CN116795985A - Network public opinion anomaly identification and processing method - Google Patents
Network public opinion anomaly identification and processing method Download PDFInfo
- Publication number
- CN116795985A CN116795985A CN202310746360.5A CN202310746360A CN116795985A CN 116795985 A CN116795985 A CN 116795985A CN 202310746360 A CN202310746360 A CN 202310746360A CN 116795985 A CN116795985 A CN 116795985A
- Authority
- CN
- China
- Prior art keywords
- public opinion
- network public
- event
- data
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003672 processing method Methods 0.000 title claims abstract description 11
- 238000000034 method Methods 0.000 claims abstract description 60
- 238000004458 analytical method Methods 0.000 claims abstract description 43
- 238000005516 engineering process Methods 0.000 claims abstract description 37
- 230000007246 mechanism Effects 0.000 claims abstract description 31
- 230000008569 process Effects 0.000 claims abstract description 22
- 238000012216 screening Methods 0.000 claims abstract description 21
- 230000010485 coping Effects 0.000 claims abstract description 18
- 238000010586 diagram Methods 0.000 claims abstract description 15
- 238000000605 extraction Methods 0.000 claims abstract description 15
- 230000019771 cognition Effects 0.000 claims abstract description 6
- 230000001364 causal effect Effects 0.000 claims description 38
- 238000010276 construction Methods 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 18
- 230000002159 abnormal effect Effects 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 17
- 230000005856 abnormality Effects 0.000 claims description 15
- 238000011160 research Methods 0.000 claims description 10
- 238000007726 management method Methods 0.000 claims description 9
- 239000003795 chemical substances by application Substances 0.000 claims description 8
- 238000011156 evaluation Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 6
- 230000010354 integration Effects 0.000 claims description 6
- 238000003062 neural network model Methods 0.000 claims description 6
- 238000009960 carding Methods 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 5
- 238000013461 design Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 5
- 230000004044 response Effects 0.000 claims description 5
- 230000003068 static effect Effects 0.000 claims description 5
- 238000012795 verification Methods 0.000 claims description 5
- 238000000265 homogenisation Methods 0.000 claims description 4
- 230000007547 defect Effects 0.000 claims description 3
- 230000008451 emotion Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 2
- 238000007499 fusion processing Methods 0.000 claims description 2
- 230000002452 interceptive effect Effects 0.000 claims description 2
- 230000008447 perception Effects 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000005070 sampling Methods 0.000 claims description 2
- 230000011218 segmentation Effects 0.000 claims description 2
- 238000012549 training Methods 0.000 claims description 2
- 238000013519 translation Methods 0.000 claims description 2
- 239000012634 fragment Substances 0.000 claims 2
- 230000003190 augmentative effect Effects 0.000 claims 1
- 230000001149 cognitive effect Effects 0.000 claims 1
- 125000004122 cyclic group Chemical group 0.000 claims 1
- 230000000694 effects Effects 0.000 claims 1
- 238000007500 overflow downdraw method Methods 0.000 claims 1
- 238000013473 artificial intelligence Methods 0.000 abstract description 3
- 238000011161 development Methods 0.000 description 4
- 230000018109 developmental process Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 238000009411 base construction Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000010845 search algorithm Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to a network public opinion anomaly identification and processing method, and belongs to the field of artificial intelligence. The invention uses the innovation points of knowledge intelligent extraction technology of network public opinion information, network public opinion anomaly identification technology based on social propagation structure, expert cognition process and strategy modeling technology of network public opinion event analysis, key element screening technology based on time relation diagram attention mechanism and the like, thereby improving the accuracy rate of network public opinion anomaly identification and the anomaly identification timeliness of network public opinion. The solution can provide good help for early warning of large-scale public opinion outbreaks, thereby providing reference for high-efficiency crisis coping of related departments.
Description
Technical Field
The invention belongs to the field of artificial intelligence, and particularly relates to a network public opinion anomaly identification and processing method.
Background
The method is a worldwide difficult problem for processing massive, multi-mode and multi-source heterogeneous data, and is particularly difficult for realizing the conversion of data to knowledge in the field of public opinion anomaly identification research with strong professionality and high quality requirements.
At present, from knowledge space construction of network public opinion data to knowledge mining and network public opinion prediction recognition, various countries take a plurality of technical routes to develop partial technical researches in respective fields, but a complete network public opinion anomaly recognition system is not formed yet, and a series of technical challenges still face the current state of the art and development trends at home and abroad, and are mainly reflected in the two aspects of insufficient network public opinion knowledge acquisition and fusion capability and insufficient network public opinion quantitative prediction recognition means. Aiming at the problem of the current network public opinion anomaly identification in China, the invention combines with the current new theoretical technical method to form a complete network public opinion anomaly identification solution.
Disclosure of Invention
First, the technical problem to be solved
The invention aims to solve the technical problems of how to provide a network public opinion anomaly identification and processing method so as to solve the problems of insufficient acquisition and fusion capability of network public opinion knowledge and insufficient quantitative prediction and identification means of network public opinion in China at present.
