Public health false information processing method based on large language model

Abstract

The invention relates to a public health false information processing method based on a large language model. The public health false information processing method comprises the following steps: acquiring text data of an image format related to public health; performing information preprocessing on the text data in the image format to obtain a text associated image of the public health event; obtaining the text associated image input by the information processing module, and identifying the text associated image through a pre-constructed LLM model to obtain a false information identification result; wherein the pre-constructed LLM model is constructed based on fusion of a false information propagation model and an epidemiological prediction model; in addition, a real-time monitoring and response mechanism is further arranged, published public health information can be monitored in real time, and response is executed immediately when false information is detected. According to the invention, the published public health information can be monitored in real time, the response is executed immediately when false information is detected, the workload of workers can be greatly reduced, and the false information identification efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field related to computer science, and in particular to a method for processing false public health information based on a large language model. Background Art

[0002] The main techniques for identifying and dealing with false information in the field of public health include the following:

[0003] 1. Rule-based content filtering systems: This type of system identifies false information through preset keywords and pattern matching. For example, specific false information indicators (such as "fake treatment", "miracle cure", etc.) are used to filter content. The limitation of this approach is its inherent rigidity and poor adaptability to emerging forms of false information.

[0004] 2. Combination of social media monitoring and manual review: Some large social platforms use a combination of automated tools and manual review to identify and manage false information. Automated tools are responsible for preliminary screening, while manual review is responsible for final judgment. Although this method improves the accuracy of processing, it consumes a lot of resources and time, and it is difficult to cope with the rapid flow of large-scale information. Summary of the invention

[0005] The purpose of the present invention is to solve at least one of the deficiencies of the prior art and to provide a method for processing public health false information based on a large language model.

[0006] In order to achieve the above object, the present invention adopts the following technical solutions:

[0007] Specifically, a method for handling false public health information based on a large language model is proposed, including the following:

[0008] Step 110: Acquire text data in image format related to public health;

[0009] Step 120: performing information preprocessing on the text data in image format to obtain a text-related image of a public health event;

[0010] Step 130: Obtain the text-related image input from the information processing module, and identify the text-related image using a pre-built LLM model to obtain a false information identification result;

[0011] Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model;

[0012] In addition, a real-time monitoring and response mechanism is set up to monitor the release of public health information in real time and respond immediately when false information is detected.

[0013] Further, specifically, obtaining text data in image format related to public health, including,

[0014] Collect multiple text data images related to any public health event uploaded by users from various online platforms to obtain a set of text data images corresponding to each public health event.

[0015] Further, specifically, the text data in image format is preprocessed to obtain a text-associated image of a public health event, including:

[0016] For each text data image set corresponding to a public health event, define the image matrix of the kth text data image in the text data image set as Qk(k), and the pixel value of the pixel point in the hkth row and lkth column in Qk(k) is expressed as Qk(k, hk, lk);

[0017] Calculate the ROI contribution rate ROI_le(k, hk, lk) between Qk(k, hk, lk) and its neighboring pixels, where ROI_le(k, hk, lk) is given by exp(Cp(k, hk, lk) 2 -avg(k,hk,lk) 2 )and

[0018] exp(max(k,hk,lk) 2 -min(k, hk, lk) 2 ) is determined by the ratio of

[0019] Wherein, max(k, hk, lk) is the maximum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), avg(k, hk, lk) and min(k, hk, lk) are the average value and minimum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), respectively, wherein the neighborhood pixel point of Qk(k, hk, lk) refers to the 8 neighborhood pixel points in the two-dimensional plane of Qk(k, hk, lk);

[0020] Calculate the ROI contribution areas of all text data images in the text data image set based on the ROI contribution rate;

[0021] Constructing an original image, wherein the original image is obtained by calculating the pixel average value of all text data images in the text data image set, that is, the pixel value of any pixel point of the original image is the pixel average value of all text data images at the same pixel point position;

[0022] The ROI contribution areas of all the text data images in the text data image set are used to replace the corresponding areas of the original image in sequence to obtain a text-associated image of the current public health event.

