Method for initial user selection for competitive influence propagation in social networks with attributes

Abstract

The application discloses an initial user selection method for competitive influence propagation in a social network with attributes, constructs a social network with attributes according to data in the social network, constructs a kernel-attribute-tree according to the social network, and obtains a candidate user set CS B of a product A according to a known initial user set I A of a competitive product B in the social network. The marginal influence propagation value of each candidate user as an initial user is calculated based on a competitive influence propagation model, the initial user with the maximum marginal influence propagation value is selected from the candidate users, and the initial user selection is continued until the number of selected initial users meets the requirement.

Description

Technical Field

[0001] This invention belongs to the field of social network technology, and more specifically, relates to an initial user selection method for the propagation of competitive influence in attribute-based social networks. Background Technology

[0002] In the real world, the connections between individuals constitute social networks. Individuals within these networks have interests, which manifest as user attributes. Such social networks are called attribute-based social networks. How to select users in attribute-based social networks to maximize competitive influence propagation is an important aspect of social network research. An attribute-based social network can be described as an undirected graph G = (V, E, P, W), where V = {v1, v2, ..., v...}. N} represents the set of individual nodes, v i Let E represent the i-th individual node, i = 1, 2, ..., N, where N represents the number of individuals, and E = {e1, e2, ..., e...} M} represents the set of edges between individuals, e m Let P represent the m-th edge in the relation set, where m = 1, 2, ..., M, and M represents the number of edges. Q} represents the set of probabilities of influence propagation between individuals, P = {p ij |i,j=1,2,…,N,i≠j} represents the set of probabilities of influence propagation between individuals, p ij Let W = {w1, w2, ..., wj} represent the probability of influence propagation between individual i and individual j. H} represents the set of attributes of an individual, w h Let h represent the h-th attribute of an individual, where h = 1, 2, ..., H, and H represents the number of attributes an individual possesses. Clearly, regarding the probability of influence propagation, the probability of influence propagation is 0 when individuals are not connected, and it takes values ​​within (0, 1) when individuals are connected. In a social network G with attributes, individuals will pass on things they are interested in to friends and friends of friends who share the same interest, ultimately ensuring that all friends interested in that thing receive the message. For example, in social network G, the number of individuals N = 10000, and the number of individuals N′ interested in mobile phones is N′ = 1000, where I... B =20 individuals have been selected as initial users of Apple phones. Now, another brand of mobile phones is to be promoted. How to select users from the remaining 998 individuals so that the number of individuals influenced by the other brand of mobile phones is the largest, which is of great significance to the spread of competitive influence on the product.

[0003] Specifically, the product competition influence propagation maximization of the attributed social network refers to that in the presence of an initial user of a certain brand, under a certain propagation mechanism, a certain number of initial users of another brand are effectively selected in the attributed social network G, and finally the influence propagation value of the another brand is maximized. As can be seen, the key of the product competition influence propagation maximization method of the attributed social network is to construct a subgraph of the attributed social network of a certain product, a competition propagation mechanism, and effectively select a set of initial user nodes of another brand in the presence of a certain brand to maximize the competition influence propagation.

[0004] The known subgraph research of the attributed social network includes a method based on node attributes and a method based on node influence and node attributes. Fang et al. <The VLDB Journal, 2019> proposed a method for searching attribute communities related to the attributes of a given keyword q. Xie et al. <Neurocomputing, 2021> proposed a method for searching communities considering the influence and attributes of nodes. This method mainly includes a pkd-truss community model, a graph simplification algorithm and a scoring function to evaluate the attribute cohesion and external influence of the obtained attribute communities. The linear threshold model and its extension and the independent cascade model and its extension are used to simulate the mechanism of competitive influence propagation. Liu et al. <Knowledge-Based Systems, 2016> extended the linear inhibition model and proposed a propagation-inhibition model. In the propagation-inhibition model, the propagation-activation probability and the inhibition-activation probability of each node are calculated based on the activation probability of the activated neighbor nodes and the influence propagation probability between nodes. Nguyen et al. <WebSci, 2012> extended the independent cascade model and the linear threshold model. In the extended model, once a user is influenced by correct information, he will not be influenced by incorrect information. In addition, if correct information and incorrect information reach a user at the same time, the user will be influenced by the correct message. In terms of selecting users to maximize influence propagation, Li Kan et al. <Invention patent, 201811037119.0, 2018> proposed an influence maximization seed set establishment method based on the hierarchical structure of the social network. This method first divides the hierarchical structure of information diffusion, and then finds the seed nodes that maximize the influence from these levels.