(II) technical scheme
In order to solve the technical problems, the invention provides a method for identifying and processing network public opinion anomalies, which comprises the following steps:
s1, constructing a network public opinion data resource base: the method for obtaining the network public opinion raw data and constructing the data resource base comprises the following steps: tracking and carding the data sources, intelligently collecting the data and constructing a data source system; constructing a data resource library comprises the following steps: processing the acquired multi-element heterogeneous original data and cataloging the processed data;
s2, constructing an online public opinion event map and analyzing a public opinion propagation path: the public opinion event map construction module processes structured and unstructured data in a public opinion data resource base to form a knowledge map containing entities, relations and causal events, and constructs an event map based on the knowledge map; the network public opinion propagation path analysis module carries out vectorization representation on the event map and identifies the public opinion propagation path through a graph searching algorithm and similarity comparison;
s3, identifying network public opinion abnormality based on social propagation structure: comprises three steps:
the first step is the representation of event word vectors, which are based on TF-IDF or word2vec, and calculating event distances;
the second step is to construct an abstract graph of the network public opinion based on social propagation, and construct the abstract graph of the network public opinion through similar event analysis based on homogenization theory and event heat analysis based on message propagation;
the third step is to identify network public opinion anomalies based on abstract event map, firstly classify the network public opinion anomalies, then predict the heat degree based on LDA model, classify the network public opinion anomalies according to the heat degree, and finally identify the anomaly type based on the network public opinion anomaly identification technology of extensible graph neural network model;
s4, a network public opinion abnormal event coping strategy based on an event propagation evolution mechanism: based on expert thinking in the field of public opinion research and a causal reasoning mechanism of an event map, verification and duplication of network public opinion anomaly identification are realized; screening key elements playing a role in determining network public opinion abnormality identification based on a screening model technology of a time sequence diagram attention mechanism; and formulating a targeted response strategy according to the network public opinion anomaly identification result.
(III) beneficial effects
The invention provides a network public opinion anomaly identification and processing method, and provides a solution for the public opinion anomaly identification and coping requirements of China, which uses innovation points such as knowledge intelligent extraction technology of network public opinion information, network public opinion anomaly identification technology based on social propagation structure, expert cognition process and strategy modeling technology of network public opinion event analysis, key element screening technology based on time relation diagram attention mechanism and the like, thereby improving the accuracy of network public opinion anomaly identification and the anomaly identification timeliness of network public opinion. The solution can provide good help for early warning of large-scale public opinion outbreaks, thereby providing reference for high-efficiency crisis coping of related departments.
Drawings
FIG. 1 is a diagram illustrating an embodiment of a method for identifying and disposing network public opinion anomalies according to the present invention.
Detailed Description
To make the objects, contents and advantages of the present invention more apparent, the following detailed description of the present invention will be given with reference to the accompanying drawings and examples.
The invention combines an artificial intelligence method and a social propagation theory to construct an accurate and efficient public opinion anomaly identification method, and forms a complete network public opinion anomaly identification solution based on the theoretical method.
The invention provides a complete network public opinion anomaly identification and treatment method which is used for solving the problems of insufficient acquisition and fusion capabilities of network public opinion knowledge and insufficient quantitative prediction and identification means of network public opinion in China at present.
In order to realize the recognition and treatment of the network public opinion abnormality, the invention is developed in the following aspects, and the specific method and steps are as follows:
s1, constructing a network public opinion data resource base: the method for obtaining the network public opinion raw data and constructing the data resource base comprises the following steps: tracking and carding the data sources, intelligently collecting the data and constructing a data source system; constructing a data resource library comprises the following steps: processing the acquired multi-element heterogeneous original data and cataloging the processed data;
s2, constructing an online public opinion event map and analyzing a public opinion propagation path: the public opinion event map construction module processes structured and unstructured data in a public opinion data resource base to form a knowledge map containing entities, relations and causal events, and constructs an event map based on the knowledge map; the network public opinion propagation path analysis module carries out vectorization representation on the event map and identifies the public opinion propagation path through a graph searching algorithm and similarity comparison;
s3, identifying network public opinion abnormality based on social propagation structure: comprises three steps:
the first step is the representation of event word vectors, which are based on TF-IDF or word2vec, and calculating event distances;
the second step is to construct an abstract graph of the network public opinion based on social propagation, and construct the abstract graph of the network public opinion through similar event analysis based on homogenization theory and event heat analysis based on message propagation;
the third step is to identify network public opinion anomalies based on abstract event map, firstly classify the network public opinion anomalies, then predict the heat degree based on LDA model, classify the network public opinion anomalies according to the heat degree, and finally identify the anomaly type based on the network public opinion anomaly identification technology of extensible graph neural network model;
s4, a network public opinion abnormal event coping strategy based on an event propagation evolution mechanism: based on expert thinking in the field of public opinion research and a causal reasoning mechanism of an event map, verification and duplication of network public opinion anomaly identification are realized; screening key elements playing a role in determining network public opinion abnormality identification based on a screening model technology of a time sequence diagram attention mechanism; and formulating a targeted response strategy according to the network public opinion anomaly identification result.
