Public health real information evaluation and early warning system based on AI technology

Abstract

The invention discloses a public health real information evaluation and early warning system based on an AI technology, and relates to the field of artificial intelligence technology application, and the system comprises a database module which is used for storing information collected by a collection module and providing powerful support for subsequent model construction and real-time response, and a data collection module which is used for collecting real-time information of public health. The invention discloses a public health data collection and processing method, which is used for collecting data related to public health from various sources (such as news, social media, professional forums and the like), including multi-modal data such as texts, pictures, videos and the like, and ensuring that the data sources cover data sources of a knowledge graph and data sources of a false information base. The information processing efficiency is greatly improved through a knowledge graph and advanced model construction, the system has strong adaptive capacity for novel false information, the system can rapidly learn and recognize a new false information mode through fusion of the knowledge graph and a large model, and the risk caused by the false information is effectively reduced.

Description

Technical Field

[0001] The present invention relates to the application field of artificial intelligence technology, and specifically to a public health real information evaluation and early warning system based on AI technology. Background Art

[0002] The public health authenticity information evaluation and early warning system based on AI technology is mainly used for the authenticity evaluation of public health information, especially in identifying and verifying the authenticity of health-related information circulated on social media and the Internet. By integrating advanced deep learning models, large-scale language model tuning (such as LoRA fine-tuning technology) and knowledge graph technology, the present invention aims to improve the accuracy and efficiency of false health information detection. In addition, this technology also includes the development of application systems for information authenticity evaluation, such as mobile applications and network services. These systems can be used by the public and professionals to promote the dissemination of correct health information and rumor-busting activities. Through this technology, we can better manage and reduce the public health risks brought by false information, especially during global health crises;

[0003] Most of the existing technologies for evaluating the authenticity of public health information can be divided into four categories. The first is a rule-based text analysis method that relies on predefined keywords and grammatical rules to identify false information. However, this method has poor adaptability to new or changing fraud strategies and is easily disturbed by complex semantic expressions. The second method uses traditional machine learning algorithms, such as support vector machines (SVMs) or random forests, to classify text content. These models are usually based on manually extracted features, such as word frequency statistics of TF-IDF weights. Although such methods can automatically process information to a certain extent, their performance depends greatly on the quality of feature selection and has limited processing capabilities for multimodal data. The third method is deep learning technology. Recent technological advances include the use of deep neural networks, especially convolutional neural networks (CNNs). ) and long short-term memory networks (LSTM) to identify false information. These models can automatically learn features from data without complex feature engineering. However, they usually require a large amount of labeled data for training and may still need further optimization when dealing with cross-modal content. The fourth verification method supported by knowledge graphs. In this method, the system uses knowledge graphs to compare information with facts to verify the authenticity of the statement. However, its limitation is that it relies on known and static knowledge bases, and it is difficult to update or process newly emerging false information in real time. The above four methods have their own advantages, but they are generally slow in processing speed, weak in adapting to new types of false information, insufficient in processing multimodal data, and not fast enough in responding to dynamic updates of emerging health information. To this end, we propose a public health real information evaluation and early warning system based on AI technology. Summary of the invention

[0004] The purpose of the present invention is to provide a public health real information evaluation and early warning system based on AI technology.

[0005] In order to solve the problems raised in the above background technology, the present invention provides the following technical solutions: a public health real information evaluation and early warning system based on AI technology, comprising:

[0006] The database module is used to store the information collected by the acquisition module and provide strong support for subsequent model construction and real-time response;

[0007] The data collection module is used to collect public health-related data from a variety of sources (such as news, social media, professional forums, etc.), including multimodal data such as text, pictures, and videos, ensuring that the data source covers the data source of the knowledge graph and the data source of the false information library;

[0008] The model building module introduces the knowledge graph, integrates it with the feature representation of the large language model, and defines deep feature conversion functions and fusion functions. It combines the knowledge graph and the representation of the large model to improve the semantic understanding and association analysis capabilities of the model in specific fields (such as medicine and epidemiology). At the same time, it defines interaction modeling functions and semantic relationship modeling functions to further enhance the semantic understanding capabilities of the model in specific fields.

[0009] The performance evaluation module establishes a system performance evaluation index system, including accuracy, timeliness, and specificity, and uses specific evaluation functions to comprehensively evaluate the performance of the large model in false information detection;

[0010] The optimization module uses the display capabilities of large modules and composite models to conduct empirical analysis, evaluate the impact of false information in crisis cases, and extract the optimal crisis response strategy by solving optimization problems;

[0011] The real-time monitoring response module can promptly detect false information in public health incidents and issue early warnings to relevant departments and the public, effectively curbing the spread of false information and ensuring public health safety.

