Rumor detection method, device, electronic device and medium

Abstract

The present invention relates to the field of artificial intelligence and discloses a rumor detection method, device, electronic device, and storage medium that can be used by insurance business personnel in the financial industry to judge current hot topics. The method includes: extracting a text set and corresponding timestamps from an event to be detected; inputting the text set into the upper network of a rumor detection model in the order of the corresponding timestamps to perform rumor detection, obtaining a first rumor detection result corresponding to each text; and simultaneously inputting the timestamp corresponding to each text into the lower network of the rumor detection model to perform rumor detection, obtaining a second rumor detection result corresponding to each timestamp; when the second rumor detection result meets a preset stop condition, stopping rumor detection in the upper network and obtaining the last rumor detection result output by the upper network as the rumor judgment result for the event to be detected. The present invention can improve the accuracy and efficiency of rumor event detection in the financial industry.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a rumor detection method, device, electronic device and readable storage medium. Background Art

[0002] Currently, common rumor detection methods are categorized as either time-series or non-time-series. Non-time-series detection lacks the ability to assess the credibility and spread of current information, resulting in low accuracy. Time-series detection, on the other hand, fails to assess the text of information at an early stage, leading to low detection efficiency. Consequently, existing rumor detection methods are unable to help financial professionals accurately assess the truthfulness of current events, reducing their accuracy in decision-making regarding major events. Summary of the Invention

[0003] The present invention provides a rumor detection method, device, electronic device and readable storage medium, the purpose of which is to improve the accuracy and efficiency of rumor event detection in the financial industry.

[0004] To achieve the above objectives, the present invention provides a rumor detection method, which includes:

[0005] Obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected;

[0006] Inputting the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, obtaining a first rumor detection result corresponding to each text, and inputting the timestamp corresponding to each text into the lower network of the rumor detection model for rumor detection, obtaining a second rumor detection result corresponding to each timestamp, wherein the upper network includes a coding layer, a graph convolution layer, a recurrent neural network layer, and a fully connected layer;

[0007] When the second rumor detection result meets a preset stop condition, stopping the rumor detection of the on-link network, and obtaining the last rumor detection result output by the on-link network as the rumor judgment result of the event to be detected;

[0008] Among them, the timestamp corresponding to each text is input into the lower network of the rumor detection model in turn for rumor detection to obtain a second rumor detection result corresponding to each timestamp, including: performing structural transformation on the timestamp to obtain a graph structure timestamp sequence; using the encoding layer in the lower network of the rumor detection model to encode the graph structure timestamp sequence to obtain a graph structure timestamp encoding vector sequence; using the graph convolution layer in the lower network of the rumor detection model to perform graph convolution operation on the graph structure timestamp encoding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence; splicing the reduced-dimensional graph structure timestamp vector sequence with the hidden layer vector sequence output by the upper network to obtain a spliced ​​vector sequence; using the fully connected layer in the lower network of the rumor detection model to calculate the classification probability of the spliced ​​vector sequence; and calculating the rumor detection result of the timestamp corresponding to each text based on the classification probability to obtain a second rumor detection result.

[0009] Optionally, the step of inputting the text set into an on-board network of a rumor detection model in the order of corresponding timestamps for rumor detection to obtain a first rumor detection result corresponding to each text includes:

[0010] Performing structural transformation on the text set to obtain a graph-structured text set;

[0011] Encode the graph-structured text set using the encoding layer in the rumor detection model on-line network to obtain a graph-structured text encoding vector sequence;

[0012] Using the graph convolution layer in the on-board network of the rumor detection model to perform a graph convolution operation on the graph structure text encoding vector sequence to obtain a dimension-reduced graph structure text vector sequence;

[0013] Utilizing the recurrent neural network layer in the on-board network of the rumor detection model to calculate the hidden layer output of the reduced-dimensional graph structure text vector sequence, to obtain a hidden layer vector sequence;

[0014] Calculate the classification probability of the hidden layer vector sequence using the fully connected layer in the on-board network of the rumor detection model;

[0015] According to the classification probability, the rumor detection result of each text in the text set is calculated to obtain a first rumor detection result.

[0016] Optionally, calculating the classification probability of the hidden layer vector sequence using a fully connected layer in the on-board network of the rumor detection model includes:

[0017] Using the first preset classification parameter, the second preset classification parameter and the third preset classification parameter in the classification layer to perform linear transformation on the hidden layer vector sequence respectively to obtain a query vector, a key vector and a numerical vector;

[0018] Multiply the query vector by the transposed vector of the key vector to obtain a similarity matrix;

[0019] Performing normalization calculation on the similarity matrix to obtain a normalized matrix;

[0020] Calculating the normalized matrix using a preset activation function to obtain an activation matrix;

[0021] The activation matrix is ​​dot-multiplied by the numerical vector to obtain the classification probability of each text rumor detection corresponding to the hidden layer vector sequence.

