An opinion information risk dynamic evaluation method fusing multi-modal data
By integrating multimodal data with a dynamic risk assessment model, the problems of information bias and poor dynamic adaptability in single-modal analysis are solved, enabling real-time monitoring and accurate prediction of public opinion risks, and improving the comprehensiveness and adaptability of public opinion analysis.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 中科天玑数据科技股份有限公司
- Filing Date
- 2025-12-19
- Publication Date
- 2026-07-10
AI Technical Summary
Existing technologies mainly rely on single-modal data for public opinion analysis, which suffers from problems such as information bias, poor dynamic adaptability, and low technical coupling, making it difficult to fully reflect the complexity of public opinion events and respond quickly to the spread of public opinion.
By integrating multimodal data (text, images, audio, and video) with a dynamic risk assessment model, and employing adaptive multimodal fusion algorithms and reinforcement learning, a dynamic risk assessment model is constructed to achieve real-time monitoring and accurate prediction of public opinion risks.
It improves the comprehensiveness, dynamic adaptability, and accuracy of public opinion analysis, enabling real-time monitoring and prediction of public opinion risks, and enhancing the system's adaptability and user participation.
Smart Images
Figure CN122364536A_ABST
Abstract
Description
Technical Field
[0001] This application belongs to the field of computer vision and big data technology, and in particular relates to a method for dynamic assessment of public opinion information risks by integrating multimodal data. Background Technology
[0002] With the popularization of internet technology, public opinion risk has become a core challenge for enterprises and government agencies. Existing technologies mainly rely on single-modal data (such as text or images) for public opinion analysis, which has the following shortcomings: information partiality: single-modal data cannot fully reflect the complexity of public opinion events; for example, text cannot capture the implicit information in images; poor dynamic adaptability: traditional static models cannot cope with the rapid evolution of public opinion dissemination, resulting in a lag in risk assessment; low technical coupling: existing systems lack the ability to collaboratively analyze cross-modal data, making it difficult to uncover the deep correlations between multimodal data.
[0003] Therefore, it is particularly important to have a method that can dynamically analyze public opinion information risks from multimodal data. Summary of the Invention
[0004] This application provides a dynamic risk assessment method for public opinion information that integrates multimodal data. This solution achieves real-time monitoring and accurate prediction of public opinion risks by fusing multimodal data with a dynamic risk assessment model. By integrating text, image, audio, and video data, combined with an adaptive multimodal fusion algorithm and a dynamic risk assessment model, this method enables real-time monitoring and accurate prediction of public opinion risks, significantly improving the accuracy, dynamic adaptability, and processing efficiency of risk assessment.
[0005] In a first aspect, embodiments of this application provide a method for dynamic risk assessment of public opinion information that integrates multimodal data, characterized in that the method includes the following steps: S1: Acquire real-time multimodal public opinion data, which includes at least one of text, image, audio, and video; extract text sentiment features, image semantic features, and audio emotional features to generate a multimodal dataset. S2: Align the temporal and spatial information of the multimodal dataset based on timestamps and geographic labels, associate the semantic information of text, images, audio, and video in the multimodal dataset, and generate a heterogeneous associated dataset with spatiotemporal alignment identifiers; S3: Construct a feature extraction model based on the multimodal fusion algorithm, and perform weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; S4: Based on the fused feature vector, construct a dynamic risk assessment model to predict the potential propagation path and emotional evolution trend of public opinion information, and generate an emotional evolution curve; According to the formula: Risk Value = α × Speed of Transmission + β × Emotional Intensity + γ × Sensitivity, α, β, and γ are preset weighting coefficients that generate a quantitative level of public opinion risk. S5: Dynamically adjust the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically update the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; S6: Trigger a graded early warning signal based on the real-time risk value.
