Social network public opinion propagation mode identification method based on directional quantitative index

Abstract

A social network public opinion propagation mode recognition method based on directional quantitative indexes comprises the steps that S1, original data of a target social network are obtained, and the original data comprise text content generated by a user, user relation data and user interaction flow data; s2, the text content is converted into a high-dimensional vector based on a semantic model so as to construct a virtual social space, and each user or topic has a corresponding coordinate in the space; s3, on the basis of the user interaction flow data and the coordinates in the social space, core indexes used for quantifying the information flow directivity are calculated, and the core indexes at least comprise an anisotropic index reflecting the flow direction distribution imbalance degree and a centripetal index reflecting the aggregation trend towards a center node; s4, according to the numerical value combination of the anisotropy index and the centripetal index, performing classification identification on the public opinion propagation mode; and S5, outputting a classification and recognition result, and exploring the characteristics and evolution rules of the classification and recognition result for public opinion monitoring and guiding.

Description

Technical Field

[0001] This invention relates to the fields of information technology and data mining technology, and in particular to a method for analyzing public opinion on social networks. Specifically, this invention is a method for quantitatively identifying and dynamically monitoring the propagation patterns of public opinion information in social networks based on high-dimensional semantic space and spatial directionality indicators. Background Technology

[0002] With the rapid development of the internet and social media, social networks have become an important platform for the public to express their opinions and disseminate information. Online public opinion exhibits characteristics such as explosiveness, complexity, and strong guidance, and its dissemination patterns have a significant impact on social governance, public opinion early warning, and platform operation. Therefore, how to efficiently and accurately identify and analyze public opinion dissemination patterns has become a core issue in current research and practice.

[0003] Existing social network sentiment analysis technologies mainly focus on the following areas: First, there are analytical methods based on graph theory and complex networks. These methods typically identify key nodes and community structure by calculating node centrality (such as degree centrality and betweenness centrality) or by performing community detection. Second, sentiment analysis techniques based on Natural Language Processing (NLP). These methods assess public opinion trends by judging the sentiment polarity of text (e.g., positive, negative, or neutral). Third, automatic classification methods based on machine learning or deep learning are used to identify topic categories or event types.

[0004] However, all of the above methods have certain limitations. Traditional graph theory methods mainly focus on the static topology of networks, making it difficult to characterize the spatial directionality of public opinion during dynamic propagation, and failing to answer key questions such as "information flows in which direction" and "whether it converges towards the center." While sentiment analysis and topic classification can reflect the state characteristics of public opinion, they struggle to quantify the "vector characteristics" of propagation (i.e., the direction and intensity of propagation), thus failing to effectively reveal the dynamic mechanism of public opinion evolution. In other words, existing technologies lack quantitative descriptions and patterned analyses of the directionality of public opinion propagation, making it difficult to predict the evolution path of public opinion and hindering targeted intervention.

[0005] Therefore, a new technical solution is needed to break through the framework of static structural analysis and innovatively quantify and model the directionality of public opinion dissemination in social networks to make up for the above-mentioned shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing social network public opinion analysis technologies, this invention provides a method for identifying social network public opinion propagation patterns based on directional quantitative indicators.

[0007] The method of this invention introduces high-dimensional semantic space modeling and spatial directionality index calculation to achieve quantitative analysis and pattern recognition of public opinion dissemination direction, thereby revealing the dissemination law and evolution characteristics of public opinion in social networks and providing a scientific basis for public opinion early warning and dissemination strategy optimization.

[0008] Specifically, the technical solution adopted by the present invention to achieve the above-mentioned objectives is as follows: A method for identifying social network public opinion propagation patterns based on directional quantitative indicators includes the following steps: S1: Obtain the raw data of the target social network, including user-generated text content, user relationship data, and user interaction traffic data; S2: Based on the semantic model, the text content is converted into a high-dimensional vector to construct a virtual social space, in which each user or topic has corresponding coordinates; S3: Based on the user interaction traffic data and the coordinates in the social space, calculate the core indicators used to quantify the directionality of information flow. The core indicators include at least the anisotropy indicator (Λ) which reflects the degree of uneven distribution of flow direction and the centripetal indicator (Γ) which reflects the tendency of flow to gather towards the central node. S4: Classify and identify public opinion dissemination patterns based on the numerical combinations of the anisotropy and centripetal indices. S5: Output classification and recognition results, explore their characteristics and evolution patterns for public opinion monitoring and guidance.

