Node importance evaluation system based on information propagation dynamics
By using a dynamic node importance assessment system for information propagation, enhanced nodes that maximize propagation gain are selected, solving the computational complexity and cost control problems of spontaneous node propagation in large-scale social networks, and achieving a significant improvement in the scope of information propagation and the reliability of the gain.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU UNIVERSITY
- Filing Date
- 2026-02-03
- Publication Date
- 2026-05-01
AI Technical Summary
Existing technologies cannot adapt to all nodes spontaneously initiating information dissemination, cannot efficiently quantify the independent contribution of enhanced nodes, and are difficult to effectively control computational costs and time complexity in large-scale social networks.
A node importance assessment system based on information propagation dynamics is adopted. Through network modeling, reverse reachability set generation, propagation gain estimation, and enhanced node screening modules, enhanced nodes that can maximize propagation gain are screened out. The node importance is assessed using an extended independent cascade propagation model and a greedy approximation algorithm.
It achieves full-node propagation adaptability, significantly improves propagation gain, increases propagation range increment by 30%-80%, and ensures the reliability and stability of propagation gain estimation, avoiding performance fluctuations of traditional algorithms.
Smart Images
Figure CN121961545A_ABST
Abstract
Description
Node Importance Assessment System Based on Information Dissemination Dynamics Technical Field
[0001] This invention relates to the field of information dissemination management, and in particular to a node importance assessment system based on information dissemination dynamics. Background Technology
[0002] With the widespread adoption of social networks, information can spread rapidly through social connections between users, becoming a core carrier for viral marketing and public information dissemination. In practical applications, enterprises or organizations often face the need to expand information coverage by enhancing the dissemination capabilities of ordinary users, without specifying a core dissemination starting point (such as spontaneous dissemination by ordinary users in public opinion events). Current research has the following technical gaps: Support for "full-node dissemination capability": Existing methods mostly rely on preset seed nodes as dissemination starting points, which cannot adapt to scenarios where "all nodes can spontaneously initiate dissemination," limiting the scope of application; Efficient quantification of the independent contribution of enhancing nodes: Directly calculating the enhancement gain in a full-node dissemination scenario requires traversing all possible dissemination paths (complexity #P hard), necessitating the use of lightweight sampling tools for efficient estimation; Adaptability to large-scale social networks: Social networks often have millions of nodes, requiring algorithms to control time and memory costs while ensuring dissemination gain, avoiding computational bottlenecks. Summary of the Invention
[0003] The purpose of this invention is to disclose a node importance assessment system based on information propagation dynamics, thereby solving the technical problems mentioned in the background art.
[0004] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides a node importance assessment system based on information propagation dynamics, comprising: a network modeling module, used to abstract a social network into a directed graph, and configure a basic influence probability and an enhanced influence probability for each directed edge in the graph, wherein the enhanced influence probability is greater than the basic influence probability; a reverse reachability set generation module, used to randomly select a root node from the directed graph, determine the activation state of each edge through probability sampling, and perform a reverse breadth-first search based on the activated edges to generate a reverse reachability set centered on the root node, while recording potential enhanced candidate nodes identified during the search process due to inactive edges and storing them in the potential enhanced candidate set; a propagation gain estimation module, used to quantitatively evaluate the expected gain of any node set on the information propagation range after enhancement by statistical calculation based on multiple independently generated reverse reachability sets and their corresponding potential enhanced candidate sets; an enhanced node screening module, used to iteratively select k target enhanced nodes that maximize the propagation gain based on the output of the propagation gain estimation module and a greedy approximation algorithm according to a preset upper limit k of the number of nodes to be enhanced; and a result output and application module, used to output the final enhanced node set.
[0005] Preferably, the propagation between nodes in the directed graph is implemented based on the extended independent cascaded propagation model. In the extended independent cascaded propagation model, all nodes in the directed graph are inactive but have potential propagation capabilities at the initial time. Nodes are divided into ordinary nodes and enhanced nodes. The probability of an enhanced node influencing its neighboring nodes through its outgoing edges is greater than the probability of an ordinary node influencing its neighboring nodes through its outgoing edges.
