A rumor propagation traceability method and system based on community clustering

Abstract

The application discloses a rumor propagation tracing method and system based on community clustering, comprising: performing community clustering on an infection subgraph to be traced to obtain a corresponding community relationship graph; performing estimation and analysis on rumor propagation possibility of each first community in the community relationship graph through an ML estimator of a node source corresponding to the community relationship graph, and determining a source community in the community relationship graph; performing correction on the source community according to a plurality of boundary nodes of the source community to obtain a corresponding community expansion graph, and performing estimation and analysis on rumor propagation possibility of each first node in the community expansion graph through an ML estimator of a node source corresponding to the community expansion graph, and determining a source node in the community expansion graph, so as to realize rumor propagation tracing of the infection subgraph. The application performs community clustering on the infection subgraph to be traced, so that the community relationship of the infection subgraph is optimized, the order of magnitude of the infection nodes needing to be traced is reduced, and the overall tracing efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of rumor information dissemination and tracing technology, and in particular to a rumor dissemination and tracing method and system based on community clustering. Background Technology

[0002] In today's world, with increasingly advanced information technology and transportation, the ways in which people and devices interact are becoming more diverse and frequent. The development of the internet and communication technologies has made the rapid and widespread dissemination of information worldwide possible, while rapid infrastructure development and more diverse, safer, and faster transportation methods have enabled communication and contact between people around the world. However, development always comes with challenges. While the convenience of communication and transportation has facilitated development, it has also provided fertile ground for the spread of rumors and viruses that have a negative impact on society. Timely detection and prevention of the spread of rumors and viruses are of great significance for maintaining social order, protecting the personal and property safety of citizens, and carrying out correct and timely disease prevention and treatment.

[0003] Currently, methods for preventing the spread of rumors can be broadly categorized into two main types: 1) Identifying the information itself through its characteristics and other information, promptly detecting it in social networks, and blocking its spread through methods such as deletion and banning to minimize losses. This can be achieved through natural language processing and manual detection; 2) Estimating the source of the information based on known network state information using information source detection algorithms, locating the source, and suppressing its spread. For the second type of method, a rumor centrality algorithm based on message passing is typically used. The rumor centrality of each node in the tree structure needs to be calculated within a certain time. Therefore, a typical tree source estimator needs to calculate a BFS spanning tree for each node in the infection graph that could potentially be the source, and then calculate the RC value of the root node in that BFS spanning tree using the RC message passing algorithm to estimate the source node of the infection graph. However, although the above rumor centrality algorithm simplifies the problem to a source tracing problem on a tree, it loses a lot of structural information in the original network, which may have a certain impact on the accuracy of the algorithm. Furthermore, when processing the graph structure, the rumor centrality algorithm calculates the BFS tree for each node and then calculates the rumor centrality of the nodes in the tree, which requires... While it has a relatively low time complexity, in actual computation, it often requires performing a calculation on all nodes in the BFS tree for each node, resulting in a huge amount of computation and limiting the efficiency of tracing the source. Summary of the Invention

[0004] This invention provides a method and system for tracing the source of rumor propagation based on community clustering, which optimizes the community and node relationships in the infection subgraph, reduces the number of infection nodes that need to be traced later, and improves the overall tracing efficiency.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for tracing the source of rumor propagation based on community clustering, including:

[0006] Community clustering is performed on the infection subgraph for source tracing to obtain the corresponding community relationship graph;

[0007] The ML estimator of the node sources corresponding to the community relationship graph is used to estimate and analyze the probability of rumor propagation in each first community in the community relationship graph, thereby determining the source community in the community relationship graph; wherein, the source community refers to the first community in the community relationship graph that is most likely to spread rumor information first; the ML estimator of the node sources corresponding to the community relationship graph is specifically as follows:

[0008]

[0009] In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score;

[0010] Based on several boundary nodes of the source community, the source community is modified to obtain a corresponding community unfolded graph. Then, using the ML estimator of the node source corresponding to the community unfolded graph, the probability of rumor propagation of each first node in the community unfolded graph is estimated and analyzed to determine the source node in the community unfolded graph, thereby achieving the tracing of rumor propagation in the infected subgraph. Here, the source node refers to the first node in the community unfolded graph most likely to transmit rumor information first. The ML estimator of the node source corresponding to the community unfolded graph is specifically as follows:

[0011]

[0012] In the formula, Community unfolding diagram China and Israel A BFS tree with the root node as its root node. Indicates traversal The order in which nodes are added to the BFS tree; Community unfolding diagram The source node in; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score.

