Opinion traceability method based on multi-source heterogeneous data

Abstract

The application discloses an opinion tracing method based on multi-source heterogeneous data, which comprises the following steps: acquiring multi-source heterogeneous data; normalizing and fusing the multi-source heterogeneous data to construct an opinion propagation network; constructing the opinion propagation network comprises constructing an opinion propagation tree set, processing the opinion propagation tree set, and constructing a multi-layer opinion propagation network based on the opinion propagation tree set; and determining an opinion propagation source in the multi-layer opinion propagation network based on a multi-layer local influence evaluation function and a shortest distance. The application can accurately locate the opinion propagation source from multi-source heterogeneous data.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method for tracing public opinion sources based on multi-source heterogeneous data. Background Technology

[0002] Current methods for tracing the source of public opinion rely on data from a single platform. However, border-related public opinion involves multiple platforms, languages, and cultural backgrounds. Given this heterogeneous data, current methods fall far short of capturing the full picture of public opinion dissemination, resulting in significant discrepancies between the tracing results and the actual source of the dissemination. This is primarily manifested in incomplete data coverage, which fails to meet the tracing needs of modern cross-platform public opinion dissemination, especially for information that spreads rapidly across multiple platforms in border areas, making it difficult to accurately identify the source and effectively manage it.

[0003] Because public opinion spreads across platforms, data tracing based on a single platform is insufficient to address this dissemination pattern, proving particularly inadequate in border public opinion governance. In contrast, multi-heterogeneous data integrates rich information from multiple platforms, encompassing data from various mainstream social media platforms, news websites, forums, and other channels. Its public opinion tracing algorithm can more comprehensively capture the dissemination path of public opinion, meeting the practical needs of cross-platform communication and providing strong technical support for the effective governance of border public opinion. Summary of the Invention

[0004] To address the issues of incomplete data coverage and low accuracy in traditional single-platform tracing methods, as well as the reliance on data from a single platform leading to significant discrepancies between the tracing results and the actual source, thus failing to meet the needs of cross-platform tracing, this application provides a public opinion tracing method based on multi-source heterogeneous data, enabling accurate tracing of the cross-platform public opinion dissemination path.

[0005] This application discloses a method for tracing public opinion sources based on multi-source heterogeneous data, which includes:

[0006] Acquire multi-source heterogeneous data; multi-source heterogeneous data includes data from social media platforms and traditional online media;

[0007] Normalize and merge multi-source heterogeneous data to construct a public opinion dissemination network; constructing a public opinion dissemination network includes constructing a public opinion dissemination tree set, processing the public opinion dissemination tree set, and constructing a multi-layer public opinion dissemination network based on the public opinion dissemination tree set;

[0008] Based on the multi-layer local impact assessment function and the shortest distance, the source of public opinion dissemination in the multi-layer public opinion dissemination network is determined.

[0009] Furthermore, the multi-source heterogeneous data originates from publicly available data on different network platforms or channels, including: data corresponding to users' direct participation in designated topics and derived sub-topics or sub-themes. This data includes the user's unique identifier ID, username, user attributes, user behavior data, and content dissemination path. Among these, user attributes include authentication type and number of followers; user behavior data includes posting frequency and interaction intensity; interaction intensity includes the number of comments, likes, and shares; and content dissemination path includes Weibo's reposting chain and TikTok's sharing data.

[0010] Furthermore, the construction of the public opinion propagation tree set includes:

[0011] The algorithm automatically identifies relevant public opinion content from multi-source heterogeneous data using topic clustering. All public opinion content under the same topic is grouped into the same set. Starting from each piece of public opinion content that directly participates in the dissemination of public opinion, a separate public opinion dissemination tree is constructed for each piece of content. During the construction of the public opinion dissemination tree, the publisher's ID and username are obtained based on the public opinion content, and the publisher is set as the root node of the public opinion dissemination tree. After the public opinion content is published, the network platform accurately pushes the public opinion content to the corresponding users according to its own preset recommendation mechanism. Users forward or comment on the received content, and users who participate in the interaction are added as new nodes to the public opinion dissemination tree. When some users spread the message, a sub-topic is generated, which in turn affects other users. Users who participate in the dissemination due to the influence of the sub-topic are also included as nodes in the public opinion dissemination tree, until no new users are added to the public opinion dissemination tree.

