Search

CN-121980303-A - Public opinion event risk assessment method and device based on multi-layer network

CN121980303ACN 121980303 ACN121980303 ACN 121980303ACN-121980303-A

Abstract

The invention relates to a public opinion event risk assessment method and device based on a multi-layer network, which belong to the technical field of event risk assessment and comprise the steps of obtaining a plurality of social platform topic event total sets, calculating text similarity, user similarity and emotion fluctuation similarity for any one social platform, conducting unified dimension and direction processing, conducting weighting calculation on final association degrees of any two topic events in the plurality of social platform topic event total sets, constructing the multi-layer network for each social platform according to the final association degrees of the topic events and any two topic events in the plurality of social platform topic event total sets, determining the multi-layer network, removing the sets to obtain an infection source set, sequentially taking each node in the infection source set as an infection source to obtain an infection state set, an infection duration set and a neighbor infection proportion set of each node, determining a risk score of each node, and determining the risk event according to the size of the risk score.

Inventors

  • LIU TING

Assignees

  • 天翼物联科技有限公司

Dates

Publication Date
20260505
Application Date
20251226

Claims (10)

  1. 1. The public opinion event risk assessment method based on the multi-layer network is characterized by comprising the following steps of: obtaining topic event sets of each social platform in a first time period, and acquiring intersections of the topic event sets to obtain topic event total sets of the social platforms; For any one social platform, calculating text similarity, user similarity and emotion fluctuation similarity for any two topic events in the total set of topic events of a plurality of social platforms from three angles of text semantics, user overlap and emotion fluctuation respectively; performing unified dimension and direction processing on the text similarity, the user similarity and the emotion fluctuation similarity; After the text similarity, the user similarity and the emotion fluctuation similarity subjected to unified dimension and direction processing are weighted, calculating the final association degree of any two topic events in the total set of topic events of the social platform; For each social platform in the plurality of social platforms, constructing a multi-layer network according to the topic event and the final association degree of any two topic events in the total collection of the topic events of the plurality of social platforms; Determining a set for the multi-layer network, and de-duplicating the set to obtain an infection source set; sequentially taking each node in the infection source set as an infection source, and respectively performing SIR algorithm simulation to obtain an infection state set, an infection duration set and a neighbor infection proportion set of each node; And determining a risk score of each node according to the infection state set, the infection duration set and the neighbor infection proportion set of each node, and determining a risk topic event according to the size of the risk score of each node.
  2. 2. The multi-layer network-based public opinion event risk assessment method of claim 1, wherein calculating text similarity, user similarity and mood fluctuation similarity for any two topic events in a total set of topic events of a plurality of social platforms from three angles of text semantics, user overlap and mood fluctuation, respectively, comprises: Selecting any two topic events And Collecting all texts related to topic events in each social platform, extracting keywords, calculating text similarity between two topic events by using cosine similarity, and the calculation formula is as follows: Selecting any two topic events And Collecting all users published texts related to topic events in each social platform, and calculating the user similarity between two topic events by using Jaccard similarity coefficients, wherein the calculation formula is as follows: Selecting any two topic events And Collecting all texts related to topic events in each social platform, counting daily emotions to form a time sequence related to the emotions, and calculating the similarity of emotion fluctuation between the two events by using a dynamic time warping algorithm, wherein the calculation formula is expressed as follows: Wherein, the And Respectively, topic events And topic event Is a vector representation of the text of (a), In the form of a dot product, And Is the modulus of vector, the model of vector is shown in the specification , ∈ ,i≠j), Representing a total set of a plurality of social platform topic events, And Respectively represent and event And events A set of related users, 1 being positive emotion, 0 being neutral emotion, -1 being negative emotion, And Respectively represent events And events Is a time series of mood swings in the air, Representing topic events in an mth social platform And topic event The degree of similarity of the text between the two, Representing topic events in an mth social platform And topic event The degree of similarity of the users between them, Representing topic events in an mth social platform And topic event The emotion fluctuation similarity between the topic events is i, i is the sequence number of the topic event, and j is the sequence number of the topic event.
  3. 3. The multi-layer network-based public opinion event risk assessment method of claim 1, wherein the formulas for unified dimension and direction processing of text similarity, user similarity and mood wave similarity are expressed as follows: Wherein, the Text similarity between any two topic events in the mth social platform, Representing the user similarity between any two topic events in the mth social platform, Representing the similarity of mood swings between any two topic events in the mth social platform, Representing the text similarity between any two topic events in the mth social platform after unified dimension and direction processing, Representing the user similarity between any two topic events in the mth social platform after unified dimension and direction processing, Representing the emotion fluctuation similarity between any two topic events in the mth social platform after unified dimension and direction processing, 0 Represents the attenuation coefficient, taken here =1。