[0013] By implementing this embodiment of the invention, community clustering is performed on the infected subgraph to be traced to obtain the corresponding community relationship graph. The community and node relationships in the infected subgraph are gradually optimized to reduce the number of infected nodes requiring tracing, avoid some redundant and ineffective calculations, and facilitate subsequent estimation and analysis of source communities and source nodes using a node-source ML estimator. This improves the overall tracing efficiency and allows for application to larger-scale real-world networks, thus expanding its applicable scenarios. Furthermore, by using a node-source ML estimator, the source communities in the community relationship graph are first determined, then the source communities are corrected, and further estimations are made on each first node in the corrected community unfolded graph. Finally, the source nodes in the community unfolded graph are determined as the result of rumor propagation tracing for the infected subgraph, rather than directly estimating and analyzing the community clustering results of the infected subgraph, fundamentally improving the accuracy of rumor propagation tracing.

[0014] As a preferred approach, the infection subgraph to be traced is subjected to community clustering to obtain the corresponding community relationship graph, specifically as follows:

[0015] Using the Louvain algorithm, each second node in the infected subgraph to be analyzed is assigned a different community number to form several second communities. Each second node is traversed, and the modular gain produced by removing the currently traversed second node from its second community and placing it in the second communities of its neighboring nodes is evaluated. Then, it is determined whether each of the modular gains corresponding to the currently traversed second node is greater than 0.

[0016] If any of the modular gains corresponding to the currently traversed second node is greater than 0, then the currently traversed second node is removed from the second community where the currently traversed second node is located and placed in the second community that produces the largest modular gain, so as to update all the second communities in the infected subgraph.

[0017] If all the modular gains corresponding to the currently traversed second node are not greater than 0, then the currently traversed second node will be kept in the second community where the currently traversed second node is located.

[0018] When all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, each of the second communities in the current infected subgraph is taken as a third node to form a corresponding weighted network, and the Louvain algorithm is used to perform community clustering on the weighted network to obtain the community relationship graph.

[0019] A preferred embodiment of the present invention introduces the Louvain algorithm to analyze the modular gain generated by node movement between communities in the infected subgraph, and maintains the node movement behavior that can generate the maximum positive modular gain. When a node movement does not generate any positive modular gain, the movement of the node is canceled, so as to gradually optimize the community relationship in the infected subgraph, realize community clustering, and reduce the number of infected nodes that need to be traced.

[0020] As a preferred embodiment, when all modular gains corresponding to all second nodes in the infected subgraph are not greater than 0, each second community in the current infected subgraph is taken as a third node to form a corresponding weighted network, and the Louvain algorithm is used to perform community clustering on the weighted network to obtain the community relationship graph, specifically including:

[0021] When all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, each of the second communities in the current infected subgraph is taken as a third node, and the weight of the link between each third node is determined according to the weight of the link between all the second nodes in each of the second communities in the current infected subgraph. Then, the corresponding weighted network is established according to all the third nodes and the weight of the link between each third node.

[0022] Using the Louvain algorithm, each third node in the weighted network is assigned a different community number to form several third communities. Each third node is traversed, and the modular gain resulting from removing the currently traversed third node from its current third community and placing the currently traversed third node in the third communities of its neighboring nodes is evaluated. Then, it is determined whether each modular gain corresponding to the currently traversed third node is greater than 0.

[0023] If any of the modular gains corresponding to the currently traversed third node is greater than 0, then the currently traversed third node is removed from the third community where the currently traversed third node is located and placed in the third community that generates the largest modular gain, so as to update the third community in the weighted network.

[0024] If all the modular gains corresponding to the currently traversed third node are not greater than 0, then the currently traversed third node will be retained in the third community where the currently traversed third node is located.

[0025] When all modular gains corresponding to all third nodes in the weighted network are not greater than 0, the current weighted network is used as the community relationship graph, and each third community in the current weighted network is used as the first community in the community relationship graph.

[0026] A preferred embodiment of the present invention is to heuristically extend the original community of the currently traversed node to the nodes adjacent to the community, and use some structural information of the original community and nearby nodes to deduce the information propagation efficiency between the original community and nearby nodes, so as to optimize its community relationship graph and reduce the amplification of the source node tracing error caused by the source community estimation error.