[0012] For all public opinion content under the same theme, construct corresponding public opinion propagation trees to obtain the final set of public opinion propagation trees for that theme.

[0013] Furthermore, the processing of the public opinion propagation tree set includes:

[0014] The system sets the degree of contribution of each node to the spread of public opinion and judges the sentiment analysis of each node in the public opinion spread tree to determine whether the role and content expressed by each node in the process of public opinion spread are positive, negative or neutral.

[0015] Furthermore, the propagation contribution of the node includes:

[0016] This involves organically integrating relevant information gathered by users along their respective opinion dissemination tree to construct a framework reflecting a node's contribution to the spread of a topic; relevant information includes the number of likes. Page views Number of comments and number of reposts ;

[0017] Assign appropriate weights to relevant information to obtain the propagation contribution of each node. ;

[0018] The node's sentiment analysis of public opinion includes:

[0019] To analyze public opinion content related to topics involving nodes, a sentiment classification method based on a sentiment dictionary is adopted from natural language processing technology. First, a sentiment dictionary needs to be built in advance. When building the sentiment dictionary, the comment characteristics of users on different social media platforms are combined, and TF-IDF is used to extract high-frequency sentiment words from each platform. Then, existing open-source dictionaries are integrated to build a general sentiment dictionary. By weighting the sentiment expressed in the public opinion content published by users, it is determined whether the opinion type of the public opinion content is positive, negative, or neutral.

[0020] Furthermore, the construction of a multi-layered public opinion dissemination network based on the public opinion dissemination tree set includes:

[0021] Based on different platforms and user IDs and usernames, the nodes in the public opinion propagation tree are linked and integrated to construct a multi-layered public opinion propagation network, so as to clearly present the propagation path and scope of influence of public opinion among different platforms and users.

[0022] The process involves associating and integrating various nodes in the public opinion dissemination tree set based on different platforms and user IDs and usernames to construct a multi-layered public opinion dissemination network, including:

[0023] This involves associating the same users across different platforms and integrating the same users within the same platform: When nodes with the same ID and username exist in the public opinion propagation tree set of the same platform, the duplicate nodes are integrated into one node, while preserving the edges between the original nodes; after all public opinion propagation trees of the same platform have been integrated, a single public opinion propagation network is constructed; when the same ID, user, or the same public opinion content appears on different platforms, the corresponding nodes are connected to associate them, representing the same user; after performing association on all platforms, multiple single public opinion propagation networks constitute a multi-layered public opinion propagation network; and the propagation contribution of nodes in the multi-layered public opinion propagation network is also considered. Adjusted to:

[0024]

[0025] in, Contribution to overall dissemination For nodes In the Types of viewpoints on the internet that spread public opinion. For nodes In the Contribution of public opinion to the spread of information on the network. This refers to the total number of layers in a multi-layered public opinion dissemination network; the opinion type includes positive, negative, or neutral.

[0026] Furthermore, the determination of the source of public opinion dissemination in the multi-layered public opinion dissemination network based on the multi-layered local impact assessment function and the shortest distance includes:

[0027] Based on the weight value of each platform With constant The product is assigned a constant to each platform. This is used to find the number of suspected source nodes for propagation on the platform;

[0028] The nodes in each multi-layered public opinion dissemination network are ranked according to their dissemination contribution, and the top nodes are selected. Each node serves as the initial set of suspected propagation source nodes;

[0029] Traverse all nodes within the two-hop region surrounding the initial suspected propagation source node set, and calculate the reference value of the initial suspected propagation source based on the multi-layer local influence evaluation function and the shortest distance. Select reference value The largest Each node constitutes a suspected propagation source node set;

[0030] From the suspected source node set Select reference value The largest node is used as the source of public opinion dissemination in a multi-layered public opinion dissemination network.

