Early misinformation detection method, apparatus, device, and medium
By segmenting the dataset of false information and performing time-series detection, and combining neural network technology to dynamically determine the detection points, the problem of insufficient accuracy and timeliness in early false information detection was solved, and more efficient false information identification was achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PING AN TECH (SHENZHEN) CO LTD
- Filing Date
- 2023-05-31
- Publication Date
- 2026-07-21
AI Technical Summary
Existing methods for detecting early misinformation suffer from low accuracy or insufficient timeliness due to improper setting of static detection points, making them unable to effectively identify early misinformation.
By segmenting the misinformation dataset into blocks, semantic features are extracted and temporal detection is performed using a bidirectional gated recurrent detection network. The neural Hawkes layer and continuous time-scale long short-term memory layer are combined to capture misinformation classification reversal information, dynamically determine detection points, and feed them back to the bidirectional gated recurrent detection network to obtain target classification results.
It improves the accuracy and timeliness of early detection of false information, ensuring accurate identification of false information within a preset time period and reducing errors caused by improper setting of detection points.
Smart Images
Figure CN117474001B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of artificial intelligence, and in particular to a method, apparatus, device, and storage medium for detecting early false information. Background Technology
[0002] False information generally refers to unverified statements or claims. Early false information typically refers to false information detected within a few hours of its creation. For example, in the financial industry, false information about a new financial product might be detected within four hours of its launch.
[0003] Existing methods for detecting early-stage misinformation involve clustering misinformation data, selecting features from the clusters to determine static misinformation detection points, and then classifying the misinformation data using a classifier based on these static detection points. However, this existing method relies on static detection points. Since different types of misinformation reach their peak online at different times, some misinformation may fail to be detected accurately if the detection points are set too early (e.g., early-stage misinformation may not be detected), while others may fail to be detected in a timely manner if the detection points are set too late, resulting in low accuracy in detecting early-stage misinformation. Summary of the Invention
[0004] This invention provides a method, apparatus, device, and storage medium for detecting early false information, the main purpose of which is to improve the accuracy of early false information detection.
[0005] To achieve the above objectives, the present invention provides an early method for detecting false information, comprising:
[0006] A dataset of false information to be detected within a preset time period is obtained. The dataset is then divided into blocks to obtain a set of false information data blocks. Semantic features of false information in the data blocks are then extracted.
[0007] The bidirectional gated recurrent detection network in the trained false information detection model is used to detect temporal false information in the semantic feature set of false information, so as to obtain the initial temporal false information classification result of the false information dataset.
[0008] The neural Hawkes layer in the detection point network of the aforementioned false information detection model is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period;
[0009] The hidden state of the false information classification inversion information is calculated using the continuous time scale long short-term memory layer in the detection point network. Based on the hidden state and the preset intensity function, the false information detection points of the false information dataset to be detected within the preset time period are determined.
[0010] The false information detection points are fed back to the bidirectional gated cyclic detection network to obtain the target false information classification result of the false information dataset.
[0011] Optionally, calculating the hidden state of the misinformation classification reversal information using the continuous-time-scale long short-term memory layer in the detection point network includes:
[0012] The activation value of the false information classification inversion information is calculated using the input gates in the continuous time-scale long short-term memory layer.
[0013] The state value of the false information classification reversal information is calculated using the forget gate in the continuous time-scale long short-term memory layer;
[0014] Based on the activation value and the state value, the hidden state of the false information classification inversion information is calculated using the output gate in the continuous time scale long short-term memory layer.
[0015] Optionally, determining the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and a preset intensity function includes:
[0016] The difference between the false information classification and reversal information at each moment within the preset time period and the final moment within the preset time period is calculated based on the hidden state and the preset intensity function.
[0017] Determine whether the difference is zero;
[0018] When the difference is not zero, the hidden state of the false information classification reversal information is recalculated, and the difference between the false information classification reversal information at the current time and the final time in the preset time region is calculated again according to the hidden state and the preset intensity function, until the difference is zero.
[0019] When the difference is zero, the time corresponding to the difference is determined as the false information detection point of the false information dataset.
[0020] Optionally, the step of capturing misinformation classification reversal information between the initial temporal misinformation classification results of every two adjacent moments within the preset time period using the neural Hawkes layer in the detection point network of the misinformation detection model includes:
[0021] The neural Hawkes layer is used to identify whether the initial false information classification result at the current moment in the preset time region is consistent with the initial false information classification result at the previous moment.
[0022] When the initial false information classification result at the current moment is consistent with the initial false information classification result at the previous moment, it is determined that the false information classification information at the current moment has not been reversed, and the current moment point that has not been reversed is assigned the first value.
[0023] When the initial misinformation classification result at the current moment is inconsistent with the initial misinformation classification result at the previous moment, it is determined that the misinformation classification information at the current moment has been reversed, and the current moment point where the reversal has occurred is assigned a second value. The next moment point where the reversal has occurred is found in the preset time area, and the second value is accumulated to obtain all the misinformation classification reversal information in the preset time area.
