Public health false information automatic identification system

Abstract

The invention discloses a public health false information automatic identification system, and relates to the technical field of information automatic identification, and the public health false information automatic identification system comprises a data collection unit, an information processing unit, a model analysis unit and a rumor reverse unit. The information processing unit can conveniently perform feature extraction on the text data and the picture information respectively, so that false information can be identified more accurately and comprehensively, and after the model analysis unit is utilized to complete information authenticity judgment, corresponding reverse information is generated aiming at the false information, so that the false information is prevented from continuously cheating network users, and the user experience is improved. And meanwhile, the page view and the forwarding amount of the reverse information are monitored in real time, so that the propagation of false information is further inhibited.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic information identification, and in particular to an automatic identification system for false public health information. Background Art

[0002] False information not only interferes with the public's judgment of real information, but also has a negative impact on social stability. Therefore, the design of an automatic false information identification system can help the public better identify and distinguish false information. The invention patent with application number 202010609485.X discloses "a social media platform false information identification system, which relates to the field of false information identification technology, including a collection module and an identification module. The collection module collects social media platform feature information and transmits it to the identification module. The identification module identifies the social media platform feature information and determines the false information output. The identification module includes a false information detection unit, and the false information detection unit includes word-level semantic sentiment analysis and sentence-level semantic sentiment analysis. The social media platform false information identification system of the present invention implements false information detection, identification and filtering by defining word-level semantic sentiment analysis and sentence-level semantic sentiment analysis on input feature information, thereby improving the accuracy of false information identification and the generalization ability of false information identification, and has a wide range of applications.";

[0003] The above-mentioned existing technology solves the problem of being unable to detect, identify and filter false information, but when the system is running, since text data, image data and video data may exist in the information at the same time, the system cannot extract features from the image data and video data, resulting in large errors in the identification process of false information, and cannot generate corresponding rebuttal information when false information appears, so that the false information can still affect the user's judgment to a large extent. Summary of the invention

[0004] The purpose of the present invention is to provide a public health false information automatic identification system to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned purpose, the present invention provides the following technical solutions: an automatic identification system for false public health information, comprising a model analysis unit;

[0006] A data collection unit, after receiving a plurality of information data to be detected, determines a data type corresponding to the information data, divides each information data according to the data type, and stores the data into a corresponding set, performs a sampling operation on the data in the video set, obtains a plurality of image information corresponding to each video data, stores the image information into an image set, obtains a plurality of information sample data, and adds a corresponding label to each sample data;

[0007] An information processing unit, wherein the information processing unit converts the received text data into a plurality of words, analyzes a text semantic feature vector according to an embedding vector corresponding to each word, extracts a feature vector after determining the height, width and number of channels corresponding to the received image information, and analyzes a corresponding fusion feature vector according to the text feature vector and the image feature vector corresponding to each information data;

[0008] The rumor refuting unit, if the rumor refuting unit determines that the news information is false information, searches in the database according to the keywords of the false information, and pushes the obtained refuting information through the information publishing platform. If the number of views and forwardings of the refutation information within 24 hours after the release does not reach the threshold, the refutation information will be pushed again on the platform.

[0009] Preferably, the data collection unit includes a mode selection module and a data division module, the mode selection module selects a data collection mode, wherein the collection mode includes real-time collection and specific time period collection. If the collection mode is real-time collection, the time interval for each data collection is set; if the collection mode is specific time period collection, the time window information for collecting data is set; after receiving multiple information data to be detected, the data division module numbers each information data and determines the data type corresponding to the information data, wherein the data type includes text data, picture data and video data, and each information data is divided according to the data type.

[0010] Preferably, the data collection unit also includes a classification storage module, a data sampling module and a sample extraction module. If the information data contains text data, the classification storage module stores the text data in a text set; if the information data contains picture data, the picture data is stored in a picture set; if the information data contains video data, the video data is stored in a video set. The data sampling module performs sampling operations on the data in the video set to obtain multiple image information corresponding to each video data, and after jointly marking the image information generated by the same video data, stores it in the image set. The sample extraction module obtains multiple information sample data, adds corresponding labels to each sample data, and stores the sample data and labels in a database.

[0011] Preferably, the information processing unit includes a text feature extraction module and an image feature extraction module. The text feature extraction module receives text data that is preprocessed and converts the text data into multiple words. The WordEmbed function is used to analyze each word to obtain a corresponding embedding vector. The text semantic feature vector is determined based on the embedding vector corresponding to each word. The image feature extraction module receives image information that is preprocessed and determines the height, width, and number of channels of the image. The image information is then scaled so that the image size is adjusted to a set size, and then the feature vector corresponding to the image information is extracted.