[0031] Furthermore, the reference value for the initial suspected propagation source is calculated based on the multi-layer local influence assessment function and the shortest distance. ,include:

[0032] The reference value for a node as the initial suspected source of propagation is obtained using the following formula. :

[0033]

[0034] in, , , Let be the weighting coefficient, satisfying , As a smoothing factor, and These represent the preset propagation contribution and the shortest path, respectively. For nodes global propagation contribution This is a multi-layered local impact assessment function. The distance between the suspected source node set and other nodes in the multi-layered public opinion propagation network The minimum value of the propagation eccentricity, For the suspected source node set;

[0035] All nodes of the multi-layered public opinion dissemination network Reference value The largest node is considered the suspected source of propagation, meaning that this node satisfies the source node evaluation function. :

[0036]

[0037] in, A set function that maximizes the parameters.

[0038] Furthermore, for multi-layered public opinion dissemination networks, a multi-layered local impact assessment function is established:

[0039]

[0040]

[0041]

[0042]

[0043] in, This is a multi-layered local impact assessment function. For suspected propagation source node set The number of nodes included. and These are the suspected propagation source node sets. The expected range of influence between one-hop neighbor nodes and two-hop neighbor nodes. For suspected propagation source node set The average local clustering coefficient of the nodes in other layers. For suspected propagation source node set The set of one-hop neighbor nodes, It is the set of all edges in a multi-layered public opinion dissemination network. For nodes and nodes The weight values ​​of the edges between them. For suspected propagation source node set The set of two-hop neighbor nodes, The preset edge weight values, For nodes exist and The number of edges between node sets As the first in a multi-layered public opinion dissemination network Layered network, This refers to the total number of layers in a multi-layered public opinion dissemination network. For nodes In the The number of edges between neighboring nodes in a hierarchical public opinion propagation network. For nodes In the The number of neighboring nodes in the hierarchical public opinion dissemination network.

[0044] Furthermore, the suspected source node set is located at a distance from nodes in other multi-layered public opinion propagation networks. The formula for calculating the minimum value R of the propagation eccentricity is:

[0045]

[0046] in, For nodes The propagation eccentricity, which is the node Distance from other nodes in the multi-layered public opinion dissemination network The maximum value of the sum of the shortest distances. For nodes and nodes The shortest distance between them The distance between the suspected source node set and other nodes in the multi-layered public opinion propagation network The node with the minimum propagation eccentricity has the highest probability of becoming a propagation source. It is a function with maximum value. A set function that minimizes parameters.

[0047] Due to the adoption of the above technical solution, this application has the following advantages:

[0048] 1. This application constructs a multi-layered public opinion dissemination network by crawling multiple heterogeneous data and using normalization and fusion operations. This eliminates the differences between heterogeneous data while retaining the characteristics of different types of data. By comprehensively considering the contribution of nodes to public opinion dissemination from the perspectives of influence and information dissemination methods, it identifies the key sources and important dissemination nodes of public opinion dissemination, providing strong support for the monitoring, early warning, and response to public opinion.

[0049] 2. This application can accurately locate the source of public opinion dissemination. Compared with traditional single-platform tracing methods, the method proposed in this application has significant advantages in terms of data coverage and tracing accuracy. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments recorded in the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings.

[0051] Figure 1 This is a flowchart illustrating a public opinion tracing method based on multi-source heterogeneous data according to an embodiment of this application.

[0052] Figure 2 This is a flowchart illustrating another method for tracing public opinion based on multi-source heterogeneous data, as described in this application. Detailed Implementation

[0053] The present application will be further described in conjunction with the accompanying drawings and embodiments. The described embodiments are only some, not all, of the embodiments of the present application. All other embodiments obtained by those skilled in the art should fall within the protection scope of the embodiments of the present application.