[0024] Optionally, the step of using the bidirectional gated recurrent detection network in the trained misinformation detection model to perform temporal misinformation detection on the semantic feature set of misinformation, and obtaining the initial temporal misinformation classification result of the misinformation dataset, includes:
[0025] The semantic feature state value of the semantic feature of the false information at each time step is calculated using the reset gate in the bidirectional gated cyclic detection network.
[0026] The semantic feature state value at each time step is updated using the update gate in the bidirectional gated cyclic detection network to obtain the initial temporal inaccuracy information classification result.
[0027] Optionally, the step of dividing the false information dataset into blocks to obtain a set of false information data blocks includes:
[0028] The initial false information data in the aforementioned false information dataset is taken as the first false information data block;
[0029] Calculate the time difference between the final false information dataset at the final moment and the second false information dataset at the second moment in the false information dataset;
[0030] The time interval between the final false information dataset and the second false information dataset is calculated based on the preset number of data blocks to be divided and the time difference.
[0031] According to the time interval, all false information data from the second false information dataset to the final false information dataset are divided into blocks to obtain an empty false information data block set and a non-empty false information data block set;
[0032] The false information data block set is obtained by summing the first false information data block, the empty false information data block set, and the non-empty false information data block set.
[0033] Optionally, extracting the set of semantic features of false information from the set of false information data blocks includes:
[0034] The false information data block set is represented as a two-dimensional matrix using a preset feature extraction model, and a convolution operation is performed on the two-dimensional matrix to obtain the feature map of the false information data block set.
[0035] Max pooling is performed on the feature map to obtain the semantic feature set of false information.
[0036] To address the aforementioned problems, the present invention also provides an early false information detection device, the device comprising:
[0037] The false information semantic feature extraction module is used to acquire a dataset of false information to be detected within a preset time period, divide the false information dataset into blocks to obtain a set of false information data blocks, and extract the set of false information semantic features from the false information data blocks.
[0038] The false information detection module is used to perform temporal false information detection on the false information semantic feature set using the bidirectional gated recurrent detection network in the trained false information detection model, so as to obtain the initial temporal false information classification result of the false information dataset;
[0039] The false information classification reversal module is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period using the neural Hawkes layer in the detection point network of the false information detection model.
[0040] The false information detection point determination module is used to calculate the hidden state of the false information classification inversion information using the continuous time scale long short-term memory layer in the detection point network, and determine the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and the preset intensity function.
[0041] The classification module is used to feed back the false information detection points to the bidirectional gated recurrent detection network to obtain the target false information classification result of the false information dataset.
[0042] To address the above problems, the present invention also provides an electronic device, the electronic device comprising:
[0043] Memory, storing at least one computer program; and
[0044] The processor executes the computer program stored in the memory to implement the aforementioned early misinformation detection method.
[0045] To address the aforementioned problems, the present invention also provides a computer-readable storage medium storing at least one computer program, which is executed by a processor in an electronic device to implement the aforementioned method for detecting early misinformation.
[0046] In this embodiment of the invention, firstly, the false information dataset is divided into blocks to obtain a set of false information data blocks, and the semantic feature set of false information in the false information data blocks is extracted. This allows for the extraction of accurate false information features, which improves the accuracy of subsequent false information classification. Secondly, by using the bidirectional gated recurrent detection network in the trained false information detection model to perform temporal false information detection on the semantic feature set of false information, the initial temporal false information classification result at each time step can be determined, which improves the accuracy of subsequent false information classification. Finally, by using the neural Hawkes layer in the detection point network of the false information detection model to capture the inversion information of false information classification, it is easier to select accurate false information detection points in the future. Furthermore, a continuous time-scale long short-term memory layer is used. Calculating the hidden state of the inverted information in the classification of misinformation can capture the inverted information of misinformation in longer data sequences. Based on the hidden state and a preset intensity function, the detection points of the misinformation dataset to be detected within a preset time period are determined. This precise determination of the detection points improves the accuracy of misinformation detection point selection. It addresses the issues that some misinformation may fail to be detected accurately due to setting static misinformation detection points too early, while others may fail to be detected in a timely manner due to setting them too late. The detection points are then fed back into a bidirectional gated recurrent detection network to obtain the target misinformation classification result of the misinformation dataset, thereby improving the accuracy of early misinformation detection. Therefore, the early misinformation detection method, apparatus, device, and storage medium proposed in this invention can improve the accuracy of early misinformation detection. Attached Figure Description
[0047] Figure 1 This is a flowchart illustrating an early false information detection method provided in an embodiment of the present invention.
[0048] Figure 2 This is a detailed flowchart illustrating one step of an early false information detection method provided in an embodiment of the present invention;
[0049] Figure 3 This is a detailed flowchart illustrating one step of an early false information detection method provided in an embodiment of the present invention;
[0050] Figure 4 This is a schematic diagram of a module for an early false information detection device provided in an embodiment of the present invention;
[0051] Figure 5 This is a schematic diagram of the internal structure of an electronic device that implements an early false information detection method according to an embodiment of the present invention.
[0052] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0053] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
[0054] This invention provides a method for detecting early misinformation. The executing entity of the early misinformation detection method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the early misinformation 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 includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.
