Identification method of important nodes in social network data, terminal and storage medium

Abstract

The invention relates to the technical field of social network analysis, and discloses a method for identifying important nodes in social network data, a terminal and a storage medium. The method comprises the following steps: acquiring social network data and calculating semantic similarity between posters; the method comprises the following steps: mapping users, stickers and topics into nodes, mapping attention, interactive behaviors and similarity relationships into edges, and constructing a multi-relationship heterogeneous graph; secondly, defining a meta path to capture high-order topological dependence, performing multi-hop neighborhood aggregation on nodes, and fusing different path characteristics by using a self-attention mechanism to obtain high-order structure characteristics; meanwhile, a unified semantic space is constructed based on a pre-training language model, and semantic embedding representation of nodes is extracted. And finally, the high-order structure features and the semantic features are deeply fused to generate joint representation, and scores are calculated based on the joint representation so as to accurately identify important nodes. According to the method, the hierarchy and potential influence of the nodes in the network can be comprehensively described, so that the accuracy and robustness of important node identification are improved.

Description

Technical Field

[0001] This invention relates to the field of social network analysis technology, specifically to a method, terminal, and storage medium for identifying important nodes in social network data. Background Technology

[0002] With the rapid development of social media and the internet, users are interacting more frequently on different platforms, forming large-scale and complex social networks. Identifying key nodes in social networks (such as key users and core disseminators) is of great significance for applications such as public opinion monitoring, information diffusion modeling, community structure analysis, and personalized recommendations.

[0003] In traditional complex network analysis methods, node importance is typically measured using centrality metrics (such as degree centrality, betweenness centrality, and compactness centrality) or the PageRank algorithm. These methods can measure a node's importance to some extent. However, these methods mainly rely on first- or second-order adjacency relationships and lack effective modeling of higher-order heterogeneous structural information in the network (such as association patterns between different types of nodes and multi-hop path structures). In recent years, with the rapid development of Graph Neural Networks (GNNs), researchers have begun to use models such as Graph Convolutional Networks (GCNs) and Graph Attention Networks (GATs) to learn representations of node features. However, existing methods usually only model homogeneous networks or only consider the direct neighborhood features of nodes, making it difficult to effectively integrate higher-order heterogeneous relationship information and semantic content features in the network. Furthermore, while traditional PageRank and its weighted variants have certain advantages in integrating topics or interest preferences, their weighting strategies often rely on manual priors or static parameters, failing to organically combine with the deep feature learning mechanism of neural networks, thus limiting the model's applicability in dynamic, multi-semantic social environments.

[0004] Therefore, existing technologies struggle to simultaneously characterize the combined influence of nodes in high-order heterogeneous topologies and semantic spaces. How to effectively integrate these two types of information to compute important nodes has become a critical problem that urgently needs to be solved in the field of social network analysis. Summary of the Invention

[0005] To address the technical problems existing in the prior art, this invention provides a method, terminal, and storage medium for identifying important nodes in social network data, which can comprehensively characterize the hierarchy and potential influence of nodes in the network, thereby improving the accuracy and robustness of important node identification.

[0006] To achieve the above objectives, the present invention provides the following technical solution: This invention discloses a method for identifying important nodes in social network data, comprising: Acquire social network data and calculate the semantic similarity between different posts; Users, posts, and topics are mapped to nodes in the graph, and followers, posting, forwarding, commenting, quoting, and semantic similarity are mapped to edges in the graph, thereby constructing a heterogeneous graph containing multiple types of nodes and multiple relationship edges; Meta-paths of heterogeneous graphs are defined to capture higher-order topological dependencies between nodes; multi-hop neighborhood aggregation is performed on the nodes corresponding to each meta-path to extract higher-order feature representations of the nodes under that meta-path; and self-attention mechanism is used to fuse higher-order feature representations under different meta-paths to obtain higher-order structural features. A unified semantic embedding representation space is constructed based on a pre-trained language model. Text inputs corresponding to different types of nodes are mapped to semantic embedding representations, and the semantic embedding representations of all nodes are fused to obtain semantic features. The higher-order structural features and the semantic features are fused to obtain a joint representation of all nodes. The importance score of each node is calculated based on the joint representation, and important nodes are identified based on the score.