[0054] See Figure 1 and Figure 2 This application provides an embodiment of a public opinion tracing method based on multi-source heterogeneous data, which includes:

[0055] Acquire multi-source heterogeneous data; multi-source heterogeneous data includes data from social media platforms and traditional online media;

[0056] Normalize and merge multi-source heterogeneous data to construct a public opinion dissemination network; constructing a public opinion dissemination network includes constructing a public opinion dissemination tree set, processing the public opinion dissemination tree set, and constructing a multi-layer public opinion dissemination network based on the public opinion dissemination tree set;

[0057] Based on the multi-layer local impact assessment function and the shortest distance, the source of public opinion dissemination in the multi-layer public opinion dissemination network is determined.

[0058] The public opinion dissemination network in this application embodiment retains the characteristics of public opinion dissemination on each platform and also achieves the integration of data structures.

[0059] Optionally, the multi-source heterogeneous data originates from publicly available data on different network platforms or channels, including: data corresponding to users' direct participation in designated topics and derived sub-topics or sub-themes. This data includes the user's unique identifier ID, username, user attributes, user behavior data, and content dissemination path. Among these, user attributes include authentication type and number of followers; user behavior data includes posting frequency and interaction intensity; interaction intensity includes the number of comments, likes, and shares; and content dissemination path includes Weibo's reposting chain and TikTok's sharing data.

[0060] This application's embodiments can capture public opinion information from multiple dimensions for specified topics or themes, ensuring the breadth and richness of data sources. It can not only access social media platforms such as Weibo, WeChat, and Douyin to capture key data in real time, including user-posted text content, posting time, number of likes, number of comments, and forwarding paths; it can also establish data connections with traditional media such as news websites, forums, and blogs to obtain authoritative news reports, in-depth analysis articles, and user comments. Furthermore, it can automatically adapt to the data structures and anti-scraping mechanisms of different websites, efficiently and stably completing data collection tasks and providing a solid data foundation for subsequent data processing and analysis. This application can achieve comprehensive coverage of cross-platform public opinion information, providing strong support for accurate source tracing.

[0061] This application's embodiments convert data collected from different platforms into a unified standard format. At the semantic level, considering the diversity and differences in language expression among users on different platforms, advanced natural language processing techniques are introduced for semantic normalization. To address the issue of dimensional differences, a normalization algorithm is used to standardize numerical data.

[0062] In this embodiment of the application, information always spreads along the shortest path, which means that during the spread of public opinion, information will prioritize the path that can reach more audiences the fastest. People with high influence are more likely to spread public opinion, that is, individuals or nodes with high influence play a key role in the spread of public opinion, and they can spread information more quickly and widely.

[0063] Optionally, constructing the public opinion propagation tree set includes:

[0064] The algorithm automatically identifies relevant public opinion content from multi-source heterogeneous data using topic clustering. All public opinion content under the same topic is grouped into the same set. Starting from each piece of public opinion content that directly participates in the dissemination of public opinion, a separate public opinion dissemination tree is constructed for each piece of content. During the construction of the public opinion dissemination tree, the publisher's ID and username are obtained based on the public opinion content, and the publisher is set as the root node of the public opinion dissemination tree. After the public opinion content is published, the network platform accurately pushes the public opinion content to the corresponding users according to its own preset recommendation mechanism. Users forward or comment on the received content, and users who participate in the interaction are added as new nodes to the public opinion dissemination tree. When some users spread the message, a sub-topic is generated, which in turn affects other users. Users who participate in the dissemination due to the influence of the sub-topic are also included as nodes in the public opinion dissemination tree, until no new users are added to the public opinion dissemination tree.

[0065] For all public opinion content under the same theme, construct corresponding public opinion propagation trees to obtain the final set of public opinion propagation trees for that theme.

