Radar text processing method and related apparatus

Abstract

This application provides a radar text processing method and related apparatus. The method includes: processing the radar text to be processed to obtain a first word vector set; extracting contextual semantic features from the first word vector set to obtain a first semantic feature set; extracting word features from the first word vector set to obtain a first part-of-speech feature set; weighting the first part-of-speech feature set with the corresponding semantic features based on importance to obtain a second part-of-speech feature set; concatenating the second part-of-speech feature set and the first word vector set according to the word order of the radar text to be processed to obtain a second word vector set; extracting contextual semantic features from the second word vector set to obtain a second semantic feature set; and performing label classification processing on the second semantic feature set to obtain a classification label corresponding to each first word vector in the first word vector set, thereby improving the accuracy of word classification processing in radar text.

Description

Technical Field

[0001] This application relates to the technical fields of data processing and artificial intelligence, specifically to a radar text processing method and related apparatus. Background Technology

[0002] In modern warfare, electronic warfare plays a crucial role. With continuous progress and development, it has evolved from an initial means of support into a vital combat force. Numerous research institutions and experts have focused on information technology and refined management, constantly proposing advanced technologies to support their work. The use of systematic big data analysis systems to assess battlefield information has become widespread.

[0003] In processing radar text, traditional methods typically involve keyword tag matching to classify words in the radar text. However, this method fails to consider the correlation between words, resulting in low accuracy in classifying words in radar text. Summary of the Invention

[0004] This application provides a radar text processing method and related apparatus. By employing a radar text processing model combined with word parts of speech and contextual features for classification processing, the accuracy of classifying words in radar text is improved.

[0005] A first aspect of this application provides a radar text processing method, which is applied to a radar text processing model. The radar text processing model includes a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module. The method includes:

[0006] The text preprocessing module processes the radar text to be processed to obtain the first set of word vectors.

[0007] The semantic feature extraction sub-model extracts contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, thus obtaining the first semantic feature set.

[0008] The part-of-speech feature extraction sub-model is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set.

[0009] The attention mechanism module performs importance weighting on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set.

[0010] The text preprocessing module concatenates the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, and obtains the second word vector set.

[0011] The semantic feature extraction sub-model is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set.

[0012] The second semantic feature set is labeled by the classification module to obtain the classification label corresponding to each first word vector in the first word vector set.

[0013] In this example, a radar text processing model is used to classify words in the radar text by combining word parts of speech and contextual features, thereby improving the accuracy of word classification in radar text.

[0014] A second aspect of this application provides a radar text processing model, which includes a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module, wherein...

[0015] The text preprocessing module is used to process the radar text to be processed to obtain the first word vector set;

[0016] A semantic feature extraction sub-model is used to extract contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, so as to obtain the first semantic feature set.

[0017] The part-of-speech feature extraction sub-model is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set.

[0018] The attention mechanism module is used to perform importance weighting processing on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set;

[0019] The text preprocessing module is used to concatenate the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, so as to obtain the second word vector set.

[0020] The semantic feature extraction sub-model is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set.

[0021] The classification module is used to perform label classification processing on the second semantic feature set to obtain the classification label corresponding to each first word vector in the first word vector set.

[0022] A third aspect of this application provides a terminal including a processor, an input device, an output device, and a memory, wherein the processor, input device, output device, and memory are interconnected, wherein the memory is used to store a computer program, the computer program including program instructions, and the processor is configured to invoke the program instructions to execute the step instructions as described in the first aspect of this application.

[0023] A fourth aspect of this application provides a computer-readable storage medium storing a computer program for electronic data interchange, wherein the computer program causes a computer to perform some or all of the steps described in the first aspect of this application.

[0024] A fifth aspect of this application provides a computer program product, wherein the computer program product includes a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the first aspect of this application. The computer program product may be a software installation package. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This application provides a schematic diagram of a radar text processing model as an embodiment;

[0027] Figure 2 This application provides a flowchart illustrating a radar text processing method.