[0022] Optionally, the encoding layer in the rumor detection model on-line network is used to encode the graph structure text set to obtain a graph structure text encoding vector sequence, including:

[0023] Vectorizing each node and connecting edge in the graph structure text set to obtain a graph structure text vector;

[0024] Performing position index coding on each character in the graph structure text set to obtain a graph structure text position code;

[0025] The graph structure text vector is added to the graph structure text position code to obtain a graph structure text code vector sequence.

[0026] Optionally, the step of performing a graph convolution operation on the graph structure text encoding vector sequence using a graph convolution layer in the on-board network of the rumor detection model to obtain a dimension-reduced graph structure text vector sequence includes:

[0027] Calculate the Laplace matrix of the graph structure text encoding vector sequence according to a preset Laplace operator derivation formula;

[0028] Comparing the preset Fourier transform characteristic function to the characteristic decomposition of the Laplace matrix, and calculating the convolution formula of the graph structure text encoding vector sequence;

[0029] The graph structure text encoding vector sequence is calculated using the convolution formula to obtain a reduced-dimensional graph structure text vector sequence.

[0030] Optionally, the step of calculating the hidden layer output of the reduced-dimensional graph structure text vector sequence using a recurrent neural network layer in the on-board network of the rumor detection model to obtain a hidden layer vector sequence includes:

[0031] Using the hidden state parameter vector in the recurrent neural network layer, a vector change calculation is performed on the first vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer first vector;

[0032] Using the hidden layer first vector to perform a vector change calculation on the second vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer second vector, and sequentially using the hidden layer second vector to perform a vector change calculation on the third vector in the reduced-dimensional graph structure text vector sequence until the vector change calculation is performed on the tail vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer tail vector;

[0033] The hidden layer vector corresponding to each vector in the dimensionality reduction graph structure text vector sequence is normalized to obtain a hidden layer vector sequence.

[0034] In order to solve the above problems, the present invention further provides a rumor detection device, comprising:

[0035] A model input acquisition module is used to obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected;

[0036] A rumor detection module is configured to input the text set into the upper network of the rumor detection model in the order of the corresponding timestamps to perform rumor detection, thereby obtaining a first rumor detection result corresponding to each text, and simultaneously input the timestamp corresponding to each text into the lower network of the rumor detection model to perform rumor detection, thereby obtaining a second rumor detection result corresponding to each timestamp. The upper network includes a coding layer, a graph convolution layer, a recurrent neural network layer, and a fully connected layer.

[0037] a rumor detection result judgment module, configured to stop rumor detection on the on-link network when the second rumor detection result meets a preset stopping condition, and obtain the last rumor detection result output by the on-link network as the rumor judgment result of the event to be detected;

[0038] Among them, the timestamp corresponding to each text is input into the lower network of the rumor detection model in turn for rumor detection to obtain a second rumor detection result corresponding to each timestamp, including: performing structural transformation on the timestamp to obtain a graph structure timestamp sequence; using the encoding layer in the lower network of the rumor detection model to encode the graph structure timestamp sequence to obtain a graph structure timestamp encoding vector sequence; using the graph convolution layer in the lower network of the rumor detection model to perform graph convolution operation on the graph structure timestamp encoding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence; splicing the reduced-dimensional graph structure timestamp vector sequence with the hidden layer vector sequence output by the upper network to obtain a spliced ​​vector sequence; using the fully connected layer in the lower network of the rumor detection model to calculate the classification probability of the spliced ​​vector sequence; and calculating the rumor detection result of the timestamp corresponding to each text based on the classification probability to obtain a second rumor detection result.

[0039] In order to solve the above problem, the present invention further provides an electronic device, comprising:

[0040] a memory storing at least one computer program; and

[0041] The processor executes the computer program stored in the memory to implement the rumor detection method described above.

[0042] In order to solve the above problems, the present invention also provides a computer-readable storage medium, in which at least one computer program is stored. The at least one computer program is executed by a processor in an electronic device to implement the rumor detection method described above.

[0043] The embodiment of the present invention performs rumor detection by sequentially inputting the text collection into the upper network of the rumor detection model in the order of the corresponding timestamps to obtain a first rumor detection result corresponding to each text, and simultaneously sequentially inputting the timestamp corresponding to each text into the lower network of the rumor detection model to perform rumor detection to obtain a second rumor detection result corresponding to each timestamp. Thus, a graph structure model is used to express input data, fusion features, and calculation results, thereby improving the accuracy of rumor detection. Furthermore, when the second rumor detection result meets a preset stop condition, the rumor detection of the upper network is stopped, and the last rumor detection result output by the upper network is obtained as the rumor judgment result of the event to be detected. This allows the text collection to obtain a rumor detection result in the early stages of rumor detection, greatly improving the efficiency of rumor detection. Therefore, the rumor detection method, device, equipment, and storage medium provided by the present invention can improve the accuracy and efficiency of rumor event detection in the financial industry, thereby helping financial workers to judge the truth or falsity of current hot topics and improving the accuracy of financial workers' decision-making on major events. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 A flowchart of a rumor detection method provided by one embodiment of the present invention;

