False information detection method and system, electronic equipment and product

Abstract

The invention belongs to the technical field of data processing, and aims to provide a false information detection method and system, electronic equipment and a product. The method comprises the following steps: capturing rumor refuting data from a specified data source, and constructing a knowledge graph according to the rumor refuting data; receiving to-be-verified data input by a user, and decomposing the to-be-verified data according to the data type to obtain sub to-be-verified data; extracting to-be-verified text data matched with the to-be-verified data from the sub to-be-verified data; and performing logic analysis processing on the to-be-verified text data by utilizing a preset large language model so as to judge whether the to-be-verified text data has a logic problem or not, if so, judging that the to-be-verified data is false information, and if not, performing false information detection processing on the to-be-verified text data based on the knowledge graph so as to obtain the to-be-verified text data. And obtaining a false information detection result of the to-be-verified text data. According to the method, automatic detection of multi-source false information can be realized, and meanwhile, the detection real-time performance is high.

Description

Technical Field

[0001] The present invention belongs to the field of data processing technology, and specifically relates to a false information detection method, system, electronic equipment and product. Background Art

[0002] In the rapid development of social networks today, the efficiency and coverage of information dissemination have reached an unprecedented level. However, the convenience of information dissemination has also led to the widespread dissemination of false information such as rumors. The spread of false information such as rumors is highly concealed and sudden. With the rapid development of artificial intelligence technology, the technology for counterfeiting multimedia content has become more mature. Relevant departments often find it difficult to detect and take measures in time at the early stage of dissemination, resulting in the rapid fermentation of false information among the public, and even causing panic and public opinion crises. Therefore, it is particularly important to strengthen the ability to detect false information.

[0003] Current false information detection methods mostly rely on manual review or analysis tools based on a single modality. However, in the process of using the prior art, the inventors found that the prior art has at least the following problems: Currently, false information is usually presented in various forms, such as text, pictures, videos and audios. These multimodal data greatly increase the complexity of false information identification and monitoring. However, existing technologies have weak detection capabilities for complex and multimodal false information and are difficult to handle complex content involving multimodal information such as text, pictures, audio and video. Especially in social networks, since the spread of false information far exceeds the spread of rumor-refuting information, traditional methods that rely on manual and single-modal analysis are difficult to meet the needs of real-time monitoring and effective response. Summary of the invention

[0004] The present invention aims to solve the above technical problems at least to a certain extent, and provides a false information detection method, system, electronic device and product.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: In a first aspect, the present invention provides a false information detection method, comprising: Capture rumor-debunking data from a specified data source, and construct a knowledge graph based on the rumor-debunking data; Receive the data to be verified input by the user, and decompose the data to be verified according to the data type to obtain sub-data to be verified; wherein the sub-data to be verified includes initial text data, initial picture data, initial audio data and / or initial video data; Extracting text data to be verified that matches the data to be verified from the sub-data to be verified; Performing logic analysis on the text data to be verified using a preset large language model to determine whether the text data to be verified has a logic problem. If so, determining that the data to be verified is false information; if not, proceeding to the next step; Based on the knowledge graph, false information detection processing is performed on the text data to be verified to obtain a false information detection result of the text data to be verified.

[0006] In a possible design, a knowledge graph is constructed based on the rumor-refuting data, including: Performing multi-source feature extraction processing on the rumor-debunking data to obtain multi-source feature data corresponding to the rumor-debunking data; wherein the multi-source feature data includes text feature data, picture feature data, audio feature data, and video feature data; Obtaining the event type of the rumor-refuting data according to the text feature data, and generating an event identifier according to the event type; The text feature data, the picture feature data, the audio feature data, and the video feature data are stored in an object storage system respectively, and storage paths corresponding to the text feature data, the picture feature data, the audio feature data, and the video feature data are generated respectively, and then the text feature data, the picture feature data, the audio feature data, and the video feature data are bound to the event identifier and the corresponding storage path respectively, and then stored in a preset text data table, picture data table, audio data table, and video data table; A knowledge graph is constructed based on the text data table, the picture data table, the audio data table and the video data table.

[0007] In a possible design, after obtaining the multi-source feature data corresponding to the rumor-refuting data, the method further includes: The image feature data, the audio feature data and the video feature data are respectively subjected to text extraction processing to obtain additional text feature data, and the additional text feature data is integrated into the text feature data to obtain final text feature data, so as to extract the event type of the rumor-refuting data from the final text feature data.

[0008] In a possible design, obtaining the event type of the rumor-refuting data according to the text feature data includes: Extracting an event feature vector of the rumor-refuting data from the text feature data; Based on the preset SVM model, the decision function value of the rumor-refuting data is calculated according to the event feature vector; wherein the decision function value is: ; In the formula, x represents the event feature vector; α i Indicates the preset i Lagrange multipliers; n Indicates the number of samples in the preset training set; y i Indicates the first i training samples x i Labels; K ( x , x i ) represents the kernel function, which is used to calculate the i training samples x i and the event feature vector x The similarity between b Represents the preset bias item; According to the decision function value, the event type of the rumor-refuting data is obtained.