[0005] In the real world, users of a social network are connected to each other, and each user has something he is interested in (hereinafter referred to as attributes). The influence of a certain product will only propagate among users who are interested in it. How to maximize the influence propagation in the attributed social network is an important aspect of social network research. In the known social network, only the product competition influence propagation maximization of the social network without attributes is considered, and the product competition influence propagation maximization of the attributed social network is not considered. SUMMARY

[0006] The present application aims at overcoming the deficiencies of the prior art, and provides an initial user selection method for influence propagation maximization in a social network with attributes, which can more effectively realize initial user selection of commodities in a competitive propagation environment.

[0007] In order to achieve the above-mentioned application purposes, the initial user selection method for influence propagation maximization in a social network with attributes comprises the following steps:

[0008] S1: constructing a non-directional graph G=(V, E, P, W) of a social network with attributes according to data in the social network, wherein V={v1, v2, …, vN} represents a set of individual nodes, vi represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e2, …, eM} represents a set of edges between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p1, p2, …, pN×N} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. N} represents a set of individual nodes, vi represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e2, …, e i} represents a set of edges between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p1, p2, …, pN×N} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. M} represents a set of edges between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p1, p2, …, pN×N} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. m} represents a set of edges between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p1, p2, …, pN×N} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. ij} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. ij} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. H} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals. h} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, and W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of an individual, h=1, 2, …, H, H represents the number of attributes of individuals.

[0009] S2: constructing a core-attribute-tree according to the non-directional graph G of the social network, comprising the following steps:

[0010] S2.1: performing k-core decomposition on the social network to obtain k-core values of each individual node vi in the social network, denoting the maximum value of the k-core values in the entire social network as F, denoting a node set with a k-core value f as C i , f=1, 2, …, F; f

[0011] S2.2: determining a connected branch G r′ of the entire social network according to the non-directional graph G of the entire social network, r'=1, 2, …, R G , R G represents the number of connected branches of the social network, then determining a connected branch C f of each node set C f,r , r=1, 2, …, R f , R f represents the number of connected branches of the node set C f ;​

[0012] S2.3: Set the root node of the kernel-attribute-tree, and connect the connected components C in the first node set C1. 1,r As the child nodes of the root node, then the set C of nodes 2 to F. F Connected component C f′,r As a child node at level f′ of the kernel-attribute-tree, f′ = 2, 3, ..., F, where the connected component C f′,r The parent node is the connected component G of the entire social network. r′ Connected components in the (f′-1)th level of the same connected component;

[0013] S2.4: Based on the undirected graph of the attributed social network, obtain the attribute dictionary D for each leaf node in the kernel-attribute-tree. f,r Attribute dictionary D f,r It contains connected component C f,r The attributes of all nodes in the array, where each attribute corresponds to the connected component C. f,r The set of nodes that contain this attribute;

[0014] S3: For product A, which requires selecting initial users, let product B be the competing product whose initial user set has already been selected, and let I be the initial user set of product B in the social network. B ;

[0015] S4: Obtain the initial user set I for product B. B In the kernel-attribute-tree, each individual node contains a set θ of its parent node, its parent nodes, its child nodes, and its child nodes' children in the connected branch. Then, from set θ, select an initial user set I that contains the attributes of the initial user for the preset product A but does not belong to product B. B The individual nodes, i.e., the candidate user set CS for product A. A ;

[0016] S5: Initialize the initial user set for product A Let the influence propagation value σ(S) corresponding to the initial user set be... A I B ) = 0;

[0017] S6: Calculate the candidate user set CS respectively. A The marginal impact propagation value Δσ(S) of each candidate user as the initial user. A ∪{v i′}, I B ), v i′ ∈CS A The specific method is as follows:

[0018] S6.1: Select candidate user v i′ The initial user set S added to product A A The initial user set S′ is obtained. A =S A ∪{v i′};

[0019] S6.2: For each individual node v in the social network i If v i ∈I B Then let individual node v i Product A - Activation Probability Product B - Activation Probability Inactive probability if Then let individual node v i Product A - Activation Probability Product B - Activation Probability Inactive probability If v i ∈VS A -I B Then let individual node v i Product A - Activation Probability Product B - Activation Probability Inactive probability

[0020] S6.3: Initialize the propagation time t = 1;

