Social network node importance ranking method based on inherent propagation mode
By classifying the dissemination model in social networks, using Monte Carlo simulation and dynamic programming methods to calculate the dissemination center, the problem of node importance evaluation in multiple dissemination model scenarios is solved, the evaluation efficiency and accuracy are improved, and the formulation of more Effective public opinion management strategy.
Patent Information
- Application Number
- CN202411155771.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-22
- Publication Date
- 2024-11-26
AI Technical Summary
The prior art ignores the role and influence of nodes in various propagation model scenarios when evaluating the importance of social network nodes, and the precise calculation time of the propagation process is relatively complex.
The importance ranking method of social network nodes based on the inherent propagation mode is adopted. By classifying the propagation model, using Monte Carlo simulation method for importance sampling and linear propagation matrix calculation, a directed acyclic graph is constructed, and a dynamic programming is used to calculate the propagation center. to achieve accurate ranking of the importance of nodes.
It provides an important node mining framework that is suitable for multiple communication model scenarios, improves algorithm efficiency, can more accurately evaluate important nodes during communication, and helps formulate better public opinion management strategies.
Smart Images

Figure CN119027257A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of social network node ranking methods, and in particular to a social network node importance ranking method based on inherent propagation patterns. Background Art
[0002] The rapid development of social networks in today's world has greatly simplified the difficulty of obtaining information, which allows key information such as commercial propaganda and official announcements to spread quickly through social networks. Since information dissemination in social networks is a complex dynamic process, tasks based on dissemination, such as rumor blocking and product promotion, are quite challenging.
[0003] Node importance, as an indicator to measure the importance of a node relative to other nodes in a network, has derived many new types, which can significantly improve the reliability and accuracy of results for specific tasks. However, most existing methods focus on analyzing the structural characteristics of networks and communities, while relatively ignoring the role and influence of nodes in various propagation model scenarios. At the same time, the precise calculation of the propagation process has a high time complexity. For this reason, one of the core contents of the present invention is to start from the classical propagation model and deeply explore the node centrality ranking problem based on propagation dynamics, aiming to provide an important node mining framework that is suitable for a variety of propagation model scenarios.
[0004] Centrality methods are a class of methods used to describe the core position of nodes in a network and their overall impact on network functions. This method provides ideas for ranking the importance of nodes in social networks. Identifying users with significant importance based on communication centrality and ranking them according to their importance is of great significance in solving these challenges. Summary of the invention
[0005] The present invention provides a method for ranking the importance of social network nodes based on inherent propagation patterns to solve the problem that the prior art ignores the role and influence of nodes in various propagation model scenarios.
[0006] In order to achieve the above object, the technical solution adopted by the present invention is:
[0007] The method for ranking the importance of social network nodes based on the inherent propagation mode includes the following steps:
[0008] Step 1: Classify the propagation models of fixed propagation modes on social networks, thereby dividing the propagation models on social networks into deterministic propagation models and uncertain propagation models;
[0009] Step 2: Perform importance sampling of the proposed distribution of event propagation on the uncertainty propagation model in step 1 by using the Monte Carlo simulation method, and calculate the linear propagation matrix of the deterministic propagation model in step 1 according to the proposed distribution of importance sampling;
[0010] Step 3: Construct a directed acyclic graph of event propagation based on the linear propagation matrix obtained in step 2;
[0011] Step 4: Based on the directed acyclic graph, the dynamic programming method is used to calculate the propagation centrality score of each node, and the importance of each node is ranked from high to low according to the propagation centrality score.
[0012] The present invention provides a method for ranking the importance of social network nodes based on inherent propagation patterns, which can mine important node users of social networks in a fixed propagation pattern in advance, and then more accurately evaluate important potential users when propagation occurs, which is helpful to formulate better public opinion management strategies. Compared with the prior art, the present invention has the following advantages:
[0013] 1. The present invention proposes a node importance measurement indicator "dissemination centrality", which can be used to quantify the importance of nodes in social networks in information dissemination scenarios.