[0009] Preferably, in step S1: Data Acquisition and Preprocessing: Obtain raw data from the target social network platform, including user-generated text content, user follow relationship data, and user interaction behavior data; clean and preprocess the text content.

[0010] Preferably, in step S2: Social space construction: Using a pre-trained semantic model (such as SBERT), the pre-processed text content is converted into a high-dimensional semantic vector. The high-dimensional semantic vector is used as the coordinates of users or topic nodes in the virtual social space to construct the mapped social space.

[0011] Preferably, in step S3: S3.1: Define the traffic from source node i to target node j based on user interaction behavior data. and the total outflow of the source node ; S3.2: In the social space, calculate the predicted average vector for each source node i: (1) in Let be the unit direction vector pointing from node i to node j; S3.3: Based on the predicted average vector, calculate the node anisotropy index, which characterizes the degree of uneven distribution of traffic direction at each node: ; (2) S3.4: Based on the anisotropy of each node, calculate the anisotropy index that characterizes the degree of uneven distribution of overall network traffic directions: (3) S3.5: Specify a reference node C, and calculate the centripetal index of the flow from each node to that reference node: , (4) in Let be the angle between the unit direction vector of this node and the vector pointing from this node to the reference node C; based on the centripetality of each node's flow towards the reference node, calculate the centripetality index, which characterizes the degree to which the overall network traffic converges towards the center: (5) S3.6: The method for selecting reference nodes is to sequentially designate each node in the network as a reference node, calculate the centripetality of the network, and select the reference node corresponding to the maximum calculated centripetality value as the final reference node.

[0012] Preferably, in step S4: Based on the calculated combination of anisotropy index Λ and centripetal index Γ, the public opinion dissemination pattern is classified into at least one of decentralized dissemination, polarized dissemination, and aggregated dissemination through threshold judgment or cluster analysis.

[0013] Preferably, in step S5: The output includes the classification results of the propagation patterns, key indicator data, and visualization charts, which can be used for public opinion dynamic monitoring, key node identification, and propagation strategy optimization.

[0014] (III) Beneficial Effects of the Invention

[0015] Compared with the prior art, the technical solution provided by the present invention has the following significant advantages: (1) It realizes the quantitative measurement of the directionality of public opinion dissemination: It innovatively introduces anisotropy and centripetal indicators, and for the first time transfers spatial directionality analysis from physical urban traffic to virtual social networks. It can accurately quantify the direction preference and concentration of information flow, filling the gap in the existing technology in the quantitative description of directionality.

[0016] (2) It reveals the deep-seated propagation patterns and evolutionary laws: By combining the two-dimensional indicators of Λ and Γ, it is possible to clearly identify three essentially different propagation patterns: dispersion, polarization and aggregation. Furthermore, by combining time series analysis, it can reveal their dynamic evolution path, thus providing a new perspective for understanding the internal dynamic mechanism of public opinion propagation.

[0017] (3) Provides precise decision support capabilities: The quantitative results and pattern recognition conclusions provided by this method can provide scientific and intuitive decision-making basis for relevant departments to conduct public opinion early warning, precise guidance and crisis intervention. At the same time, it provides an operable quantitative tool for social platforms to optimize information distribution mechanisms and evaluate dissemination effects. Attached Figure Description