[0028] Figure 3 This is a schematic diagram of the structure of a terminal provided in an embodiment of this application;

[0029] Figure 4 This application provides a schematic diagram of the structure of a radar text processing model. Detailed Implementation

[0030] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0031] The terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0032] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0033] To better understand the business product determination method provided in this application embodiment, a brief introduction to the application scenario of the radar text processing method is given below. The radar text to be processed in this application embodiment can be descriptive text about the radar, radar data text, or radar parameter text, etc. For example, the radar text to be processed could be: "The AN / APG-90 radar currently in service with the United States is an important air defense reach system, developed by Raytheon and installed on the F-16 fighter platform. It is widely used in various military equipment. Its time characteristics are long-term effectiveness, with a pulse width of 0.983 μs. Its repetition interval is 0.00531, its mode is Doppler pulse, and its frequency is 1. With its superior performance and advanced technology, it has become an important choice for many military equipment. Its spatial characteristics are regional activity patterns." This is merely an example and not a specific limitation.

[0034] After extracting the radar text, the words in the radar text need to be classified, and then the classified words are processed again for key information recognition, etc., in order to extract the core information in the radar text.

[0035] Traditional methods for processing radar text typically involve keyword tag matching to classify words. However, this method suffers from low accuracy due to the lack of consideration for word relevance. To address this issue, this application provides a radar text processing method that combines word part-of-speech tags and contextual features with a radar text processing model, thereby improving the accuracy of word classification in radar text.

[0036] The following is a brief introduction to radar text processing models that apply radar text processing methods. For example... Figure 1 As shown, the radar text processing model includes a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module.

[0037] The text preprocessing module can process the radar text to be processed, obtaining a first set of word vectors. Processing the radar text involves segmenting the original unstructured data into sentences and words to obtain the sentences and words of the radar text; then, the vocabulary and part-of-speech tags of each sentence are further extracted. For the extracted words, a pre-trained model is used to encode the words, obtaining word vectors. For the part-of-speech tags, a random encoding method is used to obtain the corresponding vector representation. Therefore, each first word vector in the first set of word vectors includes both the word vector and the part-of-speech tag vector.

[0038] The semantic feature extraction sub-model can include a bidirectional multi-layer LSTM. Specifically, when fusing part-of-speech features with other features, it is necessary to explore the LSTM hidden layer vector h. i The relationship between h and the first part-of-speech feature set P, at this time i To fully express the contextual information of words within a sentence, it's necessary to incorporate part-of-speech (POS) features. After integrating POS features, the learning process needs to acquire semantic representations of words with POS features. Therefore, traditional bidirectional LSTMs are not suitable; a reverse LSTM layer is required for learning. Furthermore, since POS features are related to each input of the forward LSTM, the two processes need to be alternated.

[0039] The multi-layer bidirectional LSTM model first uses a backward LSTM (b_1) whose input is the initialization vector set S to obtain its hidden layer vector set. Then, a forward LSTM (f_1) is constructed, whose input is the word vector q incorporating part-of-speech features. i The hidden layer vector set is obtained. The hidden layer vector sets of LSTM(b_1) and LSTM(f_1) are concatenated to obtain set H, where the hidden layer vectors h in set H are... iThis represents the contextual information of the i-th word in the sentence.

[0040] Since the fusion of part-of-speech features requires the previous hidden layer vector, when inputting the first word vector, a zero vector with the same dimension as the part-of-speech feature needs to be initialized and concatenated with the initialized word vector s1. After completing the forward LSTM, a new backward LSTM (b_2) needs to be built to learn the context information after incorporating the part-of-speech features. The hidden layer vectors of the forward LSTM (f_1) and the backward LSTM (b_2) after incorporating the part-of-speech features are extracted and concatenated. The concatenated vector is set O, which serves as the input to the subsequent fully connected layers.

[0041] During the LSTM loop computation, the impact of invalid words on model learning needs to be considered. Since the number of words read in at one time is fixed, some sentences will contain blank words. A matrix called Mask is used during the learning process, setting the positions of meaningful words to 1 and blank words to 0. Masking blank words during network learning effectively prevents the network from learning a large number of invalid features.

[0042] Furthermore, in the forward LSTM(f_1), the last hidden layer of the bidirectional RNN is concatenated and used as the hidden vector input to LSTM(f_1). This last hidden layer vector contains a preliminary understanding of the entire sentence's part-of-speech (POS) and can serve as external information to aid subsequent learning. The fusion of POS features involves combining the POS features with the initialized word vectors s. i+1 To perform the fusion, the weight matrix Score obtained from the attention mechanism is multiplied with the vectors in set P. The resulting vectors are then added together and then multiplied with s. i+1 The new input q is obtained by concatenating the two inputs. i+1 The specific formula is shown below.