[0009] In a possible design, performing false information detection processing on the text data to be verified based on the knowledge graph to obtain a false information detection result of the text data to be verified includes: The nodes in the knowledge graph are embedded into a low-dimensional vector space using a Node2Vec algorithm to obtain node embedding vectors of multiple nodes in the knowledge graph; Performing feature extraction processing on the text data to be verified to obtain feature data to be verified corresponding to the data to be verified, and obtaining an embedding vector to be verified of the feature data to be verified; Using cosine similarity, respectively calculate the similarity between the embedding vector to be verified and the node embedding vectors of multiple nodes in the knowledge graph, and select the node in the knowledge graph corresponding to the node embedding vector with the highest similarity to the embedding vector to be verified as the best matching node; Determine whether the similarity between the embedding vector to be verified and the node embedding vector of the optimal matching node is greater than a preset similarity threshold. If so, obtain the false information detection result of the text data to be verified based on the rumor-refuting data corresponding to the optimal matching node.

[0010] In one possible design, the embedding vector to be verified of the feature data to be verified includes the time embedding vector v to be verified tinput , the location embedding vector v to be verified linput , embedding vector v of the witness to be sought pinput , the embedding vector v of the evidence to be sought oinput and the event type embedding vector veinput ; The node embedding vector of any node in the knowledge graph includes the time embedding vector v tj , location embedding vector v lj , person embedding vector v pj , object embedding vector v oj and the event type embedding vector v ej ; Correspondingly, the similarity between the embedding vector to be verified and the node embedding vector of any node is: ; In the formula, , is the time embedding vector v to be verified tinput With the time embedding vector v tj The similarity between , is the embedding vector v of the location to be verified linput with the location embedding vector v lj The similarity between , is the embedding vector v of the witness to be sought pinput With the person embedding vector v pj The similarity between , is the embedding vector v of the evidence to be sought oinput With the object embedding vector v oj The similarity between Embed vector v for the event type to be verified einput with the event type embedding vector v ej The similarity between .

[0011] In one possible design, the method further includes: The data to be verified input by each user and the false information detection results matching the data are stored in the log database; Obtaining the event types corresponding to all the data to be verified within a specified time range in the log database, and summarizing the frequency of the data to be verified of the same event type to obtain the detection frequency of each event type within the specified time range; Obtain the event level of each event type, and obtain the severity of each event type based on the detection frequency and event level of each event type; For any event type, determine whether its detection frequency is greater than a preset frequency threshold, and whether its severity is greater than a preset severity threshold. If so, issue a false information warning for any event type.

[0012] In a second aspect, the present invention provides a false information detection system, comprising: A knowledge graph construction module is used to capture rumor-refuting data from a specified data source and construct a knowledge graph based on the rumor-refuting data; The data receiving module for to-be-verified is used to receive the data to-be-verified input by the user, and decompose the data to-be-verified according to the data type to obtain sub-data to-be-verified; wherein the sub-data to-be-verified includes initial text data, initial picture data, initial audio data and / or initial video data; The false information detection module is respectively communicatively connected with the knowledge graph construction module and the data to be verified receiving module, and is used to extract the text data to be verified that matches the data to be verified from the sub-data to be verified; and is also used to perform logical analysis processing on the text data to be verified using a preset large language model, so as to determine whether there is a logical problem in the text data to be verified. If so, the data to be verified is determined to be false information. If not, false information detection processing is performed on the text data to be verified based on the knowledge graph, so as to obtain a false information detection result of the text data to be verified.

[0013] In a third aspect, the present invention provides an electronic device, comprising: a memory for storing computer program instructions; and, A processor is used to execute the computer program instructions to complete the operation of a false information detection method as described in any one of the above.

[0014] In a fourth aspect, the present invention provides a computer program product, comprising a computer program or instructions, wherein the computer program or instructions, when executed by a computer, implements a false information detection method as described in any one of the above.

[0015] The beneficial effects of the present invention are: The present invention discloses a false information detection method, system, electronic device and product, which can realize automatic detection of multi-source false information, and the detection is highly real-time. Specifically, during the implementation of the present invention, first, the rumor-refuting data is captured from the designated data source, and a knowledge graph is constructed based on the rumor-refuting data; then, the data to be verified input by the user is received, and the data to be verified is decomposed and processed according to the data type to obtain sub-data to be verified; then, the text data to be verified that matches the data to be verified is extracted from the sub-data to be verified; then, the text data to be verified is logically analyzed and processed using a preset large language model to determine whether there is a logical problem in the text data to be verified. If so, the data to be verified is determined to be false information. If not, the text data to be verified is subjected to false information detection processing based on the knowledge graph to obtain a false information detection result of the text data to be verified. Based on the present invention, false information detection of various types of data to be verified can be achieved, and through automated detection, the real-time detection is higher, which is conducive to reducing social panic and misunderstanding caused by the spread of rumors, thereby ensuring that the public obtains true and reliable information and reducing the social risks brought by false information.

