New media wind control method and system based on artificial intelligence

Abstract

The invention discloses a new media risk control method and system based on artificial intelligence, and relates to the technical field of public opinion detection. Obtaining public opinion data of the target account to carry out risk control risk detection, and if there is a risk control risk, determining public opinion monitoring data; performing public opinion event extraction to obtain multiple pieces of target data, and constructing to obtain a public opinion event graph; performing feature extraction on the public opinion event graph to obtain a plurality of data storage structures and fusing the data storage structures to obtain a fused data structure; extracting the fused data structure to obtain an intrinsic mode function, converting the intrinsic mode function into a target spectrogram, and performing feature extraction on the target spectrogram to obtain actual features; and constructing according to the actual features to obtain a target event graph, evaluating the target event graph to obtain a risk score, and determining whether risk control is carried out or not. The public opinion risk is efficiently identified, data management is optimized, the accuracy and real-time performance of public opinion monitoring are improved, potential risks are effectively reduced, and stable operation of a platform is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the technical field of public opinion detection, and specifically relates to a method and system for new media public opinion risk control based on artificial intelligence. Background Art

[0002] With the rapid development of new media, the speed and scope of information dissemination are growing exponentially. After enterprises and brands carry out public opinion propaganda on new media platforms, they face a complex and changing public opinion environment. On the one hand, positive public opinion feedback can quickly amplify brand influence; on the other hand, negative information may also attract widespread attention in a short period of time, bringing potential risks. Therefore, effective public opinion monitoring has become a key link in new media operations.

[0003] Patent No.: CN118474427B, discloses a network public opinion detection method and system; obtain the heat data of the monitored account, and judge whether there is public opinion based on the heat data; if it is determined that the monitored account has public opinion, then judge the nature of the public opinion of the monitored account and determine the nature of the public opinion; take corresponding measures according to the nature of the public opinion; if it is determined that the monitored account does not have public opinion, then obtain the hot list topic data, and perform public opinion analysis on the hot list topic data to determine the public opinion content of the hot list topic, and record it as the hot list public opinion content; match the hot list public opinion content with the account positioning of the monitored account to determine whether the hot list public opinion content matches the account positioning; if the hot list public opinion content matches the account positioning, then take corresponding measures for the monitored account according to the hot list public opinion content and the account positioning.

[0004] Although the above technology has solved some problems, some problems still exist. For example, in the process of public opinion monitoring after the new media releases public opinion, the system will run slowly due to the large amount of data, which may lead to untimely monitoring, poor monitoring effect and low accuracy, resulting in insufficient risk control response speed. Summary of the invention

[0005] The purpose of the present invention is to solve the problem that in the process of public opinion monitoring after the new media public opinion propaganda, the system will run slowly due to the large amount of data, which is prone to untimely monitoring, poor monitoring effect and low accuracy, resulting in insufficient risk control response speed, and propose a method and system for new media public opinion risk control based on artificial intelligence.

[0006] In a first aspect of the present invention, a method for new media public opinion risk control based on artificial intelligence is first proposed, the method comprising: Obtain the public opinion data of the target account, and use artificial intelligence to detect whether the public opinion data has risk control risks. If there is risk control risk, determine the public opinion monitoring data; Extracting public opinion events from the public opinion data and the public opinion monitoring data to obtain a plurality of target data, and constructing a plurality of public opinion event graphs based on the target data; Extract features from each public opinion event graph to obtain multiple data storage structures, and fuse each data storage structure to obtain a fused data structure; Extracting the fused data structure through a preset model to obtain an intrinsic mode function, converting the intrinsic mode function into a target spectrum graph, and performing feature extraction on the target spectrum graph to obtain actual features; A target event graph is constructed according to the actual features, and the target event graph is evaluated to obtain a risk score, and risk control is performed on the new media according to the risk score.

[0007] Optionally, multiple public opinion event graphs are constructed based on each target data, including: Performing text extraction on the public opinion data and the public opinion monitoring data to obtain target data, and determining nodes and edges of the public opinion event graph according to the target data; the target data includes: trigger words, entity parameters and event roles; The event relationship between the public opinion data and the public opinion monitoring data is identified, the composition relationship between nodes and edges is determined according to the event relationship, and a public opinion event graph is constructed according to the composition relationship.