[0018] Figure 1 This is a flowchart of the method of the present invention; Figures 2a-2c This is a schematic diagram illustrating the calculation principle of anisotropy and centripetal indices provided in an embodiment of the present invention, wherein... Figure 2a This is a schematic diagram illustrating the principle of unit outflow vector calculation. Figure 2b This is a schematic diagram illustrating the principle of predicting the average vector. Figure 2c This is a schematic diagram of the angles required for centripetal calculation; Detailed Implementation

[0019] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0020] A method for identifying social network public opinion propagation patterns based on directional quantitative indicators includes the following steps: S1: Obtain the raw data of the target social network, including user-generated text content, user relationship data, and user interaction traffic data; S2: Based on the semantic model, the text content is converted into a high-dimensional vector to construct a virtual social space, in which each user or topic has corresponding coordinates; S3: Based on the user interaction traffic data and the coordinates in the social space, calculate the core indicators used to quantify the directionality of information flow. The core indicators include at least the anisotropy indicator (Λ) which reflects the degree of uneven distribution of flow direction and the centripetal indicator (Γ) which reflects the tendency of flow to gather towards the central node. S4: Classify and identify public opinion dissemination patterns based on the numerical combinations of the anisotropy and centripetal indices. S5: Output classification and recognition results, explore their characteristics and evolution patterns for public opinion monitoring and guidance.

[0021] Furthermore, step S1 includes the following steps: S1.1: From the dataset provided by Weibo, 30 trending events are randomly selected as experimental samples. Each event contains a root Weibo post (as source node i) and its corresponding forwarded Weibo post (as target node j).

[0022] S1.2: For each Weibo post, extract its basic attribute information, including Weibo ID, user ID, posting time, number of reposts, number of comments, and number of views; among which, the number of views or browsing popularity is used to define the source node. i To the target node j Dissemination traffic V ij .

[0023] S1.3: Based on user relationship information, construct a network of attention and interaction between nodes; record user behaviors such as following, forwarding, and commenting for subsequent propagation direction calculation.

[0024] S1.4: Clean and preprocess the acquired data, delete duplicate records, abnormal data and samples with missing content; remove stop words, emoticons and noise characters from text data, and unify the encoding format to ensure the integrity and consistency of data input.

[0025] Furthermore, step S2 includes the following steps: S2.1: The pre-trained semantic representation model Sentence-BERT (SBERT) is adopted, specifically the all-mpnet-base-v2 model. This model can map natural language text to a high-dimensional continuous semantic space, which is used to measure the semantic similarity and semantic direction relationship between different texts.

[0026] S2.2: Input the preprocessed Weibo text from step S1 and its corresponding comment text into the SBERT model sequentially to extract the semantic embedding vector for each text. Each semantic vector output by the model has a dimension of 768.

[0027] S2.3: For each user and the Weibo text they post, the generated semantic vector is used as the coordinate representation of the user or topic node in the virtual social space, thereby realizing the mapping relationship between the semantic layer and the spatial layer.

[0028] S2.4: Based on the semantic coordinates of all nodes, construct a high-dimensional semantic social space for trending events. This space is used for subsequent calculation of directional indicators and analysis of public opinion dissemination patterns.

[0029] Furthermore, step S3 includes the following steps: S3.1: To quantify the spatial orientation of public opinion flow in social networks, this invention introduces two directional quantitative indicators, anisotropy and centripetal indicators, referencing urban traffic flow research methods, to describe the directional distribution characteristics and aggregation trends of information flow.

[0030] S3.2: Please refer to Figure 2a In a high-dimensional semantic coordinate space, let the source node... i To the target node j The dissemination traffic is V ij , representing a node i To the node j Information dissemination intensity (such as Weibo read count or browsing popularity); let u ij For the node i Pointing to node j The unit direction vector.

[0031] S3.3: Please refer to Figure 2b Define nodes i The total outflow is S i , i.e., node i The sum of traffic directed to all target nodes. Based on this normalization coefficient, define the node. i The predicted average vector is as follows: (6) S3.4: Define Nodes i Anisotropy index This is used to characterize the degree of unevenness in the distribution of outflow directions at nodes, and its calculation formula is as follows: (7) in, The value range is 0 to 1. It can be seen that when all flows originating from position i... When in the same direction, Equals 1; when the flow originating from position i When space is symmetric, It equals 0. Therefore, The larger the value, the more concentrated the flow and the stronger the directionality of the nodes; conversely, the smaller the value, the weaker the anisotropy.