[0066] Optionally, the processing of the public opinion propagation tree set includes:

[0067] The system sets the degree of contribution of each node to the spread of public opinion and judges the sentiment analysis of each node in the public opinion spread tree to determine whether the role and content expressed by each node in the process of public opinion spread are positive, negative or neutral.

[0068] Optionally, the propagation contribution of the node includes:

[0069] This involves organically integrating relevant information gathered by users along their respective opinion dissemination tree to construct a framework reflecting a node's contribution to the spread of a topic; relevant information includes the number of likes. Page views Number of comments and number of reposts ;

[0070] Based on the number of likes Taking the range standardization method as an example, let... For the first The maximum number of likes among all nodes in the tree. To get the minimum value, the standardized number of likes The calculation formula is:

[0071]

[0072] Page views Number of comments and number of reposts The standardized pageviews were obtained by performing standardization using the method described above. Number of comments and number of reposts After standardization, based on the importance of each indicator in the relevant information during the propagation process, appropriate weights are assigned to them, thus determining the node's propagation contribution. The calculation formula is:

[0073]

[0074] The weight of likes is: Page views have a weighting of The weight of the number of comments is The weight of the number of reposts is ,and ;

[0075] The node's sentiment analysis of public opinion includes:

[0076] To analyze public opinion content related to topics involving nodes, a sentiment classification method based on a sentiment dictionary is adopted from natural language processing technology. First, a sentiment dictionary needs to be built in advance. When building the sentiment dictionary, the comment characteristics of users on different social media platforms are combined, and TF-IDF is used to extract high-frequency sentiment words from each platform. Then, existing open-source dictionaries are integrated to build a general sentiment dictionary. By weighting the sentiment expressed in the public opinion content published by users, it is determined whether the opinion type of the public opinion content is positive, negative, or neutral.

[0077] For example, opinion type A of public opinion content can be:

[0078]

[0079] Optionally, the construction of a multi-layered public opinion dissemination network based on the public opinion dissemination tree set includes:

[0080] Based on different platforms and user IDs and usernames, the nodes in the public opinion propagation tree are linked and integrated to construct a multi-layered public opinion propagation network, so as to clearly present the propagation path and scope of influence of public opinion among different platforms and users.

[0081] The process involves associating and integrating various nodes in the public opinion dissemination tree set based on different platforms and user IDs and usernames to construct a multi-layered public opinion dissemination network, including:

[0082] This involves associating the same users across different platforms and integrating the same users within the same platform: When nodes with the same ID and username exist in the public opinion propagation tree set of the same platform, the duplicate nodes are integrated into one node, while preserving the edges between the original nodes; after all public opinion propagation trees of the same platform have been integrated, a single public opinion propagation network is constructed; when the same ID, user, or the same public opinion content appears on different platforms, the corresponding nodes are connected to associate them, representing the same user; after performing association on all platforms, multiple single public opinion propagation networks constitute a multi-layered public opinion propagation network; and the propagation contribution of nodes in the multi-layered public opinion propagation network is also considered. Adjusted to:

[0083]

[0084] in, Contribution to overall dissemination For nodes In the Types of viewpoints on the internet that spread public opinion. For nodes In the Contribution of public opinion to the spread of information on the network. This refers to the total number of layers in a multi-layered public opinion dissemination network; the opinion type includes positive, negative, or neutral.

[0085] This application embodiment associates and integrates the nodes in the propagation tree set according to the ID and name of different platforms and users, thereby constructing a complete, three-dimensional, multi-layered public opinion propagation network to clearly present the propagation path and scope of influence of public opinion among different platforms and users.