[0008] Optionally, before extracting the fused data structure by using a preset model to obtain the intrinsic mode function, the method further includes: Establishing a data conversion model according to the data structure of the fused data structure, and inputting the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: , Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

[0009] Optionally, establishing a data conversion model according to the data structure of the fused data structure includes: Decomposing the time series data according to the frequency by empirical mode decomposition to obtain an intrinsic mode function, converting the intrinsic mode function from a time domain signal to a frequency domain signal to obtain a first intrinsic mode function, and filtering the first intrinsic mode function to delete pseudo components to obtain a second intrinsic mode function; A spectrum feature extraction model is constructed by the second intrinsic mode function, and phase difference data of each second intrinsic mode function is extracted according to a preset sampling interval; Inputting the phase difference data into the spectrum feature extraction model for training to obtain model parameters, and continuously updating the spectrum feature extraction model according to the model parameters to obtain a target spectrum feature extraction model; The fusion data structure is extracted through the preset model to obtain the intrinsic mode function, including: The fusion data structure is input into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and the target spectrum graph is obtained according to the instantaneous frequency and the instantaneous amplitude transformation.

[0010] Optionally, risk control is performed on the new media according to the risk score, including: If the first risk threshold < risk score ≤ second risk threshold, it is determined that the target account has public opinion, and risk monitoring is performed on the target account; If the second risk threshold is less than the risk score, it is determined that the target account has a public opinion risk, and risk control is performed on the target account.

[0011] In the second aspect of the implementation of the present invention, a new media public opinion risk control system based on artificial intelligence is proposed, including: a risk control judgment module, an event graph construction module, a data fusion module, a feature extraction module and a risk control module: The risk control judgment module is used to obtain the public opinion data of the target account, detect whether the public opinion data has risk control risks through artificial intelligence, and determine the public opinion monitoring data if there is risk control risk; The event graph construction module is used to extract public opinion events from the public opinion data and public opinion monitoring data to obtain multiple target data, and to construct multiple public opinion event graphs based on each target data; The data fusion module is used to extract features from each public opinion event graph to obtain multiple data storage structures, and fuse each data storage structure to obtain a fused data structure; The feature extraction module is used to extract the fusion data structure through a preset model to obtain an intrinsic mode function, convert the intrinsic mode function into a target spectrum graph, and perform feature extraction on the target spectrum graph to obtain actual features; The risk control module is used to construct a target event graph according to the actual features, evaluate the target event graph to obtain a risk score, and perform risk control on the new media according to the risk score.

[0012] Optionally, the event graph construction module includes: a public opinion event graph acquisition module and a public opinion event graph construction module: The public opinion event graph acquisition module is used to extract text from the public opinion data and the public opinion monitoring data to obtain target data, and determine the nodes and edges of the public opinion event graph according to the target data; the target data includes: trigger words, entity parameters and event roles; The public opinion event graph construction module is used to identify the event relationship between the public opinion data and the public opinion monitoring data, determine the composition relationship between nodes and edges according to the event relationship, and construct a public opinion event graph according to the composition relationship.

[0013] Optionally, the system includes: a spectral conversion module, used to establish a data conversion model according to the data structure of the fused data structure, and input the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: , Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

[0014] Optionally, the spectrum conversion module includes: a time-frequency conversion module, a phase difference data extraction module, an extraction model update module and a feature extraction module: The time-frequency conversion module is used to decompose the time series data according to the frequency through empirical mode decomposition to obtain an intrinsic mode function, convert the eigenmode function from a time domain signal to a frequency domain signal to obtain a first eigenmode function, and filter the first eigenmode function to delete pseudo components to obtain a second eigenmode function; The phase difference data extraction module is used to construct a spectrum feature extraction model through the second intrinsic mode function, and extract the phase difference data of each second intrinsic mode function according to a preset sampling interval; The extraction model updating module is used to input the phase difference data into the spectrum feature extraction model for training to obtain model parameters, and to continuously update the spectrum feature extraction model according to the model parameters to obtain a target spectrum feature extraction model; The feature extraction module is also used to input the fusion data structure into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and obtain the target spectrum diagram according to the instantaneous frequency and the instantaneous amplitude transformation.

[0015] Optionally, the risk control module includes: a first risk execution module and a second risk execution module: The first risk execution module is used to determine that the target account has public opinion if the first risk threshold < risk score ≤ second risk threshold, and perform risk monitoring on the target account; The second risk execution module is used to determine that the target account has a public opinion risk if the second risk threshold is less than the risk score, and to perform risk management on the target account.