[0055] Reference Figure 1 The flowchart shown is a schematic diagram of an early misinformation detection method provided in an embodiment of the present invention. In this embodiment, the early misinformation detection method includes the following steps S1-S5:
[0056] S1. Obtain the dataset of false information to be detected within a preset time period, divide the dataset into blocks to obtain a set of false information data blocks, and extract the set of semantic features of false information from the data blocks.
[0057] In this embodiment of the invention, the dataset of false information to be detected refers to data that needs to be determined as false information. For example, in the financial industry, it could be a rumor about a financial platform failing to pay taxes. The preset time period refers to the period from the start time to the deadline for collecting the dataset of false information, for example, within four hours after the initial false information is generated.
[0058] In this embodiment of the invention, the set of false information data blocks refers to a set of data blocks divided according to the propagation time of false information. The set of false information data blocks includes: a first set of false information data blocks, an empty set of false information data blocks, and a non-empty set of false information data blocks. The set of semantic features of false information refers to the set of semantic features of false information corresponding to the set of false information data blocks.
[0059] In one embodiment of the present invention, since a false information event may contain tens of thousands of forwarded and commented information, if tens of thousands of information are classified sequentially, it is easy to cause long-distance forgetting due to gradient vanishing in the subsequent false information classification. Therefore, by dividing the false information dataset into blocks to obtain a false information data block set, the accuracy of subsequent false information classification can be improved.
[0060] As an embodiment of the present invention, the step of dividing the false information dataset into blocks to obtain a set of false information data blocks includes:
[0061] The initial false information data in the false information dataset is taken as the first false information data block; the time difference between the final false information dataset at the final moment and the second false information dataset at the second moment is calculated; the time interval between the final false information dataset and the second false information dataset is calculated according to the preset number of data blocks to be divided and the time difference; all false information data from the second false information dataset to the final false information dataset are divided into blocks according to the time interval to obtain an empty false information data block set and a non-empty false information data block set; the first false information data block, the empty false information data block set, and the non-empty false information data block set are summarized to obtain the false information data block set.
[0062] The initial false information data can be represented as m0, and the first false information data block can be represented as b0 = {m0}.
[0063] In one embodiment of the present invention, the final false information dataset at the final moment refers to the set of false information data at the final moment within a preset time period, which can be represented as (m j , t j The second false information dataset at the second moment refers to the set of false information data from which the initial false information data began to spread within a preset time period, which can be represented as (m1, t1); then the time difference can be TD = t j -t1.
[0064] Furthermore, the time interval can be calculated using the following formula:
[0065]
[0066] Where TI represents the time interval, TD represents the time difference, and N represents the preset number of data blocks to be divided.
[0067] In this embodiment of the invention, the empty false information data block set refers to data for which no false information propagation has occurred at the corresponding time. Based on the time interval TI, all false information data from the second false information dataset to the final false information dataset are divided into blocks to obtain an empty false information data block set i and an initial non-empty false information data block set j. If the initial non-empty false information data block set is less than the number of data blocks to be divided, the time interval is shortened to obtain a new time interval. Then, based on the new time interval, all false information data from the second false information dataset to the final false information dataset are further divided into blocks until the number of the initial non-empty false information data block set is not less than the number of data blocks to be divided, thus obtaining the number of non-empty false information data block sets.
[0068] Further, the extraction of the semantic feature set of false information from the set of false information data blocks includes:
[0069] The false information data block set is represented as a two-dimensional matrix using a preset feature extraction model, and a convolution operation is performed on the two-dimensional matrix to obtain the feature map of the false information data block set; the feature map is then subjected to a max pooling operation to obtain the desired result.
[0070] The preset feature extraction model can be a CNN model, which can represent the set of false information data blocks as an n*k dimensional second-order matrix, where n is the number of text information blocks contained in the set of false information data blocks, and k is the dimension of the text semantic feature vector.
[0071] In one embodiment of the present invention, the convolution operation uses a one-dimensional matrix. Since the amount of text information contained in each set of false information data blocks is different, the matrix height of the input CNN is inconsistent. In order to ensure that the dimension of the final output data is consistent, the max pooling operation can use a 1-maxpooling window to extract the feature map, that is, each feature map only retains the maximum value feature.
[0072] S2. Using the bidirectional gated recurrent detection network in the trained false information detection model, temporal false information detection is performed on the semantic feature set of false information to obtain the initial temporal false information classification result of the false information dataset.
[0073] In this embodiment of the invention, the bidirectional gated loop detection network can be a BiGRU model, and the bidirectional gated loop detection network includes a reset gate and an update gate.
[0074] The initial temporal classification result of misinformation refers to the misinformation result corresponding to the semantic feature set of misinformation at each time step. This misinformation result includes both misinformation and non-misinformation. In other words, the initial temporal classification result of the misinformation dataset includes the misinformation type corresponding to the misinformation data, and the misinformation type includes both misinformation and non-misinformation. For example, the misinformation classification result for the disseminated event of a financial platform failing to pay taxes could be either "the financial platform failing to pay taxes is misinformation" or "the financial platform failing to pay taxes is not misinformation."