[0006] Preferably, in the extended independent cascading propagation model, the node activation rule is as follows: propagation proceeds in time steps, and for all nodes u activated at time step t-1, all their outgoing edges are traversed at time step t. If u is a non-enhancing node and v is an inactive node, then node v is activated with the base influence probability. If u is an enhancing node and v is an inactive node, then node v is activated with the enhanced influence probability. If no new node is activated in a certain time step, then propagation stops.
[0007] Preferably, the activation state of each edge is determined by probability sampling, including: for edges in a directed graph ,like For non-enhancing nodes, the base influence probability will be... As a sampling probability; if u is an enhancement node, then the enhancement influence probability will be... As the sampling probability; determined based on the sampling probability. Whether an edge is sampled as an active edge or an inactive edge in this generation.
[0008] Preferably, when performing a reverse reachable set generation module, for each incoming edge e(v,u) of the currently visited node u, the following operations are performed: if edge e(v,u) is sampled as an active edge, then node v is added to the reverse reachable set and added to the traversal queue; if edge e(v,u) is sampled as an inactive edge, then node v is only added to the potential enhancement candidate set, and its incoming edges are no longer traversed in reverse from node v.
[0009] Preferably, the gain calculation method for a single reverse reachable set R is as follows: For a given set of enhanced nodes B, if the intersection of B and the potential enhanced candidate set CR corresponding to R is not empty, and at least one node in the intersection can be connected to the root node through an activated edge after enhancement, then the gain value is defined as the number of nodes added in this connection that are not included in R; otherwise, the gain value is zero.
[0010] Preferably, a greedy approximation algorithm is used to iteratively select the k target augmentation nodes that maximize the propagation gain, including: in each iteration, calculating the estimated marginal propagation gain of each currently unselected node added to the existing augmentation node set, and selecting the node with the largest marginal gain as the target augmentation node to be added to the augmentation node set.
[0011] Preferably, the calculation process of the marginal propagation gain estimate is as follows: For node v, its corresponding marginal propagation gain estimate is... .
[0012] This represents the estimated propagation gain corresponding to the union of the enhanced node set B and v after adding node v; This represents the estimated propagation gain corresponding to the enhanced node set B.
[0013] Preferably, the reverse reachable set generation module is configured to support parallel computing and can generate multiple independent reverse reachable sets simultaneously.
[0014] Beneficial effects: Full-node propagation adaptability: By removing the seed node dependency, all nodes have potential propagation capabilities, making it suitable for scenarios without a preset propagation starting point, such as public opinion events and spontaneous marketing, significantly expanding the scope of application; Significant propagation gain: By accurately selecting the enhanced nodes with the "maximum marginal contribution," the propagation range is increased by 30%-80% compared to randomly selecting nodes, meeting the core needs of viral marketing and public information dissemination; High reliability: The Chernov bound ensures the error and confidence of the propagation gain estimation, and the greedy strategy ensures that the selected enhanced node set has a stable propagation enhancement effect, without the performance fluctuation problem of traditional heuristic algorithms (such as degree-first search). Attached Figure Description
[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0016] Figure 1 is a schematic diagram of the node importance assessment system based on information propagation dynamics according to the present invention. Detailed Implementation
[0017] 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.
[0018] As shown in Figure 1, the node importance assessment system based on information propagation dynamics of the present invention includes: a network modeling module, used to abstract the social network into a directed graph, and configure a basic influence probability and an enhanced influence probability for each directed edge in the graph, wherein the enhanced influence probability is greater than the basic influence probability; a reverse reachability set generation module, used to randomly select a root node from the directed graph, determine the activation state of each edge through probability sampling, and perform a reverse breadth-first search based on the activated edges to generate a reverse reachability set centered on the root node, while recording potential enhanced candidate nodes identified during the search process due to inactive edges and storing them in the potential enhanced candidate set; a propagation gain estimation module, used to quantitatively evaluate the expected gain of any node set on the information propagation range after enhancement by statistical calculation based on multiple independently generated reverse reachability sets and their corresponding potential enhanced candidate sets; an enhanced node screening module, used to iteratively select k target enhanced nodes that maximize the propagation gain based on the output of the propagation gain estimation module and a greedy approximation algorithm according to a preset upper limit k of the number of nodes to be enhanced; and a result output and application module, used to output the final enhanced node set.
[0019] Enhanced node set Used to implement enhancement actions (such as issuing incentives) in social networks to increase the reach of information.