[0007] As a further improvement to the above scheme, before calculating the semantic similarity between different posts, the posts are also subjected to standardization preprocessing and semantic feature extraction. The specific process is as follows: The original posts are subjected to noise removal, emoji and special symbol filtering, word segmentation and stop word removal, text regularization and standardization to ensure the consistency of the corpus at the structural and semantic levels; based on the pre-trained language model, the posts are mapped to a high-dimensional semantic embedding space to capture contextual dependencies and deep semantic relationships.

[0008] As a further improvement to the above scheme, the formula for calculating the semantic similarity between different posts is as follows: ; In the formula, This represents the semantic similarity between post i and post j; This indicates the degree of text overlap between post i and post j; This indicates the keyword repetition rate between post i and post j; Embedding of post i and post j based on a pre-trained language model and semantic similarity; , , , These are weighting coefficients used to balance the importance of different semantic dimensions.

[0009] As a further improvement to the above scheme, the meta-path includes at least the following two types: User – Post / Forward – Topic – Post / Forward / Quote – User; User – Following – User – Comment / Forward / Quote / Post – User.

[0010] As a further improvement to the above scheme, the higher-order features of a node under the meta-path are represented as follows: ; In the formula, Represents the i-th node Higher-order feature representation under the p-th meta-path; , p represents the set of nodes in the heterogeneous graph; p∈P, where P represents the set of metapaths; For the Sigmoid function; For nodes The basic feature vector; L is the aggregation hop count, i.e., the number of neighborhood layers; Represents a node The set of the l-th hop neighbors under the metapath set p For set Nodes in; and The weight matrix is ​​trainable. This represents the attention weight.

[0011] As a further improvement to the above scheme, the calculation formula for the higher-order structural features is as follows: ; ; In the formula, This refers to the higher-order structural features; Let the self-attention weights be those for the p-th meta-path; The higher-order features obtained by calculating the meta-path p are represented by all higher-order features. The set is obtained; (·) represents an exponential function with the natural constant e as its base; the superscript T is the transpose sign; LeakReLU(·) is the activation function; The weight matrix is ​​a learnable matrix; The higher-order features are calculated for all meta-paths other than p.

[0012] As a further improvement to the above scheme, the node The corresponding text input mapping to the semantic embedding representation is expressed by the following formula: ; In the formula, For nodes Semantic embedding representation; , Represents the set of nodes in a heterogeneous graph; Represents a node The corresponding text input; The embedding mapping function represents the pre-trained language model; the expression formula for the semantic features is: ; In the formula, The semantic feature is N; N is the set of nodes. The total number of nodes in the system.

[0013] As a further improvement to the above scheme, the formula for calculating the joint representation of all nodes is as follows: ; In the formula, Right now This is a joint representation of all nodes; For the Sigmoid function; This indicates vector concatenation; The weight matrix is ​​a learnable matrix; It is the bias vector; This refers to the higher-order structural features; Based on joint representation The importance score of each node is calculated using an iterative propagation algorithm. The calculation formula is as follows: ; In the formula, For nodes Importance score; This is the normalization function; For nodes The set of neighboring nodes; As weight, based on node with neighboring nodes Inter-module path features, self-attention weights, and semantic similarity are dynamically adjusted; For nodes Importance score; Sort all nodes by importance score and select the top K nodes with the highest importance score or nodes with an importance score higher than a preset threshold as important nodes.

[0014] The present invention also discloses a computer terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying important nodes in social network data as described above.

[0015] The present invention also discloses a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for identifying important nodes in social network data as described above.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. The method for identifying important nodes in social network data disclosed in this invention constructs a meta-path representation of heterogeneous social networks, extracts high-order structural features of nodes, and combines these features with the semantic embedding vectors and multi-dimensional semantic similarity (content overlap, keyword repetition rate, text embedding similarity, and sentiment consistency) to achieve joint modeling of structural and semantic information. This method can comprehensively characterize the hierarchy and potential influence of nodes in the network, thereby improving the accuracy and robustness of important node identification.

[0017] 2. This invention introduces a self-attention weighting mechanism to the higher-order structural features under different meta-paths, realizing the dynamic fusion of multiple types of higher-order information and effectively capturing the potential interaction patterns of nodes in heterogeneous networks.