[0016] Other beneficial effects of the present invention will be further described in the specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 is a flow chart of the false information detection method in Example 1; Figure 2 is a module block diagram of the false information detection system in Example 2; Figure 3 This is a module block diagram of the electronic device in Example 3. DETAILED DESCRIPTION

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in combination with the drawings and the description of the embodiments or the prior art. Obviously, the following description of the structure of the drawings is only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. 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.

[0019] Embodiment 1: This embodiment discloses a false information detection method, which can be executed by, but is not limited to, a computer device or a virtual machine with certain computing resources, such as a personal computer, a smart phone, a personal digital assistant, or a wearable device, or by a virtual machine.

[0020] like Figure 1 As shown, a false information detection method may include but is not limited to the following steps: S1. Crawl rumor-refuting data from designated data sources, and construct a knowledge graph based on the rumor-refuting data. Specifically, in this embodiment, the designated data source comes from authoritative official channels, including but not limited to the China Internet Joint Rumor-Refuting Platform, provincial and municipal local regional rumor-refuting platforms, rumor-refuting platforms in various industries, well-known news websites and portals, and official WeChat and Weibo accounts of various government departments; in addition, in this embodiment, when crawling rumor-refuting data, rumor-refuting data is crawled by deploying distributed crawler tools to deeply collect the latest rumor-refuting notifications, official notifications, and related information and other rumor-refuting data.

[0021] In step S1, a knowledge graph is constructed based on the rumor-refuting data, including: S101. Perform multi-source feature extraction on the rumor-debunking data to obtain multi-source feature data corresponding to the rumor-debunking data; wherein the multi-source feature data includes text feature data, image feature data, audio feature data, and video feature data. Specifically, in this embodiment, after the initial rumor-debunking data is captured, it is pre-cleaned to remove invalid data, duplicate data, or noise data therein, and then multi-source feature extraction is achieved by parsing the labels of the rumor-debunking data, and text feature data, image feature data, audio feature data, and video feature data can be extracted.

[0022] In step S101, after obtaining the multi-source feature data corresponding to the rumor-refuting data, the method further includes: Perform text extraction processing on the image feature data, the audio feature data and the video feature data respectively to obtain additional text feature data, and integrate the additional text feature data into the text feature data to obtain final text feature data, so as to extract the event type of the rumor-refuting data from the final text feature data, that is, enter step S102.

[0023] It should be noted that, in this embodiment, the image feature data, the audio feature data and the video feature data can be further subjected to text extraction processing to obtain additional text feature data, and the additional text feature data can be integrated into the text feature data to obtain complete text feature data corresponding to the rumor-refuting data, thereby making subsequent event type recognition results more accurate, thereby improving the accuracy of subsequent false information detection.

[0024] Specifically, in this embodiment, for the image feature data, OCR (Optical Character Recognition) technology is used to perform text extraction processing. If the text information cannot be effectively recognized, the deep learning-based visual big model can be further used to recognize and understand the content in the image feature data, and further obtain text information such as people, objects, behaviors and scenes, and then the recognized text information is evaluated for effectiveness to obtain additional text feature data after effectiveness evaluation.

[0025] For the audio feature data, a large speech recognition model obtained by training with a large amount of annotated data is used for text recognition processing to extract additional text feature data.

[0026] For the video feature data, FFmpeg (an open source, cross-platform multimedia processing tool library) can be used to extract the audio track, and extract frames at time intervals to generate a picture sequence. The extracted audio data is then processed for text recognition through a large speech recognition model, and the extracted frame pictures are processed for text extraction using OCR calculations.

[0027] S102. Obtaining the event type of the rumor-refuting data according to the text feature data, and generating an event identifier according to the event type.

[0028] Specifically, in step S102 of this embodiment, obtaining the event type of the rumor-refuting data according to the text feature data includes: S1021. Extracting the event feature vector of the rumor-debunking data from the text feature data; specifically, in this embodiment, the named entity recognition (NER) technology is applied to deeply mine the event feature vectors in the text feature data, and can further mine the feature vectors of key information such as time, place, person and thing, which are not limited here; wherein, in view of the relatively infrequent changes in detailed addresses, and the existence of complete and reliable surveying and mapping data that has been pre-trained by the large language model, the large language model can use the contextual information of the address and the known address pattern to accurately identify and extract the detailed address. Therefore, after extracting the location feature vector of the rumor-debunking data by the NER technology, this embodiment uses the large language model to accurately extract the detailed geographic location information therein.