[0021] S6.4: Calculate the v of each individual node at propagation time t. i Product A - Activation Probability The calculation formula is as follows:

[0022]

[0023] in, Represents individual node v i Among the neighboring nodes, the neighboring nodes whose activation probability of item A changes from 0 to non-zero at propagation time t-1, q A p represents the pre-defined propagation capability of the influence of product A. j′i Represents individual node v j′ To individual node v i The probability of propagation;

[0024] Calculate the v of each individual node at propagation time t. i Product B - Activation Probability The calculation formula is as follows:

[0025]

[0026] wherein, denotes the neighbor node of the individual node v i whose B-activation probability changes from 0 to non-0 at the propagation moment t-1, q B denotes the preset influence propagation ability of the product B, p j″i denotes the propagation probability from the individual node v j″ to the individual node v i ;

[0027] Then, the un-activation probability of each individual node v i at the propagation moment t is calculated respectively, P The calculation formula is as follows:

[0028]

[0029] S6.5: judging whether t < T, T represents the maximum propagation moment, if yes, entering step S6.6, otherwise, the competitive propagation ends, and entering step S6.7;

[0030] S6.6: letting t = t+1, returning to step S6.4;

[0031] S6.7: calculating the influence propagation value σ(S A ∪{v i′},I B ) of the current initial user set, the calculation formula is as follows:

[0032]

[0033] wherein, V' = V\I B , denotes the remaining individual node set of the initial user set I B of the product B in the social network from the individual node set V;

[0034] S6.8: calculating the marginal influence propagation value Δσ(S i′ ∪{v A},I B ) of the candidate user v i′ after being the initial user, the calculation formula is as follows:

[0035] Δσ(S A ∪{v i′},I B ) = σ(S A ∪{v i′},I B ) - σ(S A ,I B )

[0036] S7: selecting a candidate user v with the largest marginal influence propagation value from the candidate user set CS of the current commodity A A i* S8: judging whether the number of initial users reaches a preset threshold, if not, entering step S9, otherwise the initial user selection is finished. A A A i* B A S9: updating the candidate user set of the commodity A

[0037] S8: judging whether the number of initial users reaches a preset threshold, if not, entering step S9, otherwise the initial user selection is finished.

[0038] S9: updating the candidate user set of the commodity A S9: updating the candidate user set of the commodity A

[0039] The initial user selection method for competitive influence propagation in the social network with attributes provided by the application constructs an undirected graph of the social network with attributes according to the data in the social network, constructs a core-attribute-tree according to the undirected graph of the social network, and obtains a candidate user set CS of a commodity A under the condition that a competitive commodity B of the commodity A has an initial user set I in the social network, calculates marginal influence propagation values of each candidate user as an initial user based on a competitive influence propagation model, selects a candidate user with the largest marginal influence propagation value as an initial user, and selects the initial user in the same way until the number of selected initial users meets the requirement.

[0040] The application has the following beneficial effects:

[0041] 1) The application considers the characteristics of individuals with attributes in the social network in the real world, so that the constructed social network is more in line with the actual situation, and compared with the known method, the application not only reflects the influence propagation between individuals, but also reflects the attributes of individuals, so that the selected initial user is more reasonable.

[0042] 2) The application uses a competitive influence propagation model to simulate the influence propagation of two competitive commodities, and the competitive influence propagation model can reflect the influence of different commodities on individuals in the social network with attributes, and compared with the known propagation model, the application is more in line with the influence propagation characteristics of competitive commodities in reality.

[0043] 3) The application firstly obtains a candidate user set according to the core-attribute-tree when selecting an initial user, and then selects a candidate user with the largest marginal influence propagation value as an initial user under the competitive influence propagation model, so that the application can more quickly and effectively determine the initial user compared with the known method. BRIEF DESCRIPTION OF DRAWINGS ​​​​​​

[0044] Figure 1 is the flow chart of the embodiment of the initial user selection method of the competitive influence propagation in the attributed social network of the present application;

[0045] Figure 2 is the flow chart of the construction of the core-attribute-tree in the present application;

[0046] Figure 3 is the flow chart of the calculation of the marginal influence propagation value of the candidate user in the present application;

[0047] Figure 4 is the structure chart of the attributed social network in the embodiment;

[0048] Figure 5 is the structure chart of the core-attribute-tree constructed according to the attributed social network shown in Figure 4 . DETAILED DESCRIPTION

[0049] The detailed description of the embodiments of the present application will be described below in conjunction with the drawings, so that the skilled in the art can better understand the present application. It should be particularly noted that in the following description, when the detailed description of the known functions and designs may obscure the main content of the present application, these descriptions will be omitted here.