[0086] Optionally, determining the source of public opinion dissemination in a multi-layered public opinion dissemination network based on a multi-layered local impact assessment function and the shortest distance includes:

[0087] Based on the weight value of each platform With constant The product is assigned a constant to each platform. This is used to find the number of suspected source nodes for propagation on the platform;

[0088] The nodes in each multi-layered public opinion dissemination network are ranked according to their dissemination contribution, and the top nodes are selected. Each node serves as the initial set of suspected propagation source nodes;

[0089] Traverse all nodes within the two-hop region surrounding the initial suspected propagation source node set, and calculate the reference value of the initial suspected propagation source based on the multi-layer local influence evaluation function and the shortest distance. Select reference value The largest Each node constitutes a suspected propagation source node set;

[0090] From the suspected source node set Select reference value The largest node is used as the source of public opinion dissemination in a multi-layered public opinion dissemination network.

[0091] Optionally, the reference value for calculating the initial suspected propagation source is based on the multi-layer local influence assessment function and the shortest distance. ,include:

[0092] The reference value for a node as the initial suspected source of propagation is obtained using the following formula. :

[0093]

[0094] in, , , Let be the weighting coefficient, satisfying , As a smoothing factor, and These represent the preset propagation contribution and the shortest path, respectively. For nodes global propagation contribution This is a multi-layered local impact assessment function. The distance between the suspected source node set and other nodes in the multi-layered public opinion propagation network The minimum value of the propagation eccentricity, For the suspected source node set;

[0095] All nodes of the multi-layered public opinion dissemination network Reference value The largest node is considered the suspected source of propagation, meaning that this node satisfies the source node evaluation function. :

[0096]

[0097] in, A set function that maximizes the parameters.

[0098] Optionally, for multi-layered public opinion dissemination networks, it is necessary to consider both the local impact within the platform and the influence of nodes on other platforms. Therefore, a multi-layered local impact assessment function is established:

[0099]

[0100]

[0101]

[0102]

[0103] in, This is a multi-layered local impact assessment function. For suspected propagation source node set The number of nodes included. and These are the suspected propagation source node sets. The expected range of influence between one-hop neighbor nodes and two-hop neighbor nodes. For suspected propagation source node set The average local clustering coefficient of the nodes in other layers. For suspected propagation source node set The set of one-hop neighbor nodes, It is the set of all edges in a multi-layered public opinion dissemination network. For nodes and nodes The weight values ​​of the edges between them. For suspected propagation source node set The set of two-hop neighbor nodes, The preset edge weight values, For nodes exist and The number of edges between node sets As the first in a multi-layered public opinion dissemination network Layered network, This refers to the total number of layers in a multi-layered public opinion dissemination network. For nodes In the The number of edges between neighboring nodes in a hierarchical public opinion propagation network. For nodes In the The number of neighboring nodes in the hierarchical public opinion dissemination network.

[0104] Optionally, the suspected propagation source node set is located at a distance from nodes in other multi-layered public opinion propagation networks. The formula for calculating the minimum value R of the propagation eccentricity is:

[0105]

[0106] in, For nodes The propagation eccentricity, which is the node Distance from other nodes in the multi-layered public opinion dissemination network The maximum value of the sum of the shortest distances. For nodes and nodes The shortest distance between them The distance between the suspected source node set and other nodes in the multi-layered public opinion propagation network The node with the minimum propagation eccentricity has the highest probability of becoming a propagation source. It is a function with maximum value. A set function that minimizes parameters.

[0107] This application constructs a multi-layered public opinion dissemination network by crawling multiple heterogeneous data sets and using normalization and fusion operations. This eliminates the differences between heterogeneous data while preserving the characteristics of different data types. By considering the contribution of nodes to public opinion dissemination from the perspectives of influence and information dissemination methods, it identifies key sources and important dissemination nodes, providing strong support for the monitoring, early warning, and response to public opinion.

[0108] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them. Although this application has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of this application. Any modifications or equivalent substitutions that do not depart from the spirit and scope of this application should be covered within the protection scope of the claims of this application.