[0016] Beneficial effects of the present invention: The present invention proposes a method for risk control of new media public opinion based on artificial intelligence. The method performs risk control risk detection by acquiring the public opinion data of the target account, and determines the public opinion monitoring data if there is a risk control risk; extracts public opinion events to obtain multiple target data and constructs a public opinion event graph; extracts features from the public opinion event graph to obtain multiple data storage structures and fuses them to obtain a fused data structure; extracts the fused data structure to obtain an intrinsic mode function and converts it into a target spectrum graph, extracts features from the target spectrum graph to obtain actual features; constructs a target event graph based on the actual features, and evaluates the target event graph to obtain a risk score and determine whether to perform risk control. Artificial intelligence technology performs risk control risk detection and public opinion event extraction on new media public opinion data, constructs a public opinion event graph and fuses the data storage structure, and then uses the intrinsic mode function and spectrum graph to extract actual features, evaluates risks, and performs new media risk control. The method realizes efficient identification of public opinion risks, optimizes data management, improves the accuracy and real-time performance of public opinion monitoring, effectively reduces potential risks, and ensures the stable operation of the platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention will be further described below in conjunction with the accompanying drawings.

[0018] Figure 1 A flowchart of a method for new media public opinion risk control based on artificial intelligence is provided for an embodiment of the present invention; Figure 2 A framework diagram of another system for new media public opinion and risk control based on artificial intelligence is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The term "and / or" herein is only a description of the association relationship of the associated objects, indicating that there can be three relationships. For example, A and B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the present invention, the description of "first", "second", etc. is only used for descriptive purposes, and cannot be understood as indicating or implying its relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first" and "second" can explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but it must be based on the ability of ordinary technicians in the field to implement. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0021] The embodiment of the present invention provides a method for new media public opinion risk control based on artificial intelligence. Figure 1 , Figure 1 A flowchart of a method for new media public opinion risk control based on artificial intelligence is provided in an embodiment of the present invention. The method comprises the following steps: S101, obtaining the public opinion data of the target account, and using artificial intelligence to detect whether the public opinion data has risk control risks, and if there is risk control risks, determining the public opinion monitoring data; S102, extracting public opinion events from public opinion data and public opinion monitoring data to obtain multiple target data, and constructing multiple public opinion event graphs based on the target data; S103, extracting features from each public opinion event graph to obtain multiple data storage structures, and fusing each data storage structure to obtain a fused data structure; S104, extracting the fused data structure through a preset model to obtain an intrinsic mode function, converting the intrinsic mode function into a target spectrum graph, and performing feature extraction on the target spectrum graph to obtain actual features; S105, constructing a target event graph based on actual features and evaluating the target event graph to obtain a risk score, and performing risk control on the new media based on the risk score.

[0022] Based on the artificial intelligence-based new media public opinion risk control method provided by the embodiment of the present invention, risk control risk detection and public opinion event extraction are performed on new media public opinion data through artificial intelligence technology, a public opinion event graph is constructed and integrated with the data storage structure, and then the intrinsic mode function and spectrum graph are used to extract actual features, assess risks and perform new media risk control. Efficient identification of public opinion risks, optimization of data management, improvement of the accuracy and real-time performance of public opinion monitoring, effective reduction of potential risks, and guarantee of stable operation of the platform are achieved.

[0023] In one implementation, AI is used to detect the public opinion risk of an account, mainly using natural language processing (NLP) technology and machine learning models to perform sentiment analysis, keyword extraction, and semantic understanding on the content posted by the account, and identify whether it contains negative emotions, sensitive topics, or illegal information. At the same time, combined with the account's behavioral data (such as posting frequency, interaction mode) and user feedback, a risk assessment model is constructed to monitor and quantitatively assess the public opinion risk of the account in real time, thereby achieving rapid early warning and accurate identification of potential public opinion risks.

[0024] In one implementation, the main purpose of converting public opinion data into spectral data is to more effectively extract features from the data, reduce redundant information, and use the advantages of frequency domain analysis to improve the efficiency and accuracy of data processing. Through time series analysis, empirical mode decomposition, high-order cumulants, neural networks and other technologies, the intrinsic modal characteristics of public opinion data can be expressed in the form of a spectrum, thereby providing support for subsequent storage optimization and event analysis (risk assessment).

[0025] In one implementation, public opinion data is usually temporally dynamic, that is, the evolution of public opinion and the development of events change over time. This time series characteristic allows public opinion data to be processed by time series analysis methods. Time series analysis can capture the periodicity, trend, and random components in the data, which can be converted into frequency domain (spectral) representations through Fourier transform or other frequency domain analysis methods. EMD is a feature extraction method suitable for non-stationary and nonlinear signals, which can decompose complex time series data into multiple intrinsic mode functions (IMFs). Each IMF represents the fluctuation components of different frequency scales in the data, which can be further converted into instantaneous frequency and amplitude through Hilbert Transform to form spectral data.

