Social network rumor suppression method and system based on spherical domain index

Abstract

The invention discloses a social network rumor suppression method and system based on a sphere domain index, and relates to the technical field of social network information processing. For the NP-hard characteristic of a rumor suppression problem under a competitive independent cascade model and the defect of high calculation cost of a traditional algorithm, efficient response under a multi-query scene is achieved through three-step progressive algorithm design. The method comprises the following steps: firstly, proposing an HMP-TC algorithm, replacing a forward sampling process in mixed sampling with typical cascade, and depicting a stable structure of rumor propagation; secondly, proposing an HMP-LB algorithm, constructing a landmark index structure based on point coverage, only maintaining influence ball domain information for index nodes, and correcting deviation through a probability weight; and finally, proposing a PTC-LB algorithm, introducing a protection sphere domain index formed by protection typical cascade and protection relation samples, and realizing non-sampling rapid optimization in a query stage. According to the method, on the premise of maintaining 90%-95% of rumor blocking effect, the single query speed is increased by 10-30 times, part of scenes can reach two orders of magnitude, and the quick response problem of rumor suppression in a large-scale social network is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of social network information processing technology, specifically to a method and system for rapidly suppressing rumors in large-scale social networks, applicable to minimizing the spread of rumors in multi-query scenarios. Background Technology

[0002] With the rapid development of social networks, the efficiency of information dissemination has greatly improved, but it has also facilitated the spread of rumors and misinformation. The rapid spread of rumors can seriously harm personal lives and social stability, thus necessitating efficient rumor suppression technologies.

[0003] In the competitive independent cascade model, the rumor suppression problem is defined as follows: given a set of rumor seeds and a positive seed budget constraint, select the optimal positive seed node to minimize the scope of the rumor's impact. This problem has been proven to be NP-hard, and its objective function is #P-hard. This leads to high computational costs for traditional algorithms (such as greedy algorithms based on Monte Carlo sampling and hybrid sampling algorithms) in large-scale social network scenarios, making it difficult to respond quickly to multiple queries.

[0004] In existing technologies, research on maximizing influence has proposed backsampling algorithms and the concept of typical cascading (influence sphere), providing new ideas for rumor suppression. However, directly applying these techniques has the following drawbacks: 1) Hybrid sampling algorithms require a large amount of forward and backward sampling for each query, resulting in slow response speed; 2) Typical cascading calculations across all nodes lead to excessive index building overhead; 3) The lack of a structured characterization of the protected node set makes it impossible to completely avoid the sampling process during the query phase. Therefore, it is necessary to design an efficient rumor suppression scheme based on index optimization, which can significantly improve query response speed while ensuring suppression effectiveness. Summary of the Invention

[0005] This invention aims to overcome the aforementioned shortcomings of the prior art by providing a method and system for suppressing rumors on social networks based on a spherical index. By constructing a spherical index structure offline, it enables rapid response to online queries, thus solving the problem of rumor suppression efficiency in multi-query scenarios within large-scale social networks.

[0006] The core idea of ​​this invention is to shift the main computational overhead of rumor suppression to the offline stage through an architecture of "offline index construction + online fast query," while the online query only selects the optimal positive seed node through set operations and greedy selection. The overall process of the method includes three core steps: social network modeling, offline construction of the spherical index, and online query processing. These steps are closely linked to ensure a balance between suppression effectiveness and computational efficiency.