[0043]

[0044] The part-of-speech feature extraction sub-model can be obtained by training an initial bidirectional RNN network model using corresponding samples. In practical application, the set of hidden layer vectors of the part-of-speech feature extraction sub-model can be used as the first part-of-speech feature set P.

[0045] The attention mechanism module performs importance weighting on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set.

[0046] The core function of the attention mechanism module is to selectively focus on input features in order to more effectively capture key information in the data. In this radar text processing application, the attention mechanism module dynamically adjusts its focus on information processing by calculating the interaction strength between different features, i.e., attention weights. Specifically, the attention mechanism module receives a first set of part-of-speech features and a first set of semantic features as input, and uses a multi-head self-attention structure to evaluate the relevance and importance between the various features.

[0047] This attention mechanism module maps the input features to a high-dimensional space through a linear transformation layer to capture more complex feature relationships. In this stage, each input feature vector x is processed by a linear transformation W1x+b, where W1 is the weight matrix and b is the bias term.

[0048] The dot product attention mechanism is used to compute a weight matrix among all features. In this mechanism, the dot product of feature vectors measures their similarity, which is then normalized using a softmax function to calculate the final attention weights. Specifically, the attention weights can be determined using the following formula:

[0049]

[0050] Attention(x i ,x j ) represents the attention weights, W2 is a learnable parameter matrix specifically designed for dot product attention, and x i For the i-th element that needs attention weight determination, x j For the j-th element that requires attention weight determination, x k Let k be the element whose attention weight needs to be determined. Therefore, the above method can strengthen the influence of important features while suppressing relatively unimportant information.

[0051] The resulting weight matrix is ​​then used to weight and combine the input features to generate a weighted feature representation. This representation centrally expresses the semantic and part-of-speech information most critical in the current text processing task. Specifically, the weighted output feature is the sum of the products of the input features and their corresponding attention weights:

[0052]

[0053] Among them, z i x is the parameter output by the weighted operation. i For the i-th element that needs attention weight determination, x j This is the j-th element for which attention weights need to be determined.

[0054] Furthermore, residual connections and layer normalization steps are introduced to enhance the model's training stability and accelerate convergence. Residual connections allow direct transmission of raw input features to the output layer, helping to avoid training problems in deep networks. Layer normalization normalizes the output of each feature, ensuring that the model output has a similar distribution across different training stages.

[0055] The classification module performs label classification processing on the second semantic feature set to obtain the classification label corresponding to each first word vector in the first word vector set.

[0056] Please see Figure 2 , Figure 2 This application provides a flowchart illustrating a radar text processing method according to an embodiment. Figure 2 As shown, the method is applied to a radar text processing model, which includes a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module. The radar text processing method includes:

[0057] S201. The text preprocessing module processes the radar text to be processed to obtain the first set of word vectors.

[0058] The text preprocessing module can process the radar text to be processed to obtain a first word vector set. Processing the radar text involves segmenting the original unstructured data into sentences and words to obtain the sentences and words of the radar text; then, the vocabulary and part-of-speech tags of each sentence are further obtained. For the obtained words, a pre-trained model is used to encode the words to obtain word vectors. For the part-of-speech tags, a random encoding method is used to obtain the corresponding vector representation. Therefore, each first word vector in the first word vector set includes the word vector and the part-of-speech tag vector. The first word vector set S includes s1, s2, s3, ..., s n Among them, s i This represents the first word vector of the i-th word in the sentence.

[0059] S202. The semantic feature extraction sub-model extracts contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, thereby obtaining the first semantic feature set.

[0060] Specifically, each word vector in the first word vector set S can be processed using a multi-layer bidirectional LSTM model. The vectors corresponding to the hidden layers of the multi-layer bidirectional LSTM model are determined as the first semantic features corresponding to the first word vector, thus obtaining the first semantic feature set H. The first semantic feature set H includes h1, h2, h3, ..., hn hi represents the i-th first semantic instance, and the first semantic features include word morphological features and the contextual semantic features of the word in the sentence.

[0061] S203. Extract word features from each word vector in the first word vector set using the part-of-speech feature extraction sub-model to obtain the first part-of-speech feature set.