[0020] Specifically, social networks are represented as directed graphs. ,in A set of nodes (representing social network users). Let it be a set of edges (representing social connections between users); each edge Corresponding basic influence probability , representing a node This edge affects the node. The probability of.
[0021] Preferably, the propagation between nodes in the directed graph is implemented based on the extended independent cascaded propagation model. In the extended independent cascaded propagation model, all nodes in the directed graph are inactive but have potential propagation capabilities at the initial time. Nodes are divided into ordinary nodes and enhanced nodes. The probability of an enhanced node influencing its neighboring nodes through its outgoing edges is greater than the probability of an ordinary node influencing its neighboring nodes through its outgoing edges.
[0022] Specifically, all nodes initially possess potential propagation capabilities (no preset propagation starting point is required). Nodes are only divided into ordinary nodes (unenhanced) and enhanced nodes. The probability of affecting neighboring nodes increases to Non-enhanced nodes maintain the base influence probability .
[0023] Preferably, in the extended independent cascading propagation model, the node activation rule is as follows: propagation proceeds in time steps, and for all nodes u activated at time step t-1, all their outgoing edges are traversed at time step t. If u is a non-enhancing node and v is an inactive node, then node v is activated with the base influence probability. If u is an enhancing node and v is an inactive node, then node v is activated with the enhanced influence probability. If no new node is activated at any time step, then propagation stops. The total number of activated nodes is used as the scope of this propagation.
[0024] Preferably, the activation state of each edge is determined by probability sampling, including: for edges in a directed graph ,like For non-enhancing nodes, the base influence probability will be... As a sampling probability; if u is an enhancement node, then the enhancement influence probability will be... As the sampling probability; determined based on the sampling probability. Whether an edge is sampled as an active edge or an inactive edge in this generation.
[0025] Specifically, regarding social networks Each edge ∈ Sampling status: If node For non-enhancing nodes, based on probability Marked as "activated edge", with probability 1− Marked as "inactive edge"; if node To enhance nodes, using probability Marked as "activated edge", with probability 1− Marked as "inactive edge".
[0026] Specifically, a random number is generated for each edge and compared with the corresponding sampling probability to determine whether each edge is an active edge or an inactive edge in this generation. If the probability is greater than the sampling probability, it is an active edge.
[0027] Preferably, when performing a reverse reachable set generation module, for each incoming edge e(v,u) of the currently visited node u, the following operations are performed: if edge e(v,u) is sampled as an active edge, then node v is added to the reverse reachable set and added to the traversal queue; if edge e(v,u) is sampled as an inactive edge, then node v is only added to the potential enhancement candidate set, and its incoming edges are no longer traversed in reverse from node v.
[0028] Specifically, a reverse breadth-first search is performed based on the activated edges, including: starting from the root node. Starting from, in the opposite direction of the edge (i.e., from) arrive Iterate through the nodes, keeping only the "active edges," and collect all nodes that can be connected through the "active edges" to form a sequence. The root RRS; simultaneously recording the "potential augmentation candidate set". — That is, an unvisited node connected to an RRS node via an "inactive edge".
[0029] Furthermore, the reverse breadth-first search relationship is as follows: the initial RRS only contains the root node. Double-ended queue Store the nodes to be traversed initially. ; Mark node access status: The remaining nodes are Potential Enhancement Candidate Set
[0030] Departure Node traverse all its incoming edges (Reverse traversal, from) arrive ):like Visited, skip; if Non-activated edges: join in (This node may be able to be connected after enhancement) and (connected path), no need to continue traversing. The incoming edge; if To activate the edge, Add RRS, mark Joining the team When the queue is empty, the generated subset of nodes (RRS) and the potential augmentation candidate set ( This is the result of this sampling.
[0031] Preferably, the gain calculation method for a single reverse reachable set R is as follows: For a given set of enhanced nodes B, if the intersection of B and the potential enhanced candidate set CR corresponding to R is not empty, and at least one node in the intersection can be connected to the root node through an activated edge after enhancement, then the gain value is defined as the number of nodes added in this connection that are not included in R; otherwise, the gain value is zero.