[0007] The first aspect of this invention relates to a method for suppressing rumors on social networks based on spherical indexing, comprising the following steps: (1) Constructing a social network model: The social network is abstracted as a directed graph G=(V,E), where V is the set of nodes and E is the set of edges. Each edge (u,v)∈E is assigned an independent propagation probability p. uv A competitive independent cascade model was used to characterize the spread of rumors and positive information; (2) Construct the sphere index structure offline, including the following sub-steps: (2.1) The typical cascade index construction algorithm is adopted to sample the possible world of the social network nodes. Through the construction of strongly connected component index, propagation cascade calculation and Jaccard median solution, the typical cascade (influence sphere) of each node is obtained. (2.2) Constructing a landmark index set L based on the point coverage idea V, satisfying that for any node x∈V, either x∈L or all its outgoing neighbors. (x) L only maintains typical cascading information for nodes in L; (2.3) For each node in the landmark index set L, generate a protection candidate set sample by back sampling, solve the Jaccard median to obtain the typical protection cascade (protection sphere), and record the protection relationship sample to form a complete protection sphere index; (3) Online processing of rumor suppression query: Given the rumor seed set Sr and the positive seed budget k, the typical concatenation and protection typical concatenation of relevant landmark nodes are merged through set operation, and the optimal positive seed set S is selected based on the greedy algorithm to minimize the scope of the rumor's impact.

[0008] Preferably, step (1) specifically includes: (1.1) At the initial moment, the rumor seed set Sr is negatively activated, and the positive seed set S is positively activated; (1.2) Each activated node will have a probability p at subsequent time steps. uv Attempt to activate an inactive neighbor node, ensuring the activation status matches that of the source node; (1.3) If a node is activated by both positive and negative activations at the same time, it is determined to be negative activation; the propagation process terminates when no new nodes are activated.

[0009] Furthermore, the solution process for the typical cascade described in step (2.1) includes: (2.1.1) Sampling of social networks There are 10 possible worlds, and each possible world is solved by Kosaraju or Tarjan's algorithm to find the strongly connected components and reconstruct it into a directed acyclic graph; (2.1.2) For each node, obtain its propagation cascade in all possible worlds through breadth-first search or depth-first search; (2.1.3) Using the Jaccard median algorithm, the typical cascade C is obtained by solving the propagation cascade. v The desired Jaccard distance ρ(G,s,C) = E[d] is minimized. J (R s (G),C)], where R s Let (G) be the set of reachable nodes s in possible worlds G, and let d be the Jaccard distance. J (R s (G),C)=|R s (G)⊕C| / |R s (G)∪C|.

[0010] Furthermore, the construction of the landmark index set L in step (2.2) adopts a 2-approximation algorithm: iteratively select uncovered edges, add the two endpoints of the edges to L and delete all adjacent edges, until all edges are covered, ensuring that |L|≤2|L |, where L This is the minimum point coverage.

[0011] Furthermore, the solution process for the protection of typical cascades described in step (2.3) includes: (2.3.1) For landmark node x and its typical cascade C x In multiple possible worlds from C x Back-sampling is performed from each node to obtain the protection candidate set sample {P}. x 1 ,P x 2 , ... ,P x }; (2.3.2) Define the protection cost function ρ for landmark node s. s (P)=E[d J (P s g ,P)], where P s (g) represents the potential world g in which C can be protected. s The set of nodes; (2.3.3) Solve for the Jaccard median of the protection cost function to obtain the typical protection cascade P. x And record the protection relationship sample R x ={(z,y)|z∈P x ,y∈C x ,dis g (z,y) <dis g(x,y)}, where dis g (x,y) represents the distance x travels to y in possible world g.

[0012] Preferably, the selection process of the optimal positive seed set in step (3) includes: (3.1) For each node x in the rumor seed set Sr, if x∈L, then it is directly included in the relevant landmark set L. Sr Otherwise, move it to the neighboring area. (x) included in L Sr ; (3.2) Merge L Sr A typical cascade of nodes in the middle yields C Sr The typical cascade protection obtained by merging protections yields P. Sr The merged protection relationship samples yield R Sr ; (3.3) Regarding C Sr The weight w(z) is calculated for each node, where x∈Sr∩L and z∈C. x When w(z) = 1, otherwise for x ∈ Sr\L, y ∈ (x) and z∈C y ,

[0013] (3.4) Based on the greedy algorithm to maximize the benefit of empirical coverage, select a positive seed set S of size no more than k, where the benefit function is: .