[0029] S1022. Based on a preset SVM (Support Vector Machine) model, a decision function value of the rumor-refuting data is calculated according to the event feature vector; wherein the decision function value is: ; In the formula, x represents the event feature vector; α i Indicates the preset i Lagrange multipliers; n Indicates the number of samples in the preset training set; y i Indicates the first i training samples x i Labels; K ( x , x i ) represents the kernel function, which is used to calculate the i training samples x i and the event feature vector x The similarity between b represents the preset bias term; specifically, the Lagrange multiplier α i and the bias term b These are the parameters solved by the SVM model during the training process; about the sample label y i , for the binary classification problem, y i The value of is ±1. For multi-classification problems (such as multiple event types), the "one-to-many" or "one-to-one" strategy is usually used to convert the multi-classification problem into a binary classification problem for processing. y i Indicates i training samples x i A label that belongs to a specific category.

[0030] S1023. According to the decision function value, the event type of the rumor-refuting data is obtained. Specifically, in this embodiment, the event category described in the rumor-refuting data is determined according to the decision function value. In actual operation, the program will calculate the decision function value of the extracted event text and each category respectively, and select the category with the maximum confidence as the final event classification result. In addition, in this embodiment, a method of fine-tuning and pre-training a large language model can be used in advance to obtain a variety of event types, and manually annotate a training data set containing multi-level types such as event categories, subcategories, and small categories, such as preset [Emergency] (major category) -> [Earthquake] (subcategory) -> [Man-made earthquake\collapse earthquake\tectonic earthquake\volcanic earthquake, etc.] (small category) to divide the rumor-refuting data into event types.

[0031] S103. The text feature data, the picture feature data, the audio feature data and the video feature data are stored in an object storage system (a distributed storage technology for storing a large amount of unstructured data), and storage paths corresponding to the text feature data, the picture feature data, the audio feature data and the video feature data are generated respectively, and then the text feature data, the picture feature data, the audio feature data and the video feature data are bound to the event identifier and the corresponding storage path, and then stored in a preset text data table, picture data table, audio data table and video data table. Specifically, the text feature data is bound to the event identifier and the storage path corresponding to the text feature data, and stored in a preset text data table; the picture feature data is bound to the event identifier and the storage path corresponding to the picture feature data, and then stored in a preset picture data table; the audio feature data is bound to the event identifier and the storage path corresponding to the audio feature data, and then stored in a preset audio data table; the video feature data is bound to the event identifier and the storage path corresponding to the video feature data, and then stored in a preset video data table.

[0032] It should be noted that object storage is an object-centric storage method that stores data as objects rather than files, and has the advantages of high reliability, high scalability, and high performance. Based on this, this embodiment stores the text feature data, the image feature data, the audio feature data, and the video feature data in an object storage system, wherein the object storage system can be implemented by, but is not limited to, Alibaba Cloud OSS (a fully managed cloud storage service provided by Alibaba Cloud) and MinIO (an open source object storage system), etc., and is not limited here.

[0033] In addition, in this embodiment, the relational database MySQL is used to store the text data table, the picture data table, the audio data table and the video data table, so as to realize structured management of data; during the implementation process, all relevant text, picture, audio and video information of the corresponding event type can be easily queried through the event identifier.

[0034] Specifically, the text data tables in this embodiment include a rumor information table, a rumor refuting information table and an official notification table. During the implementation process, the text feature data is pre-segmented and refined through a natural language processing (NLP) word segmenter so as to distinguish different categories of information such as rumor information, rumor refuting statements and official notifications, and stored in the rumor information table, the rumor refuting information table and the official notification table respectively.

[0035] It should also be noted that, in this embodiment, before data is stored, it is checked in advance based on the feature vectors in the text feature data whether the text feature data exists in the text data table to avoid the problem of duplicate storage.

[0036] S104. Construct a knowledge graph based on the text data table, the picture data table, the audio data table and the video data table.

[0037] Specifically, in step S104 of this embodiment, the text data table, the picture data table, the audio data table and the video data table can be integrated to ensure effective associations between various types of data, and the integrated data can be converted into a graph data format. Then, using the Neo4j graph database (a high-performance NoSQL graph database that stores structured data on the network instead of in a table, also known as a graph database management system), a comprehensive knowledge graph containing multi-dimensional information is constructed to provide strong data support for the detection, early warning and response to false information such as rumors.

[0038] S2. Receive the data to be verified input by the user, and decompose the data to be verified according to the data type to obtain sub-data to be verified; wherein the sub-data to be verified includes initial text data, initial image data, initial audio data and / or initial video data. Specifically, in this embodiment, the data to be verified input by the user can be obtained through verification channels such as H5 (HTML5, a standard programming language for building web pages), applets and APP (Application), etc., such as rumor-refuting information verification, media verification, and rumor clue submission, which are not limited here.

[0039] In this embodiment, with the help of a multi-stage reasoning method, a complex reasoning task can be decomposed into multiple smaller sub-tasks (such as sub-data to be verified of different data types corresponding to different sub-tasks) through MQ (Message Queue, a method of communication between processes or between different threads of the same process). Each sub-task is responsible for processing a part of the information detection transaction, and finally the results of each sub-task are integrated to obtain the final result, thereby effectively reducing the complexity of the problem and improving the accuracy and reliability of the results.

