Information classification warning method, device and equipment based on multi-label causal relationship
By constructing a multi-label causal relationship information hierarchical warning model, using text encoder and tag characterization network, the problems of high error judgment rate and low accuracy rate in multi-label scenarios in the prior art are solved, and accurate identification of information emergency levels is achieved and fine-grained grading improve the efficiency and accuracy of information analysis.
Patent Information
- Application Number
- CN202411386694.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-30
- Publication Date
- 2024-12-06
- Estimated Expiration
- 2044-09-30
AI Technical Summary
The existing hierarchical early warning methods have problems such as high misjudgment rate, low accuracy and poor universality in the field of information analysis. Especially in multi-label scenarios, it is difficult to capture complex causal relationships, resulting in low recognition accuracy at the information emergency level. .
Using the information hierarchical early warning method based on multi-label causal relationship, we use multi-head attention mechanism and graph neural network to capture text and tags by constructing a model including text encoder, tag characterization network, tag obfuscation model and naive Bayes classifier. causal relationships between them, calculate simulation and prediction label distribution, reduce the misjudgment rate and improve accuracy.
Effectively improve the accuracy of information grading warning, can capture information differences at fine-grained level, reduce manpower consumption, improve the efficiency and accuracy of information analysis, distinguish it from the method of single-label correlation, and dynamically understand the multi-level tags. nuance.
Smart Images

Figure CN118916617B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of information analysis technology, and in particular to a method, device and equipment for information classification warning based on multi-label causal relationships. Background Art
[0002] Gradual warning is of great significance in the field of information analysis, especially in the context of the rapid development of information and intelligent technology and the explosion of information in the future. As society develops rapidly towards the era of informatization and intelligence, data is flooding and information is exploding, the improvement of the all-round efficiency of information analysis has become a key requirement to meet future decision-making needs. The existing graded warning method has a high misjudgment rate, so it is urgent to use advanced technology to continuously optimize the existing graded warning model to reduce the misjudgment rate and improve the recall rate.
[0003] The traditional hierarchical warning method involves using relevant private and dedicated information text data to analyze and classify information content of different urgency levels, and mark all information with corresponding levels. Then, a complete information classification specification is formed through data statistics. However, this hierarchical warning model based on manual analysis has the problems of low accuracy and time-consuming and labor-intensive. At the same time, due to the confidentiality of data and the specificity of different standards, the universality of this method is poor.
[0004] In recent years, in order to improve the accuracy of graded warning and reduce manpower consumption, machine learning-based methods have been widely used. The main technologies are as follows: one is to train a nonlinear support vector machine classifier based on a large amount of text data to achieve graded warning; the other method is to use a convolutional neural network model to learn different levels of information prototypes to characterize the characteristics of information. However, the graded warning method based on machine learning still has the problem of low recognition accuracy.
[0005] In the field of text classification, existing fine-grained classification methods mainly rely on pre-trained language models to capture subtle differences in text through contextual information and attention mechanisms. These methods perform well in single-label classification tasks, but still have significant limitations in multi-label scenarios. Most fine-grained text classification models assume that each sample corresponds to only one label, which makes it difficult to capture complex multi-label co-occurrence relationships, making them less than ideal when dealing with complex scenarios with multiple labels. Summary of the invention
[0006] Based on this, it is necessary to provide an information classification warning method, device and equipment based on multi-label causal relationships that can accurately identify information under fine-grained category conditions in response to the above technical problems.
[0007] An information classification warning method based on multi-label causal relationship, the method comprising:
[0008] Formulate information level specifications based on the preset total number of information classification warning levels and the purpose of information classification.
[0009] Real text data is preprocessed and labeled according to information level specifications to construct an annotated training dataset.
[0010] An information classification warning model based on multi-label causal relationship is constructed; the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution based on the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution based on the text representation, and the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through the dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0011] ;
[0012] in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels.
[0013] The information classification warning model is trained using a supervised model training method according to the training data set to obtain a trained information classification warning model.
[0014] The information data to be classified and the established standards are input into the trained information classification warning model to obtain the emergency level of the information to be classified.
[0015] In one embodiment, in the labeled training data set, the importance level of information increases from level 0; there is no upper limit to the importance level of information.
[0016] In one embodiment, the information classification warning model is trained using a supervised model training method according to the training data set to obtain a trained information classification warning model, including:
[0017] Input the training dataset into the text encoder to obtain text representation.