[0014] A second aspect of the present invention relates to a social network rumor suppression system based on a sphere index, comprising: Network Modeling Module: Used to abstract social networks into a directed graph model, configure edge propagation probabilities, and define competitive independent cascading propagation rules; Index building module: includes typical cascaded calculation unit, landmark index building unit and protection sphere building unit, used to generate the index structure of influence sphere and protection sphere offline; Storage module: Used to store social network topology data, propagation probability parameters, sphere index structure, and protection relationship samples; The query processing module receives the rumor seed set and budget parameters, calls the index data through set operations, and outputs the optimal positive seed set using a greedy algorithm.

[0015] This invention addresses the NP-hard nature of rumor suppression in competitive independent cascade models and the high computational cost of traditional algorithms. It employs a three-step progressive algorithm design to achieve efficient response in multi-query scenarios. While maintaining a 90%–95% rumor blocking effect, this invention increases the speed of a single query by 10–30 times, and in some scenarios by nearly two orders of magnitude, effectively solving the problem of rapid response in rumor suppression on large-scale social networks.

[0016] The advantages of this invention are: it shifts the main computational overhead of rumor suppression to the offline stage, and online queries only use set operations and greedy selection to select the optimal positive seed node, which greatly improves the query response speed. Attached Figure Description

[0017] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be described below with reference to the accompanying drawings.

[0019] Example 1

[0020] This embodiment relates to a method for suppressing rumors on social networks based on spherical indexes, such as... Figure 1 This includes the following steps: (1) Social network modeling Social networks are abstracted as directed graphs G=(V,E), where nodes V represent users or information propagation entities, and edges E represent information propagation relationships between nodes. Each edge (u,v) is assigned an independent propagation probability p_uv, and a competitive independent cascading model is used to characterize the propagation process. (1.1) At the initial moment, the rumor seed set Sr is negatively activated, and the positive seed set S is positively activated; (1.2) Each activated node will have a probability p at subsequent time steps. uv Attempt to activate an inactive neighbor node, ensuring the activation status matches that of the source node; (1.3) If a node is activated by both positive and negative activations at the same time, it is determined to be negative activation; the propagation process terminates when no new nodes are activated.

[0021] (2) Offline construction of sphere index

[0022] Index construction is the core innovation of this invention, which includes three levels: typical cascading calculation, landmark index construction, and protection sphere construction, thereby gradually optimizing index efficiency and query performance.

[0023] (2.1) Calculation of typical cascades (affected sphere)

[0024] Typical cascades are used to characterize stable structures of node propagation, and are obtained by solving using the Jaccard median. The specific process is as follows: 1. Sampling There are 10 possible worlds, each generated by randomly selecting the existence of edges; 2. Solve for strongly connected components for each possible world and reconstruct it into a directed acyclic graph to reduce computational complexity; 3. For each node, obtain its propagation cascade in all possible worlds (the set of reachable nodes); 4. Using the Jaccard median algorithm, find the set of nodes with the smallest expected Jaccard distance to all propagation cascades, and use this set as the typical cascade C for that node. v .

[0025] The advantage of typical cascading is that it can stably characterize the propagation range of nodes with only a constant number of samples, which greatly reduces the amount of data compared with direct sampling, while maintaining the representativeness of the propagation structure.

[0026] (2.2) Landmark Index Construction

[0027] To reduce the construction overhead of the full node index, a landmark index set L is constructed based on the point cover concept: 1. Iteratively construct a point cover using a 2-approximation algorithm: Select an uncovered edge, add its two endpoints to L, delete all adjacent edges, until all edges are covered; 2. The set of landmarks L satisfies: any node x∈V or x∈L, or all its outgoing neighbors. (x) L ensures that the propagation impact of non-landmark nodes can be indirectly characterized through neighboring landmark nodes; 3. Typical concatenations are calculated and stored only for nodes in the landmark set L. The influence range of non-landmark nodes is obtained by combining the typical concatenations of their neighboring landmarks.