Claims

1. A public opinion tracing method based on multi-source heterogeneous data, characterized in that, The method comprises the following steps: acquiring multi-source heterogeneous data; the multi-source heterogeneous data comprises data from social media platforms and traditional network media; normalizing and fusing the multi-source heterogeneous data to construct an opinion propagation network; constructing the opinion propagation network comprises constructing an opinion propagation tree set, processing the opinion propagation tree set, and constructing a multi-layer opinion propagation network based on the opinion propagation tree set; determining an opinion propagation source in the multi-layer opinion propagation network based on a multi-layer local influence evaluation function and a shortest distance; the processing of the opinion propagation tree set comprises: setting a propagation contribution degree of a node and judging a sentiment analysis of the node in the opinion propagation tree to determine a role played by each node in an opinion propagation process and an expression content of the node, which is positive, negative, or neutral; the constructing of the multi-layer opinion propagation network based on the opinion propagation tree set comprises: associating and integrating each node in the opinion propagation tree set according to different platforms and IDs and usernames of users to construct a multi-layer opinion propagation network to clearly present a propagation path and an influence range of the opinion among different platforms and users; the associating and integrating of each node in the opinion propagation tree set according to different platforms and IDs and usernames of users to construct the multi-layer opinion propagation network comprises: The association of the same user between different platforms and the integration of the same user in the same platform: when the same platform's public opinion propagation tree set has nodes with the same ID and username, the repeated nodes are integrated into one node while the edges between the original nodes are retained; after all the public opinion propagation trees of the same platform are integrated, a single public opinion propagation network is constructed; when the same ID, user or the same public opinion content appears in different platforms, the corresponding nodes are connected for association, which is used to represent the same user; after the association of all platforms is performed, multiple single public opinion propagation networks constitute a multi-layer public opinion propagation network; and the propagation contribution degree of the nodes in the multi-layer public opinion propagation network is adjusted to be: wherein, a global contribution degree, a node In the first a view type on the opinion propagation network of the layer, a node In the first a propagation contribution degree on the opinion propagation network of the layer, a total number of layers of the multi-layer opinion propagation network; the view type includes positive, negative, or neutral. the determining of the opinion propagation source in the multi-layer opinion propagation network based on the multi-layer local influence evaluation function and the shortest distance comprises: According to the weight value of each platform The product of the constant A constant is assigned to each platform , used to find the number of suspected source nodes of the platform Sort the nodes in each public opinion propagation network in the multi-layer public opinion propagation network by propagation contribution, and select the top nodes as the initial suspected propagation source node set; Traverse all nodes in the two-hop area around the initial suspected propagation source node set, and calculate the reference value of the initial suspected propagation source based on a multi-layer local influence evaluation function and the shortest distance , select the maximum nodes with reference value to constitute the suspected propagation source node set; from the suspect set of propagation source nodes reference value the largest node and take it as the source of public opinion in the multi-layer public opinion propagation network The reference value of the initial suspicious propagation source is calculated based on the multi-layer local influence evaluation function and the shortest distance , comprising: The reference value of a node as an initial suspicious propagation source is obtained by the following equation : wherein, , , is a weight coefficient, satisfying , is a smoothing factor, and are respectively a preset propagation contribution degree and a shortest path, is a global propagation contribution degree of a node , is a multi-layer local influence evaluation function, is a minimum value of a propagation eccentricity of a node in a multi-layer public opinion propagation network from a suspect propagation source node set, is a suspect propagation source node set; all nodes of the multi-layer public opinion propagation network are referenced the largest node as a suspected propagation source, i.e., the node satisfies a tracing node evaluation function : wherein is a set function maximizing over parameters; for the multi-layer opinion propagation network, a multi-layer local influence evaluation function is established: in, This is a multi-layered local impact assessment function. For suspected propagation source node set The number of nodes included. and These are the suspected propagation source node sets. The expected range of influence between one-hop neighbor nodes and two-hop neighbor nodes. For suspected propagation source node set The average local clustering coefficient of the nodes in other layers. For suspected propagation source node set The set of one-hop neighbor nodes, It is the set of all edges in a multi-layered public opinion dissemination network. For nodes and nodes The weight values ​​of the edges between them. For suspected propagation source node set The set of two-hop neighbor nodes, The preset edge weight values, For nodes exist and The number of edges between node sets As the first in a multi-layered public opinion dissemination network Layered network, This refers to the total number of layers in a multi-layered public opinion dissemination network. For nodes In the The number of edges between neighboring nodes in a hierarchical public opinion propagation network. For nodes In the The number of neighboring nodes in the hierarchical public opinion dissemination network.