[0018] The training data set is input into the label representation network to obtain the label representation corresponding to the text representation.
[0019] The text representation, the corresponding label representation, and the single label vector are input into the label confusion model to obtain the simulated label distribution.
[0020] The text representation is input into the Naive Bayes classifier to obtain the predicted label distribution.
[0021] The model training loss is calculated using the KL divergence method according to the simulated label distribution and the predicted label distribution. The parameters of the information classification warning model are updated according to the model training loss until the number of training reaches the preset number of iterations to obtain a trained information classification warning model.
[0022] In one embodiment, the text encoder consists of a word segmenter, a pre-trained language model, and a multi-layer feed-forward neural network.
[0023] In one embodiment, the label representation network includes a multi-layer transformer encoder, a multi-layer perceptron, and a graph neural network to fully capture the causal relationship between labels.
[0024] An information classification warning device based on multi-label causal relationship, the device comprising:
[0025] The module for constructing an annotated training data set is used to formulate information level specifications based on the total number of preset information classification warning levels and the purpose of information classification; preprocess and assign labels to real text data according to the information level specifications to construct an annotated training data set.
[0026] The information classification warning model construction module is used to construct an information classification warning model based on multi-label causal relationships; the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution based on the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution based on the text representation, and the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through the dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0027] ;
[0028] in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels.
[0029] The information classification warning model training module is used to train the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model.
[0030] The information classification warning module is used to input the information data to be classified and the formulated standards into the trained information classification warning model to obtain the emergency level of the information to be classified.
[0031] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:
[0032] Formulate information level specifications based on the preset total number of information classification warning levels and the purpose of information classification.
[0033] Real text data is preprocessed and labeled according to information level specifications to construct an annotated training dataset.
[0034] An information classification warning model based on multi-label causal relationship is constructed; the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution based on the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution based on the text representation, and the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through the dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0035] ;
[0036] in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels.
[0037] The information classification warning model is trained using a supervised model training method according to the training data set to obtain a trained information classification warning model.
[0038] The information data to be classified and the established standards are input into the trained information classification warning model to obtain the emergency level of the information to be classified.
[0039] The above-mentioned information classification warning method, device and equipment based on multi-label causal relationship, the method includes: formulating information level specifications according to the preset total number of information classification warning levels and information classification purposes; preprocessing and labeling real text data according to the information level specifications to construct a labeled training data set; constructing an information classification warning model based on multi-label causal relationship; training the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model; using the trained information classification warning model to process the information data to be classified to obtain the emergency level of the information to be classified. This method can capture the fine-grained differences of information at different levels, and combine the information classification standards to comprehensively predict the emergency level of information; through the hierarchical warning model based on multi-label causal relationship, the accuracy of information classification warning is effectively improved, while reducing manpower consumption, the efficiency and accuracy of information analysis are further improved; different from the previous warning method that only learns the correlation between data and a single label, this method dynamically captures its causal relationship with multi-level labels according to the characteristics of the instance, so as to understand the subtle differences between different levels and achieve better classification effect. BRIEF DESCRIPTION OF THE DRAWINGS
[0040] Figure 1 It is a flowchart of an information classification warning method based on multi-label causal relationship in one embodiment;
[0041] Figure 2 A schematic diagram of the structure of an information classification warning model based on multi-label causal relationships in one embodiment;
[0042] Figure 3 It is a structural block diagram of an information classification warning device based on multi-label causal relationship in one embodiment;
[0043] Figure 4 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION
[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.
[0045] In one embodiment, Figure 1As shown, a method for information classification warning based on multi-label causal relationship is provided, and the method includes the following steps:
[0046] Step 100: Formulate information level specifications based on the preset total number of information classification warning levels and information classification purposes.
[0047] Specifically, based on the total number of information classification warning levels and the purpose of information classification, information level standards are formulated and clear classification criteria are proposed.
[0048] Step 102: Preprocess and assign labels to real text data according to information level specifications to construct a labeled training data set.
[0049] Specifically, we preprocess and assign labels based on information level specifications and real text data, clean noise data, and construct a labeled training data set. In the constructed data set, the importance level of information increases from level 0; there is no upper limit to the importance level of information.