[0062] The part-of-speech feature extraction sub-model can be obtained by training an initial bidirectional RNN network model using corresponding samples. In practical application, the set of hidden layer vectors of the part-of-speech feature extraction sub-model can be used as the first part-of-speech feature set P.

[0063] The part-of-speech feature extraction sub-model can obtain the part-of-speech of each word in the corresponding radar text statement.

[0064] S204. The attention mechanism module performs importance weighting on the corresponding semantic features in the first part-of-speech feature set and the first semantic feature set to obtain the second part-of-speech feature set.

[0065] The relationship weights between the first part-of-speech feature set and the first semantic feature set can be obtained to get the first weight matrix. Then, the first weight matrix and the first part-of-speech feature set are fused to obtain the second part-of-speech feature set.

[0066] When obtaining the first weight matrix, since the dimension of the first semantic feature in the first semantic feature set is different from the dimension of the first part-of-speech feature, it is necessary to process the dimension of the first semantic feature so that the dimension of the resulting second semantic feature is the same as the dimension of the first part-of-speech feature. By introducing an attention mechanism, the first part-of-speech feature and the corresponding first semantic feature are weighted to obtain the second part-of-speech feature, thereby improving the accuracy of subsequent processing.

[0067] S205. The second part-of-speech feature in the second part-of-speech feature set and the first word vector set are concatenated according to the word order of the radar text to be processed by the text preprocessing module to obtain the second word vector set.

[0068] One method for concatenating the second part-of-speech feature in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed can be: concatenating the second part-of-speech feature with the word vector after the corresponding first word vector to obtain the second word vector.

[0069] For example, the i-th second part-of-speech feature is represented as p i The corresponding first word vector of the i-th word is s. i In this case, p is used. i With si+1 The vectors are concatenated to obtain the i-th second word vector.

[0070] S206. Extract contextual semantic features from the second word vector set using the semantic feature extraction sub-model to obtain the second semantic feature set.

[0071] The second set of word vectors is input into the semantic feature extraction sub-model for computation. The hidden layer output of the semantic feature extraction sub-model is used to determine the elements in the second semantic feature set. The second semantic feature set can be specifically represented as O, and includes o1, o2, o3, ..., o n .

[0072] S207. The second semantic feature set is processed by the classification module to obtain the classification label corresponding to each first word vector in the first word vector set.

[0073] Specifically, elements in the second semantic feature set can be mapped to the label dimension to obtain the label distribution probability of each second semantic feature. Then, the label distribution probability of each second semantic feature is used for label short-range transfer to obtain the classification label corresponding to each second semantic feature. Finally, the classification label corresponding to each second semantic feature is determined as the classification label corresponding to the first word vector, thus obtaining the classification label corresponding to each first word vector.

[0074] This paper describes radar radiation sources and the different characteristics surrounding them from three dimensions: radar system level, combat platform level, and radar radiation source level, providing a comprehensive depiction of radar radiation sources and their surrounding features. These three levels, while relatively independent, are closely interconnected, forming a comprehensive and organic knowledge network that defines the classification labels.

[0075] The specific entity relationship labels for the three levels are defined as shown in Table 1.

[0076] At the radar radiation source level: considering the characteristics of the radar itself and the project background, entity labels such as radar model, function, research and development unit, country, radar system, frequency, repetition interval, and pulse width are defined; relationship labels such as purpose, research and development, affiliation, radar system, frequency, repetition interval, pulse width, installation location, time characteristics, and spatial characteristics are defined.

[0077] At the operational platform level: considering the platform's inherent characteristics and project background, operational platforms are categorized into naval vessels and airborne aircraft. Specific entity labels such as platform model, radar model, manufacturer, and country of origin are defined; relational labels such as onboard, installed at, purpose, affiliation, and operational system are also defined.

[0078] At the operational system level: considering the characteristics of the system itself and the project background, two types of entity labels were defined: system name and platform model; relationship labels such as collaborative operations, intelligence support, and operational command were also defined.