[0032] S3.5: Calculate the overall anisotropy index Λ of the network, obtained by performing a flow-weighted average of the anisotropy values ​​of the nodes: (8) Among them, Λ is used to reflect the degree of global imbalance in the overall traffic direction distribution of the network.

[0033] S3.6: Please refer to Figure 2cTo describe the centripetal characteristics of network flow, an angle parameter needs to be introduced. Suppose there exists a common reference node C in the network, used to measure the relative direction of flow from each node to the reference center. The angle between the direction vector of node i and the unit vector of reference node C is defined as: (9) in, For nodes i Point to reference node C The unit direction vector. When Fully pointing to the reference node C hour, =0; when the directions are completely opposite. .

[0034] S3.7: Define Nodes i centripetal index This is used to measure the degree to which the outflow direction of a node points towards the reference center, and the calculation formula is as follows: (10) in The value ranges from 0 to 1. The larger the value, the more concentrated the flow of the node is on the reference node.

[0035] S3.8: Calculate the overall centripetal force Γ of the network, and calculate the outflow-weighted average of the centripetal force of all nodes: (11) S3.9: Take each node in the network as a hypothetical reference node in turn, and calculate the corresponding network centripetal index Γ; select the node that makes Γ the largest as the final center node for public opinion dissemination, i.e., reference node C.

[0036] Furthermore, step S4 includes the following steps: S4.1: Calculate the overall anisotropy index Λ and centripetal index Γ of the network obtained in step S3 to form a two-dimensional feature space for propagation pattern recognition.

[0037] S4.2: Using the feature values ​​of each hot event in the range (Λ,Γ) as input, the public opinion dissemination characteristics of different events are automatically clustered through cluster analysis or threshold division methods.

[0038] S4.3: Based on clustering results or the relationship between indicator combinations, three main patterns of public opinion dissemination were identified: Distributed dissemination pattern: Both anisotropy Λ and centripetality Γ are low, indicating that the direction of public opinion dissemination is random and lacks a centralized trend; Polarized propagation pattern: high anisotropy Λ and low centripetality Γ, indicating that the information flow is bidirectional or multi-camp differentiation; Aggregated dissemination pattern: Both anisotropy Λ and centripetality Γ are high, indicating that the information flow tends to converge towards a unified central node or mainstream viewpoint, showing a clear aggregation trend.

[0039] S4.4: Output the classification results of the propagation patterns of each hot topic event, and save the identified pattern labels and their corresponding indicator values ​​for subsequent public opinion dynamic monitoring, pattern evolution analysis and strategy optimization.

[0040] Furthermore, step S5 includes the following steps: S5.1: Select typical hot topics to demonstrate the structural characteristics of different public opinion dissemination patterns. By analyzing the numerical distribution of the anisotropy index Λ and the centripetal index Γ, three typical dissemination pattern characteristics can be identified: (1) Decentralized dissemination model: Both anisotropy Λ and centripetality Γ are relatively small. Under this model, public opinion is initially distributed in a dispersed manner and does not form a clear trend of aggregation or unification over time. The network structure exhibits multi-center and low consistency characteristics. This phenomenon usually stems from the ambiguity of the issues, the lack of authoritative information sources, or the high heterogeneity of user group opinions.

[0041] (2) Polarized propagation pattern: high anisotropy Λ and low centripetalism Γ. Public opinion may initially focus on one direction, but then gradually differentiates into opposing camps, forming a clear two-way confrontational structure. This pattern is common in controversial political and social events and is influenced by both group polarization and algorithmic recommendations.