[0014] 2. In view of the uncertain scenarios of complex propagation, the present invention proposes a method based on importance sampling to optimize the number of simulations of Monte Carlo simulation in the propagation process, improves the problem of high time cost of precise calculation, and improves the efficiency of the algorithm;
[0015] 3. The present invention designs a framework for calculating the propagation centrality for different propagation scenarios, and derives the node propagation centrality calculation to various propagation models such as uncertainty model (including SIR model, independent cascade model) and deterministic model (including linear threshold model) to adapt to various propagation scenarios. From the perspective of dynamic propagation, the importance ranking problem of nodes under various propagation models is solved. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] Figure 1 The overall framework diagram of the proposed node importance ranking method under the inherent propagation mode.
[0017] Figure 2 Flowchart of the proposed node importance ranking method.
[0018] Figure 3 An example of centrality calculation for the proposed nodes. DETAILED DESCRIPTION
[0019] The present invention is further described below in conjunction with the accompanying drawings and embodiments.
[0020] like Figure 2 As shown, this embodiment discloses a method for ranking the importance of social network nodes based on an inherent propagation mode, comprising the following steps:
[0021] Step 1: Classify the propagation models with fixed propagation modes on social networks, and divide the propagation models on social networks into deterministic propagation models and uncertain propagation models.
[0022] On social networks, user nodes will not repeatedly spread the same information to the same person, and once a user node receives a piece of information, it will always be in a state of being spread. Common propagation models in social networks include the infectious disease model, the IC propagation model, and the LT propagation model. The propagation methods of these propagation models have the following characteristics: (1) Information will not be repeatedly propagated to the same node, that is, there is no possibility of re-propagation or infection. (2) Once a node is activated (for example, receiving information or a product), it will not be activated again. (3) The propagation process is unidirectional and there is no reverse propagation. Therefore, for propagation models with different propagation modes, a propagation directed acyclic graph g is constructed. s It is very necessary that the role of user nodes on social networks will also change accordingly. The specific construction framework is as follows Figure 1 As shown in (a).
[0023] Based on the above analysis, this embodiment divides the propagation model on the social network into an uncertain propagation model and a deterministic propagation model by determining whether the occurrence of event propagation is based on probability or threshold.
[0024] Step 2: Use the Monte Carlo simulation method to perform importance sampling of the proposed distribution of event propagation on the uncertainty propagation model in step 1, and calculate the linear propagation matrix of the deterministic propagation model in step 1 based on the proposed distribution of importance sampling. The specific process is as follows:
[0025] (A) In this embodiment, the Monte Carlo simulation method is used to perform importance sampling of the proposed distribution of event propagation on the uncertainty propagation model in step 1 to obtain the most likely propagation probability of each node being propagated by the event, and then obtain the proposed distribution of importance sampling of each node.
[0026] Importance sampling is an approximate method for optimizing simulation. It changes the sampling distribution and samples more frequently those situations that contribute more to the overall simulation results. In this embodiment, this means giving priority to those nodes and paths that play a key role in network propagation. In order to enhance the sampling quality of each simulation, it is necessary to determine the proposed distribution q(·|·) of node importance sampling. For this purpose, a small sample simulation experiment is designed to find the node prior propagation probability. Here, the case where propagation starts from s is still analyzed.
[0027] In the propagation node vi The events being propagated are Where k represents the number of iterations of the event. For the initial event, it can be written as The probability that the propagation event will be carried out from the propagation starting node v0 at t0 is 1. The quality of importance sampling is determined by The probability density of an event determined, that is, if The higher the value, the corresponding proposal distribution A higher value of allows the Monte Carlo simulation to be performed fewer times and obtain a value closer to the expected value, so it is necessary to define the probability density f(·).
[0028] Therefore, this embodiment is based on the following theorem: {Theorem} In importance sampling, the higher the positive correlation between the new proposed distribution q(x) and the probability density function f(x), the lower the variance can be and the closer the sampling is to the expected value.
[0029] Proof: For the expectation μ of the original distribution, we have μ=E(f(X))=∫Df(x)p(x)dx, where D represents the domain of the distribution p in this proof. For the proposed distribution q(x)>0, in:
[0030]
[0031]
[0032] Among them, x is the propagation sampling distribution, X i is the i-th sampling distribution, p represents the probability of the event occurring, is the expectation of sampling according to the proposed distribution q, and p(x) represents the probability of sampling x.
[0033] is the variance of the estimator when using importance sampling, and Q = {x|q(x)>0) is the domain of the proposed distribution q. Smaller, can reduce (f(x)p(x)-μq(x)) 2 The size of, since p(x) is a constant and is greater than 0, the variance is reduced by making the new distribution q(x)∝f(x) so that the sampling result approaches the expected μ.