  4. 4. The public opinion event risk assessment method based on the multi-layer network of claim 1, wherein constructing the multi-layer network according to the final association degree of any two topic events in the topic event and the total collection of the topic events of the plurality of social platforms comprises constructing an undirected weighted graph by taking the topic event as a node and the final association degree between the topic events as a weight of an edge, and constructing the multi-layer network through the undirected weighted graph for the plurality of social platforms, wherein the method is represented as follows: G={ } Wherein, the , ( , ) Representing two topic events of an mth social platform And Is used to determine the degree of correlation of the final degree of correlation of (c), ( , ), And (3) representing an undirected weighted graph, wherein G represents a multi-layer network, k represents the number of nodes in the network, and n represents the number of layers of the multi-layer network.
  5. 5. The multi-layer network-based public opinion event risk assessment method of claim 1, wherein determining the set for the multi-layer network comprises: calculating the centrality of each node at each layer of the multi-layer network; Sorting the centrality of each node from large to small, and screening out the node with the largest centrality of each layer; the nodes with the greatest centrality of each layer form a set.
  6. 6. The public opinion event risk assessment method based on the multi-layer network according to claim 1, wherein the method is characterized in that each node in the infection source set is sequentially used as an infection source, and SIR algorithm simulation is respectively performed to obtain an infection state set, an infection duration set and a neighbor infection proportion set of each node, and the method comprises the following steps: For each node, calculating the total infectivity and the state transition probability; Setting the maximum iteration times, and stopping when no new infected node exists or the maximum iteration times are reached; counting the infection state, infection duration and neighbor infection proportion of each node in the network in the current infection process; And repeating the above process by taking each node in the infection source set as an infection source to obtain the infection state, the infection duration and the neighbor infection proportion of each node in the network under different infection sources, so as to form an infection state set, an infection duration set and a neighbor infection proportion set of each node.
  7. 7. The multi-layer network-based public opinion event risk assessment method of claim 1, wherein determining a risk score for each node according to the infection status set, the infection duration set, and the neighbor infection proportion set of each node, determining a risk topic event according to the magnitude of the risk score for each node, comprises: carrying out standardized treatment on an infection state set, an infection duration set and a neighbor infection proportion set of each node to unify dimensions; Weighting by using an entropy weighting method to obtain a risk score of each node; and sequencing the risk scores of each node from large to small, and inputting the first L nodes to obtain the risk topic event.
  8. 8. A public opinion event risk assessment device based on a multi-layer network, comprising: The acquisition module is used for acquiring topic event sets of each social platform in the plurality of social platforms in a first time period, and acquiring intersections of the topic event sets to obtain topic event total sets of the plurality of social platforms; The first calculation module is used for calculating text similarity, user similarity and emotion fluctuation similarity of any two topic events in the total set of topic events of the plurality of social platforms from three angles of text semantics, user overlap and emotion fluctuation for any one social platform; the processing module is used for carrying out unified dimension and direction processing on the text similarity, the user similarity and the emotion fluctuation similarity; The second calculation module is used for calculating the final association degree of any two topic events in the total collection of the topic events of the social platform after giving weight to the text similarity, the user similarity and the emotion fluctuation similarity which are subjected to unified dimension and direction processing; The construction module is used for constructing a multi-layer network according to the final association degree of any two topic events in the topic event and the topic event total set of the plurality of social platforms for each social platform in the plurality of social platforms; The first determining module is used for determining a set for the multi-layer network and de-duplicating the set to obtain an infection source set; The acquisition module is used for sequentially taking each node in the infection source set as an infection source, and respectively performing SIR algorithm simulation to acquire an infection state set, an infection duration set and a neighbor infection proportion set of each node; and the second determining module is used for determining the risk score of each node according to the infection state set, the infection duration set and the neighbor infection proportion set of each node, and determining the risk topic event according to the size of the risk score of each node.
  9. 9. An electronic device is characterized by comprising a processor and a memory; The processor is used for executing the public opinion event risk assessment method based on the multi-layer network according to any one of claims 1 to 7 by calling the program or the instructions stored in the memory.
  10. 10. A computer-readable storage medium storing a program or instructions that cause a computer to execute a multi-layered network-based public opinion event risk assessment method according to any one of claims 1 to 7.