[0040] S3. Extracting text data to be verified that matches the data to be verified from the sub-data to be verified.

[0041] Specifically, in this embodiment, the number of the sub-data to be verified is one or more, and the data type of the sub-data to be verified is text, picture, audio and / or video, which is not limited here. If the sub-data to be verified only includes initial text data, the initial text data can be directly used as the text data to be verified that matches the data to be verified; If the sub-data to be verified only includes initial picture data, initial audio data and / or initial video data, the picture data, audio data and / or video data in the sub-data to be verified are subjected to text conversion processing to obtain converted text data, and then the converted text data is used as the text data to be verified that matches the data to be verified; If the sub-data to be verified includes initial text data and also includes initial picture data, initial audio data and / or initial video data, the picture data, audio data and / or video data in the sub-data to be verified are first converted into text to obtain converted text data, and the converted text data and the initial text data are combined to obtain text data to be verified that matches the data to be verified.

[0042] In this embodiment, the method of converting the image data, audio data and / or video data into text is the same as the method of extracting text from the image feature data, the audio feature data and the video feature data in the aforementioned step S101, and will not be repeated here.

[0043] S4. Use the preset large language model to perform logical analysis on the text data to be verified in order to determine whether there is a logical problem in the text data to be verified. If so, determine that the data to be verified is false information. If not, proceed to the next step, that is, enter step S5.

[0044] As an example, in this embodiment, the large language model can be but is not limited to the Tongyi Qianwen large model, the Zidong Taichu large model, the Doubao large model or the ChatGPT series large model, which is not limited here.

[0045] As an example, for example, the data to be verified input by the user is the following text data: "On × year × month × day, × city subway line 1 was completely shut down due to technical failure, resulting in more than 5 million people stranded at the station." In this embodiment, the large language model can be used to perform logical analysis on it to check whether the causal relationship therein is reasonable. According to the large language model, it can be known that "technical failure" leading to "full line shutdown" is a reasonable causal relationship; the large language model can also further find out whether there are contradictory statements in the text data, and confirm whether the date "× year × month × day" is reasonable, and then identify whether there is implicit information in the text. For example, "causing more than 5 million people to be stranded at the station" implies a very serious impact, and the large language model usually needs to further verify the accuracy of this number. The actual passenger flow of the public data of × city subway line 1 through the information collected in the past is very different from 5 million. Therefore, based on the large language model, it can be known that there are logical problems in the text data to be verified, and the data to be verified can be determined to be false information.

[0046] S5. Perform false information detection processing on the text data to be verified based on the knowledge graph to obtain a false information detection result of the text data to be verified. Specifically, in this embodiment, the false information detection result includes that the data to be verified is false information, the data to be verified is not false information, and it is temporarily impossible to confirm whether the data to be verified is false information, etc., wherein, when the data to be verified is false information, the false information detection result may further include rumor-refuting data and official notifications that match the data to be verified, and is not limited here.

[0047] In step S5 of this embodiment, when the false information detection processing is performed on the text data to be verified based on the knowledge graph, an optimal matching algorithm is adopted to match the text data to be verified with the nodes and edges of the knowledge graph.

[0048] Specifically, in step S5, false information detection processing is performed on the text data to be verified based on the knowledge graph to obtain a false information detection result of the text data to be verified, including: S501. Embed the nodes in the knowledge graph into a low-dimensional vector space using the Node2Vec algorithm to obtain node embedding vectors of multiple nodes in the knowledge graph. The Node2Vec algorithm is an algorithm for graph embedding that combines the advantages of depth-first search and breadth-first search to generate a low-dimensional feature representation of the nodes in the graph, and can effectively capture the neighborhood structure of the nodes in the graph, thereby generating high-quality node embeddings.

[0049] S502. Perform feature extraction processing on the text data to be verified to obtain feature data to be verified corresponding to the data to be verified, and obtain an embedding vector to be verified of the feature data to be verified.

[0050] S503. Use cosine similarity to calculate the similarity between the embedding vector to be verified and the node embedding vectors of multiple nodes in the knowledge graph, and select the node in the knowledge graph corresponding to the node embedding vector with the highest similarity to the embedding vector to be verified as the optimal matching node.