[0050] EMBODIMENT

[0051] Figure 1 is the flow chart of the embodiment of the initial user selection method of the competitive influence propagation in the attributed social network of the present application. As shown in Figure 1 , the specific steps of the initial user selection method of the competitive influence propagation in the attributed social network of the present application include:

[0052] S101: constructing the undirected graph of the attributed social network:

[0053] According to the data in the social network, the undirected graph G=(V, E, P, W) of the attributed social network is drawn, wherein V={v1, v2, …, vN} represents the individual node set, vi represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e2, …, eM} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, wN} represents the attribute set of the individual, wi represents the attribute of the individual i. N} represents the individual node set, vi represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e2, …, e i} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, w M} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, w m} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, w ij} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, w ij} represents the edge set between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p| i, j=1, 2, …, N, i≠j} represents the set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, and W={w1, w2, …, wH} represents the set of attributes of an individual, w h Let h represent the h-th attribute of an individual, where h = 1, 2, ..., H, and H represents the number of attributes of the individual.

[0054] S102: Constructing the Kernel-Attribute-Tree:

[0055] Construct a kernel-attribute-tree based on the undirected graph G of the social network. Figure 2 This is a flowchart of the kernel-attribute-tree construction process in this invention. For example... Figure 2 As shown, the specific steps for constructing the kernel-attribute-tree in this invention include:

[0056] S201: k-nucleus decomposition:

[0057] Perform k-kernel decomposition on the social network to obtain the individual nodes v in the social network. i Let F be the maximum k-kernel value in the entire social network, and let C be the set of nodes with a k-kernel value of f. f f = 1, 2, ..., F.

[0058] k-kernel decomposition is a commonly used analysis method for social networks, and its specific process will not be elaborated here.

[0059] S202: Calculate connected components:

[0060] Since the kernel decomposition in step S201 involves an inclusion relationship, the resulting node set C can be... f The organization is structured as a tree, which facilitates finding the initial user node that can influence the set of individual nodes. Before determining the tree structure, the connected components G of the entire social network need to be determined based on the undirected graph G of the entire social network. r′ r′=1,2,…,R G R G This represents the number of connected branches in a social network, and then for each set of nodes C... f Determine the connected component C of the set of nodes respectively. f，r r = 1, 2, ..., R f R f Represents the set of nodes C f The number of connected components. In this embodiment, the method of disjoint sets is used to determine the connected components, and the specific process will not be described in detail here.

[0061] S203: Constructing a tree structure:

[0062] Set the root node of the kernel-attribute-tree, and connect the connected components C in the first node set C1. 1，r As the child nodes of the root node, then the set C of nodes 2 to F. F Connected component Cf′，r As the f'th layer child node of the core-attribute-tree, f'=2, 3,..., F, where the parent node of the connected branch C f′，r is the connected branch in the f'-lth layer which is in the same connected branch G r′ of the whole social network as the connected branch C

[0063] S204: Determine the attribute dictionary:

[0064] According to the attribute social network undirected graph, obtain the attribute dictionary D f，r of each leaf node in the core-attribute-tree, the attribute dictionary D f，r contains all the attributes of the nodes in the connected branch C f，r , and the value corresponding to each attribute is the node set in the connected branch C f，r which has the attribute.

[0065] S103: Determine the product information:

[0066] For the product A which needs to select the initial user, record the competing product which has selected the initial user set as product B, and record the initial user set of product B in the social network as I B . Usually the competing product is the same type of product belonging to different brands.

[0067] S104: Determine the candidate user set based on the core-attribute-tree:

[0068] According to the core-attribute-tree determined in step S102 and the initial user set I B of product B, and the pre-set attribute of the initial user of product A, determine the candidate user set CS A of product A, the specific method is: obtain the parent node of the connected branch where each individual node in the initial user set I B of product B is located in the core-attribute-tree, and the parent node, child node and child node of the child node set θ, and then select the individual node from the set θ which contains the pre-set attribute of the initial user of product A and does not belong to the initial user set I B of product B, that is, as the candidate user set CS A of product A.

[0069] S105: Initialize the initial user information of product A:

[0070] Initialize the initial user set of product A Let the influence propagation value σ(S A , I B ) corresponding to the initial user set be 0.