[0026] In one implementation, by extracting features from multiple public opinion event graphs and fusing them into a unified fusion data structure, the scattered public opinion information can be effectively integrated to avoid data redundancy and information fragmentation. This data fusion method provides a higher-quality data foundation for subsequent feature extraction, making the feature extraction process more efficient, and can more accurately capture the core features of public opinion events. It combines the public opinion data of the target account with the public opinion monitoring data of social hot spots, extracts multiple target data and constructs a public opinion event graph, which can comprehensively integrate public opinion information from both the micro (public opinion data, that is, a single account) and macro (public opinion monitoring data, that is, social hot spots) levels.

[0027] In one implementation, by obtaining the public opinion data of the target account and using artificial intelligence technology to detect risk control risks, it is possible to quickly and accurately identify whether a single account has potential risks. This data-driven risk detection method avoids the subjectivity and inefficiency of traditional manual review, can quickly locate risky accounts in massive data, provide clear goals for subsequent risk control, and significantly improve risk control efficiency.

[0028] In one implementation, the use of time series neural networks and empirical mode decomposition to extract features from public opinion monitoring data can effectively process non-stationary signals, extract more representative intrinsic mode functions, and further convert them into target spectrum graphs. This advanced feature extraction method not only improves the accuracy and stability of features, but also enhances the model's generalization ability for different types of public opinion data, enabling it to better adapt to complex and changing public opinion environments.

[0029] In one implementation, a target event graph is constructed based on the extracted actual features, and the risk score is obtained by evaluating it, which can transform the complex information of the public opinion event into an intuitive risk indicator. This visual and quantitative risk assessment method makes public opinion management more scientific and accurate, and can take targeted risk control measures according to different risk levels, effectively reducing the negative impact of public opinion risks on new media. Constructing a target event graph based on the extracted actual features is equivalent to constructing a knowledge graph. For example, event influence (I): can be measured by the following specific indicators: number of media reports (N): the number of media reports, social media popularity (S): the number of discussions, likes, and reposts on social media, spread range (R): the geographical range of event spread (such as the target area), user participation (U): the degree of user participation in the event, such as the number of comments, the number of people participating in the event, etc. Determine the scoring criteria to set weights for each indicator, and calculate the comprehensive influence score based on the actual data. For example: number of media reports (N): weight 0.3; social media popularity (S): weight 0.4; spread range (R): weight 0.2; user participation (U): weight 0.1. Data normalization, normalize the raw data of each indicator to the interval [0,1] for comprehensive calculation. For example: Number of media reports (N): 100 articles; Social media popularity (S): 100,000 discussions; Dissemination range (R): the ratio of the dissemination area to the target area (normalized value is 0.8); User engagement (U): 50,000 comments; that is: the maximum number of media reports is 200 articles, and the normalization formula is: ; The maximum value of social media popularity is 200,000, and the normalization formula is: ; The maximum value of user participation is 100,000, and the normalization formula is: ; Event Impact (I): , convert it to a ten-point system, that is, multiply by ten to get 5.6 points. Probability of risk occurrence (L): Historical data shows that the probability of similar activities being delayed due to weather is 10%. The current weather forecast shows a 50% probability of bad weather. Comprehensive assessment, assign the risk probability L=0.5. Risk controllability (P): The event organizer has an emergency plan, but it is difficult to implement. Comprehensive assessment, assign risk controllability P=0.6. Based on the risk score calculation formula: D=L×P×I, D is the risk score, that is, 0.5×0.6×5.6=1.68, the larger D is, the higher the public opinion risk.

[0030] In one embodiment, multiple public opinion event graphs are constructed based on each target data, including: Text extraction is performed on public opinion data and public opinion monitoring data to obtain target data, and nodes and edges of the public opinion event graph are determined based on the target data; the target data includes: trigger words, entity parameters and event roles; Identify the event relationships between public opinion data and public opinion monitoring data, determine the composition relationships between nodes and edges based on the event relationships, and then construct a public opinion event graph based on the composition relationships.

[0031] In one implementation, trigger words, entity parameters and event roles, trigger words: are marker words for the occurrence of an event, such as "explosion", "release", "acquisition", etc., these words indicate the core action of the event; entity parameters: are specific objects or entities involved in the event, such as "a company", "a product", etc.; event roles: are the roles of parameters in the event, such as "subject", "object", "place", etc.; corresponding relationships: these extracted contents correspond to nodes and edges in the atomic event graph respectively; the nodes in the graph are used to represent the trigger words and parameters of the event; the edges in the graph are used to represent role relationships.

[0032] In one implementation, the public opinion event graph construction task is the process of automatically constructing a public opinion event graph from documents describing public opinion events. This paper proposes a two-stage public opinion event graph construction task framework. The first step is atomic event extraction. Atomic event extraction mainly extracts the trigger words, parameters, and roles of atomic events that appear in the text, which correspond to the nodes of the public opinion event graph. The second step is to identify event relationships, which mainly identifies the relationships between atomic events, corresponding to the edges in the public opinion event graph.