[0075] In one embodiment of the present invention, since the bidirectional gated recurrent detection network can obtain the classification result of false information corresponding to the semantic feature set of false information at each moment within a preset time period, it can obtain the initial temporal false information classification result containing time series information. In one embodiment of the present invention, since the text information in false information events is interconnected and influential, and has related temporal features, the bidirectional gated recurrent detection network in the trained false information detection model is used to perform temporal false information detection on the semantic feature set of false information, thereby obtaining the initial temporal false information classification result of the false information dataset. Since the bidirectional gated recurrent detection network has effective temporal characteristic learning features, it can well maintain the long-short distance dependency relationship in the temporal data, and combine the output result of the semantic feature set of false information with past and future information to classify false information, thus initially improving the accuracy of false information classification.
[0076] As an embodiment of the present invention, the step of using a bidirectional gated recurrent detection network in the trained misinformation detection model to perform temporal misinformation detection on the semantic feature set of misinformation, and obtaining the initial temporal misinformation classification result of the misinformation dataset, includes:
[0077] The semantic feature state value of the semantic feature of the false information at each time step is calculated using the reset gate in the bidirectional gated cyclic detection network; the semantic feature state value at each time step is updated using the update gate in the bidirectional gated cyclic detection network to obtain the initial temporal false information classification result.
[0078] The reset gate is used to concatenate and activate the reset false information semantic features with the false information semantic features input at previous times, so as to remember the semantic feature state value at each time.
[0079] Furthermore, the semantic feature state value can be calculated using the following formula:
[0080] r t =σ(W r [h t-1 x t]+b r )
[0081] Where, r t W represents the semantic feature state value of the gate output at time t, where σ represents the activation function. r This indicates resetting the door's connection parameters, h t-1 x represents the state value of the BiGRU at the previous moment. t b represents the semantic features of the false information input at time t. r This indicates the parameter for resetting the door's bias term.
[0082] In one embodiment of the present invention, the update gate is used to remember or forget the semantic feature state value output by the reset gate, and the remembered semantic feature state value is saved to the current time step for updating, so as to obtain the initial time-series inaccurate information classification result.
[0083] This invention achieves the functions of the input gate and forget gate in an LSTM model through an update gate, enabling the classification of false information with fewer parameters. Furthermore, the update gate can retain long-term false information data while reducing the gradient vanishing problem, thereby improving the accuracy of false information classification.
[0084] The initial time-series inaccurate information classification result can be calculated using the following formula:
[0085] z t =σ(W z [h t-1 x t ]+b z )
[0086] Among them, z t The initial temporal misclassification result of the gate output at time t represents the value of h. t-1 x represents the state value of the BiGRU at the previous moment. t b represents the semantic feature state value input at time t. z This indicates that the bias term parameters of the updated gate are being updated.
[0087] S3. Using the neural Hawkes layer in the detection point network of the false information detection model, capture the false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period.
[0088] In this embodiment of the invention, the misclassification information refers to the information in which the initial temporal misclassification results of every two adjacent time points are reversed. For example, if the initial temporal misclassification result at time t1 is misclassified information and the initial temporal misclassification result at time t2 is not misclassified information, then the initial temporal misclassification results from time t1 to time t2 have been reversed.
[0089] In this embodiment of the invention, the neural Hawkes layer is the NHP (Neural Hawkes Process), also known as the self / mutually stimulating process, which means that historical events have a stimulating effect on the occurrence of future events, and it is assumed that the influence of historical events on future events is monotonically decreasing exponentially, and then superimposed in an additive form. The main function of the neural Hawkes in this invention is to simulate the dynamic changes of future false information events based on the occurrence of historical false information events.
[0090] For details, please refer to Figure 2 As shown, the method of using the neural Hawkes layer in the detection point network of the false information detection model to capture false information classification reversal information between the initial false information classification results of every two adjacent time points in the preset time region includes the following steps S31-S33:
[0091] S31. Use the neural Hawkes layer to identify whether the initial false information classification result at the current moment in the preset time region is consistent with the initial false information classification result at the previous moment.
[0092] S32. When the initial false information classification result at the current moment is consistent with the initial false information classification result at the previous moment, it is determined that the false information classification information at the current moment has not been reversed, and the current moment point that has not been reversed is assigned the first value.
[0093] S33. When the initial false information classification result at the current moment is inconsistent with the initial false information classification result at the previous moment, it is determined that the false information classification information at the current moment has been reversed, and the current moment point where the reversal has occurred is assigned a second value. The next moment point where the reversal has occurred is found in the preset time area, and the second value is accumulated to obtain all false information classification reversal information in the preset time area.
[0094] The neural Hawkes layer can be represented by the following formula:
[0095]
[0096] Wherein I(t) i ) represents t i Incorrect information at any given moment is reversed, I(t) i-1) represents t i-1 Inaccurate information at any moment, reverse information, y i y represents the classification result of the i-th piece of false information. i-1 This represents the classification result of false information at time i-1.