[0071] S106: Calculate the marginal influence propagation value of the candidate user:

[0072] Next, based on the competitive influence propagation model, the influence propagation process of two commodities is simulated, and the marginal influence propagation value of each candidate user in the candidate user set CS A after taking each candidate user as the initial user is calculated in turn. A ∪{v i′},I B ),v i′ ∈CS A . Figure 3 is the flowchart for calculating the marginal influence propagation value of the candidate user in the present application. As shown in the figure, the specific steps for calculating the marginal influence propagation value of the candidate user in the present application include: Figure 3

[0073] S301: The candidate user joins the initial user set:

[0074] The candidate user v i′ is added to the initial user set S A of commodity A, and the initial user set S′ A =S A ∪{v i′} is obtained.

[0075] S302: Determine the initial activation probability of the individual node:

[0076] For each individual node v i in the social network, if v i ∈I B , the A-activation probability of the individual node v i , the B-activation probability and the inactivation probability are determined. If , the A-activation probability of the individual node v i , the B-activation probability and the inactivation probability are determined. If v i ∈V-S A -I B , the A-activation probability of the individual node v i , the B-activation probability and the inactivation probability are determined.

[0077] S303: Initialize the propagation time t = 1.

[0078] S304: Calculate the activation probability of the individual node at the current time:

[0079] Calculate the individual node v i 's commodity A-activation probability at propagation time t The calculation formula is as follows:

[0080]

[0081] Wherein, represents the neighbor node of the individual node v i whose commodity A-activation probability changes from 0 to non-0 at propagation time t-1, that is, the newly added neighbor node whose commodity A-activation probability is non-0 at propagation time t-1, q A represents the pre-set influence propagation ability of commodity A, and the value range is (0, 1), p j′i represents the propagation probability from the individual node v j′ to the individual node v i . That is, at propagation time t, the individual node v i 's commodity A-activation probability is the activation probability of the individual node v i at propagation time t-1 plus the sum of the influence of the neighbor nodes in the neighbor node set of the individual node v i on the individual node v i at propagation time t-1.

[0082] Similarly, calculate the individual node v i 's commodity B-activation probability at propagation time t The calculation formula is as follows:

[0083]

[0084] Wherein, represents the neighbor node of the individual node v B whose commodity B-activation probability changes from 0 to non-0 at propagation time t-1, q j″i represents the pre-set influence propagation ability of commodity B, and the value range is (0, 1), p j″ represents the propagation probability from the individual node v i to the individual node v

[0085] Then calculate the individual node v i 's inactivation probability at propagation time t The calculation formula is as follows:

[0086]

[0087] S305: judging whether t < T, T represents the maximum propagation time, if yes, entering step S306, otherwise the competition propagation ends, entering step S307.

[0088] S306: setting t = t + 1, returning to step S304.

[0089] S307: calculating the influence propagation value of the current initial user set:

[0090] calculating the influence propagation value σ (S A ∪{v i′},I B ) of the current initial user set, the calculation formula is as follows:

[0091]

[0092] wherein, V' = V\I B , which represents the remaining individual node set of excluding the initial user set I B of the commodity B in the social network from the individual node set V.

[0093] S309: calculating the marginal influence propagation value of the candidate user:

[0094] calculating the marginal influence propagation value Δσ (S A ∪{v i′},I B ) of the candidate user v A as the initial user, the calculation formula is as follows:

[0095] Δσ (S i′ ∪{v B},I A ) = σ (S i′ ∪{v B},I A ) - σ (S B ,I A )

[0096] S107: selecting an initial user:

[0097] selecting the candidate user v i* with the largest marginal influence propagation value from the current candidate user set CS A of the commodity A, as the new initial user to join the initial user set S i* of the commodity A, i.e. setting updating the influence propagation value corresponding to the initial user set

[0098] S108: judging whether the initial user quantity reaches the preset threshold, if not, entering step S109, otherwise the initial user selection ends.

[0099] S109: update the candidate user set of commodity A:

[0100] update the candidate user set of commodity A i.e. selected user v i* is deleted from the candidate user set, and the process returns to step S106.