[0050] Step 104: Construct an information grading warning model based on multi-label causal relationships; wherein the information grading warning model includes a text encoder, a label representation network, a label confusion model, and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set, wherein the text encoder is composed of a word segmenter, a pre-trained language model, and a multi-layer feedforward neural network; the label representation network is used to extract the label representation corresponding to the text representation in the training data set. Considering that most previous methods only use fully connected neural networks, which are not enough to capture the deep relationship between labels and texts and the node relationship between different labels, the newly proposed label representation network consists of a multi-layer transformer encoder. The proposed method is composed of a decoder, a multi-layer perceptron and a graph neural network to fully capture the causal relationship between labels; the label confusion model is used to calculate and simulate label distribution based on text representation and label representation. The label confusion model is a multi-channel fusion network, which consists of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism. Regarding the network architecture of the multi-channel fusion network, a cross-attention mechanism based on multi-head attention is adopted to achieve better fusion of different channel features. The similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through the dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0051] ;
[0052] in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels.
[0053] The naive Bayes classifier is used to calculate the predicted label distribution based on the text representation, and consists of a prior probability calculation module, a conditional probability estimation module, and a prediction module. Different from the general linear layer classifier of the previous method, the Bayes classifier is more robust to the noise contained in the fine-grained information.
[0054] Specifically, the text encoder and the label representation network are used to extract the text representation of the information and the corresponding label representation respectively. The information is input into the text encoder The process of obtaining text representation can be reflected as follows: ;
[0055] in Represents the corresponding text data. The text encoder is constructed using a deep neural network to accurately learn the nonlinear features of the text.
[0056] Comprehensive label distribution is conducive to capturing the causal relationship between different labels, thereby achieving fine-grained graded warning.
[0057] The structure of the information classification warning model based on multi-label causal relationship is as follows: Figure 2 shown.
[0058] Step 106: Train the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model.
[0059] Specifically, supervised model training is performed through the obtained multi-label causal relationship to determine the model-related neural network parameters of the current data set. KL divergence is used as the model training loss here to more accurately learn the distribution difference between the predicted label and the true label.
[0060] Step 108: Input the information data to be classified and the established standards into the trained information classification warning model to obtain the emergency level of the information to be classified.
[0061] Specifically, according to the trained information classification warning model, the information data to be classified and the classification standards of information of different levels are input, the output prediction results of the emergency level of the information to be classified are obtained and sorted from high to low according to the confidence, and the highest confidence category is output as the prediction result.
[0062] Balance factor The smaller the value, the higher the sensitivity of model classification.
[0063] In the above-mentioned information classification warning method based on multi-label causal relationship, the method includes: formulating information level specifications according to the preset total number of information classification warning levels and information classification purposes; preprocessing and labeling real text data according to the information level specifications to construct a labeled training data set; constructing an information classification warning model based on multi-label causal relationship; training the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model; using the trained information classification warning model to process the information data to be classified to obtain the emergency level of the information to be classified. This method can capture the fine-grained differences of information at different levels, and combine the information classification standards to make a comprehensive prediction of the information emergency level; through the hierarchical warning model based on multi-label causal relationship, the accuracy of information classification warning is effectively improved, while reducing manpower consumption, the efficiency and accuracy of information analysis are further improved; different from the previous warning method that only learns the correlation between data and a single label, this method dynamically captures its causal relationship with multi-level labels according to the characteristics of the instance, so as to understand the subtle differences between different levels and achieve better classification effect.
[0064] In one embodiment, in the labeled training data set in step 102, the importance level of information increases from level 0; wherein the importance level of information has no upper limit.
[0065] In one embodiment, step 106 includes: inputting the training data set into a text encoder to obtain a text representation. Specifically, first use a word segmenter to split the Chinese intelligence data and map it into word embeddings, and add a [cls] classification header at the front of each intelligence. Then add position information to each word embedding and input it into the pre-trained language model Bert-Chinese, calculate the correlation between each embedding and other embeddings, and generate the representation of this layer. Finally, the representation is input into a fully connected feedforward neural network for nonlinear transformation, which consists of two linear layers and a ReLU activation function.