[0097] Specifically, for example, if the initial time series false information classification result at time t1 is false information, and the initial time series false information classification result at time t2 is also false information, then there is no false information reversal phenomenon. In this case, I(t2) at time t2 is assigned a first value, and this first value can be 0. If the initial time series false information classification result at time t3 is not false information, then there is a reversal between t2 and t3. In this case, I(t3) at time t3 is assigned a second value, and this second value can be 1, i.e., I(t3) = 1. Then, the next time point in the preset time range where reversal occurs is searched, and the second value is accumulated. That is, the subsequent time points where reversal occurs are incremented by 1.
[0098] S4. Calculate the hidden state of the false information classification inversion information using the continuous time scale long short-term memory layer in the detection point network, and determine the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and the preset intensity function.
[0099] In this embodiment of the invention, the continuous timescale long short-term memory layer is a (CTLSTM, Continuous Timescale Long-Short Term Memory) neural network. Compared with the traditional LSTM network, this network utilizes a continuous timescale for long short-term memory, which can capture longer information sequences and improve the accuracy of subsequent classification of false information.
[0100] In this embodiment of the invention, the intensity function is a weight function that describes the hidden state of the misinformation classification reversal information at each time step. The main function of the misinformation detection point is to determine at what time point the dataset to be detected is determined to be misinformation, and after the misinformation detection point is determined, it means that the misinformation dataset before the misinformation detection point has been determined to be either misinformation or not misinformation, and the misinformation classification reversal will no longer occur.
[0101] In this embodiment of the invention, a false information detection point can be determined by reversing the classification of false information. Different false information detection points can also be obtained through the above steps for different false information datasets, so that the most suitable false information detection point can be selected dynamically and in a personalized manner.
[0102] Furthermore, the target misinformation classification result refers to the final classification result of the misinformation dataset, which also includes two categories: misinformation and non-misinformation.
[0103] This invention utilizes a continuous-timescale long short-term memory layer to calculate the hidden state of misinformation classification inversion information. This allows for the capture of misinformation inversion information in longer data sequences. Based on the hidden state and a preset intensity function, it determines the misinformation detection points for the misinformation dataset to be detected within a preset time period, achieving dynamic determination of misinformation detection points. This improves the accuracy of misinformation detection point selection and addresses the issues where some misinformation cannot be detected accurately due to setting static misinformation detection points too early, while others cannot be detected in a timely manner due to setting them too late. The misinformation detection points are then fed back to a bidirectional gated recurrent detection network to obtain the target misinformation classification result of the misinformation dataset, thereby improving the accuracy of early misinformation detection.
[0104] As an embodiment of the present invention, the step of calculating the hidden state of the false information classification reversal information using the continuous time-scale long short-term memory layer in the detection point network includes:
[0105] The activation value of the false information classification reversal information is calculated using the input gate in the continuous time-scale long short-term memory layer; the state value of the false information classification reversal information is calculated using the forget gate in the continuous time-scale long short-term memory layer.
[0106] Based on the activation value and the state value, the hidden state of the false information classification inversion information is calculated using the output gate in the continuous time scale long short-term memory layer.
[0107] The activation value can be calculated using the following formula:
[0108]
[0109] in, Let represent the activation value of the i-th false information classification reversal information input at time t, and I represent the number of false information classification reversal information input. Let H represent the inverted classification information of the i-th false information input at time t, and let H represent the initial number of hidden states of the inverted classification information of the false information input by the input gate. This represents the initial hidden state of the h-th false information classification reversal information at time t-1.
[0110] In one embodiment of the present invention, the state value can be calculated using the following formula:
[0111]
[0112] in, y represents the state value of the c-th false information classification reversal information output by the forget gate at time t. t Let represent the initial hidden state of the false information classification and reversal information at time t, and g represent the activation function output by the forget gate. This represents the state value of the c-th false information category reversal information output by the forget gate at time t-1. Let represent the hidden state of the i-th false information classification reversal information at time t. This represents the activation value of the i-th false information classification reversal information input at time t.
[0113] Furthermore, the hidden state of the false information classification reversal information can be calculated using the following formula:
[0114]
[0115] Where f(t) represents the hidden state of the false information classification reversal information at time t. h represents the activation value of the i-th false information classification reversal information input at time t. t This represents the peak value of the misinformation classification reversal information at time t. This represents the state value of the c-th false information classification reversal information output by the forget gate at time t.
[0116] For details, please refer to Figure 3 As shown, determining the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and a preset intensity function includes the following steps S41-S44:
[0117] S41. Calculate the difference between the false information classification reversal information at each moment within the preset time period and the final moment within the preset time period based on the hidden state and the preset intensity function.
[0118] S42. Determine whether the difference is zero;
[0119] S43. When the difference is not zero, recalculate the hidden state of the false information classification reversal information, and calculate the difference between the false information classification reversal information at the current time and the final time in the preset time region again according to the hidden state and the preset intensity function, until the difference is zero.
[0120] S44. When the difference is zero, the time corresponding to the difference is determined as the false information detection point of the false information dataset.
[0121] The intensity function can be expressed by the following formula:
[0122] γ(t)=f(W*h(t))
[0123] Where γ(t) represents the intensity function at time t, W represents the weight matrix, h(t) represents the hidden state at time t, and f(W*h(t)) represents the weight function of the hidden state at time t.