[0034] Therefore, in this embodiment, by giving a higher probability density function f(x) a larger proposal distribution q(x), this embodiment can obtain higher quality sampling. For different propagation starting points, nodes that are sampled more frequently have a higher probability density of being propagated. Therefore, before performing the Monte Carlo simulation, the nodes are randomly sampled, a limited amount of propagation is simulated first, high-frequency nodes are recorded, and a larger proposal distribution q is assigned based on the prior probability of high-frequency nodes. Since propagation is a continuous uncertain process, when the propagation reaches node v i You can get:
[0035]
[0036]
[0037] in: Indicates that the propagation is propagated to node v at the kth iteration i The probability of Indicates that the propagation is propagated to v at the k-1th iteration i In the case of neighbor nodes, the kth propagation to v i The conditional probability of δ(x k , j, i) is the label indicator function, if the kth iteration is from v j Propagate to v i Its value is 1 when the event occurs, otherwise it is 0.
[0038] Indicates the node v in the kth iteration i Is there at least one neighbor that is propagated if node v i At the kth iteration there is at least one neighbor that is propagated, then otherwise Still by confirming that propagation has occurred and the distribution of successful sampling ∈ j,k ~Bernoulli(β)determined.
[0039] In this embodiment, the event The Monte Carlo likelihood estimate of can be expressed as:
[0040]
[0041] When the propagation iteration reaches the kth time, more nodes through which the propagation path passes need to meet the criteria of importance sampling. when And σ(x k , j, i)=1, then Represents the set of neighbor nodes of node vj, and has:
[0042]
[0043] As shown in the above formula, based on the proposed distribution obtained from a small number of simulated prior probabilities and the iterative conditional probability, we get the event Monte Carlo likelihood estimate of . It can weight each sample as the ratio of the proposed distribution and the actual transmission probability to estimate the probability of a transmission event.
[0044] Through the first few relatively small Monte Carlo simulations, the obtained Sort by The corresponding proposed distribution It will increase. Depending on the situation, the largest K distributions can be selected for sampling to improve the sampling quality and make the sampled propagation path closer to the expected value. When simulating the corresponding first K propagations, the method of updating the cumulative propagation matrix C needs to be changed. The calculation method of the updated cumulative propagation matrix C is as follows:
[0045]
[0046] The simultaneous propagation is recorded in the final propagation adjacency matrix, which can obtain the expected value of the propagation route, that is, the most likely propagation route. Thus, the cumulative propagation matrix at the last moment is obtained as the linear propagation matrix.
[0047] Compared with the SIR propagation model, the difference of the IC model propagation is that the propagation of information between nodes is not carried out according to the unique transmission rate $\beta$. Instead, each node's outgoing edge has a certain weight represents the success rate of the message spreading through this relationship. Here, the corresponding Bernoulli distribution can be changed to ∈ i,j ~Bernoulli(w ij ), w ij Indicates that v j to v i The success probability of edge propagation.
[0048] (B) In this embodiment, the calculation of the linear propagation matrix of the uncertainty propagation model in step 1 during multiple iterations in the Monte Carlo simulation is described as follows:
[0049] The state of the node is binary distributed, and the node state vector x is distributed using Bernoulli distribution for the propagation rate β i,t Generate a mask ∈ i,j (i, j represent node numbers) to determine the state vector x during the tth iteration t The element x i,t Whether it is propagated in this Monte Carlo simulation. According to the Bernoulli distribution ∈i,j ~Bernoulli(β), will∈ i,j As an indicator of whether successful propagation has occurred. Then the state vector x t The iteration formula is:
[0050]
[0051] Where: x i,t Represents node v i At the state of iteration t, x i,t =1 indicates node v i is propagated at the tth iteration, otherwise x i,t = 0. Since the node status is binary distributed, the Bernoulli distribution is used for the propagation rate β. i,t Generate a mask ∈ i,j σ(v j ) t = S, to determine the state vector x in the tth iteration t The element x i,t Whether it is propagated in this simulation. This step is equivalent to a sampling operation, according to the Bernoulli distribution ∈ i,j ~Bernoulli(β), will∈ i,j As an indicator of whether successful propagation has occurred, the state vector x is finally updated and participates in the iterative update of C.