The invention provides a method for identifying and processing network public opinion anomalies, which consists of a network public opinion data resource library construction module, a network public opinion theory map construction and public opinion propagation path analysis module, a network public opinion anomaly identification module based on a social propagation structure, a network public opinion anomaly event coping strategy generation module based on an event propagation evolution mechanism and the like, and comprises the following specific implementation steps:
the method comprises the steps of generating a network public opinion abnormal event coping strategy based on an event propagation evolution mechanism, firstly integrating the expert thinking, the network public opinion abnormal display form and a causal reasoning mechanism of an event map, verifying and replying network public opinion abnormal recognition, forming network public opinion abnormal coping knowledge, then screening key elements playing a role in determining the network public opinion abnormal recognition based on technologies such as a screening model of a time sequence diagram attention mechanism, and finally combining the verification of the network public opinion abnormal recognition and the key elements playing a role in determining the network public opinion abnormal with the network public opinion experience based on the network public opinion abnormal recognition result to realize the generation of the abnormal event coping strategy.
The details of the method for identifying and processing the network public opinion anomalies are described below.
(1) Network public opinion data resource base construction module
The module mainly comprises the following two aspects:
1) The method mainly comprises the steps of tracking and carding a data source, carrying out intelligent data acquisition based on a theme sniffing technology and constructing a data source system.
2) The method mainly comprises the steps of constructing a data resource library, and mainly comprises the steps of processing acquired multi-heterogeneous original data and cataloging processed data. The module finally outputs the organized network public opinion data resource base for the subsequent modules to use. The details of the above functional points to which this module relates are described below.
When the network public opinion raw data is obtained, the related data sources of the network public opinion event mainly comprise world-known international affair research institutions, related journal magazines, foreign government websites and other institutions. The preliminary carding of the network public opinion event data source is convenient for promoting the rapid development of intelligent data acquisition, and then the theme sniffing and automatic data tracking acquisition work is carried out, and the evaluation and feedback of the data source can be formed while the theme sniffing and the automatic data tracking acquisition work are carried out, so that the construction and the dynamic perfection of the network public opinion event data source system are promoted. The topic sniffing technology analyzes and mines given data sources through technical means to analyze potential data sources, and the technology verifies existing data sources to ensure information output stability and information quality of the existing data sources, and discovers potential new information sources from the information sources. According to the characteristics of internet data sources and sniffing forms, information acquisition intelligent agents are divided into three types, namely sniffing intelligent agents, perception intelligent agents and autonomous intelligent agents, and tracking and sniffing of information sites, search engines and interactive media information are respectively realized.
When the data source system is built, the main system planning, selecting, acquiring and organizing the data source, and the whole process of the resource system with specific functions is built.
The data resource library is constructed, the multisource organization data processing is performed, network public opinion data of different sources and different types are mainly fused, a unified data resource library is formed, and data support is provided for subsequent network public opinion map construction. In order to improve the utilization rate of data resources and the retrieval and utilization efficiency of data, the invention adopts a multi-source data fusion technology based on keywords and a network public opinion fusion processing technology based on data content.
The network public opinion event data cataloging mainly comprises: metadata design, format conversion and metadata integration, wherein the metadata design is manually completed by a field expert, the construction work of a metadata database is completed by engineering personnel, and the format and metadata integration work converts the attribute and relation information of the body content into a table structure in the database by the engineering personnel to complete the format conversion and data integration of the metadata. The constructed network public opinion data resource library is used by the subsequent modules.
(2) Network public opinion event construction and public opinion propagation path analysis module
The network public opinion event map construction and public opinion propagation path module mainly comprises two modules, namely a public opinion event map construction module and a network public opinion propagation path analysis module.
The public opinion event map construction module is mainly responsible for processing structured and unstructured data in a public opinion data resource base to form a knowledge map comprising entities, relations and causal events, and further constructing an event map based on the knowledge map;
and the network public opinion propagation path analysis module performs vectorization representation on the fact map produced in the last step, and identifies the public opinion propagation path through a graph search algorithm and similarity comparison. Finally, the module obtains a public opinion event map and a corresponding public opinion propagation path analysis result, and the result provides support for subsequent anomaly identification and strategy formulation.
In particular, the method comprises the steps of,
in the public opinion event map construction module, firstly, public opinion text data preprocessing is carried out through the methods of word segmentation, part-of-speech tagging, grammar analysis, relation rule analysis and the like, then knowledge maps are constructed through the methods of entity identification, relation extraction, event extraction, knowledge fusion and the like, and finally, event map construction is carried out through the methods of event framework construction, event type identification, event element identification, event causal relation extraction and the like, so that the event map construction is completed.
The network public opinion propagation path analysis module mainly completes the propagation path analysis of network public opinion events, and causal event extraction, causal relation extraction rule setting, causal relation structure classification and causal relation extraction are required. Firstly, for the new network public opinion event, calculating the similarity between the new network public opinion event and each node in the event map, and finding out the most similar node of the new event; and secondly, comparing the similarity with a threshold value to judge whether the propagation is continued. Specifically, given a newly-occurring public opinion event, traversing nodes in the network public opinion event map, calculating similarity with the new event, and finally comparing the maximum similarity with a given threshold, if the maximum similarity is smaller than the preset threshold, indicating that no target event corresponding node in the map cannot be propagated, otherwise, predicting possible events in reality according to subsequent nodes in the map. If a node has multiple subsequent events, the probability of possible occurrence is calculated according to the edge weight coefficient, and then the process is performed recursively until no subsequent events exist, so that one or more possible complete propagation paths can be obtained.
