A social network event key element identification method based on fusion graph calculation

Abstract

The application discloses a social network event key element identification method based on fusion graph calculation and belongs to the technical field of internet data processing. First, social network event information texts of a target topic are acquired, five types of event elements of time, place, doer, behavior and receiver are extracted, and a social network event element knowledge graph is constructed with the five types of event elements as nodes and logical relations among the elements as edges. Second, the degree centrality, betweenness centrality and PageRank value of each node in the graph are calculated, nodes ranked before a proportion threshold value are selected for each type of index, and an initial key node set is formed by merging. Finally, based on the graph and the initial set, SLPA community division is iteratively executed, a key community subset is screened, overlapping nodes are detected and merged, and the final key node set is output as event key elements until convergence. The application can effectively identify explicit and potential key elements and solves the problem of one-sided evaluation and high omission rate of existing single index methods.

Description

Technical Field

[0001] This invention relates to Internet data processing technology, and in particular to a method for identifying key elements of social network events based on fusion graph computation. Background Technology

[0002] Social network events typically involve numerous interconnected elements during their evolution, such as individuals, organizations, locations, and behaviors. Different elements exert significantly varying degrees of influence on event propagation and evolution. Accurately identifying key elements is crucial for revealing the core focus, propagation patterns, and evolutionary trends of an event. However, with the increasing prevalence of information technology and social platforms, event-related data on social networks exhibits characteristics such as massive scale, diverse sources, frequent updates, and complex structures. This makes efficiently and accurately identifying key elements from massive amounts of heterogeneous data technically challenging. Therefore, accurately identifying key elements from massive amounts of social network event-related data has become a significant research direction in the field of social networks.

[0003] Social network event data naturally possesses graph structure characteristics: various elements within an event can be considered nodes, and the relationships between elements can be considered edges. Therefore, the problem of identifying key elements in social network events can be modeled as the problem of identifying key nodes in a graph data structure.

[0004] Existing key node identification methods can be mainly divided into two categories:

[0005] The first category is data-driven learning methods, which transform the key element identification problem into a graph representation learning or graph neural network modeling problem. These methods automatically construct node importance discrimination models through supervised or semi-supervised learning on historical labeled data. For example, some studies have applied graph convolutional networks to key node identification, transforming the problem into a regression or ranking task. However, such methods heavily rely on large amounts of high-quality labeled data. In real-world social network scenarios, labeling key elements is costly, highly subjective, and the data distribution varies greatly across different events, leading to difficulties in model training, limited generalization ability, and significant constraints on engineering applications.

[0006] The second category is topology-based methods, which directly utilize the topological properties of the graph to evaluate node importance. They require no labeled data and offer strong interpretability. Classic methods include:

[0007] Degree centrality: measures the number of direct connections a node has, reflecting its local influence.

[0008] Betweenness centrality: measures how often a node is located on the shortest path between other nodes, reflecting its global control as a “bridge” or “hub”.

[0009] PageRank and its variants: measure a node's global reputation and influence through a recursive voting mechanism among neighboring nodes.

[0010] Although topology-based methods are widely used, most existing studies rely on a single metric or a single topological mechanism for evaluation, which has significant limitations:

[0011] A single evaluation dimension can easily lead to missed detections: the importance of a node can be reflected in different dimensions. For example, a node may have a low degree (few local connections) but a high betweenness (it is a key hub). If only degree centrality is used for evaluation, such key nodes will be missed.

[0012] Insufficient ability to identify potential critical nodes: There exists a special type of node in the graph that does not rank highly in traditional topological metrics (such as degree, betweenness, and PageRank), but plays a crucial role in cross-community information propagation and network connectivity due to its location at the intersection of multiple important communities. The criticality of these nodes cannot be directly reflected through explicit topological metric calculations, making it difficult for existing methods to effectively identify them. Summary of the Invention

[0013] The technical problem this invention aims to solve is that existing graph critical node identification methods based on a single topological indicator, when applied to the identification of key elements in social network events, suffer from incomplete and inaccurate identification results due to their one-sided evaluation dimensions, inability to fully characterize the importance of nodes, and difficulty in identifying potential critical nodes located at the boundaries of multiple communities that are crucial for cross-domain information dissemination. This invention provides a robust identification method that, under unsupervised or weakly supervised conditions, can integrate multi-dimensional topological information and uncover potential critical nodes in social network events that explicit indicators cannot capture.