[0101] In order to better illustrate the technical solutions and technical effects of the present application, a specific example is used to illustrate the present application. Figure 4 is a structure diagram of the attribute-bearing social network in the present embodiment. As shown in Figure 4 , v1, v2, v3, v4, v5, v6, v7, v8, v9, v 10 , v 11 , respectively represent individual 1, individual 2, individual 3, individual 4, individual 5, individual 6, individual 7, individual 8, individual 9, individual 10 and individual 11. The set on the individual node represents the attribute of the individual. For example, the set {w1, w3} on v1 represents that the individual v1 contains the attribute w1 and the attribute w3.

[0102] First, a core-attribute-tree is constructed. By using the method of k-core decomposition, the core and its corresponding node set are obtained, i.e. C1={v6, v7, v8}, C2={v5, v9, v 10 , v 11}, and C3={v1, v2, v3, v4}. Then, the connected components G1={v1, v2, v3, v4, v5, v6}, G2={v7, v8, v9, v 10 , v 11} of the entire social network are determined, and the connected components of each core are calculated, so that C1={v6, v7, v8} includes two connected components C 10 ={v6} and C 11 ={v7, v8}, C2={v5, v9, v 10 , v 11} includes two connected components C 20 ={v5} and C 21 ={v9, v 10 , v 11}, and C3={v1, v2, v3, v4} has only one connected component C 30 ={v1, v2, v3, v4}.

[0103] Given that r is the root node of the constructed core-attribute-tree, r includes two child nodes C 10 and C 11 . Since in the connected components of the entire social network, C 20 , C 30 and C10 In the same connected component, C 21 and C 11 In the same connected component, C 20 is the child node of C 10 , C 21 is the child node of C 11 , C 30 is the child node of C 20 .

[0104] Finally, the attribute dictionary of each node is added to the tree structure, where the key is the attribute and the value is the set of nodes containing the attribute. Figure 5 is the structure diagram of the core-attribute-tree built from the social network with attributes shown in Figure 4 . As shown in Figure 5 , the attribute dictionary of node C 10 is D 10 ={‘w2’:[‘v6’],‘w3‘:[‘v6’]}, and the attribute dictionary of other nodes can be obtained in the same way.

[0105] In this embodiment, the initial user set I B of product B is set to {v2, v8}, the influence propagation ability q A of product A is set to 0.6, and the influence propagation ability q B of product B is set to 0.4. In this embodiment, the initial user of product A has the attribute w1. Next, the candidate user set CS A of product A is determined, and the specific method is as follows: the positions of nodes v2 and v8 in the core-attribute-tree shown in Figure 4 are found, v2 and v8 are respectively at nodes C 30 and C 11 , and the node sets containing the attribute w1 in C 30 and C 11 are respectively S 30 ={v1, v2} and S 11 ={v8}. Next, the parent node of node C 30 and the set of child nodes and child nodes of child nodes of the parent node in the core-attribute-tree are found, and the individual nodes CS B (v2) containing the attribute w1 and not belonging to the initial user set I A of product B are selected, and CS A (v2)={v1, v4, v5}. In the same way, CS 11 (v8)={v9, v A} is obtained, and finally CS A =CS A(v8) = {v1, v4, v5, v9, v...} 11}

[0106] Next, the candidate user set CS is calculated sequentially. A The marginal impact propagation value Δσ(S) of each candidate user as the initial user. A ∪{v i′}, I B ).

[0107] Taking candidate user v1 as an example, after taking candidate user v1 as the initial user, that is, S′ A ={v1}, calculates the activation probability of each individual node in the social network.

[0108] At propagation time t=1, taking individual node v4 as an example, The influence of candidate user v1 on its direct neighbor node v4 regarding product A; Individual node v2, as the initial user of product B, influences its direct neighbor node v4 regarding product B.

[0109] Therefore, the activation probability of item A at t=1 is as follows for individual node v4. The calculation formula is as follows:

[0110]

[0111] The activation probability of item B at time t=1 for individual node v4 The calculation formula is as follows:

[0112]

[0113] The probability that individual node v4 is not activated at t=1 The calculation formula is as follows:

[0114]

[0115] Similarly, the activation probabilities of other individual nodes can be calculated. Table 1 shows the activation probabilities of each individual node in the social network at propagation time t=1 after candidate user v1 is used as the initial user in this embodiment.

[0116]

[0117] Table 1

[0118] At propagation time t=2, the activation probability calculation is explained using individual node v5 as an example. Candidate user v1 for product A influences individual node v5 through the edge v1→v4→v5. The activation probability of item A at time t=2 for individual node v5 The calculation formula is as follows:

[0119]

[0120] While the initial user v2 of the product B influences the individual node v5 through the edges v2→v4→v5, The calculation formula of the activation probability of the individual node v5 at t=1 is as follows:

[0121] The calculation formula of the non-activation probability of the individual node v5 at t=1 is as follows:

[0122]

[0123]

[0124] Similarly, the activation probabilities of other individual nodes can be calculated.

