that can be obtained with a probability of 1-δ is not less than (1-1 / e-∈) times the influence of the real most influential user set.

[0052] S3, set parameters

[0053] S4, using the preprocessing structure of subset sampling, generates random reverse reachable sets and puts them into two sets R1 and R2 until the number of reverse reachable sets contained in R1 and R2 is θ;

[0054] like Figure 3 As shown, step S5 of the brand ambassador screening method based on subset sampling in an embodiment of the present invention specifically includes:

[0055] S41, for all nodes v in the social network, set A[v] = 0, randomly select an initial node u in the social network, set A[u] = 1. Set the queue Q to be empty, and add node u to the queue;

[0056] The initial node u is uniformly randomly selected from all nodes in the graph.

[0057] S42, take out the first node u in the queue, and use the preprocessing structure to sample the neighbors of u in turn. For each neighbor v successfully sampled, first check whether A[v] is equal to 1. If so, do nothing. Otherwise, put v into queue Q and set A[v] = 1; repeat S42 until Q is empty.

[0058] The sampling process for node u is as follows: First, uniformly randomly select any entry in the row corresponding to the current row number, and read the subset T corresponding to the entry. For each group contained in the subset exist Specifically, first generate a random number r such that at this time is sampled as a candidate element. Let h = r, and then continue to generate random numbers r so that Pr[r = j] = 2 -i+1 (1-2 -i+1 ) j-1 , let h=r+h, is sampled as a candidate element, and so on, repeatedly generating random numbers r and sampling is sampled as a candidate element until For each candidate element Generate a random number r between [0,1] and judge Is it true? If so, sample For every subset in T After all sampling is completed, several sampled For each Use something like The sampling method is used to sample, and finally the incoming neighbors of the sampled node u can be obtained.

[0059] S43, all nodes v with A[v]=1 are grouped into a set S, which is a random reverse reachable set.

[0060] S5, each time using greedy selection to generate a user set S of size k based on the generated reverse reachable set set R1 * ;

[0061] Assume R = R1, and select the node u that covers the most reverse reachable sets in R to join S each time. * (That is, node u appears the most times in R. Note that u appears at most once in each reverse reachable set, so the node u that appears in the most reverse reachable sets is selected.) Then the reverse reachable set containing node u is deleted from R. Similarly, continue to use the newly generated R to select nodes until k nodes are selected.

[0062] S6, set δ1 and δ2 to be δ / (3i max ), according to S * To calculate S * The lower bound of influence and the upper bound of the influence of the real most influential user set

[0063] Specifically, Where Λ2(S * ) is S * Based on the influence of R2, equal to S * The number of reverse reachable sets in R2 that are covered is greater than the total number of reverse reachable sets in R2. in It's S * Based on the influence of R1.

[0064] S7, calculation like or i=i max Returns the current S* As the desired user set, otherwise double θ and repeat S5-S9.

[0065] Check whether the error is satisfied. If the error is satisfied, it means that the S obtained at this time is * The influence of the user set with the greatest real influence is not less than (1-1 / e-∈) times, and S * As an approximate solution, merchants and social media platforms can use S * To choose a brand ambassador.

[0066] Figure 4 A flowchart of an embodiment of the present invention provides a method for adapting to dynamic updates of a social network, including:

[0067] T1. Take the example of adding a new user link in a social network. Assume that a new edge e=(u,v) arrives in the graph structure G. First, according to p(e), add u to the first layer grouping structure of v. Assume that u is added to

[0068] like Then add u to Otherwise, if 2 -i <p(e)≤2 -i+1 Then add u to middle.

[0069] T2. Due to Change, recalculate because To change, you need to change Take it out from the original group and put it into the new group, assuming from Transfer-in

[0070] because An element has been added and needs to be recalculated

[0071] T3. Due to and Change, recalculate and And recalculate the corresponding row number in the table.