[0023] Further, specifically, the ROI contribution areas of all text data images in the text data image set are calculated based on the ROI contribution rate, including:

[0024] When k=1, that is, for the first text data image in the text data image set, traverse its image matrix Q1(1), find the two pixel points with the maximum ROI contribution rate and the second largest ROI contribution rate in Q1(1), connect the pixel point with the maximum ROI contribution rate and the pixel point with the second largest ROI contribution rate to obtain a line segment lne_sgm(1), then traverse Q1(1), select the pixel point with the maximum ROI contribution rate other than the pixel point in lne_sgm(1) and define it as pixel point H, and the area obtained by connecting H and the two endpoints of the line segment lne_sgm(1) is the ROI contribution area of ​​the first text data image;

[0025] When k>1, that is, for any text data image other than the first text data image in the text data image set, calculate the pixel average value avg(k-1) of the pixel points in the ROI contribution area of ​​Qk-1(k-1), traverse Qk(k), select any two pixel points in Qk(k) whose ROI contribution rate is not less than avg(k-1) and connect them to obtain the line segment lne_sgm(k), then traverse Qk(k), select another pixel point with the largest ROI contribution rate except the pixel point on lne_sgm(k) and define it as pixel point G, and the area obtained by connecting G and the two vertices of the line segment lne_sgm(k) is the ROI contribution area of ​​the current text data image.

[0026] Further, the specific,construction process of the pre-built LLM model includes,

[0027] The false information source identification function of the false information propagation model is defined as (F SI (x, Θ)), where (x) is the input data and (Θ) is the model parameter. The goal is to identify the origin of false information (S origin ):

[0028] [S origin =F SI (x,Θ)]

[0029] Define the propagation dynamics prediction model P spread , this model is used to predict the propagation path and scope of false information in social networks, expressed as:

[0030] [P spread =F PD (x,Φ)]

[0031] Where (Φ) represents the model parameters,

[0032] Integrate the above two models and the epidemiological prediction model (F EP ), build a composite model (M CI (F SI , F PD , F EP )); This composite model aims to analyze the combined impact of false information dissemination on public health and the development of the epidemic (I CI ):

[0033] [I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ))]

[0034] Where (y) is the input epidemiological data.

[0035] Furthermore, the method further comprises:

[0036] Defining dynamic update policies Update the pre-built LLM model, where (M) is the current LLM model, (D new ) is the newly added data, (α) is the update rate parameter, and the updated model (M′) is achieved by fusing the newly added data with the current LLM model parameters:

[0037]

[0038] Where u is a preset parameter.

[0039] Further, specific, real-time monitoring and response mechanisms include,

[0040] The monitoring mechanisms are:

[0041] Establish a real-time monitoring model (S monitor ) for monitoring public health information released;

[0042] Constructing a false information monitoring algorithm (Detection(D, A)), (Detection(D, A)) is used to automatically call method steps 110 to 130 to monitor the released public health information;

[0043] The response mechanism is:

[0044] When (Detection(D,A)) detects false information, it immediately executes the response [Response(S monitor ,Detection(D,A))] to mark false information.

[0045] The beneficial effects of the present invention are:

[0046] The present invention proposes a method for processing public health false information based on a large language model. An LLM model is constructed by integrating a false information propagation model with an epidemiological prediction model, and false information is identified on public health-related data through the LLM model. Accurate false information identification can be performed for specific areas of public health. In addition, high-quality images to be identified can be obtained by preprocessing the published false information in the form of pictures, thereby increasing the subsequent recognition accuracy. In addition, a real-time monitoring and response mechanism is constructed, which can monitor the release of public health information in real time and immediately respond when false information is detected, which can greatly reduce the workload of staff and improve the efficiency of false information identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] The above and other features of the present disclosure will become more obvious by describing in detail the embodiments shown in the accompanying drawings. The same reference numerals in the accompanying drawings of the present disclosure represent the same or similar sampling and evaluation points. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure. For those skilled in the art, other accompanying drawings can be obtained based on these accompanying drawings without creative work. In the accompanying drawings:

[0048] Figure 1 Shown is a flow chart of a method for processing false public health information based on a large language model of the present invention;

[0049] Figure 2 The figure shows the display effect after false information is identified in an example when the present invention is applied. DETAILED DESCRIPTION

[0050] The following will be combined with the embodiments and drawings to clearly and completely describe the concept, specific structure and technical effects of the present invention, so as to fully understand the purpose, scheme and effect of the present invention. It should be noted that the embodiments in this application and the features in the embodiments can be combined with each other without conflict. The same reference numerals used throughout the drawings indicate the same or similar parts.

[0051] Example 1, reference Figure 1 The present invention proposes a method for processing false public health information based on a large language model, comprising the following steps:

[0052] Step 110: Acquire text data in image format related to public health;

[0053] Step 120: performing information preprocessing on the text data in image format to obtain a text-related image of a public health event;

[0054] Step 130: Obtain the text-related image input from the information processing module, and identify the text-related image using a pre-built LLM model to obtain a false information identification result;

[0055] Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model;

[0056] In addition, a real-time monitoring and response mechanism is set up to monitor the release of public health information in real time and respond immediately when false information is detected.

[0057] In this embodiment 1, an LLM model is constructed by integrating the false information propagation model with the epidemiological prediction model, and false information is identified on public health related data through the LLM model. It is possible to accurately identify false information in specific areas of public health. In addition, by preprocessing the published false information in the image format, high-quality images to be identified can be obtained, thereby increasing the subsequent recognition accuracy. In addition, a real-time monitoring and response mechanism is constructed, which can monitor the release of public health information in real time and respond immediately when false information is detected, which can greatly reduce the workload of staff and improve the efficiency of false information identification.

[0058] As a preferred embodiment of the present invention, specifically, obtaining text data in image format related to public health includes:

[0059] Collect multiple text data images related to any public health event uploaded by users from various online platforms to obtain a set of text data images corresponding to each public health event.

[0060] As a preferred embodiment of the present invention, specifically, the text data in image format is preprocessed to obtain a text-related image of a public health event, including:

[0061] For each text data image set corresponding to a public health event, define the image matrix of the kth text data image in the text data image set as Qk(k), and the pixel value of the pixel point in the hkth row and lkth column in Qk(k) is expressed as Qk(k, hk, lk);

[0062] Calculate the ROI contribution rate ROI_le(k, hk, lk) between Qk(k, hk, lk) and its neighboring pixels, where ROI_le(k, hk, lk) is given by exp(Cp(k, hk, lk) 2 -avg(k,hk,lk) 2 )and

[0063] exp(max(k,hk,lk) 2 -min(k, hk, lk) 2 ) is determined by the ratio of

[0064] Wherein, max(k, hk, lk) is the maximum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), avg(k, hk, lk) and min(k, hk, lk) are the average value and minimum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), respectively, wherein the neighborhood pixel point of Qk(k, hk, lk) refers to the 8 neighborhood pixel points in the two-dimensional plane of Qk(k, hk, lk);

[0065] Calculate the ROI contribution areas of all text data images in the text data image set based on the ROI contribution rate;

[0066] Constructing an original image, wherein the original image is obtained by calculating the pixel average value of all text data images in the text data image set, that is, the pixel value of any pixel point of the original image is the pixel average value of all text data images at the same pixel point position;

[0067] The ROI contribution areas of all the text data images in the text data image set are used to replace the corresponding areas of the original image in sequence to obtain a text-associated image of the current public health event.

