Public health false information identification method using large language model
By adopting large language models in the field of public health, integrating false information dissemination models with epidemiological prediction models, and building and updating LLM models, the problems of insufficient adaptability and insufficient automatic update ability in the existing technology are solved. Accurate identification and management of false information on public health has been achieved.
Patent Information
- Application Number
- CN202411477507.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-22
- Publication Date
- 2025-02-18
AI Technical Summary
The prior art has problems such as insufficient adaptability, insufficient contextual understanding and insufficient automatic update capabilities in identifying false information in public health, making it difficult to effectively identify and manage false information in a rapidly changing public health crisis.
Using a large language model, a pre-constructed LLM model is constructed by integrating the disinformation dissemination model and the epidemiological prediction model, and the public health-related data is pre-processed, trained and identified through the model, and the model is dynamically updated to adapt to new disinformation. Strategy.
It has achieved accurate identification of false information in specific areas of public health, and through identification result marking and feedback, it has helped users to effectively identify false information, and improved the management and refutation capabilities of public health information.
Smart Images
Figure CN119480146A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field related to computer science, and in particular to a method for identifying false public health information using a large language model. Background Art
[0002] In the globalized information age, especially in the face of public health crises such as infectious disease outbreaks and epidemics, accurately and rapidly identifying false information and effectively managing and refuting it are essential to protecting public safety and health. The spread of false information may not only cause public panic, but also interfere with the implementation of effective disease prevention and control measures, thereby exacerbating the impact of public health crises.
[0003] Most existing false information identification methods use local OCR recognition and rely mainly on keyword filtering, user reports and expert verification to identify false information. However, these methods have problems such as large delays, low accuracy and difficulty in coping with large-scale information flows.
[0004] In the field of artificial intelligence, although large language models (such as OpenAI's GPT series) and deep learning algorithms have been applied to text recognition and processing, they still face several technical problems in practical applications:
[0005] 1. Adaptability and flexibility: Existing models often require extensive customization and training to adapt to specific types of false information, which may not be efficient enough in a rapidly changing public health crisis.
[0006] 2. Depth of contextual understanding: in cross-cultural and cross-linguistic environments.
[0007] 3. Automatic update and learning capabilities: The form and content of false information are constantly evolving, requiring the model to be able to continuously learn and self-update to adapt to new false information strategies. Summary of the invention
[0008] 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 identifying false public health information using a large language model.
[0009] In order to achieve the above object, the present invention adopts the following technical solutions:
[0010] Specifically, a method for identifying false public health information using a large language model is proposed, including the following:
[0011] Access to public health related data;
[0012] Preprocessing the public health related data to obtain a data set for model training and testing;
[0013] The data set is divided into a training set and a validation set, and a pre-built LLM model is trained based on the training set and the validation set to obtain a trained LLM model;
[0014] Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model;
[0015] The trained LLM model is used to identify the target public health information to obtain the recognition results, and false information is marked based on the recognition results.
[0016] Further, specifically, obtaining public health related data, including,
[0017] Collect public health related online information as follows:
[0018] [D={d 1 , d 2 , ..., d n}],
[0019] Among them, (D) represents the collected data set, (d i ) represents a single data sample, i∈[1,n], for each data sample (d i ), and further describe its feature set as follows:
[0020] [F i ={f i1 , f i2 , ..., f im}],
[0021] Among them, (F i ) represents the data sample (d i ) feature set, (f ij ) represents the value of a specific feature,
[0022] The choice of data source is represented by a weight function, taking into account the importance of different data sources:
[0023] [W s ={w 1 , w 2 , ..., w k}],
[0024] Among them, (W s ) represents the weights of different data sources, (w k ) represents the weight of the (k)th data source, and the final selected data set is expressed as:
[0025] [D selected ={d i |d i∈D, information related to public health that meets the selection criteria}].
[0026] Further, specifically, the public health related data is preprocessed to obtain a data set for model training and testing, including:
[0027] First, perform data cleaning to remove missing values and outliers:
[0028] [D cleaned ={d i |d i ∈D selected , meet the cleaning conditions}]
[0029] Next, we label each data sample:
[0030] [L={l 1 , l 2 , ..., l n}]
[0031] Among them, (L) represents the label set, (l i ) represents the data sample (d i ) The corresponding label, the data classification process is represented by a classification function:
[0032] [C(d i )={c 1 , c 2 , ..., c k}]
[0033] Among them, (C(d i )) represents the data sample (d i ) classification results, (c k ) represents the (k)th category,
[0034] Finally, we get the data set for model training and testing:
[0035] [D final ={(F i , l i )|d i ∈D cleanedl , F i is the sample feature, l i is the corresponding tag}].