[0006] By adopting the above scheme, the proposed method for dynamic risk assessment of public opinion information by integrating multimodal data integrates text, image, audio, and video data, combined with an adaptive multimodal fusion algorithm and a dynamic risk assessment model. This enables real-time monitoring and accurate prediction of public opinion risks. The method employs spatiotemporal alignment and semantic association to address the fragmentation problem of multimodal data, thereby improving the comprehensiveness of public opinion analysis. Furthermore, reinforcement learning and anomaly detection are used to construct a dynamic risk assessment model based on the fused feature vectors, enhancing the robustness and adaptability of the fusion algorithm. Adjustments to the dynamic risk assessment model predict public opinion evolution, improving the foresight and accuracy of risk assessment. Finally, adaptive optimization of the dynamic risk assessment model enables user feedback-driven model iteration, enhancing the system's adaptability and user engagement.
[0007] In some embodiments of the present invention, the extraction of text sentiment features, image semantic features, and audio sentiment features further includes: Text sentiment feature extraction: A pre-trained deep learning model is used to generate text semantic vectors, and the text semantic vectors are input into a long-term memory neural network to extract sentiment polarity and intensity; Image semantic feature extraction: Extracting key objects and semantic labels from images using a deep convolutional neural network model; Audio emotion feature extraction: Audio rhythm and pitch features are extracted using the Mel frequency cepstral coefficient method, and emotion types are classified by combining the Gaussian mixture model.
[0008] In some embodiments of the present invention, the generation of the multimodal dataset further includes: Text data collection: Scraping social media comments, news reports, and forum discussions, and extracting plain text using a regular expression parser; Image data acquisition: Acquire user-uploaded images and news illustrations, and extract text information using character recognition technology; Audio data acquisition: Capture live audio and short video audio, and convert them into text using speech recognition technology; Video data acquisition: Extract key frames and dynamic images from short videos, and perform multimodal annotation by combining audio and text information.
[0009] In some embodiments of the present invention, the construction of the dynamic risk assessment model further includes: By training a graph neural network model with historical public opinion data, the topology of public opinion dissemination and the influence of nodes are generated. The autoregressive integral moving average model is used to predict the sentiment polarity value at a future point in time, and the risk evolution probability is obtained by combining the current risk value.
[0010] In some embodiments of the present invention, the periodic updating of the feature extraction model and the risk assessment model includes: collecting user feedback data in real time through social media interfaces, news crawlers, and industry forum monitoring; combining the user feedback data with historical data, and using an incremental learning strategy to update the feature extraction model and the risk assessment model.
[0011] In some embodiments of the present invention, the method further includes: restricting data access permissions through API interfaces to ensure the security and privacy of real-time collected multimodal public opinion data; and desensitizing sensitive information before generating the heterogeneous correlated dataset, including at least one of face blurring and geolocation encryption.
[0012] By adopting the above-mentioned solution, this solution integrates text, image, audio and video data, and combines an adaptive multimodal fusion algorithm with a dynamic risk assessment model to achieve real-time monitoring and accurate prediction of public opinion information risks. It is applicable to fields such as corporate brand management, government public safety and social media governance.
[0013] Secondly, this application provides a dynamic risk assessment system for public opinion information that integrates multimodal data, the system comprising: Data acquisition unit: acquires real-time collected multimodal public opinion data, which includes at least one of text, image, audio, and video, extracts text sentiment features, image semantic features, and audio emotional features, and generates a multimodal dataset; Data fusion unit: Aligns the temporal and spatial information of the multimodal dataset based on timestamps and geographic tags, associates the semantic information of text, images, audio, and video in the multimodal dataset, and generates a heterogeneous associated dataset with spatiotemporal alignment identifiers; constructs a feature extraction model based on the multimodal fusion algorithm, and performs weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; Risk assessment unit: Based on the fused feature vector, a dynamic risk assessment model is constructed to predict the potential propagation path and emotional evolution trend of public opinion, and to generate an emotional evolution curve; According to the formula: Risk Value = α × Speed of Transmission + β × Emotional Intensity + γ × Sensitivity, α, β, and γ are preset weighting coefficients that generate a quantitative level of public opinion risk. Feedback adjustment unit: dynamically adjusts the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically updates the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; Early warning unit: Triggers graded early warning signals based on the real-time risk value.