[0018] 3. This invention combines node semantic features and multidimensional similarity measurement, and through an improved topic-weighted propagation algorithm, it can identify nodes with high potential influence under specific topics, thereby improving topic sensitivity and propagation value assessment capabilities.

[0019] 4. This invention forms a joint representation by vector-level fusion of high-order structural features and semantic features of nodes, which is used for subsequent node importance calculation. This takes into account both the network position influence and content semantic value of nodes, and achieves a more accurate assessment of node importance. Attached Figure Description

[0020] Figure 1 This is a flowchart of the method for identifying important nodes in social network data in Embodiment 1 of the present invention.

[0021] Figure 2 This is an architecture diagram of the method for identifying important nodes in social network data in Embodiment 1 of the present invention.

[0022] Figure 3 This is a schematic diagram of the structure of the computer terminal in Embodiment 2 of the present invention. Detailed Implementation

[0023] 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.

[0024] Example 1 Existing methods for identifying node importance, such as degree centrality, betweenness centrality, proximity centrality, and the traditional PageRank algorithm, primarily rely on first- or second-order neighborhood structure information of nodes, or global metrics calculated based on shortest paths. While these methods are effective in homogeneous networks, they struggle to capture higher-order connection patterns and semantic structural features in heterogeneous social networks containing multiple node types and numerous relational edges. Examples include multi-hop interactions between users, topics, and posts, or cross-type propagation paths. Furthermore, the characterization of potential node influence by higher-order organizational forms such as community partitioning, group structure, and co-occurrence patterns is neglected, leading to a bias in node importance assessment towards local structure.

[0025] Furthermore, although some improved models based on graph neural networks have introduced graph convolution and neighbor aggregation mechanisms, most are still limited to the propagation of first- or second-order neighborhood information. They lack deep modeling and dynamic weighting mechanisms for high-order heterogeneous path structures, making it difficult to achieve a unified characterization of multiple types of relationships and cross-semantic levels.

[0026] Traditional methods for identifying key nodes mostly focus only on network topology, neglecting the semantic features of the content layer carried by the nodes. In social network scenarios, the importance of a node is determined not only by the number of its connections and its location, but also by the significant influence of the semantic features of its published content.

[0027] For example, in public opinion dissemination networks, the high dissemination value of a node often stems from the relevance of its content to current trending topics or specific themes, rather than solely from its degree centrality. However, existing research generally fails to incorporate content features such as textual semantics, keyword co-occurrence, and sentiment bias when modeling nodes, leading to biases in importance assessment. Some existing models that incorporate textual information only perform bag-of-words vectorization or shallow topic modeling of node text, neglecting deep connections in the semantic representation space and failing to comprehensively reflect the true semantic relationships between nodes.

[0028] To address the above issues, this embodiment proposes a method for identifying important nodes in social network data. This method performs collaborative modeling from both structural and semantic levels: at the structural level, a meta-path mechanism is introduced to capture high-order association patterns between different types of nodes, and a self-attention mechanism is combined to achieve adaptive fusion of multi-path information, thereby effectively modeling high-order heterogeneous features in complex networks; at the semantic level, the text similarity of nodes (such as posts) is comprehensively considered, and the semantic association strength between nodes is quantified by calculating indicators such as content overlap, keyword repetition rate, text embedding similarity, and sentiment consistency.

[0029] Please see Figure 1 and Figure 2The method for identifying important nodes in social network data provided in this embodiment can be applied to Weibo social network data. The detailed implementation steps are as follows: Step 1: Acquisition and Preprocessing of Network Data (1) Data collection We automatically acquire social network data through Weibo's open API interface, including user node information (such as user ID, post text, number of followers, influence metrics, etc.) and user interaction relationship data (multi-type side information such as following, forwarding, commenting, and liking).

[0030] (2) Data preprocessing and semantic feature extraction An intelligent semantic modeling strategy that integrates natural language processing technology with pre-trained semantic models is adopted to perform standardized preprocessing and deep semantic feature extraction on social media texts (such as Weibo data).

[0031] First, a systematic preprocessing workflow is performed on the original text of the posts, including noise removal, emoji and special symbol filtering, word segmentation and stop word removal, text regularization, and standardization, to ensure the consistency of the corpus at the structural and semantic levels and provide high-quality input for the subsequent modeling stage. Then, based on a large-scale pre-trained language model, the text is mapped to a high-dimensional semantic embedding space to capture contextual dependencies and deep semantic relationships.