[0028] This design reduces the size of the index node from O(n) to O(2|L). |)(L (Minimum point coverage) significantly reduces the time and space overhead of index building.

[0029] (2.3) Construction of the protective sphere

[0030] To eliminate the backsampling overhead during the query phase, the concept of a typical protection cascade (protection sphere) is introduced to characterize the set of candidate forward nodes that can effectively protect the target node: 1. For each landmark node x and its typical cascade C x ,sampling A possible world; 2. From C xStarting from each node, backsampling is performed to obtain a candidate set of samples P that can protect that node. x i ; 3. Solve for the protection cost function ρ s (P)=E[d J (P s The Jaccard median of [(g),P)] is used to obtain the protection typical cascade P x ; 4. Record the protected relationship sample R x ={(z,y)|z∈P x ,y∈C x ,dis g (z,y) <dis g (x,y)} describes the correspondence between candidate nodes and protected nodes.

[0031] The construction of the protected sphere index advances the backsampling process of the query phase to the offline stage, providing support for fast queries without sampling.

[0032] (3) Online query processing

[0033] When a rumor suppression query is received (inputting a rumor seed set S_r and a positive seed budget k), the processing flow is as follows: (3.1) Determine the relevant landmark set L Sr : To S r For each node x in the array, if x∈L, it is directly included; otherwise, it is its outgoing neighbor. (x) included; (3.2) Merge index data: Merge L Sr A typical cascade of nodes in the middle yields C Sr (Potential impact of rumors), merging typical cascade protection yields P Sr (Candidate positive node set), merge the protection relationship samples to obtain R Sr ; (3.3) Calculate the node weights: For C Sr For a node z, if there exists x∈Sr∩L and z∈C x When w(z) = 1, otherwise for x ∈ Sr\L, y ∈ (x) and z∈C y , Correct index deviations; (3.4) Greedy selection of the optimal positive seed: based on experience coverage benefits Iteratively select the node with the highest marginal benefit and add it to S until the set size reaches k.

[0034] This process requires no Monte Carlo sampling and is completed solely through set operations and greedy selection, significantly improving query response speed.

[0035] Example 2

[0036] This embodiment relates to a social network rumor suppression system based on spherical indexing, used to implement the method of Embodiment 1, including: Network Modeling Module: Used to abstract social networks into a directed graph model, configure edge propagation probabilities, and define competitive independent cascading propagation rules; Index building module: includes typical cascaded calculation unit, landmark index building unit and protection sphere building unit, used to generate the index structure of influence sphere and protection sphere offline; Storage module: Used to store social network topology data, propagation probability parameters, sphere index structure, and protection relationship samples; The query processing module receives the rumor seed set and budget parameters, calls the index data through set operations, and outputs the optimal positive seed set using a greedy algorithm.

[0037] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for suppressing rumors on social networks based on spherical indexing, characterized in that, Includes the following steps: (1) Constructing a social network model: The social network is abstracted as a directed graph G=(V,E), where V is the set of nodes and E is the set of edges. Each edge (u,v)∈E is assigned an independent propagation probability p. uv A competitive independent cascade model was used to characterize the spread of rumors and positive information; (2) Construct the sphere index structure offline, including the following sub-steps: (2.1) The typical cascade index construction algorithm is adopted to sample the possible world of the social network nodes. Through the construction of strongly connected component index, propagation cascade calculation and Jaccard median solution, the typical cascade (influence sphere) of each node is obtained. (2.2) Constructing a landmark index set L based on the point coverage idea V, satisfying that for any node x∈V, either x∈L or all its outgoing neighbors. (x) L only maintains typical cascading information for nodes in L; (2.3) For each node in the landmark index set L, generate a protection candidate set sample by back sampling, solve the Jaccard median to obtain the typical protection cascade (protection sphere), and record the protection relationship sample to form a complete protection sphere index; (3) Online processing of rumor suppression query: Given the rumor seed set Sr and the positive seed budget k, the typical concatenation and protection typical concatenation of relevant landmark nodes are merged through set operation, and the optimal positive seed set S is selected based on the greedy algorithm to minimize the scope of the rumor's impact.