(3) Network public opinion anomaly identification module based on social propagation structure
The network public opinion anomaly identification module based on the social propagation structure is used for analyzing the propagation of the network public opinion and identifying the anomalies of the network public opinion, and mainly comprises the following three steps:
the first step is the representation of an event word vector, which may be based in part on the TF-IDF or word2vec event word vector representation, and calculating the event distance.
The second step is to construct an abstract graph of the network public opinion based on social propagation, and the abstract graph of the network public opinion can be constructed through similar event analysis based on homogenization theory and event heat analysis based on message propagation.
The third step is to identify network public opinion anomalies based on abstract rational graphs, firstly classify the network public opinion anomalies, then predict the heat based on LDA model, classify the network public opinion anomalies according to the heat, and finally identify the anomaly type based on the network public opinion anomaly identification technology of extensible graph neural network model.
The network public opinion anomaly identification technology is mainly applied to entity granularity representation learning technology, time sequence relation quintuple granularity representation learning technology and event granularity representation learning technology. These three techniques are described separately below.
Entity granularity representation learning technology proposes an entity representation learning model fusing entity types, which attempts to fuse information to enrich semantics in knowledge representation. The model represents entities, relationships, and entity types as low-dimensional dense vectors, respectively, and a mapping function is designed based on entity types for constructing a representation of an entity under a particular type. In a triplet containing a specific relationship, the representation of the head-to-tail entity is first mapped to a representation under the type of the relationship constraint, and then the semantic meaning expressed by the triplet is learned using translation ideas. The mapping function adopts a circumferential convolution form, so that the type vector and the entity vector are fully communicated, the characteristic association between the corresponding dimension and the non-corresponding dimension is obtained, and the quality of knowledge embedding is improved. In the negative sampling, the model limits the type of the sampled entity to be consistent with the type of the replaced entity with a certain probability, so that the difference among the entities belonging to the same type is increased, and certain similarity among the entities is considered. The method enriches semantic information contained in the knowledge representation and solves the problem of complex relationships to a certain extent.
The granularity of the time sequence relation quintuple represents a learning technology: a knowledge representation algorithm of space-time relation quintuple in knowledge graph context information is provided for solving the defect of the traditional knowledge graph space-time aspect representation model, and the algorithm takes the time compliance characteristic, time interval characteristic and space association constraint of the evolution of the relation among entities into consideration while modeling the static structural characteristics of the entities in the knowledge graph, and obtains the vector representation of the entity and the relation of the space-time association based on learning. Specifically, the method utilizes static structure context information and dynamic structure context information of five-tuple in a knowledge graph to perform representation learning, wherein the dynamic structure context information injects information of spatial-temporal evolution and spatial evolution of the relationship.
Discrete event representations, event representations based on distributed semantics and event representations based on inter-event information are proposed in event granularity representation learning technology, and the defects of single representation in the past are overcome.
The second step of the module is to construct an abstract event map based on social propagation, wherein the abstract event map is used for generalizing events, so as to find more general and universal event evolution causal relations in a certain network public opinion field. In the invention, the events with similar distances are generalized through clustering, and a plurality of specific events are represented by a more representative event, so that the network public opinion abstract situation map is constructed. The clustering sample is an event, namely a text sample, the original kmeans clustering method uses TF-IDF vectors in the process of converting the text into numerical vectors, the formed vectors lack semantic elements, and the clustering algorithm is improved by using word2vec word vectors and a distance calculation method, so that the generalization result is optimal.
The third step is a network public opinion anomaly recognition model based on an abstract rational graph, and the model further provides an extensible graph neural network model (Scaled Graph Neural Network, SGNN) which can perform network representation learning on a large-scale propagation rational graph structure. By referencing the idea of divide-and-conquer during training, the SGNN is calculated on only one small scale correlation subgraph at a time to extend the GGNN to a large scale directed loop weighted graph rather than to calculate on the entire graph. The model can be modified to be applied to other network representation learning tasks. By computing the similarity of the representation of the event and the candidate event representations in the context, the model can identify the correct post-network public opinion anomaly type.
(4) Internet public opinion abnormal event coping strategy based on event propagation evolution mechanism
The module mainly comprises: : (1) Based on expert thinking in the field of public opinion research and different forms of online public opinion anomaly display and causal reasoning mechanisms of a priori map, the online public opinion anomaly identification is verified and repeated; (2) Screening key elements playing a role in determining the network public opinion abnormality identification by adopting technologies such as screening models based on a time sequence diagram attention mechanism; (3) And formulating a targeted response strategy according to the network public opinion anomaly identification result. The relevant matters of these three aspects are described in detail below.