[0027] As a preferred embodiment, the step of modifying the source community based on several boundary nodes to obtain a corresponding community unfolding graph specifically includes:

[0028] Obtain the boundary nodes whose distance from each member node in the source community is a preset value; wherein, the boundary nodes are not in the source community;

[0029] Expand all members within the source community and all boundary nodes of the source community to obtain the corresponding community expansion diagram.

[0030] In a preferred embodiment of the present invention, after determining the source community in the infection subgraph, the source community is heuristically expanded instead of directly expanding the traced source community. This optimizes the effectiveness of the community expansion graph and makes the community expansion graph more intuitively show the information propagation relationship between the nodes in the infection subgraph.

[0031] To address the same technical problem, embodiments of the present invention also provide a rumor propagation tracing system based on community clustering, comprising:

[0032] The data acquisition module is used to perform community clustering on the infection subgraph to be traced in order to obtain the corresponding community relationship graph;

[0033] The first source tracing module is used to estimate and analyze the probability of rumor propagation in each first community in the community relationship graph by using the ML estimator of the node source corresponding to the community relationship graph, and to determine the source community in the community relationship graph; wherein, the source community refers to the first community in the community relationship graph that is most likely to spread the rumor information first; the ML estimator of the node source corresponding to the community relationship graph is specifically:

[0034]

[0035] In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score;

[0036] The second source tracing module is used to correct the source community based on several boundary nodes of the source community to obtain a corresponding community unfolded graph. Then, using the ML estimator of the node source corresponding to the community unfolded graph, it estimates and analyzes the probability of rumor propagation for each first node in the community unfolded graph to determine the source node in the community unfolded graph, thereby achieving source tracing of rumor propagation in the infected subgraph. The source node refers to the first node in the community unfolded graph most likely to transmit rumor information first. The ML estimator of the node source corresponding to the community unfolded graph is specifically:

[0037]

[0038] In the formula, Community unfolding diagram China and Israel A BFS tree with the root node as its root node. Indicates traversal The order in which nodes are added to the BFS tree; Community unfolding diagram The source node in; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score.

[0039] As a preferred embodiment, the data acquisition module specifically includes:

[0040] The first clustering unit uses the Louvain algorithm to assign different community numbers to each second node in the infected subgraph to be analyzed, forming several second communities. It then iterates through each second node, evaluating the modular gain resulting from removing the currently traversed second node from its current second community and placing it in the second communities of its neighbors. It then determines whether each modular gain corresponding to the currently traversed second node is greater than 0. If any modular gain corresponding to the currently traversed second node is greater than 0, the currently traversed second node is removed from its current second community and placed in the second community that generates the largest modular gain, thus updating all second communities in the infected subgraph. If none of the modular gains corresponding to the currently traversed second node are greater than 0, the currently traversed second node is retained in its current second community.

[0041] The second clustering unit is used to take each of the second communities in the current infected subgraph as the third node when all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, so as to form a corresponding weighted network, and use the Louvain algorithm to perform community clustering on the weighted network to obtain the community relationship graph.

[0042] As a preferred embodiment, the second clustering unit specifically includes:

[0043] The weighted network construction subunit is used to, when all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, take each of the second communities in the current infected subgraph as the third node, determine the weight of the link between each of the third nodes according to the weight of the link between all the second nodes in each of the second communities in the current infected subgraph, and then establish the corresponding weighted network according to all the third nodes and the weight of the link between each of the third nodes.

[0044] The community clustering subunit is used to assign a different community number to each third node in the weighted network using the Louvain algorithm to form several third communities. It then traverses each third node, evaluates the modularity gain resulting from removing the currently traversed third node from its current third community and placing it in the third communities of its neighbors, and then determines whether each modularity gain corresponding to the currently traversed third node is greater than 0. If any modularity gain corresponding to the currently traversed third node is greater than 0, then the currently traversed third node... Remove the third node from the third community where the currently traversed third node is located and place it in the third community that produces the largest modular gain to update the third community in the weighted network; if all the modular gains corresponding to the currently traversed third node are not greater than 0, then keep the currently traversed third node in the third community where the currently traversed third node is located; when all the modular gains corresponding to all the third nodes in the weighted network are not greater than 0, use the current weighted network as the community relationship graph, and use each of the third communities in the current weighted network as the first communities in the community relationship graph.