[0036] Further, the specific,construction process of the pre-built LLM model includes,
[0037] 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 ):
[0038] [S origin =F SI (x,Θ)]
[0039] 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:
[0040] [P spread =F PD (x,Φ)]
[0041] Where (Φ) represents the model parameters,
[0042] 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 ):
[0043] [I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ))]
[0044] Where (y) is the input epidemiological data.
[0045] Further, specifically, the data set is divided into a training set and a validation set, and the pre-built LLM model is trained based on the training set and the validation set, including:
[0046] When processing the input data, the data in the data set is divided into training data (D train ) and verification data (D val ),
[0047] The training set is used for model training, which is expressed as:
[0048] [M train =Train(D train )]
[0049] Use the validation set to perform model validation, expressed as:
[0050] [Accuracyval =Validate(M train , D val )].
[0051] Furthermore, the method further comprises:
[0052] 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 and the current LLM model parameters:
[0053]
[0054] Where u is a preset parameter.
[0055] Further, specifically, false information is marked according to the identification results, including:
[0056] When the recognition result is false information, the information interface associated with the current recognition result is marked as false information, a preset pattern stamp is added, and the associated information of the added preset pattern stamp is fed back to the user.
[0057] The beneficial effects of the present invention are:
[0058] The present invention proposes a method for identifying public health false information using 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. It can accurately identify false information in specific areas of public health, and then mark the false information according to the identification results and feed it back to the corresponding users, which can help users effectively identify false information. BRIEF DESCRIPTION OF THE DRAWINGS
[0059] 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:
[0060] Figure 1 Shown is a flow chart of a method for identifying false public health information using a large language model according to the present invention;
[0061] Figure 2 The figure shows the display effect after false information is identified in an example when the present invention is applied. DETAILED DESCRIPTION
[0062] 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.
[0063] Example 1, reference Figure 1 The present invention proposes a method for identifying false public health information using a large language model, comprising the following steps:
[0064] Step 110: Obtaining public health related data;
[0065] Step 120: preprocess the public health related data to obtain a data set for model training and testing;
[0066] Step 130: divide the data set into a training set and a validation set, and train the pre-built LLM model based on the training set and the validation set to obtain a trained LLM model;
[0067] Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model;
[0068] Step 140: Use the trained LLM model to identify the target public health information to obtain an identification result, and mark the false information according to the identification result.
[0069] 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, and then mark the false information based on the identification results and feed it back to the corresponding user, which can help users effectively identify false information.
[0070] As a preferred embodiment of the present invention, specifically, obtaining public health related data includes:
[0071] Collect public health related online information as follows:
[0072] [D={d 1 , d 2 , ..., d n}],
[0073] Among them, (D) represents the collected data set, (d i) represents a single data sample, i∈[1,n], for each data sample (d i ), and further describe its feature set as follows:
[0074] [F i ={f i1 , f i2 , ..., f im}],
[0075] Among them, (F i ) represents the data sample (d i ) feature set, (f ij ) represents the value of a specific feature,
[0076] The choice of data source is represented by a weight function, taking into account the importance of different data sources:
[0077] [W s ={w 1 , w 2 , ..., w k}],
[0078] Among them, (W s ) represents the weights of different data sources, (w k ) represents the weight of the (k)th data source, and the final selected data set is expressed as:
[0079] [D selected ={d i |d i ∈D, information related to public health that meets the selection criteria}].
[0080] As a preferred embodiment of the present invention, specifically, the public health related data is preprocessed to obtain a data set for model training and testing, including:
[0081] First, perform data cleaning to remove missing values and outliers:
[0082] [D cleaned ={d i |d i ∈D selected , meet the cleaning conditions}]
[0083] Next, we label each data sample:
[0084] [L={l 1 , l 2 , ..., l n}]
[0085] Among them, (L) represents the label set, (l i) represents the data sample (d i ) The corresponding label, the data classification process is represented by a classification function:
[0086] [C(d i )={c 1 , c 2 , ..., c k}]
[0087] Among them, (C(d i )) represents the data sample (d i ) classification results, (c k ) represents the (k)th category. Finally, we get the data set for model training and testing:
[0088] [D final ={(F i , l i )|d i ∈D cleanedl , F i is the sample feature, l i is the corresponding tag}].