[0068] In this preferred embodiment, taking into account that the quality of text data images collected by various online platforms may be poor, for the text data images of the same public health event, several more images are collected to form a text data image set, and then an ROI area is extracted from the text data image set. The area that best represents the ROI area is selected from each image and extracted, and then the corresponding area of ​​the original image is replaced. This can ensure the image quality of the text data image, and then ensure that the subsequent machine learning model can obtain the best quality image data, thereby ensuring the accuracy of subsequent results from the data source level.

[0069] As a preferred embodiment of the present invention, specifically, the ROI contribution area of ​​all text data images in the text data image set is calculated based on the ROI contribution rate, including:

[0070] When k=1, that is, for the first text data image in the text data image set, traverse its image matrix Q1(1), find the two pixel points with the maximum ROI contribution rate and the second largest ROI contribution rate in Q1(1), connect the pixel point with the maximum ROI contribution rate and the pixel point with the second largest ROI contribution rate to obtain a line segment lne_sgm(1), then traverse Q1(1), select the pixel point with the maximum ROI contribution rate other than the pixel point in lne_sgm(1) and define it as pixel point H, and the area obtained by connecting H and the two endpoints of the line segment lne_sgm(1) is the ROI contribution area of ​​the first text data image;

[0071] When k>1, that is, for any text data image other than the first text data image in the text data image set, calculate the pixel average value avg(k-1) of the pixel points in the ROI contribution area of ​​Qk-1(k-1), traverse Qk(k), select any two pixel points in Qk(k) whose ROI contribution rate is not less than avg(k-1) and connect them to obtain the line segment lne_sgm(k), then traverse Qk(k), select another pixel point with the largest ROI contribution rate except the pixel point on lne_sgm(k) and define it as pixel point G, and the area obtained by connecting G and the two vertices of the line segment lne_sgm(k) is the ROI contribution area of ​​the current text data image.

[0072] In this preferred embodiment, by calculating the ROI contribution area of ​​the text data image in the above manner, the ROI contribution areas of all text data images can be accurately obtained.

[0073] As a preferred embodiment of the present invention, specifically, the construction process of the pre-built LLM model includes:

[0074] The false information source identification function of the false information propagation model is defined as (F SI (x, Θ)), where (x) is the input data, (Θ) is the model parameter, and the goal is to identify the origin of false information (S origin ):

[0075] [S origin =F SI (x,Θ)]

[0076] Define the propagation dynamics prediction model P spread , this model is used to predict the propagation path and scope of false information in social networks, expressed as:

[0077] [P spread =F PD (x,Φ)]

[0078] Where (Φ) represents the model parameters,

[0079] Integrate the above two models and the epidemiological prediction model (F EP ), build a composite model (M CI (F SI , F PD , F EP )); This composite model aims to analyze the combined impact of false information dissemination on public health and the development of the epidemic (I CI ):

[0080] [I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ))]

[0081] Where (y) is the input epidemiological data.

[0082] In this preferred embodiment, considering the particularity of applications in the field of public health, in the model building stage, we combined advanced open source large model technologies and deep learning algorithms such as LLaMA, ChatGLM-6B and VisualGLM, and used these algorithms and models to deeply process and analyze text, images and social media metadata to enhance the model's understanding and judgment capabilities for complex false information scenarios. In order to improve the professional adaptability of the model in the field of public health, we work closely with experts in medicine and epidemiology, integrate expertise in various fields, build domain-specific corpora, and integrate knowledge graphs (KGs) to provide richer and more accurate public health-related information. Through these measures, our model is able to capture subtle changes in complex narratives and contexts related to public health, thereby more accurately identifying false information and providing strong support for public health crisis response and risk communication.

[0083] As a preferred embodiment of the present invention, the method further comprises:

[0084] Defining dynamic update policies Update the pre-built LLM model, where (M) is the current LLM model, (D new ) is the newly added data, (α) is the update rate parameter, and the updated model (M′) is achieved by fusing the newly added data with the current LLM model parameters:

[0085]

[0086] Where u is a preset parameter.