[0045] Figure 2 and Figure 3 A detailed implementation flowchart of one step in the rumor detection method provided in one embodiment of the present invention;

[0046] Figure 4 A schematic diagram of modules of a rumor detection device provided by an embodiment of the present invention;

[0047] Figure 5 A schematic diagram of the internal structure of an electronic device for implementing a rumor detection method provided by one embodiment of the present invention;

[0048] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0049] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0050] An embodiment of the present invention provides a rumor detection method. The execution subject of the rumor detection method includes but is not limited to at least one of the electronic devices such as a server and a terminal that can be configured to execute the method provided by the embodiment of the present application. In other words, the rumor detection method can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server can include an independent server, or it can include a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0051] Reference Figure 1 FIG. 1 is a flow chart of a rumor detection method according to an embodiment of the present invention. In this embodiment of the present invention, the rumor detection method includes:

[0052] S1. Obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected.

[0053] In an embodiment of the present invention, the event to be detected may be user public opinion or current hot topics that are currently widely disseminated. For example, in the financial field, the event to be detected may be user discussion of a certain insurance product. The text collection includes the original event text in the event to be detected and the forwarded / comment text of the original event text by others, for example, the original title text of a hot news in Weibo and all the text in the discussion topic area. The timestamp may be data generated using digital signature technology, and the object of the signature is the signature time information. For example, if the original event text in the text collection is released at 2022.3.24.16:00, then the timestamp of the original event text is 2022.3.24.16:00.

[0054] In an optional embodiment of the present invention, the natural language text in the text set is text data in the financial field, and the text data in the financial field may be new insurance products, etc.

[0055] In an optional embodiment of the present invention, the events to be detected can be obtained by crawling network information with a web crawler. For example, a Python crawler script is set to crawl popular events in Weibo, which reduces the time for manual query of the events to be detected, thereby improving the efficiency of rumor detection.

[0056] Furthermore, in another optional embodiment of the present invention, the required text is extracted from the event to be detected through keyword search to obtain a text set, and then the digital signature information of the text set is queried to obtain a timestamp sequence corresponding to each text in the text set, thereby ensuring that the time relationship can be incorporated into rumor detection, thereby transforming the original non-time series structure model into a time series model, so that the event to be detected can be distinguished as true or false in the early rumor detection.

[0057] S2. Input the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, and obtain the first rumor detection result corresponding to each text. At the same time, input the timestamp corresponding to each text into the lower network of the rumor detection model in sequence for rumor detection, and obtain the second rumor detection result corresponding to each timestamp.

[0058] In an embodiment of the present invention, the rumor detection model can be a temporal structure model including an upper network and a lower network, wherein the upper network includes modules such as a coding layer, a graph convolution layer, a recurrent neural network layer and a fully connected layer, and the lower network includes modules such as a coding layer, a graph convolution layer and a fully connected layer.

[0059] In an optional embodiment of the present invention, since both the upper-link model and the lower-link model of the rumor detection model contain graph convolution layers for performing convolution operations on graph structured data, when the text set is input into the upper-link network of the rumor detection model in the order of corresponding timestamps for rumor detection and the timestamps corresponding to each text are input into the lower-link network of the rumor detection model for rumor detection, it is also necessary to perform structural conversion on the text set and the timestamps, thereby adopting a graph structure model to express the input data, fusion features and calculation results, thereby improving the accuracy of rumor detection.

[0060] In an optional embodiment of the present invention, the rumor detection model can be applied to a variety of different fields, such as insurance product feedback.

[0061] Further, as an optional embodiment of the present invention, refer to Figure 2 As shown, the text set is sequentially input into the on-board network of the rumor detection model in the order of corresponding timestamps for rumor detection, and a first rumor detection result corresponding to each text is obtained, including:

[0062] S21, performing structural transformation on the text set to obtain a graph-structured text set;

[0063] S22. Encode the graph-structured text set using the encoding layer in the rumor detection model on-line network to obtain a graph-structured text encoding vector sequence;

[0064] S23. Using the graph convolution layer in the rumor detection model on-board network, perform a graph convolution operation on the graph structure text encoding vector sequence to obtain a dimensionally reduced graph structure text vector sequence;

[0065] S24. Calculate the hidden layer output of the reduced-dimensional graph structure text vector sequence using the recurrent neural network layer in the on-board network of the rumor detection model to obtain a hidden layer vector sequence;

[0066] S25. Calculate the classification probability of the hidden layer vector sequence using the fully connected layer in the on-board network of the rumor detection model;

[0067] S26. Calculate the rumor detection result of each text in the text set according to the classification probability to obtain a first rumor detection result.

[0068] In an embodiment of the present invention, the encoding layer may be a module that converts the content of a source object into another standard format content according to one standard. For example, in the Bert model, the encoding layer may convert text data into vector data. The graph convolution layer may be composed of two graph convolutional neural networks. The recurrent neural network layer may be composed of a single recurrent neural network, which is mainly used to process and predict sequence data. The function of the fully connected layer may be to map the vector sequence that has passed through the recurrent neural network layer to a specific sample space, thereby integrating it into a value, reducing the influence of the feature position of each vector sequence in the vector sequence on the classification. The first rumor detection result may indicate whether each text in the text set is a rumor.