[0072] and Similarly, and Also needs to be recalculated. The new row numbers should be:

[0073]

Claims

1. A brand ambassador screening method based on efficient subset sampling, characterized by: It consists of two parts: the first part provides brand ambassador candidate users based on social networks, and the second part handles the updates of social networks; Input a social network graph structure G = (V, E), where V is the point set, i.e., the users in the social network, and E is the edge set, representing the influence relationship between users. If there is an edge e = (u, v) ∈ E, it means that user u can influence user v. Each edge e = (u, v) has a probability p(e) ∈ [0, 1], which represents the probability that user u's brand promotion will influence user v. That is, the larger p(e) is, the more likely user u's brand promotion is to influence user v. In addition to the graph G, you also need to input the expected number of brand ambassadors k to be recruited, the error parameter ∈, and the failure probability δ. This ensures that the total number of users that the candidate set of brand ambassadors containing k users is expected to influence is no less than (1-1 / e-∈) times the total number of users that the candidate set of brand ambassadors with the greatest influence can influence, and that the failure probability is no higher than δ. When there is an update on the social network, the second part is used to update first, and then the first part is used to obtain the candidate set of brand ambassadors; The specific process of the first part is: S1, for each user node in the input graph G, build a preprocessing structure for subset sampling; S2, according to the number of graph nodes n, the number of users k included in the required most influential user set, the failure probability δ, and the error parameter ∈, sets the upper limit θ of the number of random reverse reachable sets max and the number of random reverse reachable sets generated for the first time θ0, θ=θ0 represents the current number of random reverse reachable sets; S3, set parameters Indicates the upper limit of the number of next iterations; S4, using the subset sampling preprocessing structure generated in S1, generates random reverse reachable sets and puts them into the two sets R1 and R2 until the number of reverse reachable sets contained in R1 and R2 is θ; S5, each time using greedy selection to generate a user set S of size k based on the generated reverse reachable set R1 * ; S6, set δ1 and δ2 to be δ / (3i max ), S obtained according to S5 * To calculate S * The lower bound of the influence σ l (S * ), and the real most influential user set S o The upper bound of influence σ u (S o ), where σ u (·) represents the upper bound of influence, σ l (·) indicates the lower bound of influence; S7, calculate α = σ l (S * ) / σ u (S o ),like or i=i max Returns the current S * As the desired user set, otherwise double θ and repeat S4-S7; The specific steps of S1 are as follows: S11, for each user node u, suppose it has m incoming neighbors, namely {v1,v2,…v i …,v m }, each incoming neighbor v i The probability of connecting to node u is p(v i ), first according to p(v i )Build the first layer grouping structure and get the group Each group The corresponding probability is S12, according to and Construct the second layer grouping structure and obtain m1 groups Each group The corresponding probability is S13, according to and Construct a lookup table structure and get OK, Table T of columns; The process of the second part is as follows: T1. Identify new user links in the social network. Assume that a new edge e = (u, v) is added to the graph structure G. First, according to p(e), u is added to the first layer of the grouping structure of v. Assume that u is added to T2. Due to Change, recalculate because To change, you need to change Take it out from the original group and put it into the new group, assuming from Transfer-in T3. Due to and Change, recalculate and And recalculate the corresponding row number in the table.

2. The brand ambassador screening method based on efficient subset sampling according to claim 1, characterized in that: in, The specific steps for generating a reverse reachable set in S4 are as follows. To generate multiple reverse reachable sets, repeat the following process: S41, for all nodes v in the social network, set A[v] = 0, randomly select a node u in the social network, set A[u] = 1, set the queue Q to be empty, and add node u to the queue; S42, take out the first node u in the queue, and use the preprocessing structure to sample the neighbors of u in turn. For each successfully sampled neighbor v, first check whether A[v] is equal to 1. If so, do nothing. Otherwise, put v into queue Q and set A[v] = 1. Repeat S42 until Q is empty. S43, all nodes v with A[v]=1 are grouped into a set S, which is a random reverse reachable set.

Images