[0014] In some embodiments of the present invention, the data acquisition unit further includes: a text analysis engine: generating text semantic vectors using a pre-trained deep learning model, and inputting the text semantic vectors into a long-term memory neural network to extract emotional polarity and intensity; an image analysis engine: extracting key objects and semantic labels from images using a deep convolutional neural network model; and an audio analysis engine: extracting audio rhythm and pitch features using the Mel frequency cepstral coefficient method, and classifying emotion types using a Gaussian mixture model.
[0015] In some embodiments of the present invention, the risk assessment unit further includes: training a graph neural network model using historical public opinion data to generate the topology and node influence of public opinion dissemination; predicting the sentiment polarity value at a future moment based on an autoregressive integral moving average model, and obtaining the risk evolution probability by combining the current risk value. In some embodiments of the present invention, the system further includes: restricting data access permissions through an API interface to ensure the security and privacy of real-time collected multimodal public opinion data; and desensitizing sensitive information before generating the heterogeneous correlated dataset, including at least one of face blurring and geolocation encryption.
[0016] By adopting the above-mentioned approach, this solution provides a dynamic risk assessment system for public opinion information that integrates multimodal data. It achieves the fusion of multimodal data by integrating text, image, audio, and video data, thus overcoming the limitations of single-modal analysis. It predicts the path of public opinion dissemination and the trend of emotional evolution by building a dynamic model based on real-time data, thereby achieving dynamic risk assessment. Furthermore, it improves the accuracy of assessment by dynamically adjusting model parameters through reinforcement learning and incremental learning techniques, while also achieving adaptive optimization of the assessment model.
[0017] Additional advantages, objects, and features of the invention will be set forth in part in the description which follows, and will also become apparent in part to those skilled in the art upon studying the text, or may be learned by practice of the invention. The objects and other advantages of the invention will become apparent from the description and the accompanying drawings.
[0018] Those skilled in the art will understand that the objectives and advantages achievable with the present invention are not limited to those specifically described above, and that the above and other objectives achievable with the present invention will become clearer from the following detailed description. Attached Figure Description
[0019] The accompanying drawings are provided to further understand the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof.
[0020] In the attached diagram: Figure 1 This is a schematic diagram of one implementation of the method for dynamic risk assessment of public opinion information that integrates multimodal data; Figure 2 This is a schematic diagram of a dynamic risk assessment method system for public opinion information that integrates multimodal data, provided in an embodiment of this application.
[0021] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation
[0022] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.
[0023] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.
[0024] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the term "comprising" or any other variations thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0025] Generally speaking, existing public opinion risk assessment mainly relies on single-modal data (such as text or images) for public opinion analysis, lacking the ability to adapt to multimodal data and dynamically adjust, and cannot fully meet the needs of multimodal public opinion risk prediction and rapid response.
[0026] Therefore, in order to achieve real-time updates of job data and provide highly personalized recommendation data, this application provides a method for dynamic assessment of public opinion information risk by integrating multimodal data.
[0027] Figure 1 This is a schematic diagram of a dynamic risk assessment method for public opinion information that integrates multimodal data, provided in an embodiment of this application.