[0079] Table 1. Definition of Entities and Relationships in the Radar Radiation Source Knowledge Graph

[0080]

[0081]

[0082] After obtaining the entity tags and entity relationship tags, the entity tag set can be expanded. Specifically, this can be done as follows:

[0083] Based on the given key entity and relationship tags, entities containing the aforementioned keyword tags are expanded from the general knowledge graph and added to the seed concept set. Then, the relevance evaluation index (PMI) is used to confirm whether to add them to the seed concept set. Specifically, the following formula can be used to determine whether the candidate concept d should be added to the seed concept set:

[0084] R(d,C)=max{concept d,l ,l∈C}

[0085]

[0086] Where R(d, c) represents the semantic association between the candidate concept d and the seed concept set c, and concept d,l Let P(d|l) represent the relevance of concept d to seed concept l, and let P(d|l) represent the co-occurrence probability between the candidate concept and the seed concept. PMI stands for Point Mutual Information, which measures the relevance between the candidate concept and the seed concept.

[0087] When the PMI is greater than the preset threshold, the candidate concept d is added to the seed concept set c. The seed concept set can be understood as the existing concept set, such as the entity set corresponding to radar radiation sources. The candidate concept can be an entity that includes the above keyword tags and is expanded from the general knowledge graph.

[0088] Based on the above entity relationship label definition, the following triples were designed to represent the radar radiation source knowledge graph, as shown in Table 2.

[0089] In addition to the three relatively independent triplets at each level, the radar radiation source knowledge graph connects the three levels through two triplets: radar and platform <platform model, onboard, radar> and platform and system <system name, included, platform model>. This allows for the formation of type tags.

[0090] Table 3-2 Examples of Triplets in the Knowledge Graph of Radar Radiation Sources

[0091]

[0092]

[0093]

[0094] In one possible implementation, the method of performing importance-weighted processing on the corresponding semantic features in the first part-of-speech feature set and the first semantic feature set through the attention mechanism module to obtain the second part-of-speech feature set includes:

[0095] A1. Obtain the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set through the attention mechanism module, and obtain the first weight matrix;

[0096] A2. The first weight matrix and the first part-of-speech feature set are fused through the attention mechanism module to obtain the second word feature set.

[0097] In this context, the magnitude of the first weight in the first weight matrix represents the importance of the part-of-speech tag to its semantic expression. The first weight matrix can be represented as a Score. The first part-of-speech feature set can be represented as P.

[0098] The second part-of-speech feature set can be obtained by multiplying the weight matrix Score with the first part-of-speech feature set P and then adding them together. The second part-of-speech features in the second part-of-speech feature set can represent the part-of-speech context features of words in the context of the play.

[0099] Therefore, the first part-of-speech feature set can be processed by relation weighting to obtain the second part-of-speech feature set, which improves the accuracy of subsequent processing.

[0100] In one possible implementation, the method for obtaining the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set through the attention mechanism module to obtain the first weight matrix includes:

[0101] B1. The semantic features in the first semantic feature set are transformed in dimension by the attention mechanism module to obtain a second semantic feature set. The dimension of the second semantic feature set is the same as the dimension of the first part-of-speech feature in the first part-of-speech feature set.

[0102] B2. The attention mechanism module concatenates the first part-of-speech feature in the first part-of-speech feature set with the corresponding second semantic feature to obtain the first concatenated feature set.

[0103] B3. The attention mechanism module is used to vectorize the splicing features in the first splicing feature set to obtain the splicing feature matrix.

[0104] B4. The initialization parameter matrix is ​​multiplied by the concatenated feature matrix through the attention mechanism module to obtain the reference weight matrix;

[0105] B5. The reference weight matrix is ​​normalized by the attention mechanism module to obtain the first weight matrix.

[0106] In the part-of-speech attention mechanism, a crucial first weight matrix, Score, is needed to represent the magnitude of influence on semantic understanding. Because the dimensions of `hi` and `P` do not match, a dimensionality transformation of `hi` is performed before concatenating it with `P`. The concatenated vector is then mapped through a fully connected layer, transforming it into a vector with the same dimension as the first part-of-speech feature set `P`. The first concatenated feature obtained after linear mapping is input into an activation function to obtain the concatenated feature matrix; here, the tanh function is chosen as the activation function. The initialization parameter matrix is ​​used as the parameter for calculating the weight matrix, multiplied by the concatenated feature matrix to obtain the reference weight matrix. The reference weight matrix can, to a certain extent, express the degree of influence of different parts of speech on semantics. After the reference weight matrix is ​​normalized to between 0 and 1 using a softmax function, the first weight matrix, Score, is obtained.