[0045] As a preferred embodiment, the ML estimator for the node source corresponding to the community relationship graph is specifically:

[0046]

[0047] In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score. Attached Figure Description

[0048] Figure 1 This is a flowchart illustrating a rumor propagation tracing method based on community clustering provided in Embodiment 1 of the present invention.

[0049] Figure 2: This is a schematic diagram of a rumor propagation tracing system based on community clustering provided in Embodiment 1 of the present invention. Detailed Implementation

[0050] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0051] Example 1:

[0052] Please refer to Figure 1 This invention provides a method for tracing the source of rumors based on community clustering. The method includes steps S1 to S3, each of which is detailed below:

[0053] Step S1, the infected subgraph to be traced Perform community clustering to obtain the corresponding community relationship graph. .

[0054] In this embodiment, the infection subgraph to be traced It is a collection of A weighted network of second nodes.

[0055] As a preferred embodiment, step S1 includes steps S11 to S15, and the specific details of each step are as follows:

[0056] Step S11: Using the Louvain algorithm, for each second node in the infection subgraph to be traced... Different community numbers are assigned to form several second communities. In this initial partitioning scheme, the number of nodes in the weighted network is the same as the number of communities.

[0057] It should be noted that the Louvain algorithm can find highly modular partitions in large networks in a short time, and unlike some other community detection algorithms, the limitation of network size faced by this algorithm comes from the limited storage capacity of computers, rather than the computation time.

[0058] Step S12, traverse each second node Please refer to equation (1) to evaluate whether the current traversal of the second node will be successful. Starting from the current second node, traverse... Remove the second node from the second community and iterate through the second node. Placed in the second node of the current traversal Each neighbor node The second community Modular gain generated in And determine the second node being traversed. Corresponding modular gains Is it greater than 0?

[0059] （1）

[0060] In the formula, Neighboring nodes Community The sum of the weights of the internal links, It connects to neighboring nodes. The second community The sum of the weights of the edges of the middle node. It is connected to the second node. The sum of the weights of the edges, It is a node to neighboring nodes Community The sum of the weights of the links in the middle node. It is the sum of the weights of all links in the infected subgraph. Additionally, when a node... When removed from its community, a similar expression can be used to evaluate the changes to the modularity.

[0061] It should be noted that the quality of a community's partitioning is usually determined by the partitioning's modularity. To measure modularity The greater the change (i.e., modular gain), the better the partitioning quality of the current community.

[0062] Step S13, if the current traversal is the second node any corresponding modular gain If the value is greater than 0, then the current traversal will be the second node. Starting from the current second node, traverse... Remove and place the second community that produces the largest modular gain in the second community to update all second communities in the infected subgraph.

[0063] Step S14, if the current traversal is the second node All corresponding modular gains If none of them are greater than 0, then the current traversal of the second node will be performed. Keep the second node in the current traversal It is located in the second community.

[0064] Step S15, when all second nodes in the infected subgraph are infected All corresponding modular gains When none of the values ​​are greater than 0, it indicates that none of the second nodes need to be moved, and moving any second node cannot improve modularity. At this point, the first phase of the algorithm stops, and each second community in the current infected subgraph is the community discovered in the first phase. Then, each second community in the current infected subgraph is used as the third node to form a corresponding new weighted network. The Louvain algorithm is then used to perform community clustering on the weighted network to obtain the community relationship graph.

[0065] As a preferred embodiment, step S15 includes steps S151 to S155, and the specific details of each step are as follows:

[0066] Step S151: When all modular gains corresponding to all second nodes in the infected subgraph are not greater than 0, each second community in the current infected subgraph is taken as a third node, and the weight of the link between each third node is determined according to the sum of the weights of the links between all second nodes in each second community in the current infected subgraph. Then, a corresponding weighted network is established according to all third nodes and the weights of the links between each third node.

[0067] Step S152: Using the Louvain algorithm, assign different community numbers to each third node in the weighted network to form several third communities, and traverse each third node (see Equation (1)). Substitute the third node into the second node in Equation (1). And substitute the third community where each neighbor node of the currently traversed third node is located into the second community where each neighbor node of the currently traversed second node is located in equation (1). To evaluate the modular gain resulting from removing the currently traversed third node from its own third community and placing it in the third communities of its neighboring nodes, we then determine whether each modular gain corresponding to the currently traversed third node is greater than 0.