[0066] The labels of the training dataset are input into the label representation network to obtain the label representation corresponding to the text representation. Specifically, for the label representation network, a multi-layer Transformer encoder is used to encode the labels. Through the self-attention mechanism, the contextual dependencies within the label vocabulary are captured to generate a preliminary representation of the label. Then, the hierarchical information of the label is introduced, and the label is used as a node, and the hierarchical relationship between the labels is used as an edge. The hierarchical information of the label is modeled using a graph neural network. Finally, a multi-layer MLP is used to fuse and model the multi-channel information to obtain the final label representation.
[0067] The text representation, the corresponding label representation and the single label vector are input into the label confusion model to obtain the simulated label distribution. Specifically, the structure of the label confusion model is a multi-channel fusion network. First, each channel receives the text representation, the corresponding label representation and the single label vector respectively. Next, these three types of inputs are subjected to a multi-head self-attention mechanism of a similarity layer for cross-channel feature interaction, allowing different types of inputs to influence each other under the attention mechanism, and gradually fusion of these input representations through a multi-layer self-attention network to generate a global context-related joint representation vector; the joint representation is sent to a label distribution calculation layer composed of a deep MLP network, which contains multiple hidden layers and nonlinear activation functions, aiming to map the joint representation to a multi-dimensional space, where each dimension corresponds to a possible label, and then the softmax function is used to normalize these dimensions to generate the final label probability distribution, which reflects the distribution of the input text in the label space, and the final output represents the matching probability of each label for the text.
[0068] The text representation is input into the Naive Bayes classifier to obtain the predicted label distribution. Specifically, for a given text representation x, the model calculates the likelihood P(x|y) of feature x under a certain label y and the prior probability P(y) of the label. Based on these calculated probability values, the model estimates the probability that the text belongs to each label through the Naive Bayes joint probability formula. The Naive Bayes joint probability formula is:
[0069] ;
[0070] Where P(x) is a normalization constant used to ensure that the sum of all label probabilities is 1.
[0071] The conditional probability P(y|x) of each label is calculated to obtain a multidimensional probability vector. Each element of this vector represents the predicted probability of the text representation under different labels. The final probability distribution is the predicted label distribution. The model uses the maximum a posteriori probability (MAP) principle to take the label corresponding to the maximum value in the probability distribution as the final predicted label, and can also output the confidence of the label.
[0072] The model training loss is calculated using the KL divergence method according to the simulated label distribution and the predicted label distribution. The parameters of the information classification warning model are updated according to the model training loss until the number of training reaches the preset number of iterations to obtain a trained information classification warning model.
[0073] In one embodiment, the text encoder consists of a word segmenter, a pre-trained language model, and a multi-layer feed-forward neural network.
[0074] In one embodiment, the label representation network includes a multi-layer transformer encoder, a multi-layer perceptron, and a graph neural network to fully capture the causal relationship between labels.
[0075] In one embodiment, the label confusion model is a multi-channel fusion network, which consists of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input and calculate the similarity through a dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0076] ;
[0077] in, The confusion distribution is the label confusion distribution, which can obtain the dependency between labels by calculating the similarity between instances and labels; represents the simulated label distribution; represents the true single vector label; is a balancing factor that controls the sensitivity of the algorithm to different levels.
[0078] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.
[0079] In one embodiment, Figure 3 As shown, an information classification warning device based on multi-label causal relationship is provided, including: a labeled training data set construction module, an information classification warning model construction module, an information classification warning model training module and an information classification warning module, wherein:
[0080] The module for constructing an annotated training data set is used to formulate information level specifications based on the total number of preset information classification warning levels and the purpose of information classification; preprocess and assign labels to real text data according to the information level specifications to construct an annotated training data set.
[0081] The information classification warning model construction module is used to construct an information classification warning model based on multi-label causal relationships; the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution based on the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution based on the text representation, and the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through the dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is:
[0082] ;
[0083] in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels.
[0084] The information classification warning model training module is used to train the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model.
[0085] The information classification warning module is used to input the information data to be classified and the formulated standards into the trained information classification warning model to obtain the emergency level of the information to be classified.
[0086] In one of the embodiments, in the labeled training data set in the labeled training data set construction module, the importance level of information increases from level 0; there is no upper limit to the importance level of information.
[0087] In one of the embodiments, the information classification warning model training module is also used to input the training data set into the text encoder to obtain a text representation; input the training data set into the label representation network to obtain a label representation corresponding to the text representation; input the text representation, the corresponding label representation and the single label vector into the label confusion model to obtain a simulated label distribution; input the text representation into the naive Bayes classifier to obtain a predicted label distribution; calculate the model training loss using the KL divergence method based on the simulated label distribution and the predicted label distribution, and update the parameters of the information classification warning model based on the model training loss until the number of training reaches a preset number of iterations to obtain a trained information classification warning model.