[0012] As a further solution of the present invention: the data acquisition module needs to clean and organize the collected data, remove duplicate, invalid or personal privacy information entries, and perform word segmentation, stop word removal, part-of-speech tagging and other processing for text data; for image and video data, necessary format conversion and compression are performed, and natural language processing technology is used to extract word frequency, TF-IDF value, part of speech and naming from the text data. At the same time, the word embedding technology Word2Vec is used to convert the text into a vector representation for input into the model. For image data, a convolutional neural network (CNN) is used to extract the visual features of the image, key frames can be extracted and processed using CNN, or motion, color and other features can be extracted using a video processing library. The extracted text, image and video features are fused to form a multimodal feature representation, which can be achieved through splicing and weighted summation, so that the model can simultaneously use information from multiple modes to identify false information. At the same time, data cleaning needs to remove missing values ​​and outliers, and label each data sample. The specific steps are as follows:

[0013] L={l1,l2,l3...,l n}

[0014] Among them, L represents the label set (real news, manually written news, and fake news generated by GPT), l1 represents the data sample d i Corresponding standards, the data classification process can be expressed by a classification function, as follows:

[0015] C(d1)={C1,C2,C3...,C k}

[0016] D fimal ={(F i ,l n )|d1∈Dcleaned}

[0017] Where C(d1) represents the classification result of data sample d1, c k represents the k class, D fimal ={(F i ,l n |d1∈Dcleaned} formula represents the final data set prepared for model training and testing, where D fimal Represents the final data set, (F i ,l n ) represents the pair of features of the data sample and the corresponding label, d1∈Dcleaned means that the data sample comes from the cleaned data set D fimal .

[0018] As a further solution of the present invention: the database module includes a real information storage module and a false information storage module, and the real information storage module and the false information storage module are used to store real information and false information respectively.

[0019] As a further solution of the present invention: the model building module constructs a false information identification function by mapping into the deep learning algorithm: PI (x, θ), where x is the input data and θ is the model parameter. This function converts x into a feature vector f(x) = F by applying a high-level semantic image. PI (x, θ), the model building module also introduces a false information loss function with the same specifications to guide the model to more accurately identify false information. The false information loss function L DI (f(x), y), where y is the label, this function is used to quantify the gap between the model prediction, f(x) and the true label y.

[0020] As a further solution of the present invention: the model building module optimizes the objective to minimize the false information loss function L DI (f(x),y) and the accuracy of the maximum true information Acc((f(x)), and introduce the sensitivity index Stype(F X ,T) to focus on specific types of false information, where T represents specific false information, such as disease transmission false information and vaccine false information.

[0021] As a further solution of the present invention: the performance evaluation module integrates false information source identification and propagation dynamics prediction, as well as the epidemic prediction model F EP , construct the composite model M CI (F SI , F SI , F SI ).

[0022] As a further solution of the present invention: the optimization module analyzes the comprehensive impact of false information dissemination on public health and epidemic development through a composite model. CI ,The specific conversion formula is as follows:

[0023] I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ)

[0024] Where y is epidemiological data, I CI For the combined impact.

[0025] As a further solution of the present invention: the real-time monitoring response module monitors public health information in real time and uses a constructed detection algorithm, namely R (Detection (D, A)) to detect false information during real-time monitoring. When the response mechanism R detects false information, the response mechanism is immediately executed to push warning information to relevant departments and the public.

[0026] By adopting the above technical solution, compared with the prior art, the beneficial effects of the present invention are:

[0027] 1. The present invention greatly improves the efficiency of information processing through optimized data collection and processing processes, as well as advanced model construction. It can quickly analyze and identify multimodal data, providing a strong guarantee for timely response to public health events. Secondly, the system has strong adaptability to new types of false information. Through the fusion of knowledge graphs and large models, as well as continuously optimized model parameters, the system can quickly learn and identify newly emerging false information patterns, effectively reducing the risks brought by false information. Furthermore, in multimodal data processing, the system can fully integrate various data forms such as text, pictures, and videos, comprehensively improving the accuracy and reliability of false information detection.