[0069] In an optional embodiment of the present invention, the text set is converted into a graph-structured text set and input into a rumor detection model, and the graph-structured text set is processed using various modules in the rumor detection model, thereby realizing the use of a vector sequence in the form of a graph structure to express input data, fusion features and calculation results, thereby improving the accuracy of rumor detection.

[0070] Furthermore, as an optional embodiment of the present invention, the structural transformation of the text set to obtain a graph-structured text set includes:

[0071] Extracting the relationship attributes between each text in the text collection;

[0072] A graph structure network is constructed with each text in the text set as a node and the relationship attributes as edges to obtain a graph structure text set.

[0073] In an embodiment of the present invention, the relationship attribute may be the connection between each text. For example, the connection between the initial text and the comment forwarding text may be comment forwarding, and the relationship attribute between the title text and the paragraph text may be a subordinate relationship.

[0074] In an optional embodiment of the present invention, by converting the text set into a graph-structured text set, the connection between texts is made more prominent, so that when rumor detection is performed on the text set, the correlation between texts can be considered, thereby improving the accuracy of rumor detection.

[0075] In detail, in another optional embodiment of the present invention, the coding layer in the rumor detection model on-line network is used to encode the graph structure text set to obtain a graph structure text coding vector sequence, including:

[0076] Vectorizing each node and connecting edge in the graph structure text set to obtain a graph structure text vector;

[0077] Performing position index coding on each character in the graph structure text set to obtain a graph structure text position code;

[0078] The graph structure text vector is added to the graph structure text position code to obtain a graph structure text code vector sequence.

[0079] In the embodiment of the present invention, the graph structure text vector is used to describe the global semantic information of the graph structure text set. The graph structure text position code is used to describe the position information of a single word / phrase in the graph structure text set.

[0080] In an optional embodiment of the present invention, since the semantic information carried by the characters / words appearing at different positions in the text is different, for example: "Shenzhen belongs to Guangdong Province" and "Guangdong Province belongs to Shenzhen", after the graph structure text set is vectorized, it is also necessary to add a different vector to the characters / words at different positions in the graph structure text set for distinction, so that the graph structure text encoding vector sequence can accurately represent the global semantic information and position information of the graph structure text set.

[0081] Furthermore, in an embodiment of the present invention, the graph convolution layer in the rumor detection model on-road network is used to perform a graph convolution operation on the graph structure text encoding vector sequence to obtain a reduced-dimensional graph structure text vector sequence, including:

[0082] Calculate the Laplace matrix of the graph structure text encoding vector sequence according to a preset Laplace operator derivation formula;

[0083] Comparing the preset Fourier transform characteristic function to the characteristic decomposition of the Laplace matrix, and calculating the convolution formula of the graph structure text encoding vector sequence;

[0084] The graph structure text encoding vector sequence is calculated using the convolution formula to obtain a reduced-dimensional graph structure text vector sequence.

[0085] In an optional embodiment of the present invention, the graph structure text encoding vector sequence is input into a graph convolution layer comprising two graph convolutional neural networks, thereby realizing feature extraction of the graph structure text encoding vector sequence, obtaining a reduced-dimensional graph structure text vector sequence, and improving the accuracy of the rumor detection model on-line network.

[0086] Furthermore, in an embodiment of the present invention, the calculation of the hidden layer output of the reduced-dimensional graph structure text vector sequence using the recurrent neural network layer in the on-board network of the rumor detection model to obtain the hidden layer vector sequence includes:

[0087] Using the hidden state parameter vector in the recurrent neural network layer, a vector change calculation is performed on the first vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer first vector;

[0088] Using the hidden layer first vector to perform a vector change calculation on the second vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer second vector, and sequentially using the hidden layer second vector to perform a vector change calculation on the third vector in the reduced-dimensional graph structure text vector sequence until the vector change calculation is performed on the tail vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer tail vector;

[0089] The hidden layer vector corresponding to each vector in the dimensionality reduction graph structure text vector sequence is normalized to obtain a hidden layer vector sequence.

[0090] In an embodiment of the present invention, the hidden state parameter vector may be a parameter vector obtained through multiple training and tuning in the rumor detection model.

[0091] In an optional embodiment of the present invention, by taking the result of the previous vector change as the input of the next vector change, the correlation between the various vectors in the reduced-dimensional graph structure text vector sequence is improved, thereby making the rumor detection result more accurate and reliable.

[0092] Furthermore, in an embodiment of the present invention, the calculation of the classification probability of the hidden layer vector sequence using the fully connected layer in the on-board network of the rumor detection model includes:

[0093] Using the first preset classification parameter, the second preset classification parameter and the third preset classification parameter in the classification layer to perform linear transformation on the hidden layer vector sequence respectively to obtain a query vector, a key vector and a numerical vector;

[0094] Multiply the query vector by the transposed vector of the key vector to obtain a similarity matrix;

[0095] Performing normalization calculation on the similarity matrix to obtain a normalized matrix;

[0096] Calculating the normalized matrix using a preset activation function to obtain an activation matrix;

[0097] The activation matrix is ​​dot-multiplied by the numerical vector to obtain the classification probability of each text rumor detection corresponding to the hidden layer vector sequence.