[0033] In one implementation, the atomic event extraction stage can accurately identify the atomic events and their key elements (trigger words, parameters, and roles) in the text. These elements are important components of the nodes of the public opinion event graph. The event relationship identification stage can accurately determine the relationship between atomic events, thereby reasonably constructing the edges in the graph. The data-driven extraction and identification process can more objectively and comprehensively reflect the actual situation of public opinion events, avoiding the subjective bias that may occur during manual construction. For example, for complex public opinion events, it may be difficult for humans to accurately grasp the relationship between events, and this method can identify the event relationship hidden in the text through an algorithm, so that the public opinion event graph can more completely present the full picture of the event, including the cause of the event, the development process, and the subjects involved.

[0034] In one implementation, the public opinion event graph is constructed by automatically extracting target data (such as trigger words, entity parameters, and event roles) from public opinion data and public opinion monitoring data, and identifying event relationships, thereby reducing the number of manual interventions. In the past, constructing a public opinion event graph may have required manual reading of large amounts of text data, screening out key information, and manually constructing a graph structure, which was not only time-consuming and labor-intensive, but also prone to omissions or errors due to human factors. This automated method can quickly process massive amounts of data and quickly generate a public opinion event graph, greatly improving the efficiency of public opinion analysis, enabling public opinion monitoring agencies to effectively sort out and visualize large amounts of public opinion information in a short period of time, and keep abreast of public opinion trends. In one embodiment, before extracting the fused data structure by using a preset model to obtain the intrinsic mode function, the method further includes: Establish a data conversion model according to the data structure of the fused data structure, and input the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: , Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

[0035] In one implementation, ,y(t) contains the instantaneous frequency and phase information of the original signal x(t), θ is the integral variable representing the time scale (time scale, representing all time points traversed during the integration process, that is, the global time range from the start to the end of the signal), t (the current time point, which is the target moment of analysis, indicating the specific time position where the analytical signal y(t) needs to be calculated), and the integral operation weights the signal x(θ) in the global time range through θ, and finally obtains the analytical signal y(t) at t; the data conversion model is used to extract the instantaneous frequency of the event graph feature; by introducing the energy consumption and energy efficiency characteristics P as adjustment factors, the modeling process of the time series signal y(t) is optimized. In practical applications, public opinion data often presents non-stationary and nonlinear characteristics, and is affected by external changes (for example: social environment, economic conditions and other factors). By adjusting the integral weight of the time series signal, the model can more accurately capture these complex changes, thereby improving the accuracy of public opinion monitoring.

[0036] In one implementation, the P value in the expression of the data conversion model is an adjustable parameter, which allows the model to be flexibly adjusted according to different data sets and monitoring needs. When faced with different types of hot events or different public opinion environments, researchers can optimize model performance by adjusting the P value to make it better adapt to new situations.

[0037] In one embodiment, a data conversion model is established according to the data structure of the fused data structure, including: Decomposing the time series data according to the frequency by empirical mode decomposition to obtain the intrinsic mode function, converting the intrinsic mode function from the time domain signal to the frequency domain signal to obtain the first intrinsic mode function, and filtering the first intrinsic mode function to remove the pseudo component to obtain the second intrinsic mode function; A spectrum feature extraction model is constructed by the second intrinsic mode function, and phase difference data of each second intrinsic mode function is extracted according to a preset sampling interval; The phase difference data is input into the spectrum feature extraction model for training to obtain model parameters, and the spectrum feature extraction model is continuously updated according to the model parameters to obtain a target spectrum feature extraction model; The fusion data structure is extracted through the preset model to obtain the intrinsic mode function, including: The fused data structure is input into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and the target spectrum graph is obtained according to the instantaneous frequency and instantaneous amplitude transformation.

[0038] In one implementation, empirical mode decomposition (EMD) is used to decompose complex signals into frequency-ordered intrinsic mode functions (IMFs). EMD decomposes public opinion data into multiple IMFs, each of which represents a vibration mode at a different time scale. By extracting key information from the IMF components, noise can be eliminated and the relevant parts of event extraction can be focused on. Due to the presence of noise, IMFs may contain pseudo components. To solve this problem, the Hilbert transform is applied to convert the time domain signal into a frequency domain signal to extract features such as instantaneous frequency and instantaneous amplitude. By filtering the IMF, pseudo components can be removed and the real signal can be retained, thereby ensuring the accuracy of data analysis. After IMF processing, the frequency-modulated Hilbert spectrum shift indicates frequency changes, resulting in phase differences in different frequency components to obtain phase difference data (each IMF represents a fluctuation component at a different frequency scale in the data, which can be further converted into instantaneous frequency and amplitude through Hilbert Transform to form spectral data).