[0028] Firstly, such as Figure 1 As shown in the figure, this application provides a method for dynamic assessment of public opinion information risk by integrating multimodal data. The method includes the following steps: S1: Acquire real-time multimodal public opinion data, which includes at least one of text, image, audio, and video. Perform optical character recognition to extract text information from the collected image and video data, and perform speech recognition to extract text information from the collected audio data. Generate a standardized multimodal dataset based on the text. S2: Align the temporal and spatial information of the standardized multimodal dataset based on timestamps and geographic labels, and associate the semantic information of text, images, audio, and video in the multimodal dataset to generate an associated dataset; S3: Construct a feature extraction model based on the multimodal fusion algorithm, and perform weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; Specifically, a gated recurrent unit (GRU) is used to perform weighted fusion of multimodal data features in the associated dataset; By dynamically adjusting the fusion weights of each modality feature through reinforcement learning (Q-learning), and filtering noisy data by combining the isolated forest algorithm, a fused feature vector is output. S4: Based on the fused feature vector, construct a dynamic risk assessment model to predict the potential propagation path and emotional evolution trend of public opinion information, and generate an emotional evolution curve; Specifically, the study uses graph neural networks (GNNs) to predict the propagation path and key nodes of potential public opinion; and combines ARIMA time series models to analyze the trend of sentiment evolution. According to the formula: Risk Value = α × Speed of Transmission + β × Emotional Intensity + γ × Sensitivity, α, β, and γ are preset weighting coefficients that generate a quantified level of public opinion risk. S5: Dynamically adjust the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically update the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; this step realizes real-time feedback and model optimization, dynamically adjusts the fusion weight and risk assessment model parameters through the Q-learning algorithm, and periodically updates the feature extraction model and risk assessment model based on the incremental learning strategy to adapt to changes in the data environment; S6: Trigger a graded early warning signal based on the real-time risk value; The system triggers tiered early warning signals (low risk, medium risk, and high risk) based on real-time risk values; through the "1+11+N" monitoring and early warning command system, the early warning information is distributed to five levels of response units: city, district, street, community, and grid, and the emergency response process is initiated.
[0029] Using the above scheme, the proposed method for dynamic risk assessment of public opinion information by integrating multimodal data integrates text, image, audio, and video data, combined with an adaptive multimodal fusion algorithm and a dynamic risk assessment model. This enables real-time monitoring and accurate prediction of public opinion information risks. The method employs spatiotemporal alignment and semantic association to address the fragmentation problem of multimodal data, improving the comprehensiveness of public opinion analysis. Reinforcement learning and anomaly detection are used to construct a dynamic risk assessment model based on the fused feature vectors, enhancing the robustness and adaptability of the fusion algorithm. Adjustments to the dynamic risk assessment model predict public opinion evolution, improving the foresight and accuracy of risk assessment. Adaptive optimization of the dynamic risk assessment model enables user feedback-driven model iteration, enhancing the system's adaptability and user engagement. In some embodiments of this invention, the extraction of text sentiment features, image semantic features, and audio emotion features further includes: Text sentiment feature extraction: A pre-trained deep learning model is used to generate text semantic vectors, and the text semantic vectors are input into a long-term memory neural network to extract sentiment polarity and intensity; specifically, the BERT model is used to generate text semantic vectors, and combined with an LSTM network to extract sentiment polarity (positive, negative, neutral) and intensity (low, medium, high). Image semantic feature extraction: Utilizing deep convolutional neural network models to extract key objects and semantic labels from images; for example, using the ResNet-50 model to extract key objects (such as faces and scenes) and semantic labels (such as "protest" and "accident"); Audio emotion feature extraction: Audio rhythm and pitch features are extracted using the Mel frequency cepstral coefficient (MFCC) method, and combined with Gaussian mixture model (GMM) to classify emotion types (anger, anxiety, calm).
[0030] In some embodiments of the present invention, the generation of the multimodal dataset further includes: text data acquisition: capturing social media comments, news reports, and forum discussions, and extracting plain text using a regular expression parser; image data acquisition: acquiring user-uploaded images and news illustrations, and extracting text information using character recognition (e.g., OCR) technology; audio data acquisition: capturing live audio and short video audio, and converting them into text using speech recognition technology; and video data acquisition: capturing keyframes and dynamic scenes from short videos, and combining audio and text information for multimodal annotation.
[0031] In some embodiments of the present invention, the construction of the dynamic risk assessment model further includes: training a graph neural network (e.g., GNN) model using historical public opinion data to generate the topology and node influence of public opinion dissemination; predicting the sentiment polarity value at a future moment based on an autoregressive integral moving average (e.g., ARIMA) model, and obtaining the risk evolution probability by combining the current risk value.
[0032] In some embodiments of the present invention, the periodic updating of the feature extraction model and the risk assessment model includes: collecting user feedback data in real time through social media interfaces, news crawlers, and industry forum monitoring; combining the user feedback data with historical data, and using an incremental learning strategy to update the feature extraction model and the risk assessment model.