[0098] In an embodiment of the invention, the first preset classification parameter, the second preset classification parameter, and the third preset classification parameter may be a fully connected layer parameter matrix obtained through multiple training and tuning. The preset activation function may be a softmax activation function.

[0099] In an optional embodiment of the present invention, by performing score normalization calculation on the similarity matrix, the matrix gradient is made more stable and reliable, thereby improving the calculation efficiency of the classification probability of each text rumor detection corresponding to the hidden layer vector sequence.

[0100] In an optional embodiment of the present invention, since the format of timestamps cannot well represent the relationship between each timestamp, before the timestamp corresponding to each text is input into the downlink network of the rumor detection model in turn for rumor detection, the timestamp corresponding to the initial text in the text set needs to be set to zero, and the timestamps corresponding to the remaining texts need to be changed according to the spacing between the timestamps corresponding to the initial text.

[0101] Furthermore, in an embodiment of the present invention, since this scheme adopts a graph structure model to express input data, fusion features and calculation results, when the timestamp corresponding to each text is input into the downstream network of the rumor detection model in turn for rumor detection, it is also necessary to convert the timestamp into a graph structure timestamp. The conversion method is similar to the structural transformation of the text set to obtain a graph structure text set, so it will not be repeated here.

[0102] Further, in an optional embodiment of the present invention, reference is made to Figure 3 As shown, the timestamp corresponding to each text is sequentially input into the downstream network of the rumor detection model for rumor detection, and a second rumor detection result corresponding to each timestamp is obtained, including:

[0103] S27. Perform structural transformation on the timestamp to obtain a graph structure timestamp sequence;

[0104] S28. Encode the graph structure timestamp sequence using the encoding layer in the lower network of the rumor detection model to obtain a graph structure timestamp encoding vector sequence;

[0105] S29. Using the graph convolution layer in the lower network of the rumor detection model, perform a graph convolution operation on the graph structure timestamp encoding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence;

[0106] S210: Concatenate the timestamp vector sequence of the dimensionality reduction graph structure with the hidden layer vector sequence output by the on-link network to obtain a concatenated vector sequence;

[0107] S211. Calculate the classification probability of the concatenated vector sequence using the fully connected layer in the lower network of the rumor detection model;

[0108] S212. Calculate the rumor detection result of the timestamp corresponding to each text according to the classification probability to obtain a second rumor detection result.

[0109] In an embodiment of the present invention, the second rumor detection result may represent whether to terminate rumor detection of the text collection by the on-link network.

[0110] In an embodiment of the present invention, since this scheme adopts a graph structure model to express input data, fusion features and calculation results, when the timestamp corresponding to each text is sequentially input into the downlink network of the rumor detection model for rumor detection, it is also necessary to convert the timestamp into a graph structure timestamp, wherein the structural transformation of the timestamp to obtain a graph structure timestamp sequence is similar to the structural transformation of the text set to obtain a graph structure text set, so it is not repeated here.

[0111] Furthermore, in an optional embodiment of the present invention, the coding layer in the lower network of the rumor detection model is used to encode the graph structure timestamp sequence to obtain a graph structure timestamp coding vector sequence, which is similar to the coding layer in the upper network of the rumor detection model being used to encode the graph structure text set to obtain a graph structure text coding vector sequence. The graph convolution layer in the lower network of the rumor detection model is used to perform a graph convolution operation on the graph structure timestamp coding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence, which is similar to the graph convolution layer in the upper network of the rumor detection model being used to perform a graph convolution operation on the graph structure text coding vector sequence to obtain a reduced-dimensional graph structure text vector sequence. The fully connected layer in the lower network of the rumor detection model is used to calculate the classification probability of the spliced ​​vector sequence, which is similar to the fully connected layer in the upper network of the rumor detection model being used to calculate the classification probability of the hidden layer vector sequence, so it is not repeated here.

[0112] S3. When the second rumor detection result meets the preset stop condition, stop the rumor detection of the on-road network, and obtain the last rumor detection result output by the on-road network as the rumor judgment result of the event to be detected.

[0113] In an embodiment of the present invention, the preset stopping condition may be when the second rumor detection result is displayed as an error.

[0114] In an embodiment of the present invention, in order to save rumor detection time and improve rumor detection efficiency, the first rumor detection result is selected as the rumor judgment result of the event to be detected by judging the second rumor detection result. For example, when rumor detection is performed on a text set containing ten text data, after the second rumor detection result of the timestamp corresponding to the fifth text in the text set is displayed as an error, the first rumor detection result of the fifth text in the text set containing ten text data is taken as the rumor judgment result of the event to be detected corresponding to the text set.