The network public opinion anomaly identification and duplication technology is mainly completed through three parts of expert thinking acquisition and construction, network public opinion anomaly polymorphic display and interpretive map enhancement, 1) the expert thinking acquisition and construction are mainly based on relevant theories of problem solving and cognition psychology, network public opinion anomalies in real situations are taken as research objects, relevant information is summarized from public literature, policies and intelligent libraries to determine analysis angles so as to form a questionnaire, then semi-structured interviews are carried out on the expert, process information of the expert analysis network public opinion anomalies is obtained, an integrated event analysis process is formed, the cognition process and processing characteristics of the expert analysis network public opinion anomalies are analyzed, and theoretical basis is provided for constructing an expert thinking model of network public opinion anomaly analysis. 2) In the analysis of the key element information of the network public opinion abnormal event, the polymorphic display can effectively clearly display the reasoning process of the complex public opinion abnormal event. The present invention notes that during causal reasoning there may be a series of evidence events between the causal event and the outcome event. These evidence events can serve as additional cues to help the model understand the causal mechanisms between the cause and the result and enhance the role of the interpretability and robustness of the result. 3) In order to fully utilize the role of evidence events in causal inference in the online public opinion anomaly identification part, a causal inference framework (Event Graph Knowledge Enhanced Explainable Causal Reasoning, exCAR) with enhanced knowledge of a rational atlas is used. Given a causal pair, this framework can obtain a series of causal evidence from a pre-constructed causal knowledge base.
The screening part of the network public opinion anomaly identification key elements mainly comprises a screening model based on a time sequence diagram attention mechanism and the network public opinion anomaly identification key elements. 1) In a screening model based on a time sequence diagram attention mechanism, in the process of knowledge graph construction, the original source of each knowledge segment is recorded by utilizing a URI address, so that in deduction analysis, the original data source can be traced back according to the corresponding URI of a given knowledge segment, and a fine-grained data basis is provided for decision. In the knowledge graph constructed at this time, the structure of the graph is not static any more, and the invention is also called an event knowledge graph because the entities and the links between them evolve with time and establish a connection in the form of an event. The invention abstracts the event knowledge graph to G inf . At G inf The invention has obtained edges between the predicted entities (i.e., knowledge segments) and the entire inference graph. To trace back the origin of the knowledge piece original data, the event with the largest attention score on the relevant path needs to be calculated. And meanwhile, obtaining key elements of the knowledge segments according to the arrangement of the attention scores. 2) The key elements of the network public opinion anomaly identification mainly relate to the influence of the attention layer judgment key elements after screening by using an attention mechanism and an event relation diagram.
The network public opinion anomaly coping strategy part mainly comprises network public opinion anomaly identification, network public opinion multi-element guidance and network public opinion evaluation and coping. 1) In the network public opinion anomaly identification, the network public opinion has the characteristics of freedom, interactivity, multiple, deviation, burst property and the like, can promote and change the development and trend of events by the emotion and judgment of the left and right people, is easy to be used against molecules, and becomes an important factor affecting the social stability. The invention is based on a method for constructing a rational map and extracting information, and combines the subject background and the requirement of network public opinion analysis to identify 5 types of anomalies concerned in the network public opinion analysis. The invention divides the types of the network public opinion anomalies into 5 categories, namely the network public opinion anomalies based on consciousness form, the network public opinion anomalies based on benefit appeal, the network public opinion anomalies based on emotion releasing, the network public opinion anomalies based on idea sound and the network public opinion anomalies based on social security. 2) The social participation mechanism diversification is emphasized, all levels of main bodies participate in the network public opinion emergency management together, different management modes are set for different personnel, and the front information can be transmitted through the network main stream media and other ways to guide the people to develop forward towards the network public opinion, and the diversification management education mode is utilized to play a role in the national participation. The guidance may be deployed from the following: the public opinion information disclosure of the emergency network is enhanced, and transparency given by the idea of good management is maintained. The method realizes the spatial method treatment of the network public opinion and adheres to the method treatment given by the good treatment concept. The scientific management consciousness is established, and the effectiveness given by the good management concept is maintained. 3) In the aspect of network public opinion evaluation and coping, by researching network public opinion propagation trend prediction under emergencies, an emergency mechanism capable of controlling the network public opinion to move to a correct direction is formulated aiming at a prediction result so as to aim at dynamic accurate evaluation of the network public opinion, so that accurate solution of social crisis caused by the network public opinion is realized, unnecessary influence of the network public opinion on society is avoided, and the network public opinion is expected to move towards forward development coordination propagation.
The invention provides a solution for identifying public opinion anomalies and coping with demands in China, which uses innovation points such as knowledge intelligent extraction technology of network public opinion information, network public opinion anomaly identification technology based on social propagation structure, expert cognition process and strategy modeling technology of network public opinion event analysis, key element screening technology based on time relation diagram attention mechanism and the like, thereby improving accuracy rate of network public opinion anomaly identification and anomaly identification timeliness of network public opinion. The solution can provide good help for early warning of large-scale public opinion outbreaks, thereby providing reference for high-efficiency crisis coping of related departments.
The foregoing is merely a preferred embodiment of the present invention, and it should be noted that modifications and variations could be made by those skilled in the art without departing from the technical principles of the present invention, and such modifications and variations should also be regarded as being within the scope of the invention.