[0087] In this preferred embodiment, during the model optimization phase, we use a continuous learning strategy to optimize the model parameters. The continuous learning strategy is a machine learning technique that can continuously learn from new data and update the model parameters to adapt to the rapid changes in public health information and false information scenarios. By continuously iterating and updating the model parameters, we can gradually improve the performance and accuracy of the model, making it more adaptable to the needs of the public health field.

[0088] As a preferred embodiment of the present invention, specifically, the real-time monitoring and response mechanism includes:

[0089] The monitoring mechanisms are:

[0090] Establish a real-time monitoring model (S monitor ) for monitoring public health information released;

[0091] Constructing a false information monitoring algorithm (Detection(D, A)), (Detection(D, A)) is used to automatically call method steps 110 to 130 to monitor the released public health information;

[0092] The response mechanism is:

[0093] When (Detection(D,A)) detects false information, it immediately executes the response [Response(S monitor ,Detection(D,A))] to mark false information.

[0094] In this preferred embodiment, a real-time monitoring model is constructed to monitor the public health information released, and method steps 110 to 130 are called to monitor the public health information released. When false information exists, a response [Response (S monitor ,Detection(D,A))] marks false information and informs the user.

[0095] Reference Figure 2 In actual application, after extracting the text content from the acquired public health information picture, the method proposed in the present invention is used to identify it and find that Figure 2 The content in the left picture is true, while the content in the right picture is false information, so a special logo of the platform applying the method of the present invention is marked, and the result is fed back to the user.

[0096] Although the description of the present invention has been quite detailed and specifically describes several described embodiments, it is not intended to be limited to any of these details or embodiments or any particular embodiment, but should be regarded as providing a broad possible interpretation of these claims in view of the prior art by reference to the appended claims, thereby effectively covering the intended scope of the present invention. In addition, the above description of the present invention is based on the embodiments foreseeable by the inventor, and its purpose is to provide a useful description, and those non-substantial changes to the present invention that have not yet been foreseen may still represent equivalent changes to the present invention.

[0097] The above is only a preferred embodiment of the present invention. The present invention is not limited to the above implementation. As long as the technical effect of the present invention is achieved by the same means, it should belong to the protection scope of the present invention. Within the protection scope of the present invention, its technical scheme and / or implementation method can have various modifications and changes.

Claims

1. A method for processing false public health information based on a large language model, characterized in that: Includes the following: Step 110: Acquire text data in image format related to public health; Step 120: performing information preprocessing on the text data in image format to obtain a text-related image of a public health event; Step 130: acquiring the text-associated image input from the information processing module, and identifying the text-associated image using a pre-built LLM model to obtain a false information identification result; Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model; In addition, a real-time monitoring and response mechanism is set up to monitor the release of public health information in real time and respond immediately when false information is detected.

2. The method for processing false public health information based on a large language model according to claim 1, characterized in that: Specifically, obtain text data in image format related to public health, including: Collect multiple text data images related to any public health event uploaded by users from various online platforms to obtain a set of text data images corresponding to each public health event.