[0028] 2. The present invention uses a real-time monitoring and response module to dynamically update and respond to emerging health information. The real-time monitoring and response module ensures that the system can capture new information in a timely manner, respond quickly, and issue warnings to relevant departments and the public, thereby effectively maintaining public health safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 A schematic diagram of a system flow in an embodiment of the present invention;

[0030] Figure 2 Schematic diagram of data processing flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] The specific embodiments of the present invention are further described below in conjunction with the accompanying drawings. It should be noted that the description of these embodiments is used to help understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0032] Example 1, please refer to the attached Figure 1 - Attachment Figure 2 The present invention provides a technical solution: a public health real information evaluation and early warning system based on AI technology, comprising:

[0033] The database module is used to store the information collected by the acquisition module and provide strong support for subsequent model construction and real-time response;

[0034] The data collection module is used to collect public health-related data from a variety of sources such as news, social media, and professional forums, including multimodal data such as text, images, and videos, ensuring that the data source covers the data source of the knowledge graph and the data source of the false information library;

[0035] The model building module introduces knowledge graphs, integrates them with the feature representation of large language models, and defines deep feature conversion functions and fusion functions. By combining the representation of knowledge graphs and large models, the model's semantic understanding and association analysis capabilities in specific fields such as medicine and epidemiology are improved. At the same time, interaction modeling functions and semantic relationship modeling functions are defined to further enhance the model's semantic understanding capabilities in specific fields.

[0036] The performance evaluation module establishes a system performance evaluation index system, including accuracy, timeliness, and specificity, and uses specific evaluation functions to comprehensively evaluate the performance of the large model in false information detection;

[0037] The optimization module uses the display capabilities of large modules and composite models to conduct empirical analysis, evaluate the impact of false information in crisis cases, and extract the optimal crisis response strategy by solving optimization problems;

[0038] The real-time monitoring response module can promptly detect false information in public health incidents and issue early warnings to relevant departments and the public, effectively curbing the spread of false information and ensuring public health safety.

[0039] Please refer to the attached Figure 2 , the data acquisition module needs to clean and organize the collected data, remove duplicate, invalid or personal privacy information entries, and perform word segmentation, stop word removal, part-of-speech tagging and other processing for text data; for image and video data, necessary format conversion and compression are performed, and natural language processing technology is used to extract word frequency, TF-IDF value, part of speech and naming from text data. At the same time, the word embedding technology Word2Vec is used to convert text into vector representation for input into the model. For image data, the convolutional neural network CNN is used to extract the visual features of the image. Key frames can be extracted and processed using CNN, or motion, color and other features can be extracted using the video processing library. The extracted text, image and video features are fused to form a multi-modal feature representation, which can be achieved through splicing and weighted summation, so that the model can use information from multiple modalities to identify false information at the same time. At the same time, data cleaning needs to remove missing values ​​and outliers, and label each data sample. The specific steps are as follows:

[0040] L={l1,l2,l3...,l n}

[0041] Among them, L represents the label set of real news, manually written news and fake news generated by GPT, l1 represents the data sample d i Corresponding standards, the data classification process can be expressed by a classification function, as follows:

[0042] C(d1)={C1,C2,C3...,C k}

[0043] D fimal ={(F i ,l n )|d1∈Dcleaned}

[0044] Where C(d1) represents the classification result of data sample d1, c k represents the k class, D fimal ={(F i ,l n |d1∈Dcleaned} formula represents the final data set prepared for model training and testing, where D fimal Represents the final data set, (F i ,l n ) represents the pair of features of the data sample and the corresponding label, d1∈Dcleaned means that the data sample comes from the cleaned data set D fimal The database module includes a real information storage module and a false information storage module, which are used to store real information and false information respectively. The model building module constructs a false information identification function by mapping the deep learning algorithm: F PI (x, θ), where x is the input data and θ is the model parameter, this function converts x into a feature vector by applying a high-level semantic image fx) = F PI (x,θ), the model building module also introduces the false information loss function with the same specifications to guide the model to more accurately identify false information. The false information loss function L DI (f(x), y), where y is the label. This function is used to quantify the gap between the model prediction, f(x) and the true label y. The model building module optimizes the objective to minimize the false information loss function L DI (f(x),y) and the accuracy of the maximum true information Acc((f(x)), and introduce the sensitivity index Stype(F X ,T) to focus on specific types of false information, where T represents specific false information, such as disease transmission false information and vaccine false information;

[0045] In this embodiment, the data collection module collects raw data through social media web crawlers including Twitter crawlers and Weibo crawlers and fact-checking web crawlers including Snopes website crawlers. The raw data is annotated after text data processing and image and video data processing, and transmitted to the database module for subsequent authenticity identification of public health information. The data collection module, model construction module, performance evaluation module and optimization module divide the model learning method into two types, one is supervised learning and the other is unsupervised learning. Unsupervised learning is based on the evaluation specification construction of the false health information system after the completion of supervised learning. The specific construction method is as follows;

[0046] Define the false information source identification function: F SI (X, Θ), where x is the input data and Θ is the model parameter, so as to identify the origin point X of false information origin =F SI (X, Θ), and define the propagation dynamics prediction model F PD (X,Φ), this model is used to predict the propagation path and range of false information in the network, specifically expressed as P spread =F PD (X,Φ), a composite model is constructed by integrating the above two models and the epidemiological prediction model. After the supervised learning is completed, the system enters the unsupervised learning operation and establishes real-time monitoring to monitor public health information.