[0033] The above solution employs an adaptive multimodal fusion algorithm, which adaptively optimizes the risk assessment model based on user feedback and historical data, and combines BERT, ResNet-50, and GMM to enhance the granularity and accuracy of public opinion analysis.
[0034] In some embodiments of the present invention, the method further includes: restricting data access permissions through API interfaces to ensure the security and privacy of real-time collected multimodal public opinion data; and desensitizing sensitive information before generating the heterogeneous correlated dataset, including at least one of face blurring and geolocation encryption.
[0035] The above solution addresses the risk of privacy breaches in public opinion analysis through data anonymization and access control.
[0036] The following is a description of two specific application scenarios for a dynamic risk assessment method for public opinion information based on the fusion of multimodal data and the actual needs of multimodal public opinion risk assessment: Scenario 1: Monitoring Public Opinion Regarding Government Public Policies Background: When implementing a new environmental protection policy, a municipal government needs to monitor public reactions in real time to prevent the spread of negative public opinion.
[0037] The method for dynamic risk assessment of public opinion information using multimodal data integration involves the following steps: 1) Multimodal data acquisition: Capture text comments, images, and video content from social media via API interfaces; We crawl policy discussion posts from news websites and forums and use OCR technology to extract key information from the text and images.
[0038] 2) Sentiment Analysis and Risk Assessment: The BERT+LSTM model was used to analyze the sentiment tendency of the text (positive / negative / neutral), and the score was quantified by combining the NRC sentiment dictionary. By using ResNet-50 to identify text or scenes in images and combining them with a scene classification model, the level of public opinion tension can be determined. We use GNN to model the public opinion dissemination network and predict the diffusion path and potential risk nodes of policy-related topics.
[0039] 3) Dynamic risk warning: The risk level of policy public opinion is calculated in real time based on the formula: Risk Value = α × Spread Speed + β × Emotional Intensity + γ × Sensitivity (α = 0.4, β = 0.3, γ = 0.3). When the risk value exceeds the threshold, a graded early warning signal is triggered (low risk → medium risk → high risk), and the information is distributed to the five-level response units such as district and street through the "1+11+N" system.
[0040] 4) Early warning and emergency response: Based on the early warning information, government departments quickly organized experts to interpret the policies and clarified misunderstandings through press conferences; Incremental learning strategies are used to update sentiment analysis models to adapt to new public discussions on policies.
[0041] The expected effects of the above plan are: to identify negative public opinion triggered by the policy three days in advance, to avoid large-scale protests, and to improve public understanding of the policy.
[0042] Scenario 2: Corporate Brand Reputation Management Background: After a new product launch, an electronics company discovered that some users were spreading negative videos (such as footage of product malfunctions) on social media.
[0043] The method for dynamic risk assessment of public opinion information using multimodal data integration involves the following steps: 1) Multimodal data fusion: Collect user-uploaded video content, analyze audio emotions (anger / disappointment) using the MFCC+GMM model, and calculate the intensity of emotions by combining voice intonation; By using the Transformer architecture to associate fault scenes in videos with textual comments (such as "battery explodes"), cross-modal semantic representations can be generated.
[0044] 2) Adaptive feature extraction: The text, image, and audio features are weighted and fused using a GRU model, and the weights are dynamically adjusted using Q-learning (e.g., the audio emotion weight is increased from 0.2 to 0.4). The isolated forest algorithm is used to filter out malicious spam and fake comments, while retaining genuine user feedback.
[0045] 3) Risk propagation prediction: By using a GNN model to predict the propagation path of negative videos, it was discovered that a top KOL might repost the content, triggering a high-risk warning. Using the ARIMA model to analyze the trend of sentiment evolution, it is predicted that negative public opinion may spread to 1 million users in the next 24 hours.