[0012] Preferably, the information processing unit further includes a fusion feature generation module, which determines the text feature vector and the image feature vector corresponding to each information data, and then uses a feature splicing algorithm to analyze the two to obtain a fusion feature vector corresponding to the information data. The feature splicing algorithm is specifically:

[0013]

[0014] Among them, T f Represents the text features corresponding to the information data, V f Indicates the image features corresponding to the news data, TV f Represents the fusion feature vector corresponding to the information data, R 64 Indicates that the feature vector dimension is 64.

[0015] Preferably, the model analysis unit includes a model construction module, a model training module and a result output module. The model construction module uses deep learning technology to construct a false information identification model, transfers the fused feature vector corresponding to each information sample data to the model, and after outputting the prediction result, uses the false information loss function to calculate the loss value according to the output result and the label corresponding to the sample data. The model training module updates the model parameters according to the loss value until all sample data are analyzed by the model and the model training operation is completed. After the result output module determines the model parameters and the loss value, it transfers the fused feature vector corresponding to the information data to be detected to the false information identification model for analysis to obtain the corresponding output result.

[0016] Preferably, the rumor refutation unit includes a refutation information acquisition module. If the news information is determined to be false information, the refutation information acquisition module uses natural language technology to obtain keywords in the false information, and queries the database based on the keywords. The obtained data is output as refutation information through a visual interface. If the news information is determined to be correct information, the determination result is directly output through the visual interface.

[0017] Preferably, the rumor refutation unit also includes an information monitoring module and an information pushing module. After the information monitoring module determines the publishing platform of the false information, the refutation information is pushed through the same platform, and the number of views and forwardings after the refutation information is released is counted. The information pushing module sets a threshold value according to the number of views and forwardings of the corresponding false information before the refutation information is released. If the number of views and forwardings of the refutation information within 24 hours after its release do not reach the threshold, the refutation information is pushed on the platform again.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The present invention divides information data according to data types through a data collection unit, which is convenient for an information processing unit to extract features of text data and image information respectively, so that the identification of false information can be more accurate and comprehensive. After the model analysis unit completes the judgment of the authenticity of the information, corresponding rebuttal information will be generated for the false information to prevent the false information from continuing to deceive network users. At the same time, the number of views and forwardings of the refutation information is monitored in real time, further suppressing the spread of false information. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Provides a schematic diagram of the overall system flow for an embodiment of the present invention;

[0021] Figure 2 An internal module block diagram of a data collection unit provided by an embodiment of the present invention;

[0022] Figure 3 An internal module block diagram of an information processing unit provided by an embodiment of the present invention;

[0023] Figure 4 An internal module block diagram of a model analysis unit provided in an embodiment of the present invention;

[0024] Figure 5 This is a block diagram of the internal modules of the rumor refutation unit provided in an embodiment of the present invention.

[0025] In the figure: 1. Data collection unit; 101. Method selection module; 102. Data division module; 103. Classification storage module; 104. Data sampling module; 105. Sample extraction module; 2. Information processing unit; 201. Text feature extraction module; 202. Image feature extraction module; 203. Fusion feature generation module; 3. Model analysis unit; 301. Model construction module; 302. Model training module; 303. Result output module; 4. Rumor refutation unit; 401. Refutation information acquisition module; 402. Information monitoring module; 403. Information push module. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] See also Figure 1-5 ,The present invention provides a technical solution: a system for automatically identifying false public health information, comprising a model analysis unit 3;

[0028] The data collection unit 1 receives a plurality of information data to be detected, determines the data type corresponding to the information data, divides each information data according to the data type, and stores them in a corresponding set, performs sampling operation on the data in the video set, obtains a plurality of image information corresponding to each video data, stores them in the image set, obtains a plurality of information sample data, and adds a corresponding label to each sample data;

[0029] Information processing unit 2, information processing unit 2 converts the received text data into multiple words, analyzes the text semantic feature vector according to the embedding vector corresponding to each word, extracts the feature vector after determining the height, width and number of channels corresponding to the received image information, and analyzes the corresponding fusion feature vector according to the text feature vector and image feature vector corresponding to each information data;

[0030] Rumor rebuttal unit 4: If the news information is judged to be false information, the rumor rebuttal unit 4 will query the database according to the keywords of the false information, and the obtained rebuttal information will be pushed through the information release platform. If the number of views and forwarding of the rebuttal information within 24 hours after release does not reach the threshold, the rebuttal information will be pushed on the platform again.

[0031] The data collection unit 1 includes a mode selection module 101 and a data division module 102. The mode selection module 101 selects a data collection mode, wherein the collection mode includes real-time collection and specific time period collection. If the collection mode is real-time collection, the time interval for each data collection is set. If the collection mode is specific time period collection, the time window information for data collection is set. After receiving a plurality of information data to be detected, the data division module 102 numbers each information data, determines the data type corresponding to the information data, wherein the data type includes text data, picture data and video data, and divides each information data according to the data type.