[0124] In one embodiment of the present invention, the difference can be calculated using the following formula:
[0125]
[0126] in, This represents the difference between the inaccurate information classification and reversal information at each moment within a preset time period and at the final moment within the preset time period; t ∞ The final time is represented by γ(t), and users can customize the final time according to their actual needs; γ(t) represents the intensity function at time t.
[0127] Specifically, the difference has been calculated using the above formula. When the difference is 0, it indicates that the false information detection points in the false information dataset within the preset time period have reached a stable state, meaning there is no longer any false information classification reversal. When the difference is not 0, it indicates that the false information dataset within the preset time period may still experience false information classification reversal in the future, meaning the false information detection points are in an unstable state. The final time selection needs to be shifted backward, and the hidden state of the false information classification reversal information needs to be recalculated. The difference between the current time and the final time in the preset time region is calculated again based on the hidden state and the preset intensity function until the difference is zero, meaning the false information detection points are shifted backward until they reach a stable state.
[0128] S5. Feed back the false information detection points to the bidirectional gated cyclic detection network to obtain the target false information classification result of the false information dataset.
[0129] In this embodiment of the invention, the target misinformation classification result refers to the final misinformation classification result of the misinformation dataset, and also includes the misinformation type corresponding to the misinformation data. The misinformation type includes misinformation and non-misinformation.
[0130] In one embodiment of the present invention, the false information detection points are fed back to the bidirectional gated cyclic detection network. The bidirectional gated cyclic detection network is used to detect false information in the semantic feature set of false information at the time corresponding to the false information detection points, so as to obtain the target false information classification result of the false information dataset.
[0131] In this embodiment of the invention, firstly, the false information dataset is divided into blocks to obtain a set of false information data blocks, and the semantic feature set of false information in the false information data blocks is extracted. This allows for the extraction of accurate false information features, which improves the accuracy of subsequent false information classification. Secondly, by using the bidirectional gated recurrent detection network in the trained false information detection model to perform temporal false information detection on the semantic feature set of false information, the initial temporal false information classification result at each time step can be determined, which improves the accuracy of subsequent false information classification. Finally, by using the neural Hawkes layer in the detection point network of the false information detection model to capture the inversion information of false information classification, it is easier to select accurate false information detection points in the future. Furthermore, a continuous time-scale long short-term memory layer is used. Calculating the hidden state of the inverted information in the misinformation classification allows for the capture of inverted misinformation information in longer data sequences. Based on the hidden state and a preset intensity function, the detection points for the misinformation dataset to be detected within a preset time period are determined. This precise identification of the detection points improves the accuracy of misinformation detection point selection. It addresses the issues of some misinformation being detected too early (resulting in inaccurate detection) and others being detected too late (resulting in inadequate timeliness). The detection points are then fed back into a bidirectional gated recurrent detection network to obtain the target misinformation classification result for the misinformation dataset, thereby improving the accuracy of early misinformation detection. Therefore, the early misinformation detection method proposed in this invention can improve the accuracy of early misinformation detection.
[0132] like Figure 4 As shown, the early false information detection device 100 of the present invention can be installed in an electronic device. Depending on the functions implemented, the early false information detection device may include a false information semantic feature extraction module 101, a false information detection module 102, a false information classification reversal module 103, a false information detection point determination module 104, and a classification module 105. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, stored in the memory of the electronic device.
[0133] In this embodiment, the functions of each module / unit are as follows:
[0134] In detail, each module in the early false information detection device 100 described in this embodiment of the invention adopts the same characteristics as described above during use. Figures 1 to 3 The method uses the same techniques as the early misinformation detection methods described above and can produce the same technical effects, so it will not be elaborated here.
[0135] In this embodiment of the invention, firstly, the false information dataset is divided into blocks to obtain a set of false information data blocks, and the semantic feature set of false information in the false information data blocks is extracted. This allows for the extraction of accurate false information features, which improves the accuracy of subsequent false information classification. Secondly, by using the bidirectional gated recurrent detection network in the trained false information detection model to perform temporal false information detection on the semantic feature set of false information, the initial temporal false information classification result at each time step can be determined, which improves the accuracy of subsequent false information classification. Finally, by using the neural Hawkes layer in the detection point network of the false information detection model to capture the inversion information of false information classification, it is easier to select accurate false information detection points in the future. Furthermore, a continuous time-scale long short-term memory layer is used. Calculating the hidden state of the inverted information in the misinformation classification allows for the capture of inverted misinformation information in longer data sequences. Based on the hidden state and a preset intensity function, the detection points for the misinformation dataset to be detected within a preset time period are determined. This precise determination of the detection points improves the accuracy of misinformation detection point selection. It addresses the issues of some misinformation being detected too early (resulting in inaccurate detection) and others being detected too late (resulting in inadequate timeliness). The detection points are then fed back into a bidirectional gated recurrent detection network to obtain the target misinformation classification result for the misinformation dataset, thereby improving the accuracy of early misinformation detection. Therefore, the early misinformation detection device proposed in this embodiment can improve the accuracy of early misinformation detection.