3. The method for processing false public health information based on a large language model according to claim 2, characterized in that: Specifically, the text data in image format is preprocessed to obtain a text-related image of a public health event, including: For each text data image set corresponding to a public health event, define the image matrix of the kth text data image in the text data image set as Qk(k), and the pixel value of the pixel point in the hkth row and lkth column in Qk(k) is expressed as Qk(k, hk, lk); Calculate the ROI contribution rate ROI_le(k, hk, lk) between Qk(k, hk, lk) and its neighboring pixels, where ROI_le(k, hk, lk) is given by exp(Cp(k, hk, lk) 2 -avg(k,hk,lk) 2 ) and exp(max(k, hk, lk) 2 -min(k, hk, lk) 2 ) is determined by the ratio of Wherein, max(k, hk, lk) is the maximum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), avg(k, hk, lk) and min(k, hk, lk) are the average value and minimum value of the pixel value of the neighborhood pixel point of Qk(k, hk, lk), respectively, wherein the neighborhood pixel point of Qk(k, hk, lk) refers to the 8 neighborhood pixel points in the two-dimensional plane of Qk(k, hk, lk); Calculate the ROI contribution areas of all text data images in the text data image set based on the ROI contribution rate; Constructing an original image, wherein the original image is obtained by calculating the pixel average value of all text data images in the text data image set, that is, the pixel value of any pixel point of the original image is the pixel average value of all text data images at the same pixel point position; The ROI contribution areas of all the text data images in the text data image set are used to replace the corresponding areas of the original image in sequence to obtain a text-associated image of the current public health event.

4. The method for processing false public health information based on a large language model according to claim 1, characterized in that: Specifically, the ROI contribution areas of all text data images in the text data image set are calculated based on the ROI contribution rate, including: When k=1, that is, for the first text data image in the text data image set, traverse its image matrix Q1(1), find the two pixel points with the maximum ROI contribution rate and the second largest ROI contribution rate in Q1(1), connect the pixel point with the maximum ROI contribution rate and the pixel point with the second largest ROI contribution rate to obtain a line segment lne_sgm(1), then traverse Q1(1), select the pixel point with the maximum ROI contribution rate other than the pixel point in lne_sgm(1) and define it as pixel point H, and the area obtained by connecting H and the two endpoints of the line segment lne_sgm(1) is the ROI contribution area of ​​the first text data image; When k>1, that is, for any text data image other than the first text data image in the text data image set, calculate the pixel average value avg(k-1) of the pixel points in the ROI contribution area of ​​Qk-1(k-1), traverse Qk(k), select any two pixel points in Qk(k) whose ROI contribution rate is not less than avg(k-1) and connect them to obtain the line segment lne_sgm(k), then traverse Qk(k), select another pixel point with the largest ROI contribution rate except the pixel point on lne_sgm(k) and define it as pixel point G, and the area obtained by connecting G and the two vertices of the line segment lne_sgm(k) is the ROI contribution area of ​​the current text data image.

5. The method for processing false public health information based on a large language model according to claim 4, characterized in that: Specifically, the construction process of the pre-built LLM model includes: The false information source identification function of the false information propagation model is defined as (F SI (x, Θ)), where (x) is the input data, (Θ) is the model parameter, and the goal is to identify the origin of false information (S origin ): [S origin =F SI (x,Θ)] Define the propagation dynamics prediction model P spread , this model is used to predict the propagation path and scope of false information in social networks, expressed as: [P spread =F PD (x,Φ)] Where (Φ) represents the model parameters, Integrate the above two models and the epidemiological prediction model (F EP ), build a composite model (M CI (F SI , F PD , F EP )); This composite model aims to analyze the combined impact of false information dissemination on public health and the development of the epidemic (I CI ): [I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ))] Where (y) is the input epidemiological data.

6. The method for processing false public health information based on a large language model according to claim 5, characterized in that: The method further comprises, Defining dynamic update policies Update the pre-built LLM model, where (M) is the current LLM model, (D new ) is the newly added data, (α) is the update rate parameter, and the updated model (M′) is achieved by fusing the newly added data with the current LLM model parameters: Where u is a preset parameter.

7. The method for processing false public health information based on a large language model according to claim 1, characterized in that: Specific,real-time monitoring and response mechanisms include, The monitoring mechanisms are: Establish a real-time monitoring model (S monitor ) for monitoring public health information released; Constructing a false information monitoring algorithm (Detection(D, A)), (Detection(D, A)) is used to automatically call method steps 110 to 130 to monitor the released public health information; The response mechanism is: When (Detection(D,A)) detects false information, execute the response immediately [Response(S monitor ,Detection(D,A))] to mark false information.

Images