[0039] In one implementation, the data spectrum is filtered and adjusted by setting appropriate threshold conditions to eliminate or weaken interfering elements and highlight key features. The filtering strategy can be adaptively adjusted according to specific data conditions to obtain better spectral feature extraction results.

[0040] In one implementation, the spectrum feature extraction model is also a time series neural network model. The time series neural network can model the temporal dynamics of public opinion data and capture the dynamic evolution of events and the development of public opinion. Through the feature extraction capability of the neural network, the time series data can be mapped to a feature space, and the data in these feature spaces can be further converted into spectral data through frequency domain analysis.

[0041] In one implementation, the target spectrum feature extraction model is obtained by training the spectrum feature extraction model, and the target spectrum feature extraction model is used to capture the dynamic characteristics of the time series. In the model parameter training stage, the high-order cumulative filtering technology is used to improve the accuracy and robustness of parameter estimation. By analyzing the signal, the instantaneous frequency and instantaneous amplitude of each IMF are calculated, which can reveal the local dynamic characteristics of the time series. The target spectrum graph (spectrogram) is obtained based on the instantaneous frequency and instantaneous amplitude transformation. Through these instantaneous features, a time-frequency distribution graph, namely the target spectrum graph, can be constructed. The spectrum graph shows the frequency distribution and amplitude changes of public opinion data at different time points.

[0042] In one embodiment, risk control of new media is performed based on risk scores, including: If the first risk threshold < risk score ≤ second risk threshold, it is determined that the target account has public opinion, and risk monitoring is performed on the target account; If the second risk threshold is less than the risk score, the target account is judged to have a public opinion risk, and risk control is performed on the target account.

[0043] In one implementation, by setting a first risk threshold and a second risk threshold, the risk scores of the target accounts are classified and processed, and the public opinion risks can be divided into different levels. This grading method makes public opinion management no longer a simple binary judgment, but can take differentiated response measures according to the degree of risk. For target accounts with risk scores in the middle range, only risk monitoring is carried out to avoid excessive intervention; for accounts with higher risk scores, they directly enter the risk control link to ensure that limited resources can be reasonably allocated to the accounts that need the most attention, thereby improving the efficiency and pertinence of public opinion management.

[0044] In one implementation, a dynamic assessment mechanism based on risk scoring can reflect the public opinion status of the target account in real time, so that the management strategy can be adjusted in time according to the actual risk level of the account. For example, when the risk score of an account is close to the threshold, early warning can be issued and measures can be taken to prevent further escalation of the risk; for accounts with lower risk scores, unnecessary intervention can be reduced, reducing management costs.

[0045] In one implementation, high-risk accounts can be proactively identified and control measures can be taken in a timely manner through clear risk assessment criteria and hierarchical control measures. This proactive management approach helps to intervene before public opinion risks fully erupt, thus avoiding further spread and deterioration of public opinion events. At the same time, this hierarchical control mechanism also provides a clear operating process for public opinion management, reduces management errors caused by inaccurate human judgment, and improves the overall effect of public opinion risk control.

[0046] Based on the same inventive concept, the embodiment of the present invention also provides a new media public opinion risk control system based on artificial intelligence. Figure 2 , Figure 2 A schematic diagram of the structure of a new media public opinion risk control system based on artificial intelligence provided by an embodiment of the present invention includes: a risk control judgment module, an event graph construction module, a data fusion module, a feature extraction module and a risk control module: The risk control judgment module is used to obtain the public opinion data of the target account and detect whether there is risk control risk in the public opinion data through artificial intelligence. If there is risk control risk, the public opinion monitoring data is determined; An event graph construction module is used to extract public opinion events from public opinion data and public opinion monitoring data to obtain multiple target data, and to construct multiple public opinion event graphs based on each target data; A data fusion module is used to extract features from each public opinion event graph to obtain multiple data storage structures, and fuse each data storage structure to obtain a fused data structure; A feature extraction module is used to extract the fusion data structure through a preset model to obtain an intrinsic mode function, convert the intrinsic mode function into a target spectrum graph, and perform feature extraction on the target spectrum graph to obtain actual features; The risk control module is used to construct a target event graph based on actual characteristics and evaluate the target event graph to obtain a risk score, and perform risk control on new media based on the risk score.