[0014] The technical solution adopted by this invention to solve the above-mentioned technical problems is a method for identifying key elements of social network events based on fusion graph calculation, comprising the following steps:

[0015] Steps for constructing a knowledge graph of social network event elements: Obtain social network event information text for the target topic, and extract five types of event elements: time, location, agent, behavior, and victim; Use the five types of event elements as nodes and the logical relationships between event elements as edges to construct a knowledge graph of social network event elements;

[0016] The key node selection steps based on multiple indicators are as follows: Calculate the degree centrality, betweenness centrality, and PageRank value of each node in the social network event element knowledge graph; for each indicator, sort the nodes from high to low according to the indicator value, and select the nodes ranked before the corresponding proportion threshold; take the union of the nodes selected based on the three indicators to form the initial key node set.

[0017] The steps for mining potential key nodes based on overlapping community analysis are as follows: Based on the knowledge graph of social network event elements and the initial set of key nodes, iterative community overlap detection and key node set updates are performed until the key node set converges, and the final set of key nodes is output; the nodes in the key node set correspond to the key elements of social network events.

[0018] This invention first comprehensively evaluates node criticality using three topological features: degree centrality, betweenness centrality, and PageRank, to initially identify important nodes. Based on this, community discovery is used to filter the community structures of the identified critical nodes, and overlapping nodes between these community structures are analyzed to obtain the final set of critical nodes. Furthermore, multi-indicator evaluation and overlapping community detection are employed to achieve in-depth mining of potential critical nodes that cannot be identified by explicit topological indicators, thus yielding more accurate results compared to existing methods.

[0019] Specifically, the steps for mining potential key nodes based on overlapping community analysis include the following sub-steps:

[0020] Community segmentation sub-step: The community discovery algorithm is used to segment the knowledge graph of the social network event elements into overlapping communities to obtain a community set;

[0021] Key community selection sub-step: Select communities from the community set that have intersection with the current key node set to form a community subset, wherein the current key node set is initially the initial key node set;

[0022] Overlapping node detection sub-step: For each node in the community subset, calculate its community overlap strength, where the community overlap strength is the number of communities to which the node belongs; include nodes whose community overlap strength is not less than a preset overlap strength threshold into the candidate overlapping key node set.

[0023] Set update sub-step: Merge the candidate overlapping key node set into the current key node set to obtain the updated key node set;

[0024] Iteration and convergence judgment sub-step: Repeat the key community screening sub-step, overlapping node detection sub-step and set update sub-step until the key node set before and after the update no longer changes, and output the key node set at this time as the final key node set.

[0025] The beneficial effects of this invention are that by integrating multiple key node identification algorithms, a node importance evaluation system based on multiple topological indicators is constructed, and implicit key nodes are identified using a community overlap detection algorithm, resulting in more comprehensive and accurate node identification results. Specifically:

[0026] 1. By introducing multiple topological indicators such as degree centrality, betweenness centrality, and PageRank value, the importance of nodes is comprehensively evaluated. This measures the importance of nodes from multiple different levels, avoiding the omissions that may be caused by evaluation with a single indicator, and making the node identification results more comprehensive and accurate.

[0027] 2. By performing overlap detection on the communities where the identified key nodes are located, the system can supplement the identification of potential key nodes, making up for the shortcomings of relying solely on topological index calculations, and further improving the completeness and accuracy of key node identification.

[0028] 3. By combining multi-indicator evaluation with community overlap detection analysis, this method can adapt to graph data of different structures and scales, effectively identify all explicit and implicit key nodes, thereby improving its versatility and reliability in complex network environments. Attached Figure Description

[0029] Figure 1 This is a diagram illustrating the association of single-event elements.