[0051] Specifically, in this embodiment, the feature data to be verified obtained by performing feature extraction processing on the text data to be verified includes time feature data t input , location feature data l input , Personal characteristics data p input , object feature data o input and event type feature data e input The embedding vector to be verified of the feature data to be verified includes the time embedding vector v to be verified tinput , the location embedding vector v to be verified linput , embedding vector v of the witness to be sought pinput , the embedding vector v of the evidence to be sought oinput and the event type embedding vector v einput ; Any node in the knowledge graph includes a time node t j , Location Node l j , Human Node p j , Object Node o j and event type nodes e j , the node embedding vector of any node in the knowledge graph includes the time embedding vector v tj , location embedding vector v lj , person embedding vector v pj , object embedding vector v oj and the event type embedding vector v ej ; Correspondingly, in step S503, the similarity between the embedding vector to be verified and the node embedding vector of any node is: ; In the formula, , is the time embedding vector v to be verified tinput With the time embedding vector vtj The similarity between , is the embedding vector v of the location to be verified linput with the location embedding vector v lj The similarity between , is the embedding vector v of the witness to be sought pinput With the person embedding vector v pj The similarity between , is the embedding vector v of the evidence to be sought oinput With the object embedding vector v oj The similarity between Embed vector v for the event type to be verified einput with the event type embedding vector v ej It should be understood that ||·|| represents the norm of a vector, which is used to measure the size of a vector.

[0052] It should be noted that the similarity between the embedding vector to be verified and the node embedding vector of any node is a comprehensive similarity, based on which the optimal matching node can be extracted.

[0053] S504. Determine whether the similarity between the embedding vector to be verified and the node embedding vector of the optimal matching node is greater than a preset similarity threshold. If so, obtain the false information detection result of the text data to be verified based on the rumor-refuting data corresponding to the optimal matching node.

[0054] In addition, in step S504, when determining whether the similarity between the embedding vector to be verified and the node embedding vector of the optimal matching node is greater than a preset similarity threshold, if not, when the sub-data to be verified includes initial image data, image processing and forgery detection are performed on it through a preset AI (Artificial Intelligence) model, and when it is detected as a forged image, the data to be verified is determined to be false information.

[0055] Specifically, in this embodiment, the open source project tool DeepFakeDefenders is used to perform image forgery detection and processing. This tool is an open source AI model developed by the VisionRush team of the Institute of Automation, Chinese Academy of Sciences. Its main function is to identify and defend against forged images and videos generated by Deepfake technology. By analyzing tiny pixel changes in media content, this model can effectively detect Deepfake, help users distinguish between authenticity and falsehood, and reduce the spread of false information and potential risks of abuse.

[0056] Furthermore, if it is impossible to confirm whether the text data to be verified is false information based on the knowledge graph and the image forgery detection, and the time contained in the verified content is within 48 hours of the preset server time, then the distributed crawler technology can be further used to conduct a real-time in-depth search for hot news, and the hot news content and the verification data can be matched through the reasoning ability of the large language model to determine whether there is similar rumor-refuting information and official notification. If it is consistent with the marked rumor-refuting information, the data to be verified is determined to be false information, and the output is a rumor status and the official notification information is given to ensure that the identification result is authoritative and reliable.

[0057] If it is impossible to confirm whether the data to be verified is false information based on the knowledge graph, image forgery detection, and real-time in-depth search of hot news, a detection result that temporarily cannot confirm whether the data to be verified is false information is output, and rumor-refuting data and official notifications with the highest similarity to the data to be verified are output for users to refer to and confirm whether the data to be verified is false information.

[0058] In this embodiment, the method further includes: A1. Store the data to be verified input by each user and the false information detection results matching it in the log database; A2. Obtain the event type corresponding to all the data to be verified within the specified time range in the log database, and summarize the frequency of the data to be verified for the same event type to obtain the detection frequency of each event type within the specified time range; A3. Obtain the event level of each event type, and obtain the severity of each event type based on the detection frequency and event level of each event type; A4. For any event type, determine whether its detection frequency is greater than a preset frequency threshold, and whether its severity is greater than a preset severity threshold. If so, issue a false information warning for any event type.

[0059] It should be noted that in this embodiment, when each user performs false information detection, the data to be verified input by the user and the false information detection results matching the data will be stored in the log database. This embodiment can realize automatic early warning of false information by analyzing the user detection situation within a specified time range, thereby realizing early identification and effective prevention and control of the risk of false information propagation, which will help the Internet information department to quickly identify false information such as rumors in the early stages of the spread, and effectively curb the further spread of false information such as rumors.

[0060] This embodiment can realize the automatic detection of multi-source false information, and the detection is highly real-time. Specifically, in the implementation process of this embodiment, first, the rumor-refuting data is captured from the designated data source, and a knowledge graph is constructed based on the rumor-refuting data; then the data to be verified input by the user is received, and the data to be verified is decomposed and processed according to the data type to obtain sub-data to be verified; then, the text data to be verified that matches the data to be verified is extracted from the sub-data to be verified; then, the text data to be verified is logically analyzed and processed using a preset large language model to determine whether there is a logical problem in the text data to be verified. If so, the data to be verified is determined to be false information. If not, the text data to be verified is subjected to false information detection processing based on the knowledge graph to obtain the false information detection result of the text data to be verified. Based on this embodiment, false information detection of multiple types of data to be verified can be realized, and through automated detection, the real-time detection is higher, which is conducive to reducing social panic and misunderstanding caused by the spread of rumors, thereby ensuring that the public obtains true and reliable information and reduces the social risks brought by false information.