[0089] As a preferred embodiment of the present invention, specifically, the construction process of the pre-built LLM model includes:
[0090] 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 ):
[0091] [S origin =F SI (x,Θ)]
[0092] 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:
[0093] [P spread =F PD (x,Φ)]
[0094] Where (Φ) represents the model parameters,
[0095] 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 (ICI ):
[0096] [I CI =M CI (F SI (x,Θ),F PD (x,Φ),F EP (y,Ψ))]
[0097] Where (y) is the input epidemiological data.
[0098] 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.
[0099] As a preferred embodiment of the present invention, specifically, the data set is divided into a training set and a validation set, and the pre-built LLM model is trained based on the training set and the validation set, including:
[0100] When processing the input data, the data in the data set is divided into training data (D train ) and verification data (D val ),
[0101] The training set is used for model training, which is expressed as:
[0102] [M train =Train(D train )]
[0103] Use the validation set to perform model validation, expressed as:
[0104] [Accuracy val =Validate(M train , D val )].
[0105] As a preferred embodiment of the present invention, the method further comprises:
[0106] 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 and the current LLM model parameters:
[0107]
[0108] Where u is a preset parameter.
[0109] 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.
[0110] As a preferred embodiment of the present invention, specifically, marking false information according to the recognition result includes:
[0111] When the recognition result is false information, the information interface associated with the current recognition result is marked as false information, a preset pattern stamp is added, and the associated information of the added preset pattern stamp is fed back to the user.
[0112] Specifically, refer to 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.
[0113] 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.
[0114] 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 identifying false public health information using a large language model, characterized in that: Includes the following: Access to public health-related data; Preprocessing the public health related data to obtain a data set for model training and testing; The data set is divided into a training set and a validation set, and a pre-built LLM model is trained based on the training set and the validation set to obtain a trained LLM model; Among them, the pre-built LLM model is constructed based on the fusion of false information propagation model and epidemiological prediction model; The trained LLM model is used to identify the target public health information to obtain the recognition results, and false information is marked based on the recognition results.
2. According to claim 1, a method for identifying false public health information using a large language model is characterized in that: Specifically, obtain public health related data, including: Collect public health related online information as follows: [D={d1,d2,...,d n }], Among them, (D) represents the collected data set, (d i ) represents a single data sample, i∈[1,n], for each data sample (d i ), and further describe its feature set as follows: [F i ={f i1 ,f i2 ,...,f im }], Among them, (F i ) represents the data sample (d i ) feature set, (f ij ) represents the value of a specific feature, The choice of data source is represented by a weight function, taking into account the importance of different data sources: [IN s ={w1,w2,...,w k }], Among them, (W s ) represents the weights of different data sources, (w k ) represents the weight of the (k)th data source, and the final selected data set is expressed as: [D selected ={d i |d i ∈D, information related to public health that meets the selection criteria}].
3. A method for identifying false public health information using a large language model according to claim 1, characterized in that: Specifically, the public health related data is preprocessed to obtain a data set for model training and testing, including: First, perform data cleaning to remove missing values and outliers: [D cleaned ={d i |d i ∈D selected , meet the cleaning conditions}] Next, we label each data sample: [L＝{l1，l2，...，l n}] Among them, (L) represents the label set, (l i ) represents the data sample (d i ) The corresponding label, the data classification process is represented by a classification function: [C(d i )={c1,c2,...,c k }] Among them, (C(d i )) represents the data sample (d i ) classification results, (c k ) represents the (k)th category, Finally, we get the data set for model training and testing: [D final ={(F i , l i )|d i ∈D cleanedl , F i is the sample feature, l i is the corresponding tag}].
4. A method for identifying false public health information using a large language model according to claim 1, 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,Φ))] Where (y) is the input epidemiological data.
5. A method for identifying false public health information using a large language model according to claim 4, characterized in that: Specifically, the data set is divided into a training set and a validation set, and the pre-built LLM model is trained based on the training set and the validation set. include, When processing the input data, the data in the data set is divided into training data (D train ) and verification data (D val ), The training set is used for model training, which is expressed as: [M train =Train(D train )] Use the validation set to perform model validation, expressed as: [Accuracy val =Validate(M train ,D val )]。 6. A method for identifying false public health information using 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 and the current LLM model parameters: Where u is a preset parameter.
7. A method for identifying false public health information using a large language model according to claim 5, characterized in that: Specifically, false information is marked based on the identification results, including: When the recognition result is false information, the information interface associated with the current recognition result is marked as false information, a preset pattern stamp is added, and the associated information of the added preset pattern stamp is fed back to the user.