[0107] In one possible implementation, the method further includes:

[0108] C1. Mark the blank words in the sample data using the Mask matrix to obtain the marked sample data;

[0109] C2. The initial model is trained using the labeled sample data to obtain the semantic feature extraction sub-model.

[0110] The semantic feature extraction sub-model can be a multi-layer bidirectional LSTM model. For example... Figure 1 As shown, the semantic feature extraction sub-model first obtains its hidden layer vector set through a backward LSTM (b_1), whose input is the initialization vector set S. Then, a forward LSTM (f_1) is constructed, whose input is the word vector q incorporating part-of-speech features. i The hidden layer vector set is obtained. The hidden layer vector sets of LSTM(b_1) and LSTM(f_1) are concatenated to obtain set H, where the hidden layer vectors h in set H are... i This represents the contextual information of the i-th word in the sentence.

[0111] Since the fusion of part-of-speech features requires the previous hidden layer vector, when inputting the first word vector, a zero vector with the same dimension as the part-of-speech feature needs to be initialized and concatenated with the initialized word vector s1. After completing the forward LSTM, a new backward LSTM (b_2) needs to be built to learn the context information after incorporating the part-of-speech features. The hidden layer vectors of the forward LSTM (f_1) and the backward LSTM (b_2) after incorporating the part-of-speech features are extracted and concatenated. The concatenated vector is set O, which serves as the input to the subsequent fully connected layers.

[0112] During the LSTM loop computation, the impact of invalid words on model learning needs to be considered. Since the number of words read in at one time is fixed, some sentences will contain blank words. A mask matrix is ​​used during the learning process, setting the positions of meaningful words to 1 and blank words to 0. By masking blank words during the semantic feature extraction sub-model's learning process, it is possible to effectively prevent the semantic feature extraction sub-model from learning a large number of invalid features.

[0113] Furthermore, in the forward LSTM(f_1), the last hidden layer of the bidirectional RNN is concatenated and used as the hidden vector input to LSTM(f_1). This last hidden layer vector contains a preliminary understanding of the entire sentence's part-of-speech (POS) and can serve as external information to aid subsequent learning. The fusion of POS features involves combining the POS features with the initialized word vectors s. i+1 To perform the fusion, the weight matrix Score obtained from the attention mechanism is multiplied with the vectors in set P. The resulting vectors are then added together and then multiplied with s. i+1 The new input q is obtained by concatenating the two inputs. i+1 The specific formula is shown below.

[0114]

[0115] In one possible implementation, the step of performing label classification processing on the second semantic feature set through a classification module to obtain the classification label corresponding to each first word vector in the first word vector set includes:

[0116] D1. The classification module maps the second semantic features in the second semantic feature set to the label dimension to obtain the label distribution probability of each second semantic feature;

[0117] D2. The label distribution probability of each second semantic feature is transferred by the classification module to obtain the classification label corresponding to each second semantic feature.

[0118] D3. The classification module determines the classification label corresponding to each second semantic feature as the classification label corresponding to the first word vector.

[0119] During feature mapping, the classification module uses two fully connected layers to map the second semantic features to the label dimension, obtaining the label distribution probability for each second semantic feature. The classification module then inputs the label distribution probability of each second semantic feature into the CRF layer. The CRF layer constrains the label short-range transfer method, thereby obtaining the classification label corresponding to each second semantic feature.

[0120] For examples consistent with the above embodiments, please refer to... Figure 3 , Figure 3 A schematic diagram of a terminal structure provided in an embodiment of this application is shown in the figure. It includes a processor, an input device, an output device, and a memory. The processor, input device, output device, and memory are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to call the program instructions. The program includes instructions for performing the following steps.

[0121] The text preprocessing module processes the radar text to be processed to obtain the first set of word vectors.

[0122] The semantic feature extraction sub-model extracts contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, thus obtaining the first semantic feature set.

[0123] The part-of-speech feature extraction sub-model is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set.

[0124] The attention mechanism module performs importance weighting on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set.

[0125] The text preprocessing module concatenates the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, and obtains the second word vector set.

[0126] The semantic feature extraction sub-model is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set.

[0127] The second semantic feature set is labeled by the classification module to obtain the classification label corresponding to each first word vector in the first word vector set.