[0032] Specifically, propagation gain is estimated based on RRS: The RRS gain function is defined. For a single RRSR and a potential enhancement candidate set If the node is enhanced And after enhancement With the root node By activating edge connectivity, then To "enhance the root node" "Number of new nodes reachable in reverse", otherwise ;sampling Group of independent RRSs, calculate propagation gain estimates based on RRS ensemble: in Guarantee through the Chernov boundary Compared with true propagation gain The error does not exceed the preset threshold, and the confidence level meets the preset requirements.
[0033] By using multiple independent RRSs, a correlation between "enhanced node contribution and propagation gain" is established to achieve efficient estimation: Single RRS gain function: for a single RRS and its corresponding ,definition Used for quantization enhancement sets For the root node Contribution to propagation range: if nodes exist And enhance behind (become an active edge). With the root node By activating edge connectivity: then To "enhance the root node" The number of newly added nodes reachable in the reverse direction (i.e., the number of connected nodes added after enhancement that were not originally covered by the RRS); other cases (no enhanced nodes in...). (or still not connected after enhancement). ; Expected propagation gain calculation: due to the root node Random selection, propagation gain The expected value is equal to the average of multiple RRS gain functions (the sample mean can approximate the overall expected value); sampling estimation: sampling For group-independent RRS, the estimated propagation gain is: Based on the Chernov bound, when When the scale is large enough, Compared with true propagation gain The error can be controlled within a preset threshold. Within, and with a confidence level not lower than ( (Confidence parameter).
[0034] Preferably, a greedy approximation algorithm is used to iteratively select the k target augmentation nodes that maximize the propagation gain, including: in each iteration, calculating the estimated marginal propagation gain of each currently unselected node added to the existing augmentation node set, and selecting the node with the largest marginal gain as the target augmentation node to be added to the augmentation node set.
[0035] Specifically, for the k-augmentation problem: given an upper limit on the number of augmentation nodes... Searching for augmented sets and This results in an increased range of information dissemination. Maximize; where To enhance the collection The expected propagation range under the influence (the expected total number of nodes activated after propagation terminates). This represents the expected propagation range without any node enhancements.
[0036] The process of selecting boosting nodes includes: executing the RR-Boost algorithm: based on the RRS set. A greedy strategy is used to iteratively select enhancement nodes—initializing the enhancement set. Each time from Select to enable: The node with the largest (marginal gain) join in until .
[0037] The pseudocode for the RR-Boost algorithm is as follows: Input: Figure Increase the number of nodes ,error Confidence parameter ℓ; Output: Enhanced node set .
[0038] 1. Calculate the number of samples:
[0039] 2. Generate A set of RRS: =
[0040] 3. Initialization
[0041] 4. Repeat Next: In In the selection process, choose the node that covers the most RRS.
[0042] Will join in
[0043] 5. Return .
[0044] OPT represents the optimal propagation gain, and ϵ is the error parameter, ensuring that the confidence level of the propagation gain estimate is not lower than [value missing]. .
[0045] RRS sampling quantity: based on the number of nodes With the number of enhanced nodes Dynamic adjustments are made to ensure that error and confidence requirements are met.
[0046] Error parameter ϵ: Set according to accuracy requirements, generally between 0.3 and 0.5; Confidence parameter ℓ: Generally between 1 and 2, ensuring the confidence level is not lower than... .
[0047] Preferably, the calculation process of the marginal propagation gain estimate is as follows: For node v, its corresponding marginal propagation gain estimate is... .
[0048] This represents the estimated propagation gain corresponding to the union of the enhanced node set B and v after adding node v; This represents the estimated propagation gain corresponding to the enhanced node set B.
[0049] Preferably, the reverse reachable set generation module is configured to support parallel computing and can generate multiple independent reverse reachable sets simultaneously.
[0050] This invention proposes a method and system for enhancing social network information dissemination based on reverse reachable sets. By reconstructing the dissemination model of "full-node dissemination capability," optimizing the RRS generation logic (including potential enhancement candidate sets), and designing a greedy selection algorithm, it solves the core problem of "how to efficiently select enhancement nodes when there is no preset dissemination starting point." This method and system can be widely applied in social network marketing, public information dissemination, and other fields, providing an efficient and reliable technical solution for expanding the scope of information dissemination. In the future, it can be further extended to dynamic social networks (where user relationships change over time) and various enhancement strategies (such as content optimization and incentive tiering), improving the system's versatility and flexibility.