[0046] 4) Real-time feedback and optimization: Enterprises collect user feedback through social media interfaces and update the BERT model to identify newly emerging negative keywords (such as "security risks"). By using reinforcement learning to adjust the parameters of the risk assessment model, the response speed to similar events can be improved.
[0047] The expected effect of the above plan is that if the company issues a recall statement and provides a compensation plan within 2 hours, the scope of negative public opinion will be reduced by 60% and the loss of brand reputation will be reduced by 80%.
[0048] In its implementation, this solution provides a dynamic risk assessment method for public opinion information that integrates multimodal data. By integrating text, image, audio, and video data, it achieves the fusion of multimodal data and overcomes the limitations of single-modal analysis. By building a dynamic model based on real-time data, it predicts the path of public opinion dissemination and the trend of emotional evolution, thereby achieving dynamic risk assessment. Through reinforcement learning and incremental learning techniques, it dynamically adjusts model parameters to improve assessment accuracy and simultaneously achieves adaptive optimization of the assessment model.
[0049] Secondly, such as Figure 2 As shown in the figure, this application provides a dynamic risk assessment system for public opinion information that integrates multimodal data. The system includes: Data acquisition unit S11: Acquires real-time collected multimodal public opinion data, which includes at least one of text, image, audio, and video, extracts text sentiment features, image semantic features, and audio emotional features, and generates a multimodal dataset; Data fusion unit S12: Aligns the temporal and spatial information of the multimodal dataset based on timestamps and geographic labels, associates the semantic information of text, images, audio, and video in the multimodal dataset, and generates a heterogeneous associated dataset with spatiotemporal alignment identifiers; constructs a feature extraction model according to the multimodal fusion algorithm, and performs weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; Risk assessment unit S13: Based on the fused feature vector, construct a dynamic risk assessment model to predict the potential propagation path and emotional evolution trend of public opinion, and generate an emotional evolution curve; Based on the formula: Risk Value = α × Spread Speed + β × Emotional Intensity + γ × Sensitivity (where α, β, and γ are preset weighting coefficients), a quantitative public opinion risk level is generated; Feedback adjustment unit S14: dynamically adjusts the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically updates the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; Early warning unit S15: Triggers a graded early warning signal based on the real-time risk value.
[0050] In some embodiments of the present invention, the data acquisition unit further includes: Text analysis engine: It uses a pre-trained deep learning model to generate text semantic vectors, and inputs the text semantic vectors into a long-term memory neural network to extract sentiment polarity and intensity; it also extracts sentiment features based on the BERT+LSTM model. Image analysis engine: Utilizes a deep convolutional neural network model to extract key objects and semantic labels from images; employs a ResNet-50 model to extract semantic labels; Audio analysis engine: Extracts audio rhythm and pitch features using the Mel frequency cepstral coefficient method, combines Gaussian mixture model to classify emotion types, and uses MFCC+GMM model to classify emotion types.
[0051] In some embodiments of the present invention, the risk assessment unit further includes: training a graph neural network (e.g., GNN) model using historical public opinion data to generate the topology and node influence of public opinion dissemination; predicting the sentiment polarity value at a future moment based on an autoregressive integral moving average (e.g., ARIMA) model, and obtaining the risk evolution probability by combining the current risk value.
[0052] In some embodiments of the present invention, the feedback adjustment unit further includes: collecting user feedback data in real time through social media interfaces, news crawlers, and industry forum monitoring; combining user feedback data with historical data, and updating the feature extraction model and risk assessment model using an incremental learning strategy. Specifically, it includes: 1) a reinforcement learning module: performing weighted fusion of text, image, and audio features in the data through a gated recurrent unit (GRU); optimizing the AMF fusion weights through a Q-learning algorithm; 2) an anomaly detection module: filtering noisy data using an isolated forest algorithm and outputting a high-confidence fusion feature vector; 3) a model update module: periodically updating the feature extraction and risk assessment models based on an incremental learning strategy.
[0053] In some embodiments of the present invention, the system further includes: an early warning and emergency response unit that triggers graded early warning signals (low risk, medium risk, high risk) based on real-time risk values; and distributes early warning information to five-level response units (city, district, street, community, grid) through a “1+11+N” monitoring and early warning command system to initiate emergency response procedures.