[0088] In one of the embodiments, the text encoder in the information classification warning model training module is composed of a word segmenter, a pre-trained language model and a multi-layer feedforward neural network.
[0089] In one of the embodiments, the label representation network in the information classification warning model training module includes a multi-layer transformer encoder, a multi-layer perceptron and a graph neural network to fully capture the causal relationship between labels.
[0090] For the specific limitations of the information classification warning device based on multi-label causal relationships, please refer to the limitations of the information classification warning method based on multi-label causal relationships above, which will not be repeated here. Each module in the above-mentioned information classification warning device based on multi-label causal relationships can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0091] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, an information classification warning method based on multi-label causal relationship is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.
[0092] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0093] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps in the above method embodiment when executing the computer program.
[0094] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0095] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.
Claims
1. An information classification warning method based on multi-label causal relationship, characterized in that: The method comprises: Formulate information level specifications based on the total number of information level warning levels and the purpose of information level classification; Preprocessing and labeling real text data according to the information level specification to construct a labeled training data set; An information classification warning model based on multi-label causal relationship is constructed; wherein the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution according to the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution according to the text representation; the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through a dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is: ; in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels; The information classification warning model is trained using a supervised model training method according to the training data set to obtain a trained information classification warning model; The information data to be classified and the established standards are input into the trained information classification warning model to obtain the emergency level of the information to be classified.
2. The method according to claim 1, characterized in that In the labeled training data set, the importance level of information increases from level 0; there is no upper limit to the importance level of information.
3. The method according to claim 1, characterized in that: The information classification warning model is trained using a supervised model training method according to the training data set to obtain a trained information classification warning model, including: Inputting the training data set into the text encoder to obtain text representation; Inputting the training data set into the label representation network to obtain a label representation corresponding to the text representation; Inputting the text representation, the corresponding label representation and the single label vector into the label confusion model to obtain a simulated label distribution; Inputting the text representation into the Naive Bayes classifier to obtain a predicted label distribution; The model training loss is calculated using the KL divergence method according to the simulated label distribution and the predicted label distribution, and the parameters of the information classification warning model are updated according to the model training loss until the number of training reaches a preset number of iterations to obtain a trained information classification warning model.
4. The method according to claim 1, characterized in that: The text encoder consists of a word segmenter, a pre-trained language model and a multi-layer feedforward neural network.
5. The method according to claim 1, characterized in that The label representation network includes a multi-layer transformer encoder, a multi-layer perceptron and a graph neural network to comprehensively capture the causal relationship between labels.
6. An information classification warning device based on multi-label causal relationship, characterized in that: The device comprises: A labeled training data set construction module is used to formulate information level specifications according to the total number of preset information classification warning levels and information classification purposes; pre-process and label the real text data according to the information level specifications to construct a labeled training data set; An information classification warning model construction module is used to construct an information classification warning model based on multi-label causal relationships; wherein the information classification warning model includes a text encoder, a label representation network, a label confusion model and a naive Bayes classifier; the text encoder is used to extract the text representation of the training data set; the label representation network is used to extract the label representation corresponding to the text representation in the training data set; the label confusion model is used to calculate the simulated label distribution according to the text representation and the label representation; the naive Bayes classifier is used to calculate the predicted label distribution according to the text representation; the label confusion model is a multi-channel fusion network, which is composed of a similarity layer and a label distribution calculation layer using a multi-head attention mechanism; the similarity layer is used to take the label representation and the current instance representation as input, and calculate the similarity through a dot product function; the label distribution calculation layer is used to obtain the label confusion distribution using a neural network with an activation layer; the label confusion distribution is: ; in, is the label confusion distribution; represents the simulated label distribution; represents the true single vector label; is the balance factor, which is a number between 0 and 1 and is used to control the sensitivity of the algorithm to different levels; An information classification warning model training module is used to train the information classification warning model using a supervised model training method according to the training data set to obtain a trained information classification warning model; The information classification warning module is used to input the information data to be classified and the formulated standards into the trained information classification warning model to obtain the emergency level of the information to be classified.
7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 5 is implemented.