[0047] Based on the artificial intelligence-based new media public opinion risk control system provided by the embodiment of the present invention, the new media public opinion data is subjected to risk control risk detection and public opinion event extraction through artificial intelligence technology, a public opinion event graph is constructed and integrated with the data storage structure, and then the intrinsic mode function and spectrum graph are used to extract actual features, assess risks and perform new media risk control. It realizes efficient identification of public opinion risks, optimizes data management, improves the accuracy and real-time performance of public opinion monitoring, effectively reduces potential risks, and ensures the stable operation of the platform.

[0048] In one embodiment, the event graph construction module includes: a public opinion event graph acquisition module and a public opinion event graph construction module: The public opinion event graph acquisition module is used to extract text from public opinion data and public opinion monitoring data to obtain target data, and determine the nodes and edges of the public opinion event graph based on the target data; the target data includes: trigger words, entity parameters and event roles; The public opinion event graph construction module is used to identify the event relationship between public opinion data and public opinion monitoring data, determine the composition relationship of nodes and edges based on the event relationship, and construct the public opinion event graph based on the composition relationship.

[0049] In one embodiment, the system includes: a spectral conversion module, for establishing a data conversion model according to the data structure of the fused data structure, and inputting the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: , Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

[0050] In one embodiment, the spectrum conversion module includes: a time-frequency conversion module, a phase difference data extraction module, an extraction model update module and a feature extraction module: The time-frequency conversion module is used to decompose the time series data according to the frequency through empirical mode decomposition to obtain the intrinsic mode function, convert the intrinsic mode function from the time domain signal to the frequency domain signal to obtain the first intrinsic mode function, and filter the first intrinsic mode function to remove the pseudo component to obtain the second intrinsic mode function; A phase difference data extraction module is used to construct a spectrum feature extraction model through the second intrinsic mode function, and extract the phase difference data of each second intrinsic mode function according to a preset sampling interval; An extraction model updating module is used to input the phase difference data into the spectrum feature extraction model for training to obtain model parameters, and to continue updating the spectrum feature extraction model according to the model parameters to obtain a target spectrum feature extraction model; The feature extraction module is also used to input the fusion data structure into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and obtain the target spectrum diagram according to the instantaneous frequency and instantaneous amplitude transformation.

[0051] In one embodiment, the risk control module includes: a first risk execution module and a second risk execution module: A first risk execution module is used to determine that there is public opinion on the target account and perform risk monitoring on the target account if the first risk threshold < risk score ≤ second risk threshold; The second risk execution module is used to determine that the target account has public opinion risk and perform risk management on the target account if the second risk threshold is less than the risk score.

[0052] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for new media public opinion risk control based on artificial intelligence, characterized in that: The method comprises: Obtain the public opinion data of the target account, and use artificial intelligence to detect whether the public opinion data has risk control risks. If there is risk control risk, determine the public opinion monitoring data; Extracting public opinion events from the public opinion data and the public opinion monitoring data to obtain a plurality of target data, and constructing a plurality of public opinion event graphs based on the target data; Extract features from each public opinion event graph to obtain multiple data storage structures, and fuse each data storage structure to obtain a fused data structure; Extracting the fused data structure through a preset model to obtain an intrinsic mode function, converting the intrinsic mode function into a target spectrum graph, and performing feature extraction on the target spectrum graph to obtain actual features; A target event graph is constructed according to the actual features, and the target event graph is evaluated to obtain a risk score, and risk control is performed on the new media according to the risk score.

2. According to the method of new media public opinion risk control based on artificial intelligence in claim 1, it is characterized in that: According to each target data, multiple public opinion event graphs are constructed, including: Performing text extraction on the public opinion data and the public opinion monitoring data to obtain target data, and determining nodes and edges of the public opinion event graph according to the target data; the target data includes: trigger words, entity parameters and event roles; The event relationship between the public opinion data and the public opinion monitoring data is identified, the composition relationship between nodes and edges is determined according to the event relationship, and a public opinion event graph is constructed according to the composition relationship.

3. The method for new media public opinion risk control based on artificial intelligence according to claim 1 is characterized in that: Before extracting the fused data structure through a preset model to obtain an intrinsic mode function, the method further includes: Establishing a data conversion model according to the data structure of the fused data structure, and inputting the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: ， Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

4. The method for new media public opinion risk control based on artificial intelligence according to claim 3 is characterized in that: Establishing a data conversion model according to the data structure of the fused data structure includes: Decomposing the time series data according to the frequency by empirical mode decomposition to obtain an intrinsic mode function, converting the intrinsic mode function from a time domain signal to a frequency domain signal to obtain a first intrinsic mode function, and filtering the first intrinsic mode function to delete pseudo components to obtain a second intrinsic mode function; A spectrum feature extraction model is constructed by the second intrinsic mode function, and phase difference data of each second intrinsic mode function is extracted according to a preset sampling interval; Inputting the phase difference data into the spectrum feature extraction model for training to obtain model parameters, and continuously updating the spectrum feature extraction model according to the model parameters to obtain a target spectrum feature extraction model; The fusion data structure is extracted through the preset model to obtain the intrinsic mode function, including: The fusion data structure is input into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and the target spectrum graph is obtained according to the instantaneous frequency and the instantaneous amplitude transformation.