[0030] Figure 2 A schematic diagram of a knowledge graph of social network event elements;

[0031] Figure 3 A process for identifying key elements of social network events. Detailed Implementation

[0032] The overall process for social network event element identification proposed in this invention is as follows: Figure 1 As shown:

[0033] First, relevant event information for the target topic is collected from social networks, and element identification and extraction are performed to obtain a knowledge graph of social network event element associations. Then, three types of topological feature indicators—degree centrality, betweenness centrality, and PageRank value—are used to identify initial explicit key nodes. Next, a community detection algorithm is used to partition the knowledge graph into communities, analyzing the potential key nodes overlapping within the communities of the identified key nodes, thus obtaining the final key node identification results.

[0034] This invention abstracts social network events into knowledge graphs. ,in Represents a collection of social network event element nodes. This represents the set of edges that indicate the relationships between elements in a social network event. `i` is the node index, defining any node... The degree centrality is Betweenness centrality is PageRank value Knowledge graph The results of the club division are as follows N is the total number of clubs, and k is the club index. Using... Indicates the first The set of key nodes identified after the iteration.

[0035] The principle of this invention is as follows: In a social network event element knowledge graph, various event elements are interconnected through relationships, forming complex and hierarchical topological connections. The position of a node in the graph structure and its topological characteristics can intuitively reflect its importance and influence in the entire graph data. Identifying key nodes in the social network event element knowledge graph involves finding nodes that fall within a preset ranking range in the topological feature evaluation results, as well as nodes located in special topological positions. Therefore, the method proposed in this invention can effectively solve the problem of identifying key elements of social network events.

[0036] Specifically, the knowledge graph of social network event elements is constructed as follows:

[0037] Social network events refer to a collection of information related to a specific theme or public issue that is widely disseminated and continuously evolves on social media platforms. The information carrier is primarily text data, but also includes heterogeneous data from multiple sources such as images and videos. Given these expressive and dissemination characteristics, this invention primarily focuses on the text data. To effectively identify key elements in social network events, we need to perform structured modeling of these events. The modeling process includes the following steps:

[0038] First, a set of keywords related to the target topic is extracted, and event information from social networking platforms is retrieved and collected based on these keywords to obtain corresponding social network event information text. Second, the obtained social network event information text is preprocessed and semantically analyzed. Natural language processing technologies such as large language models and named entity recognition are used to extract five categories of event elements from the event text: time, location, agent, behavior, and victim. Specifically, large language models can directly output structured five categories of event elements based on deep semantic understanding of the text; named entity recognition technology can locate and extract corresponding time, location, agent, and victim entity elements from the text through predefined entity types (such as person names, place names, and organization names) and rules, and then combine dependency parsing or pattern matching to identify behavior elements. These two technologies can be used individually or in combination to automate the extraction of event elements.

[0039] Finally, a knowledge graph is constructed using the five elements of time, place, agent, behavior, and patient as nodes, and the inherent logical relationships between event elements as edges.

[0040] Figure 1 This diagram illustrates the relationships between core elements in a single social network event, presented in a topological structure. The circular nodes represent the six core elements of the event: time, location, agent, action, and recipient. Nodes are connected by edges, reflecting the inherent logical connections between elements. The action node 'a' is also connected to the recipient node 'a', while the location node 'a' and the time node 'a' are connected to the agent node 'a'. The edges between the agent node 'a' and the action node 'a' reflect the transmission of action. The edges between the action node 'a' and the recipient node 'a' reflect the transmission of action.

[0041] When a set of events all belong to the same target topic, there will be some overlap in the event elements between different events; that is, common event elements will appear among different events. Based on these common event elements, effective connections can be established between the element relationship graphs of different events. Ultimately, the element relationship graphs corresponding to each event are interconnected through shared common event elements, gradually converging to form a social network event element knowledge graph oriented towards the target topic, such as... Figure 2 As shown in the diagram, multiple locally densely connected sub-regions (i.e., potential community structures) can be seen, such as the sub-regions where behavior a, behavior b, and behavior c are located. Each sub-region corresponds to a set of highly related event elements. Different sub-regions are connected across regions through common element nodes, such as the relationship between the recipient a and the location b.