[0042] (3) Aggregated dissemination model: Both anisotropy (Λ) and centripetal force (Γ) are high. Although public opinion is initially scattered, it gradually converges towards core nodes or mainstream viewpoints over time, eventually forming a highly consistent dissemination network. This model is often seen in events where authoritative media intervene or where the factual basis of the issue is clear (such as public health topics).

[0043] S5.2: Perform time-slice visualization analysis on typical events to reveal the dynamic evolution of public opinion dissemination patterns.

[0044] In the decentralized dissemination model, the distribution of public opinion always shows a uniform diffusion characteristic with no significant directional change, similar to the random dissemination of "Brownian motion".

[0045] In a polarized communication model, opinions present an opposing structure from the initial stage, with multiple dense opinion clusters gradually forming stable camps that differentiate, and the degree of opposition intensifies over time.

[0046] In the aggregated dissemination model, scattered small opinion groups gradually move toward the central node, forming a "gravity model"-like aggregation characteristic, and the attractiveness of the central node increases with the intensity of dissemination.

[0047] S5.3: By comparing the characteristic changes of the three dissemination modes over time, it was found that the Λ and Γ indicators in the early stage of public opinion dissemination have predictive significance. This characteristic can be used to judge the subsequent trend of public opinion dissemination, providing an effective time window for public opinion early warning, guidance and intervention.

[0048] S5.4: Conduct group segmentation analysis on the social network structure of typical events to reveal differences in public opinion flow under different models: In a decentralized communication model, the flow of opinions among different groups is random and has no clear dominant direction; In a polarized communication model, opinions flow between groups tend to be in opposition, with a clear dominant group distribution. In the aggregated dissemination model, the flow of the group is concentrated towards the central node, which is consistent with the characteristics of public opinion aggregation.

[0049] S5.5: Based on the analysis results of this embodiment, public opinion monitoring personnel can promptly initiate early warning and guidance strategies according to the identified pattern type (such as "polarized propagation"); platform operators can conduct precise communication or key monitoring based on the core nodes (reference center C) identified by the centripetal index; researchers can conduct further propagation dynamics analysis based on the quantitative results output by this method.

[0050] The embodiments described in this specification are merely examples of implementations of the inventive concept. The scope of protection of this invention should not be considered as limited to the specific forms stated in the embodiments. The scope of protection of this invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A method for identifying social network public opinion propagation patterns based on directional quantitative indicators, characterized in that: Includes the following steps: S1: Obtain the raw data of the target social network, including user-generated text content, user relationship data, and user interaction traffic data; S2: Based on the semantic model, the text content is converted into a high-dimensional vector to construct a virtual social space, in which each user or topic has corresponding coordinates; S3: Based on the user interaction traffic data and the coordinates in the social space, calculate the core indicators used to quantify the directionality of information flow. The core indicators include at least anisotropy indicators that reflect the degree of uneven distribution of flow direction and centripetal indicators that reflect the trend of flow towards the central node. S4: Classify and identify public opinion dissemination patterns based on the numerical combinations of the anisotropy and centripetal indices. S5: Output classification and recognition results, explore their characteristics and evolution patterns for public opinion monitoring and guidance.

2. The method for identifying social network public opinion propagation patterns based on directional quantitative indicators as described in claim 1, characterized in that: Step S1 specifically includes: S1.1: Obtain raw data related to public opinion from the target social network platform, including text content posted by users, user relationship data, and interaction behavior data; S1.2: Extract the basic attribute information of the text, including text identifier, user identifier, publication time and interaction statistics, wherein the interaction statistics are used to define the propagation traffic between the source node and the target node; S1.3: Based on user relationships and interactive behaviors, construct a social connection network between nodes for the calculation of subsequent propagation directional indicators; S1.4: Clean and preprocess the raw data, delete duplicate or abnormal records, remove noise symbols and stop words from the text content, and ensure the accuracy and consistency of the data input.