[0054] In some embodiments of the present invention, the system further includes: restricting data access permissions through API interfaces to ensure the security and privacy of real-time collected multimodal public opinion data; and desensitizing sensitive information before generating the heterogeneous correlated dataset, including at least one of face blurring and geolocation encryption.
[0055] The above-mentioned solution has achieved technological breakthroughs in multimodal data fusion, dynamic risk assessment, real-time feedback optimization, adaptive system design and application scenario expansion. It solves the problems of information partiality, poor dynamic adaptability and low technology coupling of existing technologies, and has significant technological progress and application value. It is applicable to fields such as enterprise brand management, government public safety and social media governance.
[0056] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.
[0057] like Figure 3 As shown, this application provides an electronic device, which includes a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the above-mentioned method for dynamic risk assessment of public opinion information that integrates multimodal data.
[0058] The electronic device may include a processor 1201 and a memory 1202 storing computer program instructions.
[0059] Specifically, the processor 1201 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.
[0060] Memory 1202 may include mass storage for data or instructions. For example, and not limitingly, memory 1202 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 1202 may include removable or non-removable (or fixed) media. Where appropriate, memory 1202 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 1202 is non-volatile solid-state memory.
[0061] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.
[0062] The processor 1201 reads and executes computer program instructions stored in the memory 1202 to implement any of the battery thermal runaway parameter determination methods in the above embodiments.
[0063] In one example, the electronic device may also include a communication interface 1203 and a bus 1210. For example, Figure 3 As shown, the processor 1201, memory 1202, and communication interface 1203 are connected through bus 1210 and complete communication with each other.
[0064] The communication interface 1203 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.
[0065] Bus 1210 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 1210 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.
[0066] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.
[0067] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.
[0068] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.
[0069] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.
[0070] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.
Claims
1. A method for dynamic assessment of public opinion information risk by integrating multimodal data, characterized in that, The method includes the following steps: S1: Acquire real-time multimodal public opinion data, which includes at least one of text, image, audio, and video; extract text sentiment features, image semantic features, and audio emotional features to generate a multimodal dataset. S2: Align the temporal and spatial information of the multimodal dataset based on timestamps and geographic labels, associate the semantic information of text, images, audio, and video in the multimodal dataset, and generate a heterogeneous associated dataset with spatiotemporal alignment identifiers; S3: Construct a feature extraction model based on the multimodal fusion algorithm, and perform weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; S4: Based on the fused feature vector, construct a dynamic risk assessment model to predict the potential propagation path and emotional evolution trend of public opinion information, and generate an emotional evolution curve; According to the formula: Risk Value = α × Speed of Transmission + β × Emotional Intensity + γ × Sensitivity, α, β, and γ are preset weighting coefficients that generate a quantified level of public opinion risk. S5: Dynamically adjust the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically update the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; S6: Trigger a graded early warning signal based on the real-time risk value.
2. The method for dynamic risk assessment of public opinion information by integrating multimodal data according to claim 1, characterized in that, The extraction of text sentiment features, image semantic features, and audio emotion features also includes: Text sentiment feature extraction: A pre-trained deep learning model is used to generate text semantic vectors, and the text semantic vectors are input into a long-term memory neural network to extract sentiment polarity and intensity; Image semantic feature extraction: Extracting key objects and semantic labels from images using a deep convolutional neural network model; Audio emotion feature extraction: Audio rhythm and pitch features are extracted using the Mel frequency cepstral coefficient method, and emotion types are classified by combining the Gaussian mixture model.
3. The method for dynamic risk assessment of public opinion information by integrating multimodal data according to claim 1, characterized in that, The generation of the multimodal dataset also includes: Text data collection: Scraping social media comments, news reports, and forum discussions, and extracting plain text using a regular expression parser; Image data acquisition: Acquire user-uploaded images and news illustrations, and extract text information using character recognition technology; Audio data acquisition: Capture live audio and short video audio, and convert them into text using speech recognition technology; Video data acquisition: Extract key frames and dynamic images from short videos, and perform multimodal annotation by combining audio and text information.