[0061] Embodiment 2: This embodiment discloses a false information detection system for implementing the false information detection method in Embodiment 1; Figure 2 As shown, the false information detection system includes: A knowledge graph construction module is used to capture rumor-refuting data from a specified data source and construct a knowledge graph based on the rumor-refuting data; The data receiving module for to-be-verified is used to receive the data to-be-verified input by the user, and decompose the data to-be-verified according to the data type to obtain sub-data to-be-verified; wherein the sub-data to-be-verified includes initial text data, initial picture data, initial audio data and / or initial video data; The false information detection module is respectively communicatively connected with the knowledge graph construction module and the data to be verified receiving module, and is used to extract the text data to be verified that matches the data to be verified from the sub-data to be verified; and is also used to perform logical analysis processing on the text data to be verified using a preset large language model, so as to determine whether there is a logical problem in the text data to be verified. If so, the data to be verified is determined to be false information. If not, false information detection processing is performed on the text data to be verified based on the knowledge graph, so as to obtain a false information detection result of the text data to be verified.

[0062] It should be noted that the working process, working details and technical effects of the false information detection system provided in this embodiment 2 can be found in embodiment 1 and will not be repeated here.

[0063] Embodiment 3: Based on Embodiment 1 or 2, this embodiment discloses an electronic device, which may be a smart phone, a tablet computer, a laptop computer, or a desktop computer. The electronic device may be referred to as a user terminal, a portable terminal, a desktop terminal, etc. Figure 3 As shown, the electronic equipment includes: a memory for storing computer program instructions; and, A processor is used to execute the computer program instructions to complete the operation of a false information detection method as described in any one of Example 1.

[0064] Specifically, the processor 301 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 301 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 301 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 301 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen.

[0065] The memory 302 may include one or more computer-readable storage media, which may be non-transitory. The memory 302 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 302 is used to store at least one instruction, which is used to be executed by the processor 301 to implement the false information detection method provided in Example 1 of the present application.

[0066] In some embodiments, the terminal may further optionally include: a communication interface 303 and at least one peripheral device. The processor 301, the memory 302 and the communication interface 303 may be connected via a bus or a signal line. Each peripheral device may be connected to the communication interface 303 via a bus, a signal line or a circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 304, a display screen 305 and a power supply 306.

[0067] The communication interface 303 may be used to connect at least one peripheral device related to I / O (Input / Output) to the processor 301 and the memory 302. In some embodiments, the processor 301, the memory 302, and the communication interface 303 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 301, the memory 302, and the communication interface 303 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

[0068] The radio frequency circuit 304 is used to receive and transmit RF (Radio Frequency) signals, also known as electromagnetic signals. The radio frequency circuit 304 communicates with a communication network and other communication devices through electromagnetic signals.

[0069] The display screen 305 is used to display a UI (User Interface). The UI may include any combination of graphics, text, icons, and videos.

[0070] The power supply 306 is used to supply power to various components in the electronic device.

[0071] Embodiment 4: Based on any one of Embodiments 1 to 3, this embodiment discloses a computer program product, including a computer program or instructions, which, when executed by a computer, implements a false information detection method as described in any one of Embodiments 1. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0072] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, 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. Thus, the present invention is not limited to any specific combination of hardware and software.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A false information detection method, characterized in that: include: Capture rumor-debunking data from a specified data source, and construct a knowledge graph based on the rumor-debunking data; Receive the data to be verified input by the user, and decompose the data to be verified according to the data type to obtain sub-data to be verified; wherein the sub-data to be verified includes initial text data, initial picture data, initial audio data and / or initial video data; Extracting text data to be verified that matches the data to be verified from the sub-data to be verified; Performing logic analysis on the text data to be verified using a preset large language model to determine whether the text data to be verified has a logic problem. If so, determining that the data to be verified is false information; if not, proceeding to the next step; Based on the knowledge graph, false information detection processing is performed on the text data to be verified to obtain a false information detection result of the text data to be verified.

2. A false information detection method according to claim 1, characterized in that: A knowledge graph is constructed based on the rumor-refuting data, including: Performing multi-source feature extraction processing on the rumor-debunking data to obtain multi-source feature data corresponding to the rumor-debunking data; wherein the multi-source feature data includes text feature data, picture feature data, audio feature data, and video feature data; Obtaining the event type of the rumor-refuting data according to the text feature data, and generating an event identifier according to the event type; The text feature data, the picture feature data, the audio feature data, and the video feature data are stored in an object storage system respectively, and storage paths corresponding to the text feature data, the picture feature data, the audio feature data, and the video feature data are generated respectively, and then the text feature data, the picture feature data, the audio feature data, and the video feature data are bound to the event identifier and the corresponding storage path respectively, and then stored in a preset text data table, picture data table, audio data table, and video data table; A knowledge graph is constructed based on the text data table, the picture data table, the audio data table and the video data table.