Description

Public opinion event risk assessment method and device based on multi-layer network Technical Field The invention belongs to the technical field of event risk assessment, and particularly relates to a public opinion event risk assessment method and device based on a multi-layer network. Background With the development of diversification of social platforms, the supervision of a single platform cannot fully reflect the full view of public opinion information, so that it is necessary to integrate information from multiple platforms to achieve more comprehensive and accurate evaluation of public opinion events. Meanwhile, public opinion events often exist in a non-isolated form, and it is necessary to analyze the association between events so as to timely identify potential high-risk public opinion and reduce the possible harm caused by the public opinion events. Disclosure of Invention In view of the shortcomings of the prior art, the invention aims to provide a public opinion event risk assessment method and device based on a multi-layer network, which take topic events as main bodies, take the association degree among the topic events as edges, construct a multi-layer complex network from the angles of a plurality of social platforms, identify network key nodes, assess the potential risk of each node through an SIR algorithm of the multi-layer network, and mine high risk nodes to realize risk assessment. The first aspect of the present invention provides a public opinion event risk assessment method based on a multi-layer network, including: obtaining topic event sets of each social platform in a first time period, and acquiring intersections of the topic event sets to obtain topic event total sets of the social platforms; For any one social platform, calculating text similarity, user similarity and emotion fluctuation similarity for any two topic events in the total set of topic events of a plurality of social platforms from three angles of text semantics, user overlap and emotion fluctuation respectively; performing unified dimension and direction processing on the text similarity, the user similarity and the emotion fluctuation similarity; After the text similarity, the user similarity and the emotion fluctuation similarity subjected to unified dimension and direction processing are weighted, calculating the final association degree of any two topic events in the total set of topic events of the social platform; For each social platform in the plurality of social platforms, constructing a multi-layer network according to the topic event and the final association degree of any two topic events in the total collection of the topic events of the plurality of social platforms; Determining a set for the multi-layer network, and de-duplicating the set to obtain an infection source set; sequentially taking each node in the infection source set as an infection source, and respectively performing SIR algorithm simulation to obtain an infection state set, an infection duration set and a neighbor infection proportion set of each node; And determining a risk score of each node according to the infection state set, the infection duration set and the neighbor infection proportion set of each node, and determining a risk topic event according to the size of the risk score of each node. Further, in the above-mentioned public opinion event risk assessment method based on a multi-layer network, calculating text similarity, user similarity and emotion fluctuation similarity for any two topic events in a total set of topic events of a plurality of social platforms from three angles of text semantics, user overlap and emotion fluctuation respectively, including: Selecting any two topic events AndCollecting all texts related to topic events in each social platform, extracting keywords, calculating text similarity between two topic events by using cosine similarity, and the calculation formula is as follows: Selecting any two topic events AndCollecting all users published texts related to topic events in each social platform, and calculating the user similarity between two topic events by using Jaccard similarity coefficients, wherein the calculation formula is as follows: Selecting any two topic events AndCollecting all texts related to topic events in each social platform, counting daily emotions to form a time sequence related to the emotions, and calculating the similarity of emotion fluctuation between the two events by using a dynamic time warping algorithm, wherein the calculation formula is expressed as follows: Wherein, the AndRespectively, topic eventsAnd topic eventIs a vector representation of the text of (a),In the form of a dot product,AndIs the modulus of vector, the model of vector is shown in the specification,∈,i≠j),Representing a total set of a plurality of social platform topic events,AndRespectively represent and eventAnd eventsA set of related users, 1 being positive emotion, 0 being neutral emotion, -1