[0032] The data collection unit 1 also includes a classification storage module 103, a data sampling module 104 and a sample extraction module 105. If the information data contains text data, the classification storage module 103 stores the text data in a text set; if the information data contains picture data, the picture data is stored in a picture set; if the information data contains video data, the video data is stored in a video set. The data sampling module 104 performs sampling operations on the data in the video set to obtain multiple image information corresponding to each video data, and after jointly marking the image information generated by the same video data, stores it in the image set. The sample extraction module 105 obtains multiple information sample data, adds corresponding labels to each sample data, and stores the sample data and labels in a database.

[0033] The information processing unit 2 includes a text feature extraction module 201 and an image feature extraction module 202. The text feature extraction module 201 receives text data and converts them into multiple words after preprocessing. The WordEmbed function is used to analyze each word to obtain a corresponding embedding vector. The text semantic feature vector is determined according to the embedding vector corresponding to each word. The image feature extraction module 202 receives image information and receives image information and determines the height, width and number of channels of the image. The image information is scaled so that the image size is adjusted to a set size, and then the feature vector corresponding to the image information is extracted.

[0034] The information processing unit 2 also includes a fusion feature generation module 203. After the fusion feature generation module 203 determines the text feature vector and the image feature vector corresponding to each information data, it uses a feature splicing algorithm to analyze the two to obtain a fusion feature vector corresponding to the information data. The feature splicing algorithm is specifically:

[0035]

[0036] Among them, T f Represents the text features corresponding to the information data, V f Indicates the image features corresponding to the news data, TV f Represents the fusion feature vector corresponding to the information data, R 64 Indicates that the feature vector dimension is 64;

[0037] The model analysis unit 3 includes a model construction module 301, a model training module 302 and a result output module 303. The model construction module 301 uses deep learning technology to construct a false information recognition model, transmits the fused feature vector corresponding to each information sample data to the model, and after outputting the prediction result, uses the false information loss function to calculate the loss value according to the output result and the label corresponding to the sample data. The model training module 302 updates the model parameters according to the loss value until all sample data are analyzed by the model and the model training operation is completed. After the result output module 303 determines the model parameters and the loss value, it transmits the fused feature vector corresponding to the information data to be detected to the false information recognition model for analysis to obtain the corresponding output result. The false information loss function is specifically:

[0038] L(y i ,f(x i ))=(y i -f(x i )) 2

[0039] Among them, f(x i ) represents the model prediction result, y i represents the label, L(y i ,f(x i )) represents the false information loss function;

[0040] The rumor rebuttal unit 4 includes a rebuttal information acquisition module 401. If the news information is determined to be false information, the rebuttal information acquisition module 401 uses natural language technology to obtain keywords in the false information, searches the database according to the keywords, and outputs the obtained data as rebuttal information through a visual interface. If the news information is determined to be correct information, the determination result is directly output through the visual interface;

[0041] The rumor rebuttal unit 4 also includes an information monitoring module 402 and an information pushing module 403. After the information monitoring module 402 determines the publishing platform of the false information, it pushes the rebuttal information through the same platform, and counts the number of views and forwardings after the rebuttal information is released. The information pushing module 403 sets a threshold value according to the number of views and forwardings of the corresponding false information before the rebuttal information is released. If the number of views and forwardings of the rebuttal information within 24 hours after its release does not reach the threshold, the rebuttal information will be pushed on the platform again.

[0042] Working principle: The present invention selects the data collection method through the method selection module 101 in the data collection unit 1, divides each information data according to the data type by the data division module 102, stores different types of data in the corresponding set by the classification storage module 103, samples the data in the video set by the data sampling module 104, adds corresponding labels to each sample data by the sample extraction module 105, determines the text semantic feature vector of the information data by the text feature extraction module 201 in the information processing unit 2, extracts the feature vector corresponding to the image information by the image feature extraction module 202, fuses the two by the fusion feature generation module 203 to obtain the corresponding fusion feature vector, constructs the false information recognition model by the model construction module 301 in the model analysis unit 3, and after the model training module 302 updates the model parameters, analyzes the information data to be detected by the result output module 303 to obtain the corresponding output result, uses the rebuttal information acquisition module 401 in the rumor rebuttal unit 4 to generate corresponding rebuttal information according to the false information, and the information monitoring module 402 counts the number of views and forwardings after the rebuttal information is released, and performs a push operation according to the obtained data by the information push module 403.