Claims (10)
1. A network public opinion anomaly identification and processing method is characterized by comprising the following steps:
s1, constructing a network public opinion data resource base: the method for obtaining the network public opinion raw data and constructing the data resource base comprises the following steps: tracking and carding the data sources, intelligently collecting the data and constructing a data source system; constructing a data resource library comprises the following steps: processing the acquired multi-element heterogeneous original data and cataloging the processed data;
s2, constructing an online public opinion event map and analyzing a public opinion propagation path: the public opinion event map construction module processes structured and unstructured data in a public opinion data resource base to form a knowledge map containing entities, relations and causal events, and constructs an event map based on the knowledge map; the network public opinion propagation path analysis module carries out vectorization representation on the event map and identifies the public opinion propagation path through a graph searching algorithm and similarity comparison;
s3, identifying network public opinion abnormality based on social propagation structure: comprises three steps:
the first step is the representation of event word vectors, which are based on TF-IDF or word2vec, and calculating event distances;
the second step is to construct an abstract graph of the network public opinion based on social propagation, and construct the abstract graph of the network public opinion through similar event analysis based on homogenization theory and event heat analysis based on message propagation;
the third step is to identify network public opinion anomalies based on abstract event map, firstly classify the network public opinion anomalies, then predict the heat degree based on LDA model, classify the network public opinion anomalies according to the heat degree, and finally identify the anomaly type based on the network public opinion anomaly identification technology of extensible graph neural network model;
s4, a network public opinion abnormal event coping strategy based on an event propagation evolution mechanism: based on expert thinking in the field of public opinion research and a causal reasoning mechanism of an event map, verification and duplication of network public opinion anomaly identification are realized; screening key elements playing a role in determining network public opinion abnormality identification based on a screening model technology of a time sequence diagram attention mechanism; and formulating a targeted response strategy according to the network public opinion anomaly identification result.
2. The method for identifying and processing the network public opinion anomaly according to claim 1, wherein when the network public opinion raw data is acquired, the data sources of the network public opinion events comprise world-known international affair research institutions, related journals and foreign government websites, the network public opinion event data sources are primarily combed, then subject sniffing and automatic data tracking acquisition work is performed, and evaluation and feedback of the data sources are formed while the subject sniffing and automatic data tracking acquisition work is performed, so that construction and dynamic perfection of a network public opinion event data source system are promoted; the topic sniffing technology analyzes and mines given data sources, analyzes potential data sources, classifies information acquisition agents into three types according to characteristics and sniffing forms of internet data sources, namely sniffing agents, perception agents and autonomous agents, and respectively realizes tracking and sniffing of information sites, search engines and interactive media information; when the data source system is built, the data sources are systematically planned, selected, acquired and organized to manage, and the whole process of the resource system with specific functions is built.
3. The network public opinion anomaly identification and processing method of claim 2, wherein when constructing a data resource base, the multi-source institution data processing fuses different sources and different types of network public opinion data to form a unified data resource base, and provides data support for subsequent network public opinion map construction; comprising the following steps: a multi-source data fusion technology based on keywords and a network public opinion fusion processing technology based on data content; the network public opinion event data cataloging includes: metadata design, format conversion and metadata integration, wherein the metadata design is manually completed by a field expert, the construction work of a metadata database is completed by engineering personnel, the format and metadata integration work is to convert the attribute and relation information of the body content into a table structure in the database by the engineering personnel, the format conversion and data integration of the metadata are completed, and the constructed network public opinion data resource library is used for subsequent modules.
4. The network public opinion anomaly identification and processing method of claim 3, wherein in the public opinion event map construction module, public opinion text data preprocessing is firstly performed through word segmentation, part-of-speech tagging, grammar analysis and relationship rule analysis methods, then knowledge maps are constructed through entity identification, relationship extraction, event extraction and knowledge fusion methods, and finally event map construction is performed through event framework construction, event type identification, event element identification and event causal relationship extraction methods, so far, the event map construction is completed.
5. The network public opinion anomaly identification and processing method of claim 4, wherein the network public opinion propagation path analysis module completes the propagation path analysis of network public opinion events, and causal event extraction, causal relation extraction rule setting, causal relation structure classification and causal relation extraction are required; firstly, calculating the similarity between the new network public opinion event and each node in the event map, and finding out the most similar node of the new event; secondly, comparing the similarity with a threshold value to judge whether to continue to propagate; specifically, given a newly-occurring public opinion event, traversing nodes in a network public opinion event map, calculating similarity with the new event, finally comparing the maximum similarity with a given threshold, if the maximum similarity is smaller than the preset threshold, indicating that no target event corresponds to the node in the map, and failing to propagate, otherwise, presuming that an event possibly occurs in reality according to a subsequent node in the map, if a certain node has a plurality of subsequent events, calculating the probability of possibly occurring according to an edge weight coefficient, and then recursively executing the process until no subsequent event exists, thus obtaining one or more possible complete propagation paths.