[0051] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. A node importance assessment system based on information propagation dynamics, characterized in that, include: The network modeling module is used to abstract the social network into a directed graph and configure a basic influence probability and an enhanced influence probability for each directed edge in the graph, with the enhanced influence probability being greater than the basic influence probability. The reverse reachability set generation module is used to randomly select a root node from the directed graph, determine the activation state of each edge through probability sampling, and perform a reverse breadth-first search based on the activated edges to generate a reverse reachability set centered on the root node. At the same time, it records and stores the potential enhanced candidate nodes identified during the search process due to inactive edges into the potential enhanced candidate set. The propagation gain estimation module is used to quantitatively evaluate the expected gain of any set of nodes on the information propagation range after being enhanced, based on multiple independently generated reverse reachable sets and their corresponding potential enhancement candidate sets, through statistics and calculation. The enhancement node selection module is used to iteratively select k target enhancement nodes that maximize the propagation gain based on the output of the propagation gain estimation module, according to a preset upper limit k of the number of nodes to be enhanced. The results output and application module are used to output the final set of enhanced nodes.
2. The node importance assessment system based on information propagation dynamics according to claim 1, characterized in that, Propagation between nodes in a directed graph is achieved based on the extended independent cascaded propagation model. In the extended independent cascaded propagation model, all nodes in the directed graph are initially inactive but have the potential to propagate. Nodes are divided into ordinary nodes and enhanced nodes. The probability of an enhanced node influencing its neighboring nodes through its outgoing edges is greater than that of an ordinary node influencing its neighboring nodes through its outgoing edges.
3. The node importance assessment system based on information propagation dynamics according to claim 2, characterized in that, In the extended independent cascade propagation model, the node activation rule is as follows: propagation proceeds in time steps. For all nodes u activated at time step t-1, all their outgoing edges are traversed at time step t. If u is a non-enhanced node and v is an inactive node, then node v is activated with the base influence probability; if u is an enhanced node and v is an inactive node, then node v is activated with the enhanced influence probability. If no new node is activated at a certain time step, the propagation stops.
4. The node importance assessment system based on information propagation dynamics according to claim 1, characterized in that, The activation state of each edge is determined by probability sampling, including: for edges in a directed graph ,like For non-enhancing nodes, the base influence probability will be... As a sampling probability; if u is an enhancement node, then the enhancement influence probability will be... As the sampling probability; determined based on the sampling probability. Whether an edge is sampled as an active edge or an inactive edge in this generation.
5. The node importance assessment system based on information propagation dynamics according to claim 4, characterized in that, When performing a reverse breadth-first search, the reverse reachable set generation module performs the following operations for each incoming edge e(v,u) of the currently visited node u: if edge e(v,u) is sampled as an active edge, then node v is added to the reverse reachable set and the traversal queue; if edge e(v,u) is sampled as an inactive edge, then node v is only added to the potential enhancement candidate set, and its incoming edges are no longer traversed in reverse from node v.
6. The node importance assessment system based on information propagation dynamics according to claim 2, characterized in that, The gain for a single reverse reachable set R is calculated as follows: For a given set of enhanced nodes B, if the intersection of B and the potential enhanced candidate set CR corresponding to R is not empty, and at least one node in the intersection can be connected to the root node through an activated edge after enhancement, then the gain value is defined as the number of nodes added in this connection that are not included in R; otherwise, the gain value is zero.
7. The node importance assessment system based on information propagation dynamics according to claim 1, characterized in that, A greedy approximation algorithm is used to iteratively select the k target augmentation nodes that maximize the propagation gain. This includes: in each iteration, calculating the estimated marginal propagation gain of each currently unselected node when added to the existing augmentation node set, and selecting the node with the largest marginal gain as the target augmentation node to be added to the augmentation node set.
8. The node importance assessment system based on information propagation dynamics according to claim 7, characterized in that, The calculation process for the marginal propagation gain estimate is as follows: For node v, its corresponding marginal propagation gain estimate is... 。 This represents the estimated propagation gain corresponding to the union of the enhanced node set B and v after adding node v; This represents the estimated propagation gain corresponding to the enhanced node set B.
9. The node importance assessment system based on information propagation dynamics according to claim 1, characterized in that, The reverse reachable set generation module is configured to support parallel computing and can generate multiple independent reverse reachable sets simultaneously.