[0115] In an embodiment of the present invention, by taking a timestamp sequence as input, inputting it into the downlink network of rumor detection, and splicing it with the vector sequence of the text set, a transformation of the model structure is achieved, and the non-time series structure model is converted into a time series structure model. Therefore, when the second rumor detection result meets the preset stop condition, the rumor detection of the uplink network is stopped, and the last rumor detection result output by the uplink network is obtained as the rumor judgment result of the event to be detected, so that the text set can obtain the rumor detection result in the early rumor detection, which greatly improves the efficiency of rumor detection.

[0116] The embodiment of the present invention performs rumor detection by sequentially inputting the text collection into the upper network of the rumor detection model in the order of the corresponding timestamps to obtain a first rumor detection result corresponding to each text, and simultaneously sequentially inputting the timestamp corresponding to each text into the lower network of the rumor detection model to perform rumor detection to obtain a second rumor detection result corresponding to each timestamp. Thus, a graph structure model is used to express input data, fusion features, and calculation results, thereby improving the accuracy of rumor detection. Furthermore, when the second rumor detection result meets a preset stop condition, rumor detection in the upper network is stopped, and the last rumor detection result output by the upper network is obtained as the rumor judgment result of the event to be detected. This allows the text collection to obtain rumor detection results in early rumor detection, greatly improving the efficiency of rumor detection. Therefore, the rumor detection method provided by the present invention can improve the accuracy and efficiency of rumor detection events in the financial industry, thereby helping financial workers to judge the truth or falsity of current hot topics and improving the accuracy of financial workers' decision-making on major events.

[0117] like Figure 4 FIG. 1 is a functional module diagram of the rumor detection device of the present invention.

[0118] The rumor detection device 100 described in the present invention can be installed in an electronic device. Depending on the functionality implemented, the rumor detection device 100 may include a model input acquisition module 101, a rumor detection module 102, and a rumor detection result determination module 103. A module, also referred to as a unit, is a series of computer program segments that can be executed by an electronic device processor and perform a fixed function, and is stored in the electronic device's memory.

[0119] In this embodiment, the functions of each module / unit are as follows:

[0120] The model input acquisition module 101 is used to acquire an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected.

[0121] The rumor detection module 102 is used to input the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, and obtain the first rumor detection result corresponding to each text, and at the same time input the timestamp corresponding to each text into the lower network of the rumor detection model for rumor detection, and obtain the second rumor detection result corresponding to each timestamp.

[0122] The rumor detection result judgment module 103 is used to stop the rumor detection of the on-road network when the second rumor detection result meets the preset stopping condition, and obtain the last rumor detection result output by the on-road network as the rumor judgment result of the event to be detected.

[0123] like Figure 5 FIG. 1 is a schematic diagram of the structure of an electronic device for implementing the rumor detection method of the present invention.

[0124] The electronic device may include a processor 10, a memory 11, a communication bus 12 and a communication interface 13, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a rumor detection program.

[0125] The memory 11 includes at least one type of readable storage medium, including a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., SD or DX memory), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 11 may be an internal storage unit of the electronic device, such as a mobile hard disk of the electronic device. In other embodiments, the memory 11 may also be an external storage device of the electronic device, such as a plug-in mobile hard disk, a smart memory card (SMC), a secure digital (SD) card, a flash memory card, etc. equipped on the electronic device. Furthermore, the memory 11 may include both an internal storage unit of the electronic device and an external storage device. The memory 11 can be used not only to store application software and various types of data installed in the electronic device, such as the code of a rumor detection program, but also to temporarily store data that has been output or is about to be output.

[0126] In some embodiments, the processor 10 may be comprised of an integrated circuit, such as a single packaged integrated circuit or multiple packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips. The processor 10 is the control core (control unit) of the electronic device, connecting the various components of the entire electronic device using various interfaces and circuits. It executes programs or modules (such as a rumor detection program) stored in the memory 11 and accesses data stored in the memory 11 to perform various functions of the electronic device and process data.

[0127] The communication bus 12 may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. This bus can be divided into an address bus, a data bus, a control bus, etc. The communication bus 12 is configured to enable communication between the memory 11 and at least one processor 10. For ease of illustration, the figure uses only a single thick line, but this does not indicate that there is only one bus or only one type of bus.

[0128] Figure 5 Only the electronic device with components is shown, and it can be understood by those skilled in the art that Figure 5 The structure shown does not limit the electronic device, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0129] For example, although not shown, the electronic device may further include a power source (e.g., a battery) to power various components. Preferably, the power source may be logically connected to the at least one processor 10 via a power management device, thereby enabling functions such as charge management, discharge management, and power consumption management via the power management device. The power source may further include any of one or more DC or AC power sources, a recharging device, a power failure detection circuit, a power converter or inverter, a power status indicator, and the like. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, and the like, which are not further detailed here.

[0130] Optionally, the communication interface 13 may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device and other electronic devices.

[0131] Optionally, the communication interface 13 may further include a user interface, which may be a display or an input unit (such as a keyboard). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. The display may also be appropriately referred to as a display screen or display unit, and is used to display information processed in the electronic device and to display a visual user interface.