5. The method for new media public opinion risk control based on artificial intelligence according to claim 1 is characterized in that: Perform risk control on new media based on the risk score, including: If the first risk threshold < risk score ≤ second risk threshold, it is determined that the target account has public opinion, and risk monitoring is performed on the target account; If the second risk threshold is less than the risk score, it is determined that the target account has a public opinion risk, and risk control is performed on the target account.

6. A new media public opinion risk control system based on artificial intelligence, characterized in that: The system includes: a risk control judgment module, an event graph construction module, a data fusion module, a feature extraction module and a risk control module: The risk control judgment module is used to obtain the public opinion data of the target account, detect whether the public opinion data has risk control risks through artificial intelligence, and determine the public opinion monitoring data if there is risk control risk; The event graph construction module is used to extract public opinion events from the public opinion data and public opinion monitoring data to obtain multiple target data, and to construct multiple public opinion event graphs based on each target data; The data fusion module is used to extract features from each public opinion event graph to obtain multiple data storage structures, and fuse each data storage structure to obtain a fused data structure; The feature extraction module is used to extract the fusion data structure through a preset model to obtain an intrinsic mode function, convert the intrinsic mode function into a target spectrum graph, and perform feature extraction on the target spectrum graph to obtain actual features; The risk control module is used to construct a target event graph according to the actual features, evaluate the target event graph to obtain a risk score, and perform risk control on the new media according to the risk score.

7. The system for new media public opinion and risk control based on artificial intelligence according to claim 6 is characterized in that: The event graph construction module includes: a public opinion event graph acquisition module and a public opinion event graph construction module: The public opinion event graph acquisition module is used to extract text from the public opinion data and the public opinion monitoring data to obtain target data, and determine the nodes and edges of the public opinion event graph according to the target data; the target data includes: trigger words, entity parameters and event roles; The public opinion event graph construction module is used to identify the event relationship between the public opinion data and the public opinion monitoring data, determine the composition relationship between nodes and edges according to the event relationship, and construct a public opinion event graph according to the composition relationship.

8. The system for new media public opinion and risk control based on artificial intelligence according to claim 6 is characterized in that: The system comprises: a spectral conversion module, for establishing a data conversion model according to the data structure of the fused data structure, and inputting the fused data structure into the data conversion model to obtain time series data; The expression of the data conversion model is: ， Among them, y(t) represents the instantaneous frequency of the event graph feature at time t, x(θ) is the original time series signal, the integral symbol ∫ represents the integral of θ, and P is the energy consumption of the input signal x(θ). is a mathematical constant, and θ is an integral variable representing the time scale.

9. The system for new media public opinion and risk control based on artificial intelligence according to claim 8 is characterized in that: The spectrum conversion module includes: a time-frequency conversion module, a phase difference data extraction module, an extraction model update module and a feature extraction module: The time-frequency conversion module is used to decompose the time series data according to the frequency through empirical mode decomposition to obtain an intrinsic mode function, convert the eigenmode function from a time domain signal to a frequency domain signal to obtain a first eigenmode function, and filter the first eigenmode function to delete pseudo components to obtain a second eigenmode function; The phase difference data extraction module is used to construct a spectrum feature extraction model through the second intrinsic mode function, and extract the phase difference data of each second intrinsic mode function according to a preset sampling interval; The extraction model updating module is used to input the phase difference data into the spectrum feature extraction model for training to obtain model parameters, and to continuously update the spectrum feature extraction model according to the model parameters to obtain a target spectrum feature extraction model; The feature extraction module is also used to input the fusion data structure into the target spectrum feature extraction model to obtain the instantaneous frequency and instantaneous amplitude, and obtain the target spectrum diagram according to the instantaneous frequency and the instantaneous amplitude transformation.

10. The system for new media public opinion and risk control based on artificial intelligence according to claim 6, characterized in that: The risk control module includes: a first risk execution module and a second risk execution module: The first risk execution module is used to determine that the target account has public opinion if the first risk threshold < risk score ≤ second risk threshold, and perform risk monitoring on the target account; The second risk execution module is used to determine that the target account has a public opinion risk if the second risk threshold is less than the risk score, and to perform risk management on the target account.

Images