2. The method of claim 1, wherein, the multi-source heterogeneous data is obtained from data disclosed by different network platforms or channels, and comprises data corresponding to a user directly participating in a specified topic and a subtopic or subtheme derived from the specified topic, the data comprising a unique identifier ID of the user, a username of the user, user attributes, user behavior data, and a content propagation path; wherein the user attributes comprise an authentication type and a number of fans; the user behavior data comprises a publishing frequency and an interaction intensity; the interaction intensity comprises a number of comments, a number of likes, and a number of shares; and the content propagation path comprises a retweet chain of a microblog and sharing data of TikTok.

3. The method of claim 1, wherein, the constructing of the opinion propagation tree set comprises: identifying, by a topic clustering algorithm, opinion content with correlation from the multi-source heterogeneous data, grouping all opinion content under a same topic into a same set, taking each opinion content directly participating in opinion propagation as a starting point to construct an opinion propagation tree for each opinion content, obtaining an ID and a username of a publisher of the opinion content as a root node of the opinion propagation tree during the construction of the opinion propagation tree, pushing the opinion content to corresponding users according to a recommendation mechanism of a network platform after the opinion content is published, adding users participating in interaction as new nodes to the opinion propagation tree when the users forward or comment on the received content, adding users participating in propagation as nodes to the opinion propagation tree when a subtheme is derived from part of the users during the propagation of the message, and adding users participating in propagation as nodes to the opinion propagation tree when the users are affected by the subtheme until no new user joins the opinion propagation tree. The same theme is respectively constructed for all the public opinion contents to obtain a corresponding public opinion propagation tree, and finally a set of public opinion propagation trees of the theme is obtained.

4. The method of claim 1, wherein, The propagation contribution degree of the node comprises: The related information harvested by the user on the public opinion propagation tree in which the user is located is organically integrated to construct a node contribution degree to the theme propagation; the related information includes a number of likes , a number of views , a number of comments , and a number of forwards ; Assigning corresponding weights to the relevant information to obtain a propagation contribution degree of the node ; The sentiment analysis of the node to the public opinion comprises: The public opinion content of the topic participated by the node is analyzed, a sentiment classification method based on a sentiment dictionary in a natural language processing technology is adopted, a sentiment dictionary needs to be constructed in advance, in the construction of the sentiment dictionary, the high-frequency sentiment words of each platform are extracted by using TF-IDF in combination with the comment features of the users of different social media platforms, and an existing open source dictionary is fused to construct a general sentiment dictionary; the sentiment expressed by the public opinion content published by the user is weighted and calculated to determine whether the view type of the public opinion content is positive, negative or neutral.

5. The method of claim 1, wherein, The suspected propagation source node set distance other nodes in a multi-layer public opinion propagation network The calculation formula of the minimum value R of the propagation eccentricity of the nodes in the multi-layer public opinion propagation network is: in, For nodes The propagation eccentricity, which is the node Distance from other nodes in the multi-layered public opinion dissemination network The maximum value of the sum of the shortest distances. For nodes and nodes The shortest distance between them The distance between the suspected source node set and other nodes in the multi-layered public opinion propagation network The node with the minimum propagation eccentricity has the highest probability of becoming a propagation source. It is a function with maximum value. A set function that minimizes parameters.

Images