[0068] Step S153: If any modular gain corresponding to the currently traversed third node is greater than 0, then remove the currently traversed third node from the third community where the currently traversed third node is located and place it in the third community that produces the largest modular gain, so as to update the third community in the weighted network.

[0069] Step S154: If all modular gains corresponding to the currently traversed third node are not greater than 0, then the currently traversed third node is kept in the third community where the currently traversed third node is located.

[0070] Step S155: When all modular gains corresponding to all third nodes in the weighted network are not greater than 0, the current weighted network is used as a community relationship graph. And each third community in the current weighted network is used as a community relationship diagram. The various first communities in the middle.

[0071] in, This represents the set of community nodes after completing the community clustering in step S1. This represents the set of edges between communities after completing step S1's community clustering. The edges between clustered communities often have certain weights, which are used to represent the number of edges between two communities before clustering or the degree of connection between communities.

[0072] It should be noted that after steps S141 to S145, the number of the original community (i.e., the second community set in step S11) will decrease in each transmission, so most of the calculation time is used in the first transmission.

[0073] Step S2: Using the ML estimator of the node source corresponding to the community relationship graph, the probability of rumor propagation of each first community in the community relationship graph is estimated and analyzed to determine the source community in the community relationship graph; where the source community refers to the first community in the community relationship graph that is most likely to spread the rumor information first.

[0074] As a preferred option, the ML estimator for the node source corresponding to the community relationship graph mentioned in step S2 is specifically referred to in equation (2):

[0075] （2）

[0076] In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of.

[0077] Step S3: Based on several boundary nodes of the source community, the source community is corrected to obtain the corresponding community unfolded graph. Then, through the ML estimator of the node source corresponding to the community unfolded graph, the probability of rumor propagation of each first node in the community unfolded graph is estimated and analyzed to determine the source node in the community unfolded graph, so as to realize the source of rumor propagation in the infected subgraph. Here, the source node refers to the first node in the community unfolded graph that is most likely to spread the rumor information first.

[0078] As a preferred embodiment, step S3 includes steps S31 to S33, and the specific details of each step are as follows:

[0079] Step S31: Obtain the boundary nodes whose distance from each member node in the source community is a preset value; wherein, the boundary nodes are not in the source community.

[0080] In this embodiment, the preset value is 1.

[0081] Step S32: Expand all members within the source community and all boundary nodes of the source community to obtain the corresponding community expansion graph. .

[0082] To mitigate the amplified errors in tracing the source node caused by miscalculations in the source community estimation, this embodiment employs a heuristic approach during the community expansion process. As the original node propagates information throughout the community, some nodes within or adjacent to the original community act as propagators or bridges, such as boundary nodes. Therefore, considering only the node structure of the original community to trace the source node is insufficient. This embodiment expands both the source community itself and nodes closely related to it, resulting in a corrected community expansion diagram. .

[0083] Step S33: Using the ML estimator of the node source corresponding to the community unfolded graph, the probability of rumor propagation of each first node in the community unfolded graph is estimated and analyzed to determine the source node in the community unfolded graph, so as to trace the source of rumor propagation in the infected subgraph; where the source node refers to the first node in the community unfolded graph that is most likely to spread rumor information first.

[0084] As a preferred option, the ML estimator for the node source corresponding to the community relationship graph mentioned in steps S3 and S33 is specifically referred to in equation (3):

[0085] （3）

[0086] In the formula, Community unfolding diagram China and Israel A BFS tree with the root node as its root node. Indicates traversal The order in which nodes are added to the BFS tree; Community unfolding diagram The source node in; Indicates from the root node Start to get The probability of.

[0087] Please refer to Figure 2This is a schematic diagram of a rumor propagation tracing system based on community clustering provided by an embodiment of the present invention. The system includes a data acquisition module M1, a first tracing module M2, and a second tracing module M3. The specific details of each module are as follows:

[0088] The data acquisition module M1 is used to perform community clustering on the infection subgraph to be traced in order to obtain the corresponding community relationship graph;

[0089] The first source tracing module M2 is used to estimate and analyze the probability of rumor propagation in each first community in the community relationship graph by using the ML estimator of the node source corresponding to the community relationship graph, and to determine the source community in the community relationship graph; wherein, the source community refers to the first community in the community relationship graph that is most likely to spread the rumor information first.