[0032] To construct similarity relationships between post nodes, this invention calculates semantic similarity between homogeneous nodes based on high-dimensional embedding representations of nodes. During the similarity calculation process, multiple semantic feature indicators are comprehensively considered. These include: content overlap, which measures the proportion of content similarity between texts; keyword repetition rate, reflecting the consistency of core semantic keywords; text embedding similarity, which calculates semantic closeness through cosine similarity in the semantic vector space; and sentiment consistency, which characterizes the convergence of nodes in terms of emotional attitude.

[0033] The formula for calculating the semantic similarity between different posts is: ; In the formula, This represents the semantic similarity between post i and post j; This indicates the degree of text overlap between post i and post j; This indicates the keyword repetition rate between post i and post j; Embedding of post i and post j based on a pre-trained language model and semantic similarity; , , , These are weighting coefficients used to balance the importance of different semantic dimensions.

[0034] (3) Construction of multi-relation heterogeneous graph structure Based on the aforementioned data preprocessing and feature extraction results, a multi-relation heterogeneous graph structure is constructed, which is generally represented as follows: Where V is the set of nodes, E is the set of edges, and R is the set of edge types (relation types). Each cleaned and semantically embedded Weibo user and post is mapped to a single node in the graph: ; Construct multiple relationship types, including at least the following categories: follow, forward, comment, quote, and similar.

[0035] Step 2: High-order heterogeneous information modeling and fusion (1) Metapath construction For heterogeneous social networks with multiple node types and multiple relationship edges, we define typical meta-paths of the network to capture higher-order topological dependencies and potential interaction patterns between nodes of different types.

[0036] This invention uses a meta-path mechanism to abstractly represent network relationships. A meta-path is a directed path defined on a sequence of node and edge types, used to characterize semantic associations and multi-hop dependencies between nodes. This path can characterize higher-order interaction patterns between nodes of different types.

[0037] This invention considers the following two types of metapaths: Users – Posts / Forwards – Topics – Posts / Forwards / Quotes – Users: Indirectly linked through shared participation in posts on a particular topic, depicting potential connections and higher-level interactions based on interests or topic preferences.

[0038] User – Follow – User – Comment / Forward / Quote / Post – User: Through follow relationships and various interactive behaviors (commenting, forwarding, quoting or posting), a high-level dissemination link is formed with the post, reflecting the information diffusion and influence transmission pattern.

[0039] (2) High-order feature extraction For each meta-path, multi-hop neighborhood aggregation is performed on the node representation to extract the node's influence features under higher-order structures. The higher-order features of a node under a meta-path are represented as follows: ; In the formula, Represents the i-th node Higher-order feature representation under the p-th meta-path; ; p∈P, where P represents the set of metapaths; For the Sigmoid function; For nodes The basic feature vector; L is the aggregation hop count, i.e., the number of neighborhood layers; Represents a node The set of the l-th hop neighbors under the metapath set p For set Nodes in; and The weight matrix is ​​trainable. This represents the attention weight.

[0040] (3) Fusion of self-attention mechanisms Higher-order features from different meta-paths are weighted and fused using a self-attention mechanism to achieve adaptive integration of multi-type, higher-order path information. The calculation formula for the higher-order structural features is as follows: ; ; In the formula, This refers to the higher-order structural features; Let the self-attention weights be those for the p-th meta-path; The higher-order features obtained by calculating the meta-path p are represented by all higher-order features. The set is obtained; (·) represents an exponential function with the natural constant e as its base; the superscript T is the transpose sign; LeakReLU(·) is the activation function; The weight matrix is ​​a learnable matrix; The higher-order features are calculated for all meta-paths other than p.

[0041] Step 3: Semantic Alignment Representation Method for Multiple Node Types For the text, phrases, or attribute descriptions contained in different types of nodes (such as user nodes, post nodes, and topic nodes), a unified semantic embedding representation space is constructed based on a pre-trained Large Language Model (LLM) to achieve semantic alignment across node types. The corresponding text input mapping to the semantic embedding representation is expressed by the following formula: ; In the formula, For nodes Semantic embedding representation; Represents a node The corresponding text input; The embedding mapping function represents the pre-trained language model; the expression formula for the semantic features is: ; In the formula, The semantic feature is N; N is the set of nodes. The total number of nodes in the system.