3. A false information detection method according to claim 2, characterized in that: After obtaining the multi-source feature data corresponding to the rumor-refuting data, the method further includes: The image feature data, the audio feature data and the video feature data are respectively subjected to text extraction processing to obtain additional text feature data, and the additional text feature data is integrated into the text feature data to obtain final text feature data, so as to extract the event type of the rumor-refuting data from the final text feature data.

4. A false information detection method according to claim 2, characterized in that: The event type of the rumor-refuting data is obtained according to the text feature data, including: Extracting an event feature vector of the rumor-refuting data from the text feature data; Based on the preset SVM model, the decision function value of the rumor-refuting data is calculated according to the event feature vector; wherein the decision function value is: ； In the formula, x represents the event feature vector; α i Indicates the preset i Lagrange multipliers; n Indicates the number of samples in the preset training set; y i Indicates the first i training samples x i Labels; K ( x , x i ) represents the kernel function, which is used to calculate the i training samples x i and the event feature vector x The similarity between b Represents the preset bias item; According to the decision function value, the event type of the rumor-refuting data is obtained.

5. A false information detection method according to claim 1, characterized in that: Performing false information detection processing on the text data to be verified based on the knowledge graph to obtain a false information detection result of the text data to be verified, including: The nodes in the knowledge graph are embedded into a low-dimensional vector space using a Node2Vec algorithm to obtain node embedding vectors of multiple nodes in the knowledge graph; Performing feature extraction processing on the text data to be verified to obtain feature data to be verified corresponding to the data to be verified, and obtaining an embedding vector to be verified of the feature data to be verified; Using cosine similarity, respectively calculate the similarity between the embedding vector to be verified and the node embedding vectors of multiple nodes in the knowledge graph, and select the node in the knowledge graph corresponding to the node embedding vector with the highest similarity to the embedding vector to be verified as the best matching node; Determine whether the similarity between the embedding vector to be verified and the node embedding vector of the optimal matching node is greater than a preset similarity threshold. If so, obtain the false information detection result of the text data to be verified based on the rumor-refuting data corresponding to the optimal matching node.

6. A false information detection method according to claim 5, characterized in that: The embedding vector to be verified of the feature data to be verified includes a time embedding vector v to be verified. tinput , the location embedding vector v to be verified linput , embedding vector v of the witness to be sought pinput , the embedding vector v of the evidence to be sought oinput and the event type embedding vector v einput ; The node embedding vector of any node in the knowledge graph includes the time embedding vector v tj , location embedding vector v lj , person embedding vector v pj , object embedding vector v oj and the event type embedding vector v ej ; Correspondingly, the similarity between the embedding vector to be verified and the node embedding vector of any node is: ； In the formula, , is the time embedding vector v to be verified tinput With the time embedding vector v tj The similarity between , is the embedding vector v of the location to be verified linput with the location embedding vector v lj The similarity between , is the embedding vector v of the witness to be sought pinput With the person embedding vector v pj The similarity between , is the embedding vector v of the evidence to be sought oinput With the object embedding vector v oj The similarity between Embed vector v for the event type to be verified einput with the event type embedding vector v ej The similarity between .

7. A false information detection method according to claim 1, characterized in that: The method further comprises: The data to be verified input by each user and the false information detection results matching the data are stored in the log database; Obtaining the event types corresponding to all the data to be verified within a specified time range in the log database, and summarizing the frequency of the data to be verified of the same event type to obtain the detection frequency of each event type within the specified time range; Obtain the event level of each event type, and obtain the severity of each event type based on the detection frequency and event level of each event type; For any event type, determine whether its detection frequency is greater than a preset frequency threshold, and whether its severity is greater than a preset severity threshold. If so, issue a false information warning for any event type.

8. A false information detection system, characterized in that: include: A knowledge graph construction module is used to capture rumor-refuting data from a specified data source and construct a knowledge graph based on the rumor-refuting data; The data receiving module for to-be-verified is used to receive the data to-be-verified input by the user, and decompose the data to-be-verified according to the data type to obtain sub-data to-be-verified; wherein the sub-data to-be-verified includes initial text data, initial picture data, initial audio data and / or initial video data; The false information detection module is respectively communicatively connected with the knowledge graph construction module and the data to be verified receiving module, and is used to extract the text data to be verified that matches the data to be verified from the sub-data to be verified; and is also used to perform logical analysis processing on the text data to be verified using a preset large language model, so as to determine whether there is a logical problem in the text data to be verified. If so, the data to be verified is determined to be false information. If not, false information detection processing is performed on the text data to be verified based on the knowledge graph, so as to obtain a false information detection result of the text data to be verified.

9. An electronic device, characterized in that: include: a memory for storing computer program instructions; as well as, A processor, configured to execute the computer program instructions to complete the operation of a false information detection method as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program or instructions, characterized in that When the computer program or the instruction is executed by a computer, the false information detection method as claimed in any one of claims 1 to 7 is implemented.

Images