[0128] The above mainly describes the solutions of the embodiments of this application from the perspective of the method execution process. It is understood that, in order to achieve the above functions, the terminal includes the corresponding hardware structure and / or software modules for executing each function. Those skilled in the art should readily recognize that, in conjunction with the units and algorithm steps of the various examples described in the embodiments provided herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0129] This application embodiment can divide the terminal into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0130] For those consistent with the above, please refer to Figure 4 , Figure 4 This application provides a schematic diagram of the structure of a radar text processing model. For example... Figure 4 As shown, the radar text processing model includes a text preprocessing module 401, a semantic feature extraction sub-model 402, a part-of-speech feature extraction sub-model 403, an attention mechanism module 404, and a classification module 405, wherein...

[0131] The text preprocessing module 401 is used to process the radar text to be processed to obtain a first set of word vectors;

[0132] The semantic feature extraction sub-model 402 is used to extract contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, so as to obtain the first semantic feature set.

[0133] The part-of-speech feature extraction sub-model 403 is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set.

[0134] The attention mechanism module 404 is used to perform importance weighting processing on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set;

[0135] The text preprocessing module 401 is used to concatenate the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, so as to obtain the second word vector set.

[0136] Semantic feature extraction sub-model 402 is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set;

[0137] The classification module 405 is used to perform label classification processing on the second semantic feature set to obtain the classification label corresponding to each first word vector in the first word vector set.

[0138] In one possible implementation, the attention mechanism module 404 is specifically used for:

[0139] Obtain the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the first weight matrix;

[0140] The first weight matrix and the first part-of-speech feature set are fused together to obtain the second word feature set.

[0141] In one possible implementation, the attention mechanism module 404 is specifically used to obtain the relationship weights between the corresponding semantic features in the first part-of-speech feature set and the first semantic feature set, and to obtain the first weight matrix, in order to:

[0142] The semantic features in the first semantic feature set are transformed in dimension to obtain the second semantic feature set, and the dimension of the second semantic feature set is the same as the dimension of the first part-of-speech feature in the first part-of-speech feature set.

[0143] The first part-of-speech feature in the first part-of-speech feature set is concatenated with the corresponding second semantic feature to obtain the first concatenated feature set.

[0144] The splicing features in the first splicing feature set are vectorized to obtain the splicing feature matrix;

[0145] Multiply the initialization parameter matrix by the concatenated feature matrix to obtain the reference weight matrix;

[0146] The reference weight matrix is ​​normalized to obtain the first weight matrix.

[0147] In one possible implementation, the model is also used for:

[0148] The blank words in the sample data are marked using the Mask matrix to obtain the marked sample data.

[0149] The initial model is trained using the labeled sample data to obtain the semantic feature extraction sub-model.

[0150] In one possible implementation, the classification module 405 is specifically used for:

[0151] Map the second semantic features in the second semantic feature set to the label dimension to obtain the label distribution probability of each second semantic feature;

[0152] The label distribution probability of each second semantic feature is transferred by label proximity to obtain the classification label corresponding to each second semantic feature;

[0153] The classification label corresponding to each second semantic feature is determined as the classification label corresponding to the first word vector.

[0154] This application also provides a computer storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the radar text processing methods described in the above method embodiments.

[0155] This application also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program that causes a computer to perform some or all of the steps of any of the radar text processing methods described in the above method embodiments.

[0156] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0157] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0158] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0159] The units described as separate components may or may not be physically separate. The components shown as units 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0160] Furthermore, the functional units in the various embodiments of the application 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 as a software program module.

[0161] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0162] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0163] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A radar text processing method, characterized in that, The method is applied to a radar text processing model, which includes a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module. The method includes: The text preprocessing module processes the radar text to be processed to obtain the first set of word vectors. The semantic feature extraction sub-model extracts contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, thus obtaining the first semantic feature set. The part-of-speech feature extraction sub-model is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set. The attention mechanism module performs importance weighting on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set. The text preprocessing module concatenates the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, and obtains the second word vector set. The semantic feature extraction sub-model is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set. The second semantic feature set is labeled by the classification module to obtain the classification label corresponding to each first word vector in the first word vector set.