[0090] The second source tracing module M3 is used to modify the source community based on several boundary nodes of the source community to obtain a corresponding community unfolded graph. It then uses the ML estimator of the node source corresponding to the community unfolded graph to estimate and analyze the rumor propagation probability of each first node in the community unfolded graph, and determines the source node in the community unfolded graph to achieve source tracing of rumor propagation in the infected subgraph. The source node refers to the first node in the community unfolded graph that is most likely to spread rumor information first.

[0091] As a preferred embodiment, the data acquisition module M1 specifically includes a first clustering unit 11 and a second clustering unit 12, with each unit as follows:

[0092] The first clustering unit 11 is used to assign different community numbers to each second node in the infected subgraph to be analyzed using the Louvain algorithm to form several second communities. It then traverses each second node, evaluates the modular gain resulting from removing the currently traversed second node from its current second community and placing it in the second communities of its neighboring nodes, and then determines whether each of the modular gains corresponding to the currently traversed second node is greater than 0. If any of the modular gains corresponding to the currently traversed second node is greater than 0, then the currently traversed second node is removed from its current second community and placed in the second community that produces the largest modular gain, thereby updating all second communities in the infected subgraph. If none of the modular gains corresponding to the currently traversed second node are greater than 0, then the currently traversed second node is retained in its current second community.

[0093] The second clustering unit 12 is used to take each of the second communities in the current infection subgraph as third nodes to form a corresponding weighted network when all the modular gains corresponding to all the second nodes in the infection subgraph are not greater than 0, and to use the Louvain algorithm to perform community clustering on the weighted network to obtain the community relationship graph.

[0094] As a preferred embodiment, the second clustering unit 12 specifically includes a weighted network construction subunit 121 and a community clustering subunit 122, with each unit as follows:

[0095] The weighted network construction subunit 121 is used to, when all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, take each of the second communities in the current infected subgraph as the third node, determine the weight of the link between each of the third nodes according to the weight of the link between all the second nodes in each of the second communities in the current infected subgraph, and then establish the corresponding weighted network according to all the third nodes and the weight of the link between each of the third nodes.

[0096] Community clustering subunit 122 is used to assign different community numbers to each third node in the weighted network using the Louvain algorithm to form several third communities. It then traverses each third node, evaluates the modularity gain resulting from removing the currently traversed third node from its current third community and placing it in the third communities of its neighbors, and then determines whether each modularity gain corresponding to the currently traversed third node is greater than 0. If any modularity gain corresponding to the currently traversed third node is greater than 0, then the currently traversed third node is removed from its current third community. A node is removed from the third community where the currently traversed third node is located and placed in the third community that produces the largest modular gain, thereby updating the third community in the weighted network; if all the modular gains corresponding to the currently traversed third node are not greater than 0, then the currently traversed third node is retained in the third community where the currently traversed third node is located; when all the modular gains corresponding to all third nodes in the weighted network are not greater than 0, the current weighted network is used as the community relationship graph, and each of the third communities in the current weighted network is used as the first community in the community relationship graph.

[0097] As a preferred embodiment, the ML estimator for the node source corresponding to the community relationship graph is specifically:

[0098]

[0099] In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of.

[0100] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the system described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0101] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0102] This invention proposes a method and system for tracing the source of rumors based on community clustering. It obtains a corresponding community relationship graph by performing community clustering on the infected subgraph to be traced. The community and node relationships in the infected subgraph are progressively optimized to significantly reduce the number of network nodes required for tracing, avoid some redundant and ineffective calculations, and alleviate memory pressure during computation. This facilitates subsequent estimation and analysis of source communities and source nodes using a node-source ML estimator, thereby improving the overall tracing efficiency. This method can be applied to larger-scale real-world networks, expanding its applicability. Furthermore, by using a node-source ML estimator, the source communities in the community relationship graph are first determined, then the source communities are corrected, and further estimations are made on each first node in the corrected community unfolded graph. Finally, the source nodes in the community unfolded graph are determined as the result of rumor tracing for the infected subgraph, rather than directly estimating and analyzing the community clustering results of the infected subgraph, fundamentally improving the accuracy of rumor tracing.

[0103] Furthermore, by heuristically extending the original community of the currently traversed node to the nodes adjacent to the community, and utilizing some structural information of the original community and nearby nodes, the information propagation efficiency between the original community and nearby nodes is derived to optimize its community relationship graph, thereby reducing the amplification of the source node tracing error caused by the source community estimation error.