[0043] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0044] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A public health false information automatic identification system, comprising a model analysis unit (3), characterized in that: A data collection unit (1), after receiving a plurality of information data to be detected, the data collection unit (1) determines the data type corresponding to the information data, divides each information data according to the data type, and stores the data in a corresponding set, performs a sampling operation on the data in the video set, obtains a plurality of image information corresponding to each video data, stores the image information in the image set, obtains a plurality of information sample data, and adds a corresponding label to each sample data; An information processing unit (2), wherein the information processing unit (2) converts the received text data into a plurality of words, analyzes a text semantic feature vector according to an embedding vector corresponding to each word, extracts a feature vector after determining the height, width and number of channels corresponding to the received image information, and analyzes a corresponding fusion feature vector according to the text feature vector and the image feature vector corresponding to each information data; A rumor refuting unit (4), wherein if the rumor refuting unit (4) determines that the news information is false information, a query is performed in the database based on the keywords of the false information, and the obtained refuting information is pushed through the information publishing platform. If the number of views and forwardings of the refuting information within 24 hours after the release does not reach a threshold, the refuting information is pushed again on the platform.

2. The public health false information automatic identification system according to claim 1, characterized in that: The data collection unit (1) comprises a mode selection module (101) and a data division module (102). The mode selection module (101) selects a data collection mode, wherein the collection mode comprises real-time collection and specific time period collection. If the collection mode is real-time collection, a time interval for each data collection is set. If the collection mode is specific time period collection, time window information for data collection is set. After receiving a plurality of information data to be detected, the data division module (102) numbers each information data, determines a data type corresponding to the information data, wherein the data type comprises text data, picture data and video data, and divides each information data according to the data type.

3. The public health false information automatic identification system according to claim 2, characterized in that: The data collection unit (1) also includes a classification storage module (103), a data sampling module (104) and a sample extraction module (105). If the information data contains text data, the classification storage module (103) stores the text data in a text set; if the information data contains image data, the image data is stored in a picture set; if the information data contains video data, the video data is stored in a video set. The data sampling module (104) performs a sampling operation on the data in the video set to obtain multiple image information corresponding to each video data, and after jointly marking the image information generated by the same video data, stores it in the image set. The sample extraction module (105) obtains multiple information sample data, adds a corresponding label to each sample data, and stores the sample data and the label in a database.

4. The public health false information automatic identification system according to claim 1, characterized in that: The information processing unit (2) comprises a text feature extraction module (201) and an image feature extraction module (202). The text feature extraction module (201) receives text data and converts the text data into a plurality of words after preprocessing, analyzes each word using the WordEmbed function to obtain a corresponding embedding vector, and determines a text semantic feature vector according to the embedding vector corresponding to each word. The image feature extraction module (202) receives image information and receives image information and determines the height, width and number of channels of the image, and then performs a scaling operation on the image information so that the image size is adjusted to a set size, and then extracts a feature vector corresponding to the image information.

5. The public health false information automatic identification system according to claim 4, characterized in that: The information processing unit (2) also includes a fusion feature generation module (203). After determining the text feature vector and image feature vector corresponding to each information data, the fusion feature generation module (203) uses a feature splicing algorithm to analyze the two to obtain a fusion feature vector corresponding to the information data.

6. The public health false information automatic identification system according to claim 1, characterized in that: The model analysis unit (3) comprises a model construction module (301), a model training module (302) and a result output module (303). The model construction module (301) uses deep learning technology to construct a false information recognition model, transmits the fused feature vector corresponding to each information sample data to the model, and after outputting the prediction result, uses the false information loss function to calculate the loss value according to the output result and the label corresponding to the sample data. The model training module (302) updates the model parameters according to the loss value until all the sample data are analyzed by the model and the model training operation is completed. After the result output module (303) determines the model parameters and the loss value, it transmits the fused feature vector corresponding to the information data to be detected to the false information recognition model for analysis to obtain the corresponding output result.

7. The public health false information automatic identification system according to claim 1, characterized in that: The rumor rebuttal unit (4) includes a rebuttal information acquisition module (401). If the news information is determined to be false information, the rebuttal information acquisition module (401) uses natural language technology to obtain keywords in the false information, searches the database based on the keywords, and outputs the obtained data as rebuttal information through a visual interface. If the news information is determined to be correct information, the determination result is directly output through the visual interface.

8. The public health false information automatic identification system according to claim 7, characterized in that: The rumor rebuttal unit (4) also includes an information monitoring module (402) and an information pushing module (403). After the information monitoring module (402) determines the publishing platform of the false information, it pushes the rebuttal information through the same platform and counts the number of views and forwardings of the rebuttal information after it is published. The information pushing module (403) sets a threshold value according to the number of views and forwardings of the corresponding false information before the rebuttal information is published. If the number of views and forwardings of the rebuttal information within 24 hours after it is published does not reach the threshold value, the rebuttal information is pushed on the platform again.

Images