[0047] During use, the efficiency of information processing is greatly improved through optimized data collection and processing processes, as well as advanced model construction. It can quickly analyze and identify multimodal data, providing a strong guarantee for timely response to public health events. Secondly, the system has strong adaptability to new types of false information. Through the fusion of knowledge graphs and large models, as well as continuously optimized model parameters, the system can quickly learn and identify newly emerging false information patterns, effectively reducing the risks brought by false information. Furthermore, in multimodal data processing, the system can fully integrate multiple data forms such as text, pictures, and videos, comprehensively improving the accuracy and reliability of false information detection.

[0048] Example 2, please refer to the attached Figure 1 - Attachment Figure 2 The present invention provides a technical solution: a public health real information evaluation and early warning system based on AI technology, comprising:

[0049] The database module is used to store the information collected by the acquisition module and provide strong support for subsequent model construction and real-time response;

[0050] The data collection module is used to collect public health-related data from a variety of sources such as news, social media, and professional forums, including multimodal data such as text, images, and videos, ensuring that the data source covers the data source of the knowledge graph and the data source of the false information library;

[0051] The model building module introduces knowledge graphs, integrates them with the feature representation of large language models, and defines deep feature conversion functions and fusion functions. By combining the representation of knowledge graphs and large models, the model's semantic understanding and association analysis capabilities in specific fields such as medicine and epidemiology are improved. At the same time, interaction modeling functions and semantic relationship modeling functions are defined to further enhance the model's semantic understanding capabilities in specific fields.

[0052] The performance evaluation module establishes a system performance evaluation index system, including accuracy, timeliness, and specificity, and uses specific evaluation functions to comprehensively evaluate the performance of the large model in false information detection;

[0053] The optimization module uses the display capabilities of large modules and composite models to conduct empirical analysis, evaluate the impact of false information in crisis cases, and extract the optimal crisis response strategy by solving optimization problems;

[0054] The real-time monitoring response module can promptly detect false information in public health incidents and issue early warnings to relevant departments and the public, effectively curbing the spread of false information and ensuring public health safety.

[0055] Please see attached Figure 1 The performance evaluation module is based on accuracy, timeliness, specificity and the evaluation function E SYS ={Acc(F DI ),Timeliness(Mcp),Specificity(F PD )} constructs the evaluation criteria for false information, where F DI represents the function of forecast accuracy, Mcp represents the composite model considering timeliness, and F PD The performance evaluation module integrates the false information source identification and propagation dynamics prediction, as well as the epidemic prediction model F EP , construct the composite model M CI (F SI , F SI , F SI ), the optimization module analyzes the comprehensive impact of false information dissemination on public health and epidemic development through a composite model. CI , the specific conversion formula is as follows:

[0056] I CI =M CI(F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ)

[0057] Where y is epidemiological data, I CI For comprehensive impact, the real-time monitoring response module monitors public health information in real time and uses the constructed detection algorithm, namely R (Detection (D, A)) to detect false information in real-time monitoring. When the response mechanism R detects false information, the response mechanism is immediately executed to push warning information to relevant departments and the public.

[0058] In this embodiment, in the detection algorithm R (Detection (D, A)), D represents the content to be detected, A is a composite model, and R is the specific action taken after the false information is detected. The purpose of the response mechanism is to respond in a timely manner to reduce the impact of false information. The formula describes a complete process from real-time monitoring to detecting false information and then taking countermeasures. This model aims to improve the reliability and timeliness of public health information and prevent false information from causing harm to public health. When in use, the real-time monitoring response module dynamically updates the response to emerging health information. The real-time monitoring response module ensures that the system can capture new information in a timely manner, respond quickly, and issue early warnings to relevant departments and the public, effectively maintaining public health safety.

[0059] Those skilled in the art should understand that the modules or steps of the present invention described above can be implemented by a general-purpose computer device, or alternatively, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. The present invention is not limited to any specific combination of hardware and software.

[0060] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

[0061] Although the above describes the specific implementation mode of the present invention in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art on the basis of the technical solution of the present invention without creative work are still within the scope of protection of the present invention.