[0132] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0133] The rumor detection program stored in the memory 11 of the electronic device is a combination of multiple computer programs. When running in the processor 10, it can achieve the following:

[0134] Obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected;

[0135] Inputting the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, obtaining a first rumor detection result corresponding to each text; and inputting the timestamp corresponding to each text into the lower network of the rumor detection model for rumor detection, obtaining a second rumor detection result corresponding to each timestamp;

[0136] When the second rumor detection result meets the preset stop condition, the rumor detection of the on-link network is stopped, and the last rumor detection result output by the on-link network is obtained as the rumor judgment result of the event to be detected.

[0137] Specifically, the specific implementation method of the processor 10 for the above computer program can refer to Figure 1 The description of the relevant steps in the corresponding embodiments will not be repeated here.

[0138] Furthermore, if the module / unit integrated into the electronic device is implemented as a software functional unit and sold or used as a standalone product, it may be stored on a computer-readable storage medium. The computer-readable medium may be non-volatile or volatile. The computer-readable medium may include any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).

[0139] An embodiment of the present invention may further provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program. When the computer program is executed by a processor of an electronic device, the computer program may implement:

[0140] Obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected;

[0141] Inputting the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, obtaining a first rumor detection result corresponding to each text; and inputting the timestamp corresponding to each text into the lower network of the rumor detection model for rumor detection, obtaining a second rumor detection result corresponding to each timestamp;

[0142] When the second rumor detection result meets the preset stop condition, the rumor detection of the on-link network is stopped, and the last rumor detection result output by the on-link network is obtained as the rumor judgment result of the event to be detected.

[0143] Furthermore, the computer-usable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0144] In the several embodiments provided herein, it should be understood that the disclosed electronic devices, apparatuses, and methods may be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For example, the module division is merely a logical functional division, and actual implementation may employ other division methods.

[0145] The modules described as separate components may or may not be physically separate, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed across multiple network elements. Some or all of the modules may be selected to achieve the purpose of the solution of this embodiment according to actual needs.

[0146] In addition, the functional modules in various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or hardware plus software functional modules.

[0147] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0148] Therefore, the embodiments should be considered in all respects as illustrative and non-restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes that come within the meaning and range of equivalents of the claims are intended to be embraced therein. Any reference to a figure in a claim should not be construed as limiting the claim to which it relates.

[0149] Blockchain, as used in this article, refers to a novel application model for computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Blockchain is essentially a decentralized database, a series of data blocks generated using cryptographic methods. Each block contains information about a batch of online transactions, used to verify the validity of this information (to prevent counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, the platform product service layer, and the application service layer.

[0150] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices recited in a system claim may also be implemented by a single unit or device through software or hardware. Second-order terms are used to indicate names and do not imply any particular order.

[0151] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A rumor detection method, characterized in that: The method comprises: Obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected; Inputting the text set into the upper network of the rumor detection model in the order of corresponding timestamps for rumor detection, obtaining a first rumor detection result corresponding to each text, and inputting the timestamp corresponding to each text into the lower network of the rumor detection model for rumor detection, obtaining a second rumor detection result corresponding to each timestamp, wherein the upper network includes a coding layer, a graph convolution layer, a recurrent neural network layer, and a fully connected layer; When the second rumor detection result meets a preset stop condition, stopping the rumor detection of the on-link network, and obtaining the last rumor detection result output by the on-link network as the rumor judgment result of the event to be detected; Among them, the timestamp corresponding to each text is input into the lower network of the rumor detection model in turn for rumor detection to obtain a second rumor detection result corresponding to each timestamp, including: performing structural transformation on the timestamp to obtain a graph structure timestamp sequence; using the encoding layer in the lower network of the rumor detection model to encode the graph structure timestamp sequence to obtain a graph structure timestamp encoding vector sequence; using the graph convolution layer in the lower network of the rumor detection model to perform graph convolution operation on the graph structure timestamp encoding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence; splicing the reduced-dimensional graph structure timestamp vector sequence with the hidden layer vector sequence output by the upper network to obtain a spliced ​​vector sequence; using the fully connected layer in the lower network of the rumor detection model to calculate the classification probability of the spliced ​​vector sequence; and calculating the rumor detection result of the timestamp corresponding to each text based on the classification probability to obtain a second rumor detection result.

2. The rumor detection method according to claim 1, wherein: The step of inputting the text set into the on-board network of the rumor detection model in the order of corresponding timestamps to perform rumor detection, and obtaining a first rumor detection result corresponding to each text, includes: Performing structural transformation on the text set to obtain a graph-structured text set; Encode the graph-structured text set using the encoding layer in the rumor detection model on-line network to obtain a graph-structured text encoding vector sequence; Using the graph convolution layer in the on-board network of the rumor detection model to perform a graph convolution operation on the graph structure text encoding vector sequence to obtain a dimension-reduced graph structure text vector sequence; Utilizing the recurrent neural network layer in the on-board network of the rumor detection model to calculate the hidden layer output of the reduced-dimensional graph structure text vector sequence, to obtain a hidden layer vector sequence; Calculate the classification probability of the hidden layer vector sequence using the fully connected layer in the on-board network of the rumor detection model; According to the classification probability, the rumor detection result of each text in the text set is calculated to obtain a first rumor detection result.