2. The method as described in claim 1, characterized in that, Step (1) specifically includes: (1.1) At the initial moment, the rumor seed set Sr is negatively activated, and the positive seed set S is positively activated; (1.2) Each activated node will have a probability p at subsequent time steps. uv Attempt to activate an inactive neighbor node, ensuring the activation status matches that of the source node; (1.3) If a node is activated by both positive and negative activations at the same time, it is determined to be negative activation; the propagation process terminates when no new nodes are activated.

3. The method according to claim 1, characterized in that, The solution process for the typical cascade described in step (2.1) includes: (2.1.1) Sampling of social networks There are 10 possible worlds, and each possible world is solved by Kosaraju or Tarjan's algorithm to find the strongly connected components and reconstruct it into a directed acyclic graph; (2.1.2) For each node, obtain its propagation cascade in all possible worlds through breadth-first search or depth-first search; (2.1.3) Using the Jaccard median algorithm, the typical cascade C is obtained by solving the propagation cascade. v The desired Jaccard distance ρ(G,s,C) = E[d] is minimized. J (R s (G),C)], where R s Let (G) be the set of reachable nodes s in possible worlds G, and let d be the Jaccard distance. J (R s (G),C)=|R s (G)⊕C| / |R s (G)∪C|.

4. The method according to claim 3, characterized in that, The construction of the landmark index set L in step (2.2) adopts a 2-approximation algorithm: iteratively select uncovered edges, add the two endpoints of the edges to L and delete all adjacent edges until all edges are covered, ensuring that |L|≤2|L |, where L This is the minimum point coverage.

5. The method according to claim 3, characterized in that, The solution process for the typical protection cascade described in step (2.3) includes: (2.3.1) For landmark node x and its typical cascade C x In multiple possible worlds from C x Back-sampling is performed from each node to obtain the protection candidate set sample {P}. x 1 ,P x 2 , ... ,P x }; (2.3.2) Define the protection cost function ρ for landmark node s. s (P)=E[d J (P s g ,P)], where P s (g) represents the potential world g in which C can be protected. s The set of nodes; (2.3.3) Solve for the Jaccard median of the protection cost function to obtain the typical protection cascade P. x And record the protection relationship sample R x ={(z,y)|z∈P x ,y∈C x ,dis g (z,y) <dis g (x,y)}, where dis g (x,y) represents the distance x travels to y in possible world g.

6. The method according to claim 1, characterized in that, The selection process for the optimal positive seed set in step (3) includes: (3.1) For each node x in the rumor seed set Sr, if x∈L, then it is directly included in the relevant landmark set L. Sr Otherwise, move it to the neighboring area. (x) included in L Sr ; (3.2) Merge L Sr A typical cascade of nodes in the middle yields C Sr The typical cascade protection obtained by merging protections yields P. Sr The merged protection relationship samples yield R Sr ; (3.3) Regarding C Sr The weight w(z) is calculated for each node, where x∈Sr∩L and z∈C. x When w(z) = 1, otherwise for x ∈ Sr\L, y ∈ (x) and z∈C y , (3.4) Based on the greedy algorithm to maximize the benefit of empirical coverage, select a positive seed set S of size no more than k, where the benefit function is: .

7. A social network rumor suppression system based on spherical indexing, characterized in that, include: Network Modeling Module: Used to abstract social networks into a directed graph model, configure edge propagation probabilities, and define competitive independent cascading propagation rules; Index building module: includes typical cascaded calculation unit, landmark index building unit and protection sphere building unit, used to generate the index structure of influence sphere and protection sphere offline; Storage module: Used to store social network topology data, propagation probability parameters, sphere index structure, and protection relationship samples; The query processing module receives the rumor seed set and budget parameters, calls the index data through set operations, and outputs the optimal positive seed set using a greedy algorithm.

Images