[0125] Table 2 is the activation probability of each individual node in the social network at the propagation time t=2 after the candidate user v1 is taken as the initial user in the embodiment.

[0126]

[0127] Table 2

[0128] In the embodiment, it is assumed that the maximum propagation time T=2, and the competitive influence propagation ends. Then the product A-activation probability of the individual node other than the initial user set I B of the product B is summed, and the influence propagation value of the candidate user v1 is obtained.

[0129] Similarly, the influence propagation values of other candidate users are obtained. Table 3 is the influence propagation value and the marginal influence propagation value of each candidate user when the initial user is selected for the first time in the embodiment.

[0130]

[0131] Table 3

[0132] Since the initial user set S when the initial user is selected for the first time, the influence propagation value σ(S A , I B ) corresponding to the initial user set is 0, and thus the marginal influence propagation value of each candidate user is the influence propagation value of the candidate user, that is, the marginal influence propagation value of the candidate user v4 is the largest, and the candidate user v4 is added to the initial user set, that is, the initial user set S A is updated to {v4}, and the influence propagation value σ(S A , I B ) corresponding to the initial user set is 1.48.​​

[0133] The second time the initial users are selected, the influence propagation value of each candidate user is recalculated based on the initial user set S A ={v4}, the influence propagation value of each candidate user is recalculated. Table 4 is the influence propagation value and marginal influence propagation value of each candidate user when the initial users are selected the second time in this embodiment.

[0134] <![CDATA[v2]]> <![CDATA[v5]]> <![CDATA[v9]]> v 11 ]]> <![CDATA[σ(S A ∪{v i′ },I B )]]> 2.24 2.24 2.66 2.66 A ∪{v i′}, I B )- σ(S A , I B )]]> ​ 0.76 0.76 1.18 1.18

[0135] Table 4

[0136] As shown in Table 4, the marginal influence propagation value of candidate user v9 or v 11 is the largest, and candidate user v9 is randomly selected to join the initial user set, i.e. the initial user set S A ={v4, v9} is updated, and the influence propagation value corresponding to the initial user set is σ(S A , I B )=2.66.

[0137] In this embodiment, the number of initial users is set to 2, and thus the selection of initial users is completed.

[0138] Although the above describes the specific embodiments of the present application in detail in order to facilitate the understanding of the present application by those skilled in the art, it should be clear that the present application is not limited in the scope of the specific embodiments, and all the applications and creations using the concept of the present application are within the scope of protection as long as various changes are obvious to those skilled in the art within the spirit and scope of the present application defined and determined by the appended claims.