[0104] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for tracing the source of rumor propagation based on community clustering, characterized in that, include: Community clustering is performed on the infection subgraph for source tracing to obtain the corresponding community relationship graph; The ML estimator of the node sources corresponding to the community relationship graph is used to estimate and analyze the probability of rumor propagation in each first community in the community relationship graph, thereby determining the source community in the community relationship graph; wherein, the source community refers to the first community in the community relationship graph that is most likely to spread rumor information first; the ML estimator of the node sources corresponding to the community relationship graph is specifically as follows: In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score; Based on several boundary nodes of the source community, the source community is modified to obtain a corresponding community unfolded graph. Then, using the ML estimator of the node source corresponding to the community unfolded graph, the probability of rumor propagation of each first node in the community unfolded graph is estimated and analyzed to determine the source node in the community unfolded graph, thereby achieving the tracing of rumor propagation in the infected subgraph. Here, the source node refers to the first node in the community unfolded graph most likely to transmit rumor information first. The ML estimator of the node source corresponding to the community unfolded graph is specifically as follows: In the formula, Community unfolding diagram China and Israel A BFS tree with the root node as its root node. Indicates traversal The order in which nodes are added to the BFS tree; Community unfolding diagram The source node in; Indicates from the root node Start to get The probability of; Indicates in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score.

2. The rumor propagation tracing method based on community clustering as described in claim 1, characterized in that, The process of performing community clustering on the infection subgraph to be traced to obtain the corresponding community relationship graph is as follows: Using the Louvain algorithm, each second node in the infected subgraph to be analyzed is assigned a different community number to form several second communities. Each second node is traversed, and the modular gain produced by removing the currently traversed second node from its second community and placing it in the second communities of its neighboring nodes is evaluated. Then, it is determined whether each of the modular gains corresponding to the currently traversed second node is greater than 0. If any of the modular gains corresponding to the currently traversed second node is greater than 0, then the currently traversed second node is removed from the second community where the currently traversed second node is located and placed in the second community that produces the largest modular gain, so as to update all the second communities in the infected subgraph. If all the modular gains corresponding to the currently traversed second node are not greater than 0, then the currently traversed second node will be kept in the second community where the currently traversed second node is located. When all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, each of the second communities in the current infected subgraph is taken as a third node to form a corresponding weighted network, and the Louvain algorithm is used to perform community clustering on the weighted network to obtain the community relationship graph.

3. The rumor propagation tracing method based on community clustering as described in claim 2, characterized in that, When all modular gains corresponding to all second nodes in the infected subgraph are not greater than 0, each second community in the current infected subgraph is taken as a third node to form a corresponding weighted network. The Louvain algorithm is then used to perform community clustering on the weighted network to obtain the community relationship graph. Specifically, this includes: When all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, each of the second communities in the current infected subgraph is taken as a third node, and the weight of the link between each third node is determined according to the weight of the link between all the second nodes in each of the second communities in the current infected subgraph. Then, the corresponding weighted network is established according to all the third nodes and the weight of the link between each third node. Using the Louvain algorithm, each third node in the weighted network is assigned a different community number to form several third communities. Each third node is traversed, and the modular gain resulting from removing the currently traversed third node from its current third community and placing the currently traversed third node in the third communities of its neighboring nodes is evaluated. Then, it is determined whether each modular gain corresponding to the currently traversed third node is greater than 0. If any of the modular gains corresponding to the currently traversed third node is greater than 0, then the currently traversed third node is removed from the third community where the currently traversed third node is located and placed in the third community that generates the largest modular gain, so as to update the third community in the weighted network. If all the modular gains corresponding to the currently traversed third node are not greater than 0, then the currently traversed third node will be retained in the third community where the currently traversed third node is located. When all modular gains corresponding to all third nodes in the weighted network are not greater than 0, the current weighted network is used as the community relationship graph, and each third community in the current weighted network is used as the first community in the community relationship graph.

4. The rumor propagation tracing method based on community clustering as described in claim 1, characterized in that, The step of modifying the source community based on several boundary nodes to obtain a corresponding community unfolding graph specifically includes: Obtain the boundary nodes whose distance from each member node in the source community is a preset value; wherein, the boundary nodes are not in the source community; Expand all members within the source community and all boundary nodes of the source community to obtain the corresponding community expansion diagram.