3. The method for identifying social network public opinion propagation patterns based on directional quantitative indicators as described in claim 1, characterized in that: Step S2 specifically includes: S2.1: Using a pre-trained semantic representation model, the pre-processed text content in step S1 is vectorized to obtain the corresponding semantic embedding representation; S2.2: The semantic embedding result is used as the coordinate representation of user nodes or topic nodes in the virtual social space to reflect the distribution relationship of text semantics in high-dimensional space; S2.3: Based on the semantic coordinates of all nodes, construct a high-dimensional semantic space for the social network to achieve the mapping between the semantic layer and the spatial layer, providing basic data support for subsequent calculation of directional indicators.

4. The method for identifying social network public opinion propagation patterns based on directional quantitative indicators as described in claim 1, characterized in that: Step S3 specifically includes: S3.1: Based on user interaction behavior data, define the source node i To the target node j Dissemination traffic and the total outflow from the source node. ; S3.2: In the social space, calculate each source node i The unit outflow vector is used to characterize the overall directional trend of the outflow towards other nodes. The unit outflow vector is defined as follows: (1) in Let be the unit direction vector pointing from node i to node j; S3.3: Based on the aforementioned unit outflow vector, calculate the node anisotropy index, which characterizes the degree of uneven distribution of flow direction at each node: （2） Where, λ i The value range is [0,1]. The larger the value, the more concentrated the flow of the node and the stronger the directionality. S3.4: Using the traffic of each node as weight, perform a weighted average of the above anisotropy to obtain a global anisotropy index that characterizes the degree of uneven distribution of the overall directional distribution of the network: （3） Among them, Λ reflects the overall directional concentration of online public opinion flow in spatial distribution. S3.5: Specify a reference node C, and calculate the centripetal index of the flow from each node to that reference node: ， （4） in The angle between the unit direction vector of the node and the vector pointing from the node to the reference node C; S3.6: By traversing each node in the social network and using them as reference nodes in turn, the centripetal force of the network is calculated, and the node with the maximum centripetal force is selected as the final reference center.

5. The method for identifying social network public opinion propagation patterns based on directional quantitative indicators as described in claim 1, characterized in that: Step S4 specifically includes: S4.1: Based on the calculation results of step S3, obtain the anisotropy index Λ and centripetal index Γ of the network as a whole, and construct a propagation feature space with (Λ, Γ) as the two-dimensional feature. S4.2: Using the feature values ​​of each hot topic event in the feature space as input variables, and using clustering analysis algorithms or threshold-based partitioning methods, the public opinion dissemination characteristics of different events are automatically classified. S4.3: Based on the distribution results of cluster centers or feature intervals, identify and label the propagation patterns of events, including: Distributed dissemination pattern: Both anisotropy Λ and centripetality Γ are low, indicating that the direction of public opinion dissemination is random and lacks a centralized trend; Polarized propagation pattern: high anisotropy Λ and low centripetality Γ, indicating that the information flow is bidirectional or multi-camp differentiation; Aggregated dissemination pattern: Both anisotropy Λ and centripetality Γ are high, indicating that the information flow tends to converge towards a unified central node or mainstream viewpoint, showing a clear aggregation trend.

6. The method for identifying social network public opinion propagation patterns based on directional quantitative indicators as described in claim 1, characterized in that: Step S5 specifically includes: S5.1: Output the propagation pattern recognition results and corresponding directional quantification indicators described in step S4; S5.2: Based on the identified propagation patterns, generate visual analysis charts, including event propagation structure diagrams, indicator change curves, and node distribution diagrams; S5.3: The analysis results are used for public opinion dynamic monitoring, key node identification and dissemination strategy optimization, so as to achieve quantitative analysis and real-time early warning of public opinion evolution trends.

7. A method for identifying social network public opinion propagation patterns based on directional quantitative indicators, characterized in that, Public opinion monitoring personnel can promptly initiate early warning and guidance strategies based on the identified pattern types; platform operators can conduct precise communication or key monitoring based on the core nodes identified by the centripetal index; researchers can use this method to implement the social network public opinion dissemination pattern identification method based on directional quantitative indicators as described in any one of claims 1-6.

Images