6. The network public opinion anomaly identification and processing method of claim 5, wherein the social propagation structure-based network public opinion anomaly identification uses entity granularity representation learning techniques, time sequence relation quintuple granularity representation learning techniques and event granularity representation learning techniques;
an entity representation learning model integrating entity types is provided in an entity granularity representation learning technology, and the model tries to integrate information so as to enrich semantics in knowledge representation; the model respectively represents the entity, the relation and the entity type as vectors with low dimension and density, and designs a mapping function based on the entity type for constructing the representation of the entity under the specific type; in a triplet containing a specific relation, firstly mapping the representation of a head entity and a tail entity into the representation under the type limited by the relation, and then learning the semantics expressed by the triplet by adopting a translation idea; the mapping function adopts a circumferential convolution form, so that the type vector and the entity vector are fully communicated, the characteristic association between the corresponding dimension and the non-corresponding dimension is obtained, and the knowledge embedding quality is improved; in the negative sampling, the model limits the type of the sampled entity to be consistent with the type of the replaced entity with a certain probability, so that the difference among the entities belonging to the same type is increased, and certain similarity among the entities is considered;
the granularity of the time sequence relation quintuple represents a learning technology: performing representation learning by utilizing static structure context information and dynamic structure context information of five-tuple in a knowledge graph, wherein the dynamic structure context information is injected with information of space-time evolution and space evolution of the relationship;
discrete event representations, event representations based on distributed semantics and event representations based on inter-event information are proposed in event granularity representation learning technology, and the defects of single representation in the past are overcome.
7. The method for identifying and processing network public opinion anomalies according to claim 6, wherein in the second step of identifying network public opinion anomalies based on a social propagation structure, namely, in the construction of an abstract event map based on the social propagation, the abstract event map generalizes events so as to find more general and universal event evolution causal relationships in a certain network public opinion field; generalizing the events with similar distances through clustering, and representing a plurality of specific events by using a more representative event, thereby constructing an abstract map of network public opinion; the clustering samples are events, namely text samples, and the clustering algorithm is improved by using word2vec word vectors and a distance calculation method, so that the generalization result is optimal.
8. The method for identifying and processing network public opinion anomalies according to claim 6, characterized in that the third step of identifying network public opinion anomalies based on a social propagation structure uses a network public opinion anomaly identification model based on an abstract rational graph, the model further provides an extensible graph neural network model (ScaledGraphNeuralNetwork, SGNN), and network representation learning is performed on a large-scale propagation rational graph structure; by referencing the idea of divide-and-conquer in the training process, the SGNN only calculates on one small-scale related subgraph at a time to extend the GGNN to a large-scale directed cyclic weighted graph instead of calculating on the whole graph; the model identifies the correct post-network public opinion anomaly type by computing the similarity of the representation of the event and the candidate event representations in the context.
9. The method for identifying and processing network public opinion anomalies according to any one of claims 6-8, wherein the verification and duplication of the network public opinion anomalies is accomplished through three parts, namely expert thinking acquisition and construction, network public opinion anomaly polymorphic display and interpretive causal reasoning with augmented incident map;
the expert thinking acquisition and construction of relevant theory based on problem solving and cognitive psychology takes network public opinion abnormality under real situation as a research object, relevant information is summarized from public literature, policies and intelligent libraries to determine analysis angles so as to form a questionnaire, and then, the expert is subjected to semi-structured interview to acquire process information of expert analysis of the network public opinion abnormality so as to form an integrated event analysis process, the expert analysis of cognitive process and processing characteristics of the network public opinion abnormality is analyzed, and theoretical basis is provided for construction of an expert thinking model of the network public opinion abnormality analysis;
in the part of the network public opinion abnormal multi-form display, the multi-form display can effectively and clearly display the reasoning process of the complex public opinion abnormal events, a series of evidence events possibly exist between cause events and result events in the causal reasoning process, and the evidence events can serve as additional clues to help a model understand causal mechanisms between the cause and the result and enhance the effect of the interpretability and the robustness of the result;
in order to fully exploit the role of evidence events in causal inference, a causal inference framework (EventGraph KnowledgeEnhancedExplainableCausalReasoning, exCAR) with knowledge enhancement of the event atlas is used, which can obtain a series of causal evidence from a pre-constructed causal knowledge base, given causal pairs.