3. The rumor detection method according to claim 2, wherein: The calculating the classification probability of the hidden layer vector sequence by using the fully connected layer in the on-board network of the rumor detection model includes: Using the first preset classification parameter, the second preset classification parameter and the third preset classification parameter in the classification layer to perform linear transformation on the hidden layer vector sequence respectively to obtain a query vector, a key vector and a numerical vector; Multiply the query vector by the transposed vector of the key vector to obtain a similarity matrix; Performing normalization calculation on the similarity matrix to obtain a normalized matrix; Calculating the normalized matrix using a preset activation function to obtain an activation matrix; The activation matrix is ​​dot-multiplied by the numerical vector to obtain the classification probability of each text rumor detection corresponding to the hidden layer vector sequence.

4. The rumor detection method according to claim 2, wherein: The coding layer in the rumor detection model on-road network is used to encode the graph structure text set to obtain a graph structure text coding vector sequence, including: Vectorizing each node and connecting edge in the graph structure text set to obtain a graph structure text vector; Performing position index coding on each character in the graph structure text set to obtain a graph structure text position code; The graph structure text vector is added to the graph structure text position code to obtain a graph structure text code vector sequence.

5. The rumor detection method according to claim 2, wherein: The method of performing a graph convolution operation on the graph structure text encoding vector sequence using the graph convolution layer in the on-road network of the rumor detection model to obtain a dimension-reduced graph structure text vector sequence includes: Calculate the Laplace matrix of the graph structure text encoding vector sequence according to a preset Laplace operator derivation formula; Comparing the preset Fourier transform characteristic function to the characteristic decomposition of the Laplace matrix, and calculating the convolution formula of the graph structure text encoding vector sequence; The graph structure text encoding vector sequence is calculated using the convolution formula to obtain a reduced-dimensional graph structure text vector sequence.

6. The rumor detection method according to claim 2, wherein: The step of calculating the hidden layer output of the reduced-dimensional graph structure text vector sequence using the recurrent neural network layer in the on-board network of the rumor detection model to obtain the hidden layer vector sequence includes: Using the hidden state parameter vector in the recurrent neural network layer, a vector change calculation is performed on the first vector in the dimensionality reduction graph structure text vector sequence to obtain the hidden layer first vector; Using the hidden layer first vector to perform a vector change calculation on the second vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer second vector, and sequentially using the hidden layer second vector to perform a vector change calculation on the third vector in the reduced-dimensional graph structure text vector sequence until the vector change calculation is performed on the tail vector in the reduced-dimensional graph structure text vector sequence to obtain a hidden layer tail vector; The hidden layer vector corresponding to each vector in the dimensionality reduction graph structure text vector sequence is normalized and calculated to obtain the hidden layer vector sequence.

7. A rumor detection device, characterized in that: The device comprises: A model input acquisition module is used to obtain an event to be detected, and extract a text set and a timestamp corresponding to each text from the event to be detected; A rumor detection module is configured to input the text set into the upper network of the rumor detection model in the order of the corresponding timestamps to perform rumor detection, thereby obtaining a first rumor detection result corresponding to each text, and simultaneously input the timestamp corresponding to each text into the lower network of the rumor detection model to perform rumor detection, thereby obtaining a second rumor detection result corresponding to each timestamp. The upper network includes a coding layer, a graph convolution layer, a recurrent neural network layer, and a fully connected layer. a rumor detection result judgment module, configured to stop rumor detection on the on-link network when the second rumor detection result meets a preset stopping condition, and obtain the last rumor detection result output by the on-link network as the rumor judgment result of the event to be detected; Among them, the timestamp corresponding to each text is input into the lower network of the rumor detection model in turn for rumor detection to obtain a second rumor detection result corresponding to each timestamp, including: performing structural transformation on the timestamp to obtain a graph structure timestamp sequence; using the encoding layer in the lower network of the rumor detection model to encode the graph structure timestamp sequence to obtain a graph structure timestamp encoding vector sequence; using the graph convolution layer in the lower network of the rumor detection model to perform graph convolution operation on the graph structure timestamp encoding vector sequence to obtain a reduced-dimensional graph structure timestamp vector sequence; splicing the reduced-dimensional graph structure timestamp vector sequence with the hidden layer vector sequence output by the upper network to obtain a spliced ​​vector sequence; using the fully connected layer in the lower network of the rumor detection model to calculate the classification probability of the spliced ​​vector sequence; and calculating the rumor detection result of the timestamp corresponding to each text based on the classification probability to obtain a second rumor detection result.

8. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor to enable the at least one processor to perform the rumor detection method according to any one of claims 1 to 6.

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the rumor detection method according to any one of claims 1 to 6 is implemented.

Images