Claims

1. Public health real information assessment and early warning system based on AI technology, characterized by: include The database module is used to store the information collected by the acquisition module and provide strong support for subsequent model construction and real-time response; The data collection module is used to collect public health-related data from a variety of sources (such as news, social media, professional forums, etc.), including multimodal data such as text, pictures, and videos, ensuring that the data source covers the data source of the knowledge graph and the data source of the false information library; The model building module introduces the knowledge graph, integrates it with the feature representation of the large language model, and defines deep feature conversion functions and fusion functions. It combines the knowledge graph and the representation of the large model to improve the semantic understanding and association analysis capabilities of the model in specific fields (such as medicine and epidemiology). At the same time, it defines interaction modeling functions and semantic relationship modeling functions to further enhance the semantic understanding capabilities of the model in specific fields. The performance evaluation module establishes a system performance evaluation index system, including accuracy, timeliness, and specificity, and uses specific evaluation functions to comprehensively evaluate the performance of the large model in false information detection; The optimization module uses the display capabilities of large modules and composite models to conduct empirical analysis, evaluate the impact of false information in crisis cases, and extract the optimal crisis response strategy by solving optimization problems; The real-time monitoring response module can promptly detect false information in public health incidents and issue early warnings to relevant departments and the public, effectively curbing the spread of false information and ensuring public health safety.

2. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The data collection module needs to clean and organize the collected data, remove duplicate, invalid or personal privacy information entries, and perform word segmentation, stop word removal, part-of-speech tagging and other processing on text data; For image and video data, necessary format conversion and compression are performed, and natural language processing technology is used to extract word frequency, TF-IDF value, part of speech and naming from text data. At the same time, the word embedding technology Word2Vec is used to convert text into vector representation for input into the model. For image data, convolutional neural network (CNN) is used to extract visual features of images. Key frames can be extracted and processed using CNN, or motion, color and other features can be extracted using video processing library. The extracted text, image and video features are fused to form a multimodal feature representation. This can be achieved through splicing and weighted summation, so that the model can use information from multiple modes to identify false information at the same time. At the same time, data cleaning needs to remove missing values ​​and outliers, and label each data sample. The specific steps are as follows: L＝{l1,l2,l3...,l n} Among them, L represents the label set (real news, manually written news, and fake news generated by GPT), l1 represents the data sample d i Corresponding standards, the data classification process can be expressed by a classification function, as follows: C(d1)={C1,C2,C3...,C k } D fimal ={(F i ,l n )|d1∈Dcleaned} Where C(d1) represents the classification result of data sample d1, c k represents the k class, D fimal ={(F i ,l n |d1∈Dcleaned} formula represents the final data set prepared for model training and testing, where D fimal Represents the final data set, (F i ,l n ) represents the pair of features of the data sample and the corresponding label, d1∈Dcleaned means that the data sample comes from the cleaned data set D fimal .

3. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The database module includes a real information storage module and a false information storage module, and the real information storage module and the false information storage module are used to store real information and false information respectively.

4. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The model building module constructs a false information identification function by mapping into the deep learning algorithm: PI (x, θ), where x is the input data and θ is the model parameter. This function converts x into a feature vector f(x) = F by applying a high-level semantic image. PI (x, θ), the model building module also introduces a false information loss function with the same specifications to guide the model to more accurately identify false information. The false information loss function L DI (f(x), y), where y is the label, this function is used to quantify the gap between the model prediction, f(x) and the true label y.

5. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The model building module optimizes the objective of minimizing the false information loss function L DI (f(x),y) and the accuracy of the maximum true information Acc((f(x)), and introduce the sensitivity index Stype(F X ,T) to focus on specific types of false information, where T represents specific false information, such as disease transmission false information and vaccine false information.

6. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The performance evaluation module is based on accuracy (Acc), timeliness (Timeliness), specificity (Specificity) and the evaluation function E SYS ={Acc(F DI ),Timeliness(Mcp),Specificity(F PD )} constructs the evaluation criteria for false information, where F DI represents the function of forecast accuracy, Mcp represents the composite model considering timeliness, and F PD Represents the specificity-related function.

7. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The performance evaluation module integrates false information source identification and propagation dynamics prediction, as well as the epidemic prediction model F EP , construct the composite model M CI (F SI , F SI , F SI ).

8. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The optimization module analyzes the comprehensive impact of false information dissemination on public health and epidemic development through a composite model. CI , the specific conversion formula is as follows: I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ)) Where y is epidemiological data, I CI For the combined impact.

9. The public health real information evaluation and early warning system based on AI technology according to claim 1 is characterized by: The real-time monitoring response module monitors public health information in real time and uses the constructed detection algorithm, namely R (Detection (D, A)) to detect false information during real-time monitoring. When the response mechanism R detects false information, the response mechanism is immediately executed to push warning information to relevant departments and the public.

Images