[0042] Step 4: Fusion of higher-order structure and semantic features, and calculation of node importance. (1) Joint representation construction: combining high-order structural features With semantic features Perform vector-level fusion to obtain the joint representation of all nodes: ; In the formula, Right now This is a joint representation of all nodes; For the Sigmoid function; This indicates vector concatenation; The weight matrix is ​​a learnable matrix; It is the bias vector; This refers to the higher-order structural features.

[0043] (2) Node importance calculation: based on joint representation The importance score of each node is calculated using an iterative propagation algorithm (such as an improved PageRank or neural network propagation model). : ; In the formula, For nodes Importance score; This is the normalization function; For nodes The set of neighboring nodes; As weight, based on node with neighboring nodes Inter-module path features, self-attention weights, and semantic similarity are dynamically adjusted to achieve unified quantification of node importance in both structural and semantic spaces. For nodes Importance score.

[0044] Step 5: Node Importance Scoring and Result Analysis Sort the node importance scores output in step four, and select nodes with scores higher than a preset threshold or those in the top-K as important nodes; The technical solution designed in the above steps effectively utilizes high-order network structure information and text topic semantic information, and proposes a node importance identification method applicable to social network data such as Weibo, which can achieve more accurate and effective node influence analysis.

[0045] Example 2 This embodiment provides a computer terminal, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the method for identifying important nodes in social network data as described in Embodiment 1.

[0046] like Figure 3 As shown, the computer terminal provided in this embodiment includes: at least one processor 101, and a memory 102 connected to at least one processor 101. This embodiment does not limit the specific connection medium between the processor 101 and the memory 102. Figure 3 The example shown is the connection between processor 101 and memory 102 via bus 100. Bus 100 is... Figure 3 The connections between other components are shown in bold lines and are for illustrative purposes only, not as limiting information. Bus 100 can be divided into address bus, data bus, control bus, etc., for ease of representation. Figure 3 The bus is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. Alternatively, the processor 101 may also be called a controller; there is no restriction on the name.

[0047] In this embodiment, the memory 102 stores instructions that can be executed by at least one processor 101. The at least one processor 101 can execute the aforementioned method by executing the instructions stored in the memory 102.

[0048] The processor 101 is the control center of the device. It can connect to various parts of the control device through various interfaces and lines. By running or executing instructions stored in memory 102 and calling data stored in memory 102, the processor can perform various functions and process data, thereby monitoring the device as a whole.

[0049] In one possible design, processor 101 may include one or more processing units. Processor 101 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into processor 101. In some embodiments, processor 101 and memory 102 may be implemented on the same chip; in some embodiments, they may also be implemented on separate chips.

[0050] Processor 101 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor, application-specific integrated circuit, field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method for identifying important nodes in social network data disclosed in Embodiment 1 can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules in processor 101.

[0051] Memory 102, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. Memory 102 may include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type memory, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic storage, magnetic disk, optical disk, etc. Memory 102 can be any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In this embodiment, memory 102 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0052] By designing and programming the processor 101, the code corresponding to the method for identifying important nodes in social network data described in the foregoing embodiments can be embedded into the chip, thereby enabling the chip to execute the code during operation. Figure 1 The steps of the method for identifying important nodes in the social network data shown are as follows. How to design and program the processor 101 is a technique well-known to those skilled in the art and will not be described further here.

[0053] Example 3 This embodiment provides a computer-readable storage medium storing a computer program thereon. When the program is executed by a processor, it implements the steps of the method for identifying important nodes in social network data as described in Embodiment 1.

[0054] The computer-readable storage medium may include flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the storage medium may be an internal storage unit of a computer device, such as the hard disk or memory of the computer device. In other embodiments, the storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart memory card, secure digital card, flash memory card, etc., provided on the computer device. Of course, the storage medium may include both internal storage units and external storage devices of the computer device. In this embodiment, the memory is typically used to store the operating system and various application software installed on the computer device. In addition, the memory can also be used to temporarily store various types of data that have been output or will be output.