2. The radar text processing method according to claim 1, characterized in that, The second part-of-speech feature set is obtained by weighting the corresponding semantic features in the first semantic feature set with the first part-of-speech feature set through the attention mechanism module, including: The attention mechanism module obtains the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set, thus obtaining the first weight matrix; The attention mechanism module fuses the first weight matrix with the first part-of-speech feature set to obtain the second word feature set.

3. The radar text processing method according to claim 2, characterized in that, The step of obtaining the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set through the attention mechanism module to obtain the first weight matrix includes: The attention mechanism module transforms the dimensions of the semantic features in the first semantic feature set to obtain a second semantic feature set, wherein the dimensions of the second semantic features are the same as the dimensions of the first part-of-speech features in the first part-of-speech feature set. The attention mechanism module concatenates the first part-of-speech feature in the first part-of-speech feature set with the corresponding second semantic feature to obtain the first concatenated feature set. The attention mechanism module performs vectorization processing on the splicing features in the first splicing feature set to obtain the splicing feature matrix; The attention mechanism module multiplies the initialization parameter matrix with the concatenated feature matrix to obtain the reference weight matrix. The first weight matrix is ​​obtained by normalizing the reference weight matrix through the attention mechanism module.

4. The radar text processing method according to any one of claims 1-3, characterized in that, The method further includes: The blank words in the sample data are marked using the Mask matrix to obtain the marked sample data. The initial model is trained using the labeled sample data to obtain the semantic feature extraction sub-model.

5. The radar text processing method according to claim 4, characterized in that, The step of performing label classification processing on the second semantic feature set through the classification module to obtain the classification label corresponding to each first word vector in the first word vector set includes: The classification module maps the second semantic features in the second semantic feature set to the label dimension to obtain the label distribution probability of each second semantic feature; The classification module performs label proximity transfer on the label distribution probability of each second semantic feature to obtain the classification label corresponding to each second semantic feature; The classification module determines the classification label corresponding to each second semantic feature as the classification label corresponding to the first word vector.

6. A radar text processing model, comprising a text preprocessing module, a semantic feature extraction sub-model, a part-of-speech feature extraction sub-model, an attention mechanism module, and a classification module, wherein, The text preprocessing module is used to process the radar text to be processed to obtain the first word vector set; A semantic feature extraction sub-model is used to extract contextual semantic features from each word vector in the first word vector set according to the word order of the radar text to be processed, so as to obtain the first semantic feature set. The part-of-speech feature extraction sub-model is used to extract word features from each word vector in the first word vector set to obtain the first part-of-speech feature set. The attention mechanism module is used to perform importance weighting processing on the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the second part-of-speech feature set; The text preprocessing module is used to concatenate the second part-of-speech features in the second part-of-speech feature set with the first word vector set according to the word order of the radar text to be processed, so as to obtain the second word vector set; The semantic feature extraction sub-model is used to extract contextual semantic features from the second word vector set to obtain the second semantic feature set. The classification module is used to perform label classification processing on the second semantic feature set to obtain the classification label corresponding to each first word vector in the first word vector set.

7. The radar text processing model according to claim 6, characterized in that, The attention mechanism module is specifically used for: Obtain the relationship weights between the first part-of-speech feature set and the corresponding semantic features in the first semantic feature set to obtain the first weight matrix; The first weight matrix and the first part-of-speech feature set are fused together to obtain the second word feature set.

8. The radar text processing model according to claim 7, characterized in that, In obtaining the relationship weights between the corresponding semantic features in the first part-of-speech feature set and the first semantic feature set to obtain the first weight matrix, the attention mechanism module is specifically used for: The semantic features in the first semantic feature set are transformed in dimension to obtain the second semantic feature set, and the dimension of the second semantic feature set is the same as the dimension of the first part-of-speech feature in the first part-of-speech feature set. The first part-of-speech feature in the first part-of-speech feature set is concatenated with the corresponding second semantic feature to obtain the first concatenated feature set. The splicing features in the first splicing feature set are vectorized to obtain the splicing feature matrix; Multiply the initialization parameter matrix by the concatenated feature matrix to obtain the reference weight matrix; The reference weight matrix is ​​normalized to obtain the first weight matrix.

9. A terminal, characterized in that, The system includes a processor, an input device, an output device, and a memory, which are interconnected. The memory is used to store a computer program, which includes program instructions. The processor is configured to invoke the program instructions to perform the method as described in any one of claims 1-5.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1-5.

Images