[0042] Specifically, the method for assessing the importance of multiple indicators is as follows:

[0043] To overcome the problem that a single indicator evaluation may lead to an incomplete characterization of node importance and result in the omission of key elements, this invention comprehensively evaluates the overall importance of nodes based on topological structural features from three indicator dimensions: degree centrality, betweenness centrality, and PageRank value.

[0044] Degree centrality is a topological characteristic metric used to characterize the degree of direct connectivity between a node and other nodes in a network. It reflects a node's local connectivity and direct influence within the overall network. In a social network event element knowledge graph, the higher the degree of a node, the more direct connections it establishes with other event elements, and the more significant its role in information dissemination, element association, and structural support. For graphs... any node in Its degree centrality Defined as the ratio of the number of connected edges of a node to the maximum possible number of connections in the network, the formula is:

[0045] ;

[0046] in, Represents a node The degree, that is, the number of edges directly connected to the node. This represents the total number of nodes in the network. By calculating the degree centrality of each node, the importance of local connectivity of event elements can be quantitatively evaluated.

[0047] Betweenness centrality measures how much a node acts as a hub for the shortest paths between other nodes in a network, reflecting its importance in information transmission, path connectivity, and network control. In social network event element knowledge graphs, event elements typically propagate and interact through multi-level associations. If a node lies on the shortest paths between a large number of other node pairs, it plays a crucial bridging or hub role in event information transmission, element association, and maintaining network connectivity. Even if the number of its direct connections is low, it can still significantly impact the overall network structure and information flow. For graphs... any node in Its betweenness centrality For any two nodes and The nodes passed through in the shortest path between The sum of the proportions is calculated using the following formula:

[0048] ;

[0049] in, Represents a node With nodes The total number of shortest paths between them. This indicates the nodes passed through in the shortest path. The number of paths. By calculating and comparing the betweenness centrality of each node in the knowledge graph of social network event elements, key event elements that play an important bridging role in the global structure can be effectively identified.

[0050] PageRank (PR) is a topological metric used to measure the global importance of nodes based on network link relationships. Its core idea is to recursively evaluate node importance through connections and mutual voting mechanisms. In a social network event element knowledge graph, if an event node connects to multiple important nodes, that node also has significant influence in the overall network structure. Therefore, PageRank can reflect node importance from a global perspective. For graphs... any node in Its PageRank value The calculation formula is:

[0051] ;

[0052] in, The damping coefficient has a value ranging from 0 to 1. The total number of nodes. Pointing to a node Incoming neighbor node The set, For incoming neighbor nodes The out-degree of the event element is determined by iteratively calculating the above formula until convergence. This yields a stable PageRank value for each event element node, thus enabling a quantitative assessment of the global importance of the event element.

[0053] After calculating and detecting the degree centrality, betweenness centrality, and PageRank value of each node in the knowledge graph of social network event elements, a corresponding proportional threshold was set for each type of indicator. Sort the index values ​​of each node from highest to lowest, and select the nodes that rank highest under the corresponding index. The nodes are used as the initial explicit key nodes, and they are included in the initial key node set. In the middle, superscript Equivalent to superscript Indicates the iteration number. This represents the initial stage. Specifically, the nodes selected based on the three categories of indicators are combined to form the initial set of key nodes. The event elements corresponding to these nodes are the initial explicit key elements, thus achieving preliminary key element identification based on multiple indicators.

[0054] Specifically, the key elements identification methods for social network events are as follows:

[0055] After completing the identification of key elements based on multiple indicators, the results of the community discovery algorithm are effectively filtered, and overlap detection is performed on the filtered community structure to obtain potential key nodes, such as... Figure 3 As shown:

[0056] The community discovery algorithm takes as input a knowledge graph of social network event elements. And the initial set of key nodes consisting of key element display nodes. First, let's look at the diagram. The SLPA community discovery algorithm is used to obtain the community set of overlapping community partitions in the knowledge graph of social network event elements. In each iteration, based on the set of key nodes From the community gathering Filter out the set of current key nodes Clubs that intersect form a key club subset; the club subset corresponding to the key node is then selected. This means retaining only community structures containing at least one key node. This effectively suppresses nodes that may appear in multiple community structures but have limited practical significance, thus preventing them from interfering with overlap detection and key element supplementation processes, and improving the accuracy of key element identification. Community overlap detection is then performed on the selected community subset. We define a node... Community overlap strength for:

[0057] ;

[0058] in, Let be an indicator function, satisfying:

[0059] ;

[0060] Set the overlap strength threshold to For each subset of communities Calculate the nodes overlap strength And define nodes that satisfy the following conditions as the set of candidate overlapping key nodes:

[0061] ;

[0062] This condition ensures that only when the node... Satisfying stable membership at least Only when a node is identified as a community is it considered to have potential cross-community criticality, thus avoiding the generalization of critical nodes caused by weakly overlapping nodes. Finally, the newly identified overlapping critical nodes are merged into the critical node set to form a new set of critical nodes. Then update the iteration count t = t + 1. Then, using the updated set of key nodes, repeat the above operation. The algorithm converges when the set of key nodes stops being updated. The event elements corresponding to the nodes in the network constitute the key elements of the target social network event.

Claims

1. A method for identifying key elements of social network events based on fusion graph computation, characterized in that, Includes the following steps: Steps for constructing a knowledge graph of social network event elements: Obtain the text of social network event information for the target topic, and extract five types of event elements: time, location, agent, behavior, and victim. Using the five types of event elements as nodes and the logical relationships between event elements as edges, a social network event element knowledge graph is constructed. The key node selection steps based on multiple indicators are as follows: Calculate the degree centrality, betweenness centrality, and PageRank value of each node in the social network event element knowledge graph; for each indicator, sort the nodes from high to low according to the indicator value, and select the nodes ranked before the corresponding proportion threshold; take the union of the nodes selected based on the three indicators to form the initial key node set. The steps for mining potential key nodes based on overlapping community analysis are as follows: Based on the knowledge graph of social network event elements and the initial set of key nodes, iterative community overlap detection and key node set updates are performed until the key node set converges, and the final set of key nodes is output; the nodes in the key node set correspond to the key elements of social network events. The steps for identifying potential key nodes based on overlapping community analysis include the following sub-steps: Community segmentation sub-step: The community discovery algorithm is used to segment the knowledge graph of the social network event elements into overlapping communities to obtain a community set; Key community selection sub-step: Select communities from the community set that have intersection with the current key node set to form a community subset, wherein the current key node set is initially the initial key node set; Overlapping node detection sub-step: For each node in the community subset, calculate its community overlap strength, where the community overlap strength is the number of communities to which the node belongs; include nodes whose community overlap strength is not less than a preset overlap strength threshold into the candidate overlapping key node set. Set update sub-step: Merge the candidate overlapping key node set into the current key node set to obtain the updated key node set; Iteration and convergence judgment sub-step: Repeat the key community screening sub-step, overlapping node detection sub-step and set update sub-step until the key node set before and after the update no longer changes, and output the key node set at this time as the final key node set.

2. The method as described in claim 1, characterized in that, The i-th node Degree centrality for: ； in, Represents a node The degree, This represents the total number of nodes in the social network event element knowledge graph.

3. The method as described in claim 1, characterized in that, The i-th node betweenness centrality for: ； in, Represents any two nodes With nodes The total number of shortest paths between them. This indicates the nodes passed through in the shortest path. The number of paths.

4. The method as described in claim 1, characterized in that, The i-th node PageRank value for: ； in, The damping coefficient has a value ranging from 0 to 1. This represents the total number of nodes in the social network event element knowledge graph. Pointing to a node Incoming neighbor node The set, For incoming neighbor nodes The degree of departure.

5. The method as described in claim 1, characterized in that, In the social network event element knowledge graph construction step, the five types of event elements are extracted from the social network event information text through large language models and / or named entity recognition technology.

6. The method as described in claim 1, characterized in that, In the key node screening step based on multiple indicators, the ratio threshold is set independently for the three types of indicators: degree centrality, betweenness centrality, and PageRank value.

7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method of claim 1.

Images