[0055] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for identifying important nodes in social network data, characterized in that, include: Acquire social network data and calculate the semantic similarity between different posts; Users, posts, and topics are mapped to nodes in the graph, and followers, posting, forwarding, commenting, quoting, and the semantic similarity are mapped to edges in the graph, thereby constructing a heterogeneous graph containing multiple types of nodes and multiple relational edges; Define meta-paths in heterogeneous graphs to capture higher-order topological dependencies between nodes; For each meta-path, multi-hop neighborhood aggregation is performed on the nodes to extract the higher-order feature representation of the nodes under that meta-path. The higher-order feature representations under different meta-paths are then fused using a self-attention mechanism to obtain higher-order structural features. A unified semantic embedding representation space is constructed based on a pre-trained language model. Text inputs corresponding to different types of nodes are mapped to semantic embedding representations, and the semantic embedding representations of all nodes are fused to obtain semantic features. The higher-order structural features and the semantic features are fused to obtain a joint representation of all nodes. The importance score of each node is calculated based on the joint representation, and important nodes are identified based on the score.

2. The method of claim 1, wherein, Before calculating the semantic similarity between different posts, the posts undergo standardization preprocessing and semantic feature extraction. The specific process is as follows: The original posts are subjected to noise removal, emoji and special symbol filtering, word segmentation and stop word removal, text regularization and standardization to ensure the consistency of the corpus at the structural and semantic levels; based on the pre-trained language model, the posts are mapped to a high-dimensional semantic embedding space to capture contextual dependencies and deep semantic relationships.

3. The method of claim 2, wherein, The formula for calculating the semantic similarity between different posts is: wherein, represents the semantic similarity of post i and post j; represents the text content overlap of post i and post j; represents the keyword group repetition rate of post i and post j; is the semantic similarity of post i and post j based on pre-trained language model embedding and ; , , , is a weighting coefficient for balancing the importance of different semantic dimensions.

4. The method of claim 1, wherein, The metapath includes at least the following two types: User – Post / Forward – Topic – Post / Forward / Quote – User; User – Following – User – Comment / Forward / Quote / Post – User.

5. The method of claim 1, wherein, The higher-order features of a node under the metapath are represented as follows: In the formula, represents the ith node High-order feature representation under the pth meta-path; , represents a set of nodes in a heterogeneous graph; p ∈ P, P denotes a set of meta-paths; is a Sigmoid function; is the base feature vector of node ; L is the aggregated hop number, i.e., the number of neighborhood layers; denotes the set of the l-hop neighbors of node under meta-path p, is the node in set ; and are trainable weight matrices; is the attention weight.

6. The method of claim 5, wherein, The formula for calculating the higher-order structural features is as follows: wherein, is the high-order structural feature; is the self-attention weight under the p-th meta-path; represents the high-order feature calculated by the p-th meta-path, and all high-order features are represented by ; (·) is an exponential function with a natural constant e as the base; the superscript T is the transpose symbol; LeakReLU(·) is an activation function; is a learnable weight matrix; is the high-order feature calculated by the meta-path other than the p-th meta-path.

7. The method for identifying important nodes in social network data according to claim 1, characterized in that, Node The expression formula of the corresponding text input of the node is as follows: In the formula, is a semantic embedding representation of a node . , represents a set of nodes in a heterogeneous graph; represents a corresponding text input of a node . represents an embedding mapping function of a pre-trained language model; and an expression formula of the semantic feature is: In the formula, The semantic feature is N; N is the set of nodes. The total number of nodes in the system.

8. The method of claim 7, wherein, The formula for calculating the joint representation of all nodes is: wherein i.e. is a joint representation for all nodes; is a Sigmoid function; denotes vector concatenation; is a learnable weight matrix; is a bias vector; is the high-order structure feature; Based on joint representation The importance score of each node is calculated by an iterative propagation algorithm, and the calculation formula is: wherein, is an importance score of a node ; is a normalization function; is a set of neighbor nodes of a node ; is a weight dynamically adjusted according to the meta-path features between a node and a neighbor node , the self-attention weight, and the semantic similarity; is an importance score of a node ; Sort all nodes by importance score and select the top K nodes with the highest importance score or nodes with an importance score higher than a preset threshold as important nodes.

9. A computer terminal, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method for identifying important nodes in social network data as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps of the method for identifying important nodes in social network data as described in any one of claims 1 to 8.

Images