[0136] like Figure 5 The diagram shown is a structural schematic of an electronic device that implements the method for detecting early false information according to the present invention.
[0137] 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 an early misinformation detection program.
[0138] The memory 11 includes at least one type of medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, local disk, optical disk, etc. In some embodiments, the memory 11 can be an internal storage unit of an electronic device, such as a portable hard drive. In other embodiments, the memory 11 can be an external storage device of the electronic device, such as a plug-in portable hard drive, Smart Media Card (SMC), Secure Digital (SD) card, Flash Card, etc. Furthermore, the memory 11 can include both internal and external storage units of the electronic device. The memory 11 can be used not only to store application software and various types of data installed on the electronic device, such as the code of an early misinformation detection program, but also to temporarily store data that has been output or will be output.
[0139] In some embodiments, the processor 10 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 10 is the control unit of the electronic device, connecting various components of the entire electronic device through various interfaces and lines. It executes programs or modules stored in the memory 11 (e.g., early false information detection programs) and calls data stored in the memory 11 to perform various functions of the electronic device and process data.
[0140] The communication bus 12 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. 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, etc. For ease of illustration, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus.
[0141] Figure 5 Only electronic devices with components are shown; it will be understood by those skilled in the art that... Figure 5The structure shown does not constitute a limitation on the electronic device and may include fewer or more components than shown, or combine certain components, or have different component arrangements.
[0142] For example, although not shown, the electronic device may also include a power supply (such as a battery) to power the various components. Preferably, the power supply can be logically connected to the at least one processor 10 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The electronic device may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.
[0143] 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 typically used to establish communication connections between the electronic device and other electronic devices.
[0144] Optionally, the communication interface 13 may further include a user interface, which may be a display, an input unit (such as a keyboard), or, optionally, a standard wired or 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, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the electronic device and to display a visual user interface.
[0145] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.
[0146] The early misinformation detection program stored in the memory 11 of the electronic device is a combination of multiple computer programs that, when run in the processor 10, can achieve the following:
[0147] A dataset of false information to be detected within a preset time period is obtained. The dataset is then divided into blocks to obtain a set of false information data blocks. Semantic features of false information in the data blocks are then extracted.
[0148] The bidirectional gated recurrent detection network in the trained false information detection model is used to detect temporal false information in the semantic feature set of false information, so as to obtain the initial temporal false information classification result of the false information dataset.
[0149] The neural Hawkes layer in the detection point network of the aforementioned false information detection model is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period;
[0150] The hidden state of the false information classification inversion information is calculated using the continuous time scale long short-term memory layer in the detection point network. Based on the hidden state and the preset intensity function, the false information detection points of the false information dataset to be detected within the preset time period are determined.
[0151] The false information detection points are fed back to the bidirectional gated cyclic detection network to obtain the target false information classification result of the false information dataset.
[0152] Specifically, the processor 10's implementation method of the above-mentioned computer program can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.
[0153] Furthermore, if the modules / units integrated into the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable medium. The computer-readable medium can 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 portable hard drive, a magnetic disk, an optical disk, a computer memory, or a read-only memory (ROM).
[0154] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor of an electronic device, can perform the following:
[0155] A dataset of false information to be detected within a preset time period is obtained. The dataset is then divided into blocks to obtain a set of false information data blocks. Semantic features of false information in the data blocks are then extracted.
[0156] The bidirectional gated recurrent detection network in the trained false information detection model is used to detect temporal false information in the semantic feature set of false information, so as to obtain the initial temporal false information classification result of the false information dataset.
[0157] The neural Hawkes layer in the detection point network of the aforementioned false information detection model is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period;
[0158] The hidden state of the false information classification inversion information is calculated using the continuous time scale long short-term memory layer in the detection point network. Based on the hidden state and the preset intensity function, the false information detection points of the false information dataset to be detected within the preset time period are determined.
[0159] The false information detection points are fed back to the bidirectional gated cyclic detection network to obtain the target false information classification result of the false information dataset.
[0160] Furthermore, the computer-readable storage medium may primarily include a program storage area and a data storage area, wherein the program storage area may store the operating system, at least one application required for a function, etc.; and the data storage area may store data created based on the use of blockchain nodes, etc.
[0161] In the several embodiments provided by this invention, it should be understood that the disclosed media, devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.
[0162] The modules described as separate components may or may not be physically separate. 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 units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.
[0163] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.
[0164] 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.
[0165] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.
[0166] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.
[0167] 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. The term "second class" is used to indicate names and does not indicate any specific order.
[0168] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.