5. A rumor propagation tracing system based on community clustering, characterized in that, Also includes: The data acquisition module is used to perform community clustering on the infection subgraph to be traced in order to obtain the corresponding community relationship graph; The first source tracing module is used to estimate and analyze the probability of rumor propagation in each first community in the community relationship graph by using the ML estimator of the node source corresponding to the community relationship graph, and to determine the source community in the community relationship graph; wherein, the source community refers to the first community in the community relationship graph that is most likely to spread the rumor information first; the ML estimator of the node source corresponding to the community relationship graph is specifically: In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates that in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score; The second source tracing module is used to correct the source community based on several boundary nodes of the source community to obtain a corresponding community unfolded graph. Then, using the ML estimator of the node source corresponding to the community unfolded graph, it estimates and analyzes the probability of rumor propagation for each first node in the community unfolded graph to determine the source node in the community unfolded graph, thereby achieving source tracing of rumor propagation in the infected subgraph. The source node refers to the first node in the community unfolded graph most likely to transmit rumor information first. The ML estimator of the node source corresponding to the community unfolded graph is specifically: In the formula, Community unfolding diagram China and Israel A BFS tree with the root node as its root node. Indicates traversal The order in which nodes are added to the BFS tree; Community unfolding diagram The source node in; Indicates from the root node Start to get The probability of; Indicates that in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score.

6. A rumor propagation tracing system based on community clustering as described in claim 5, characterized in that, The data acquisition module specifically includes: The first clustering unit uses the Louvain algorithm to assign different community numbers to each second node in the infected subgraph to be analyzed, forming several second communities. It then iterates through each second node, evaluating the modular gain resulting from removing the currently traversed second node from its current second community and placing it in the second communities of its neighbors. It then determines whether each modular gain corresponding to the currently traversed second node is greater than 0. If any modular gain corresponding to the currently traversed second node is greater than 0, the currently traversed second node is removed from its current second community and placed in the second community that generates the largest modular gain, thus updating all second communities in the infected subgraph. If none of the modular gains corresponding to the currently traversed second node are greater than 0, the currently traversed second node is retained in its current second community. The second clustering unit is used to take each of the second communities in the current infection subgraph as the third node when all the modular gains corresponding to all the second nodes in the infection subgraph are not greater than 0, so as to form a corresponding weighted network, and use the Louvain algorithm to perform community clustering on the weighted network to obtain the community relationship graph.

7. A rumor propagation tracing system based on community clustering as described in claim 6, characterized in that, The second clustering unit specifically includes: The weighted network construction subunit is used to, when all the modular gains corresponding to all the second nodes in the infected subgraph are not greater than 0, take each of the second communities in the current infected subgraph as the third node, determine the weight of the link between each of the third nodes according to the weight of the link between all the second nodes in each of the second communities in the current infected subgraph, and then establish the corresponding weighted network according to all the third nodes and the weight of the link between each of the third nodes. The community clustering subunit is used to assign a different community number to each third node in the weighted network using the Louvain algorithm to form several third communities. It then traverses each third node, evaluates the modularity gain resulting from removing the currently traversed third node from its current third community and placing it in the third communities of its neighbors, and then determines whether each modularity gain corresponding to the currently traversed third node is greater than 0. If any modularity gain corresponding to the currently traversed third node is greater than 0, then the currently traversed third node... Remove the third node from the third community where the currently traversed third node is located and place it in the third community that produces the largest modular gain to update the third community in the weighted network; if all the modular gains corresponding to the currently traversed third node are not greater than 0, then keep the currently traversed third node in the third community where the currently traversed third node is located; when all the modular gains corresponding to all the third nodes in the weighted network are not greater than 0, use the current weighted network as the community relationship graph, and use each of the third communities in the current weighted network as the first communities in the community relationship graph.

8. A rumor propagation tracing system based on community clustering as described in claim 5, characterized in that, The ML estimator for the node source corresponding to the community relationship graph is as follows: In the formula, Representing a community relationship diagram China and Israel A BFS tree with the root node as the root node; Indicates traversal The order in which nodes are added to the BFS tree; Representing a community relationship diagram The source community in the middle; Indicates from the root node Start to get The probability of; Indicates that in BFS tree with root node The above is calculated based on the rumor centrality algorithm. The rumor centrality score.

Images