4. The method for dynamic risk assessment of public opinion information by integrating multimodal data according to claim 1, characterized in that, The construction of the dynamic risk assessment model also includes: By training a graph neural network model with historical public opinion data, the topology of public opinion dissemination and the influence of nodes are generated. The autoregressive integral moving average model is used to predict the sentiment polarity value at a future point in time, and the risk evolution probability is obtained by combining the current risk value.
5. The method for dynamic risk assessment of public opinion information by integrating multimodal data according to claim 2, characterized in that, The periodically updated feature extraction model and risk assessment model include: User feedback data is collected in real time through social media interfaces, news crawlers, and industry forum monitoring. By combining user feedback data with historical data, an incremental learning strategy is used to update the feature extraction model and the risk assessment model.
6. The method for dynamic risk assessment of public opinion information by integrating multimodal data according to claim 1, characterized in that, The method further includes: By restricting data access permissions through API interfaces, the security and privacy of real-time collected multimodal public opinion data are ensured. Before generating the heterogeneous associated dataset, sensitive information is desensitized, including at least one of face blurring and geolocation encryption.
7. A dynamic risk assessment system for public opinion information that integrates multimodal data, characterized in that, The system also includes: Data acquisition unit: acquires real-time collected multimodal public opinion data, which includes at least one of text, image, audio, and video, extracts text sentiment features, image semantic features, and audio emotional features, and generates a multimodal dataset; Data fusion unit: Aligns the temporal and spatial information of the multimodal dataset based on timestamps and geographic tags, associates the semantic information of text, images, audio, and video in the multimodal dataset, and generates a heterogeneous associated dataset with spatiotemporal alignment identifiers; constructs a feature extraction model based on the multimodal fusion algorithm, and performs weighted fusion of the multimodal data features in the heterogeneous associated dataset based on the feature extraction model to generate a fused feature vector; Risk assessment unit: Based on the fused feature vector, a dynamic risk assessment model is constructed to predict the potential propagation path and emotional evolution trend of public opinion, and to generate an emotional evolution curve; According to the formula: Risk Value = α × Speed of Transmission + β × Emotional Intensity + γ × Sensitivity, α, β, and γ are preset weighting coefficients that generate a quantitative level of public opinion risk. Feedback adjustment unit: dynamically adjusts the fusion weight parameters and risk assessment model parameters in the multimodal fusion based on historical data and real-time monitoring data, and periodically updates the feature extraction model and risk assessment model based on the fusion weight parameters and risk assessment model parameters; Early warning unit: Triggers graded early warning signals based on the real-time risk value.
8. The dynamic risk assessment system for public opinion information that integrates multimodal data according to claim 7, characterized in that, The data acquisition unit further includes: Text analysis engine: It uses a pre-trained deep learning model to generate text semantic vectors, and inputs the text semantic vectors into a long-term memory neural network to extract sentiment polarity and intensity; Image analysis engine: Utilizes deep convolutional neural network models to extract key objects and semantic labels from images; Audio analysis engine: Extracts audio rhythm and pitch features using the Mel frequency cepstral coefficient method, and classifies emotion types using a Gaussian mixture model.
9. The dynamic risk assessment system for public opinion information that integrates multimodal data according to claim 7, characterized in that, The risk assessment unit also includes: By training a graph neural network model with historical public opinion data, the topology of public opinion dissemination and the influence of nodes are generated. The autoregressive integral moving average model is used to predict the sentiment polarity value at a future point in time, and the risk evolution probability is obtained by combining the current risk value.
10. The dynamic risk assessment system for public opinion information that integrates multimodal data according to claim 7, characterized in that, The system also includes: By restricting data access permissions through API interfaces, the security and privacy of real-time collected multimodal public opinion data can be ensured. Before generating the heterogeneous associated dataset, sensitive information is desensitized, including at least one of face blurring and geolocation encryption.