Claims

1. A method for initial user selection in competition influence propagation in a social network with attributes, characterized in that, Comprising the following steps: S1: Construct a directed graph G=(V, E, P, W) of the social network with attributes according to the data in the social network, wherein V={v1, v2, …, vN} represents a set of individual nodes, vi represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e2, …, eM} represents a set of edges between individuals, em represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p1, p2, …, pN(N-1)} represents a set of influence propagation probabilities between individuals, pi,j represents the influence propagation probability between individual i and individual j, i, j=1, 2, …, N, i≠j, W={w1, w2, …, wH} represents a set of attributes of individuals, wh represents the hth attribute of the individual, h=1, 2, …, H, H represents the number of attributes of the individual. N} represents a set of individual nodes, v i represents the ith individual node, i=1, 2, …, N, N represents the number of individuals, E={e1, e M} represents a set of edges between individuals, e m represents the mth edge in the relationship set, m=1, 2, …, M, M represents the number of edges, P={p ij |i,j=1,2,…,N,i≠j} represents a set of influence propagation probabilities between individuals, p ij represents the influence propagation probability between individual i and individual j, W={w1, w2, …, w H} represents a set of attributes of individuals, w h represents the hth attribute of the individual, h=1, 2, …, H, H represents the number of attributes of the individual. S2: Constructing a core-attribute-tree according to the undirected graph G of the social network, comprising the following steps: S2.1: k-core decomposition is performed on the social network to obtain the k-core value of each individual node v in the social network, and the maximum value of the k-core value in the entire social network is recorded as F, and the node set with the k-core value f is recorded as C i , f = 1, 2,..., F f ; S2.2: determining connected components G of the entire social network from the undirected graph G of the entire social network r′ , r' = 1, 2,..., R G , R G denotes the number of connected components of the social network, then for each node set C f , respectively, the connected components C f,r , r = 1, 2,..., R f , R f denotes the number of connected components of the node set C f ; S2.3: Set the root node of the core-attribute-tree to the connected component C 1,r as the child node of the root node, then set the connected components C F as the child nodes of the 2nd to Fth node set C f′,r as the f' layer child nodes of the core-attribute-tree, f' = 2, 3,..., F, where the parent node of the connected component C f′,r is the connected component in the (f'-l)th layer which is in the same connected component G r′ in the whole social network as the connected component C S2.4: According to the social network undirected graph with attributes, obtaining the attribute dictionary D of each leaf node in the core-attribute-tree f,r , the attribute dictionary D f,r contains the attributes of all nodes in the connected branch C f,r , and the value corresponding to each attribute is the node set in the connected branch C f,r where the attribute exists; S3: For the commodity A which needs to select the initial user, record the competing commodity which has selected the initial user set as commodity B, and record the initial user set of commodity B in the social network as I B ; S4: Obtain the initial user set I of the commodity B B The parent node of the connected branch in which each individual node in the core-attribute-tree and the parent node of the parent node, the child node and the child node of the child node of the individual node are located, and then select the attribute containing the initial user of the preset commodity A from the set θ and not belonging to the initial user set I of the commodity B B The individual node of the commodity A, that is, the candidate user set CS of the commodity A A ; S5: initialize the initial user set of commodity A Let the influence propagation value σ(S A ,I B ) = 0; S6: calculate the marginal influence propagation value Δσ(S A A ∪{v i′},I B ), v i′ ∈CS A , specifically as follows:​ S6.1: Add candidate user v i′ to the initial user set S of commodity A A , and get the initial user set S' A = S A ∪{v i′} S6.2: For each individual node v i if v i ∈ I B then let the A-activation probability of individual node v i the B-activation probability of individual node v the inactivation probability of individual node v if v i ∈ S' A then let the A-activation probability of individual node v i the B-activation probability of individual node v the inactivation probability of individual node v if v i ∈ V - S A - I B then let the A-activation probability of individual node v i the B-activation probability of individual node v the inactivation probability of individual node v ​​​ S6.3: Initializing the propagation time t=1; S6.4: Calculate the commodity A-activation probability of each individual node v at the propagation time t, respectively i S6.4: Calculate the commodity A-activation probability of each individual node v at the propagation time t, respectively The calculation formula is as follows: wherein, denotes the neighbor node of the individual node v i in which the activation probability of the commodity A changes from 0 to non-0 at the propagation moment t-1, q A denotes the preset influence propagation ability of the commodity A, p j′i denotes the propagation probability from the individual node v j′ to the individual node v i ; The product B-activation probability of each individual node v at the propagation time t is calculated respectively i The calculation formula is as follows:​ wherein, denotes the neighbor node of the individual node v i in which the activation probability of the commodity B changes from 0 to non-0 at the propagation moment t-1, q B denotes the preset influence propagation ability of the commodity B, p j″i denotes the propagation probability from the individual node v j″ to the individual node v i ; Then the inactivation probability of each individual node v i at the propagation time t is calculated respectively The calculation formula is as follows: S6.5: Judging whether t S6.6: Let t=t+1, and return to step S6.4; S6.7: Calculate the impact propagation value σ(S) of the current initial user set. A ∪{v i′ },I B The calculation formula is as follows: V' = V \ I B , denotes the remaining set of individual nodes from the set V of individual nodes excluding the initial set I of users of the social network for the good B B . S6.8: Calculate candidate users v i′ The marginal influence propagation value Δσ(S A ∪{v i′},I B ) is calculated as follows: AS A ∪{v i′},I B ) = AS A ∪{v i′},I B - AS A , I B ​ S7: selecting a candidate user v with the largest marginal influence propagation value from the candidate user set CS of the current commodity A A as a new initial user to join the initial user set S of the commodity A i* A i.e. updating the influence propagation value corresponding to the initial user set ​ S8: Judging whether the initial user quantity reaches a preset threshold, if not, entering step S9, otherwise, the initial user selection ends; S9: update the candidate user set of commodity A Return to step S6.

Images