Claims
1. A method for detecting early-stage misinformation, characterized in that, The method includes: A dataset of false information to be detected within a preset time period is obtained. The dataset is then divided into blocks to obtain a set of false information data blocks. Semantic features of false information in the data blocks are then extracted. The bidirectional gated recurrent detection network in the trained false information detection model is used to detect temporal false information in the semantic feature set of false information, so as to obtain the initial temporal false information classification result of the false information dataset. The neural Hawkes layer in the detection point network of the aforementioned false information detection model is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period; The hidden state of the false information classification inversion information is calculated using the continuous time scale long short-term memory layer in the detection point network. Based on the hidden state and the preset intensity function, the false information detection points of the false information dataset to be detected within the preset time period are determined. The false information detection points are fed back to the bidirectional gated cyclic detection network to obtain the target false information classification result of the false information dataset; The step of determining the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and a preset intensity function includes: The difference between the false information classification reversal information at each moment within the preset time period and the final moment within the preset time period is calculated based on the hidden state and the preset intensity function. Determine whether the difference is zero; When the difference is not zero, the hidden state of the false information classification reversal information is recalculated, and the difference between the false information classification reversal information at the current time and the final time in the preset time region is calculated again according to the hidden state and the preset intensity function, until the difference is zero. When the difference is zero, the time corresponding to the difference is determined as the false information detection point of the false information dataset; The method of capturing misinformation classification reversal information between the initial temporal misinformation classification results of every two adjacent moments within the preset time period using the neural Hawkes layer in the detection point network of the misinformation detection model includes: The neural Hawkes layer is used to identify whether the initial false information classification result at the current moment in the preset time region is consistent with the initial false information classification result at the previous moment; When the initial false information classification result at the current moment is consistent with the initial false information classification result at the previous moment, it is determined that the false information classification information at the current moment has not been reversed, and the current moment point that has not been reversed is assigned the first value. When the initial misinformation classification result at the current moment is inconsistent with the initial misinformation classification result at the previous moment, it is determined that the misinformation classification information at the current moment has been reversed, and the current moment point where the reversal has occurred is assigned a second value. The next moment point where the reversal has occurred is found in the preset time region, and the second value is accumulated to obtain all the misinformation classification reversal information in the preset time region.
2. The method for detecting early false information as described in claim 1, characterized in that, The step of calculating the hidden state of the misinformation classification reversal information using the continuous-time-scale long short-term memory layer in the detection point network includes: The activation value of the false information classification inversion information is calculated using the input gates in the continuous time-scale long short-term memory layer. The state value of the false information classification reversal information is calculated using the forget gate in the continuous time-scale long short-term memory layer; Based on the activation value and the state value, the hidden state of the false information classification inversion information is calculated using the output gate in the continuous time scale long short-term memory layer.
3. The method for detecting early false information as described in claim 1 or 2, characterized in that, The bidirectional gated recurrent detection network in the trained misinformation detection model is used to detect temporal misinformation in the semantic feature set of misinformation, resulting in an initial temporal misinformation classification result for the misinformation dataset, including: The semantic feature state value of the semantic feature of the false information at each time step is calculated using the reset gate in the bidirectional gated cyclic detection network. The semantic feature state value at each time step is updated using the update gate in the bidirectional gated cyclic detection network to obtain the initial temporal inaccuracy information classification result.
4. The method for detecting early false information as described in claim 1 or 2, characterized in that, The step of dividing the false information dataset into blocks to obtain a set of false information data blocks includes: The initial false information data in the aforementioned false information dataset is taken as the first false information data block; Calculate the time difference between the final false information dataset at the final moment and the second false information dataset at the second moment in the false information dataset; The time interval between the final false information dataset and the second false information dataset is calculated based on the preset number of data blocks to be divided and the time difference. According to the time interval, all false information data from the second false information dataset to the final false information dataset are divided into blocks to obtain an empty false information data block set and a non-empty false information data block set; The false information data block set is obtained by summing the first false information data block, the empty false information data block set, and the non-empty false information data block set.
5. The method for detecting early false information as described in claim 1 or 2, characterized in that, The step of extracting the semantic feature set of false information from the set of false information data blocks includes: The false information data block set is represented as a two-dimensional matrix using a preset feature extraction model, and a convolution operation is performed on the two-dimensional matrix to obtain the feature map of the false information data block set. Max pooling is performed on the feature map to obtain the semantic feature set of false information.
6. An early false information detection device, used to implement the early false information detection method as described in any one of claims 1-5, characterized in that, The device includes: The false information semantic feature extraction module is used to acquire a dataset of false information to be detected within a preset time period, divide the false information dataset into blocks to obtain a set of false information data blocks, and extract the set of false information semantic features from the false information data blocks. The false information detection module is used to perform temporal false information detection on the false information semantic feature set using the bidirectional gated recurrent detection network in the trained false information detection model, so as to obtain the initial temporal false information classification result of the false information dataset; The false information classification reversal module is used to capture false information classification reversal information between the initial temporal false information classification results of every two adjacent moments within the preset time period using the neural Hawkes layer in the detection point network of the false information detection model. The false information detection point determination module is used to calculate the hidden state of the false information classification inversion information using the continuous time scale long short-term memory layer in the detection point network, and determine the false information detection points of the false information dataset to be detected within the preset time period based on the hidden state and the preset intensity function. The classification module is used to feed back the false information detection points to the bidirectional gated recurrent detection network to obtain the target false information classification result of the false information dataset.
7. An electronic device, characterized in that, The electronic device includes: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the early misinformation detection method as described in any one of claims 1 to 5.
8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the early misinformation detection method as described in any one of claims 1 to 5.