[0052] In order to record the propagation path, a cumulative propagation matrix C is needed to record the path changes of each state iteration. The initial state of C is a full zero matrix. After each propagation iteration, if the node v i Successfully propagated to node v j , then C ij Increase by 1. This can be expressed by the following expression:
[0053]
[0054] Where: δ(x t , i, j) represents the state, which is 1 when propagation occurs and 0 otherwise. C is the required linear propagation matrix, and δ is a label indicator function. Since the calculation of the cumulative propagation matrix C adopts the cumulative mode, it is necessary to remove the nodes that have been propagated in the previous iteration during iteration, so that the final C will not accumulate invalid propagation.
[0055] Step 3: Construct the directed acyclic graph g of event propagation based on the linear propagation matrix calculated in step 2 s , the specific method is:
[0056] Set a specific pre-order collection data structure At the same time, set the successor set S and the propagation node queue Q.
[0057] For the node v that receives the propagated message i , all pairs of v i The nodes that contribute to the propagation of will be added to v if the number of iterations is consistent. i The preorder set This method avoids the i When the number of neighbors n(i) is large, paths with similar expected values are ignored due to uneven sampling.
[0058] After the expected propagation path is determined, a parallel breadth-first search is performed on the expected propagation graph of the origin node s, the number of node iterations is recorded, and the propagation paths with the same expectation are uniformly added to the pre-order set. When accumulating the propagation range, these nodes will receive the same number of propagation marks.
[0059] Directed Acyclic Graph g s It is easier to get s and q, for example, there is a g s It contains 4 nodes, and the edges are a->b, b->c, and b->d. Then, s of a is {b,c,d}, s of b is {c,d}, and s of c and d are empty sets. Q represents the ordered set of nodes where the propagation occurs, which is {a, b, c, d} in the previous example. This order facilitates the reverse traversal in step 4 based on the stack principle.
[0060] Step 4: Directed acyclic graph g based on event propagation s , Preorder Set Successor Set And the propagation node queue Q, use the dynamic programming method to calculate the propagation centrality score of each node, and rank the importance of each node from high to low according to the propagation centrality score. The specific process is as follows:
[0061] First, traverse each node in the network as the source node, use the breadth-first search strategy to build the expected propagation path graph, and dynamically update the successor set of the node, so as to accumulate the propagation centrality of the node. Dynamic programming calculation node v i Importance score The state transition formula is:
[0062]
[0063] like Figure 3 As shown, Figure 3The example shows the node propagation centrality ranking of a small social network. Through the calculation of the above steps of this embodiment, the specific ranking can be obtained (the propagation model is the SIR model, the infection rate β=0.9, and the recovery rate γ=1.0) as shown in Table 1:
[0064] Table 1 Ranking table
[0065] 1:(2,6) 2:(4) 3:(0,3,5,8) 4:(1,7)
[0066] The preferred embodiments of the present invention are described in detail above in conjunction with the accompanying drawings. The embodiments described in the present invention are merely descriptions of the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. The various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction, and such combinations should also be regarded as the contents disclosed in the present disclosure as long as they do not violate the concept of the present invention. In order to avoid unnecessary repetition, the present invention will not further describe various possible combinations.
[0067] The present invention is not limited to the specific details of the above-mentioned embodiments. Within the technical concept of the present invention and without departing from the design concept of the present invention, various modifications and improvements made to the technical solution of the present invention by technical personnel in this field should fall within the protection scope of the present invention. The technical contents for which protection is sought in the present invention have been fully recorded in the claims.
Claims
1. A social network node importance ranking method based on inherent propagation mode, characterized in that: The following steps are involved: Step 1: Classify the propagation models of fixed propagation modes on social networks, thereby dividing the propagation models on social networks into deterministic propagation models and uncertain propagation models; Step 2: Perform importance sampling of the proposed distribution of event propagation on the uncertainty propagation model in step 1 by using the Monte Carlo simulation method, and calculate the linear propagation matrix of the deterministic propagation model in step 1 according to the proposed distribution of importance sampling; Step 3: Construct a directed acyclic graph of event propagation based on the linear propagation matrix obtained in step 2; Step 4: Based on the directed acyclic graph, the dynamic programming method is used to calculate the propagation centrality score of each node, and the importance of each node is ranked from high to low according to the propagation centrality score.