10. The method for identifying and processing Internet public opinion anomalies according to claim 9, wherein,
when screening key elements for determining the network public opinion abnormality identification, a screening model based on a time sequence diagram attention mechanism and the key elements for network public opinion abnormality identification are included;
in a screening model based on a time sequence diagram attention mechanism, in the process of knowledge graph construction, the original source of each knowledge segment is recorded by utilizing a URI address, so that in deduction analysis, the original data source can be traced back according to the corresponding URI of a given knowledge segment, and a fine-grained data basis is provided for decision making; in the knowledge graph constructed at this time, the structure of the graph is not static any more, and because the entities and the links between them evolve with time and establish a connection in the form of an event, the knowledge graph is also called an event knowledge graph; abstracting event knowledge graph as G inf At G inf If the source of the original data of the knowledge fragment needs to be traced back, the event with the largest attention score on the relevant path is required to be calculated, and meanwhile, key elements of the knowledge fragment are obtained according to the arrangement of the attention score;
key elements for identifying network public opinion abnormality: judging the influence of the key elements after screening by using an attention mechanism and an event relation diagram attention layer;
the network public opinion anomaly coping strategy part includes: network public opinion anomaly identification, network public opinion multi-element guidance, and network public opinion evaluation and response;
in the network public opinion anomaly identification, 5 types of anomalies focused in network public opinion analysis are identified based on a method for constructing a situation map and extracting information, wherein the 5 types of anomalies are respectively network public opinion anomalies based on consciousness forms, network public opinion anomalies based on interest appeal, network public opinion anomalies based on emotion release, network public opinion anomalies based on idea sound sheets and network public opinion anomalies based on social public security;
the social participation mechanism diversification is emphasized, so that all levels of main bodies participate in the network public opinion emergency management together, different management modes are set for different personnel, the front message is transmitted through the network main stream media, the people are guided to develop forward towards the network public opinion, and the participation of the whole people is exerted by utilizing the diversification management education mode;
in the aspect of network public opinion evaluation and coping, an emergency mechanism capable of controlling the network public opinion to go to a correct direction is formulated aiming at a prediction result by researching network public opinion propagation trend prediction under an emergency.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310746360.5A CN116795985A (en) | 2023-06-25 | 2023-06-25 | Network public opinion anomaly identification and processing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310746360.5A CN116795985A (en) | 2023-06-25 | 2023-06-25 | Network public opinion anomaly identification and processing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116795985A true CN116795985A (en) | 2023-09-22 |
Family
ID=88034212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310746360.5A Pending CN116795985A (en) | 2023-06-25 | 2023-06-25 | Network public opinion anomaly identification and processing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116795985A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117035087A (en) * | 2023-10-09 | 2023-11-10 | 北京壹永科技有限公司 | Method, device, equipment and medium for generating a rational map for medical reasoning |
| CN117422063A (en) * | 2023-12-18 | 2024-01-19 | 四川省大数据技术服务中心 | Big data processing method applying intelligent auxiliary decision and intelligent auxiliary decision system |
-
2023
- 2023-06-25 CN CN202310746360.5A patent/CN116795985A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117035087A (en) * | 2023-10-09 | 2023-11-10 | 北京壹永科技有限公司 | Method, device, equipment and medium for generating a rational map for medical reasoning |
| CN117035087B (en) * | 2023-10-09 | 2023-12-26 | 北京壹永科技有限公司 | Method, device, equipment and medium for generating a rational map for medical reasoning |
| CN117422063A (en) * | 2023-12-18 | 2024-01-19 | 四川省大数据技术服务中心 | Big data processing method applying intelligent auxiliary decision and intelligent auxiliary decision system |
| CN117422063B (en) * | 2023-12-18 | 2024-02-23 | 四川省大数据技术服务中心 | Big data processing method applying intelligent auxiliary decision and intelligent auxiliary decision system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wu et al. | A survey of human-in-the-loop for machine learning | |
| CN110489395B (en) | Method for automatically acquiring knowledge of multi-source heterogeneous data | |
| US20170200125A1 (en) | Information visualization method and intelligent visual analysis system based on text curriculum vitae information | |
| CN116795985A (en) | Network public opinion anomaly identification and processing method | |
| CN111930906A (en) | Knowledge graph question-answering method and device based on semantic block | |
| CN112883286A (en) | BERT-based method, equipment and medium for analyzing microblog emotion of new coronary pneumonia epidemic situation | |
| Cabrio et al. | Abstract dialectical frameworks for text exploration | |
| Liu et al. | An overview of event extraction and its applications | |
| Yuan et al. | Dual-level attention based on a heterogeneous graph convolution network for aspect-based sentiment classification | |
| Deng et al. | Beyond word embeddings: Heterogeneous prior knowledge driven multi-label image classification | |
| Jin et al. | Fintech key-phrase: a new Chinese financial high-tech dataset accelerating expression-level information retrieval | |
| Tong et al. | Multimedia network public opinion supervision prediction algorithm based on big data | |
| Kan et al. | A Composable Generative Framework based on Prompt Learning for Various Information Extraction Tasks | |
| Chang et al. | Multi-information preprocessing event extraction with BiLSTM-CRF attention for academic knowledge graph construction | |
| Al-Tameemi et al. | Multi-model fusion framework using deep learning for visual-textual sentiment classification | |
| Qiu et al. | NeuroSPE: A neuro‐net spatial relation extractor for natural language text fusing gazetteers and pretrained models | |
| CN117236676A (en) | RPA process mining method and device based on multi-mode event extraction | |
| Zhao et al. | Network media public opinion and social governance supported by the internet-of-things big data | |
| Gan et al. | A multimodal fusion network with attention mechanisms for visual–textual sentiment analysis | |
| Zhu et al. | PlanGPT: Enhancing Urban Planning with Tailored Language Model and Efficient Retrieval | |
| Sun et al. | ASRC: A Knowledge Graph Relation Construction Model based on Active Learning and Semantic Recognition | |
| Sun et al. | Knowledge based machine reading comprehension | |
| Divya et al. | Machine Learning Techniques and Frameworks for Heterogeneous Data Fusion in Big Data Analytics | |
| Ranjbar-Khadivi et al. | Persian topic detection based on Human Word association and graph embedding | |
| Harzig et al. | Image captioning with clause-focused metrics in a multi-modal setting for marketing |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |