Low-quality data emotion metaphor identification method and system based on fuzzy granular ball modeling

Abstract

The invention discloses a low-quality data emotion metaphor identification method and system based on fuzzy granular ball modeling, and relates to the field of natural language processing, the method comprises the following steps: S1, obtaining low-quality text data, and carrying out labeling and semantic preprocessing to obtain text data containing linguistic information; s2, inputting text data containing linguistic information into the word embedding model to obtain sentiment metaphor word vectors and form a word vector matrix; s3, generating a granularity list meeting an inclusion degree threshold value by using fuzzy granular ball calculation, and performing feature reduction on the word vector matrix by using a variable precision dependency function to obtain a reduction matrix; and S4, dividing the reduction matrix into a training set and a test set, and training and predicting the convolutional neural network to obtain an emotion metaphor recognition result. According to the method, feature selection is carried out through fuzzy particle ball calculation, redundant features are deleted, the efficiency of feature extraction of the convolutional neural network model is improved, and the problem that noise information is doped in acquisition of a large amount of text information is solved.

Description

Technical Field

[0001] The present invention relates to the field of natural language processing, and in particular to a method and system for identifying low-quality data emotional metaphors based on fuzzy granular sphere modeling. Background Art

[0002] Metaphor is a linguistic phenomenon used for rhetoric and a cognitive phenomenon. It exists in all aspects of human thinking and language expression and is one of the core issues of artificial intelligence, especially computational linguistics. In order for machines to understand human language and thus better perform intelligent information processing, it is necessary to deeply explore and analyze the characteristics of human language. Metaphor is the main means for humans to describe abstract concepts. In this context, the recognition and understanding of metaphors have important theoretical significance and practical application value. Understanding metaphors will help solve many natural language processing tasks, such as search engines, chatbots, machine translation, and public opinion analysis. In recent years, with the development of third-generation platform technologies such as smart terminals, mobile Internet, and the Internet of Things, especially the rapid rise of Internet media such as WeChat and online shopping platforms, people have widely participated in comments in various fields of social life, generating a large amount of language information with implicit emotional tendencies. These comments often contain a large number of metaphors. Only by analyzing this information quickly and accurately can the needs of public opinion monitoring, intelligent data analysis, etc. be met. In life, emotions are abstract, vague, and difficult to describe. People often use metaphors to express emotions. Therefore, emotional metaphor computing is an important field of artificial intelligence and the key to promoting the development of emotional computing to a deeper level. As a result, a wave of research on metaphors and emotional metaphors has been set off in the academic and industrial circles at home and abroad, and a large number of emotional metaphor computing models have emerged. As a result, emotional metaphor computing has begun a new exploration and achieved many valuable results, but it also faces huge difficulties and challenges.

[0003] With the development of science and technology and the explosive growth of data, large-scale, high-dimensional data sets have emerged. High-dimensional data contains rich semantic information, but often there are a large number of irrelevant and redundant features, which brings great challenges to data analysis. On the one hand, redundant features increase the risk of overfitting of classifiers and easily lead to poor generalization ability of training models. On the other hand, a large number of irrelevant features will lead to high computational complexity. Feature selection is a feature reduction technique. The goal is to find an optimal feature subset to predict the sample category while minimizing information loss. Feature selection can solve the curse of dimensionality, reduce computational costs, and improve the interpretability of learning models. In recent years, feature selection has played a vital role in fields such as data mining, machine learning, and data analysis. Summary of the invention

[0004] In order to solve the problem that the acquired text data contains noise information, resulting in inaccurate emotion metaphor recognition results, the problem that the existence of redundant information in high-dimensional data sets leads to low recognition efficiency, and the problem that traditional machine learning models have weak feature extraction capabilities, the present invention proposes a low-quality data emotion metaphor recognition method and system based on fuzzy granular sphere modeling, which solves the technical problem that the existing emotion metaphor recognition technology obtains a large amount of text data with rich emotional tendencies through crawlers containing noise information or redundant information, resulting in inaccurate emotion metaphor recognition results and insufficient recognition efficiency.

[0005] On the one hand, the low-quality data emotional metaphor recognition method based on fuzzy granular sphere modeling has the following specific steps:

[0006] S1, obtain low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information;

[0007] S2, input the text data containing linguistic information into the word embedding model to obtain the emotional metaphor word vector and form a word vector matrix;

[0008] S3, using fuzzy granular ball calculation to generate a granularity list that meets the inclusion threshold, and using a variable precision dependency function based on the granularity list to perform feature simplification on the word vector matrix to obtain a simplified matrix;

[0009] S4, divide the reduced matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify the metaphor information and emotion categories in the test set in parallel to obtain the emotion metaphor recognition result.

[0010] Preferably, the word embedding model is a Skip-Gram model in the Word2Vec framework.

[0011] Preferably, the granularity list satisfying the inclusion threshold is generated by using fuzzy granular sphere calculation, specifically as follows:

[0012] Fuzzy spheres are calculated according to the inclusion threshold to obtain a sphere list: based on the current feature set, k-means clustering is performed until the inclusion of each sphere reaches the inclusion threshold, and spheres that do not meet the conditions are deleted. After multiple iterations, a sphere list is obtained; the feature set is composed of column vectors of the word vector matrix.

[0013] Preferably, the variable precision dependency function based on the granularity list is used to perform feature simplification on the word vector matrix to obtain the optimal simplification matrix, which is specifically as follows:

[0014] The word vector matrix is ​​used as the data set to calculate the initial particle ball, and the initial particle ball list GB0 and initial dependency are obtained based on the feature set C. The dependency is the value of the variable precision dependency function;

[0015] Perform feature selection and filter feature a in feature set C s :like Remove feature a from feature set C s , that is, let C = C-{a s}, otherwise, retain feature a s ;

[0016] Perform feature selection on each feature in the feature set C to obtain a feature selection result, and restore a reduced matrix based on the feature selection result;

[0017] Where C represents the feature set, which is composed of the column vectors of the word vector matrix; D represents the decision set, D = (d1, d2), d1 represents metaphor information, and d2 represents the emotion category; Represents the set C-{a s}dependency.

[0018] Preferably, the variable precision dependency function of the decision set D with respect to the feature subset B is expressed as The details are as follows:

[0019]

[0020] Among them, |·| represents the cardinality of the set; feature subset C represents the feature set; U represents the sample set, which is composed of the row vectors of the word vector matrix; D represents the decision set, D = (d1, d2), d1 represents metaphor information, and d2 represents the emotion category; Represents the variable precision positive domain of the decision set D relative to the feature subset B; Indicates D t Variable precision lower approximation relative to B; Indicates that the inclusion degree is satisfied GB q The set of samples contained; x represents a sample, x∈U; ∪ represents a union; GB represents a ball list; GB q represents the qth ball, l represents the number of balls, q = 1, 2, ..., l; Indicates the degree of inclusion, represents the granular fuzzy neighborhood of sample x, represents the tth fuzzy decision, r represents the number of decisions, t=1,2,…,r; D t represents the tth sample decision; α represents the inclusion threshold.

[0021] Preferably, for a sample x∈GB q , exist The inclusion degree in It is expressed as:

[0022]

[0023] Among them, |·| represents the cardinality of the set; ∩ represents the intersection.

[0024] Preferably, the particle-sphere fuzzy neighborhood is a fuzzy set, and the particle-sphere fuzzy neighborhood of sample x is Contains all samples y and the corresponding membership of y It is expressed as:

[0025]

[0026] Where y∈U,R B (x,y) represents the fuzzy similarity between sample x and sample y, which is calculated based on the normalized value.

[0027] Preferably, the convolutional neural network includes a convolutional layer, a pooling layer, two fully connected layers and two softmax layers; the convolutional layer is connected to the pooling layer, and then fully connected with two parallel fully connected layers, each fully connected layer is respectively connected to a softmax layer, one softmax layer is used for metaphor category division, and the other softmax layer is used for emotion category division.

[0028] Preferably, in S4, a ten-fold cross validation method is used to divide the reduced matrix into a training set and a test set.

[0029] On the other hand, the low-quality data emotional metaphor recognition system based on fuzzy granular sphere modeling includes the following:

[0030] The data acquisition and preprocessing module is used to acquire low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information;

[0031] The word vector matrix acquisition module is used to input text data containing linguistic information into the word embedding model to obtain the emotional metaphor word vector and form a word vector matrix;

[0032] The feature reduction module is used to generate a granularity list that meets the inclusion threshold using fuzzy granular ball calculation, and to perform feature reduction on the word vector matrix using a variable precision dependency function based on the granularity list to obtain a reduced matrix;

[0033] The model training and testing module is used to divide the reduced matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify the metaphor information and emotion categories in the test set in parallel to obtain the emotion metaphor recognition result.

[0034] Compared with the prior art, the present invention has the following beneficial effects:

[0035] The present invention obtains vector representation by embedding low-quality data, combines the vectors into a matrix, uses fuzzy granular ball processing to obtain a granular ball list that meets the inclusion threshold, determines whether a candidate feature can be reduced by calculating a variable precision dependency function, outputs a matrix after feature selection, and uses it for training and prediction of convolutional neural networks; in the process of obtaining the granular ball list, noise data is removed, and the feature selection process deletes redundant features, leaving features with strong identification capabilities. The efficiency of feature extraction of the convolutional neural network model is improved, the problem of noise information mixed in the acquisition of a large amount of text information is solved, and a higher accuracy emotional metaphor recognition result is obtained, providing a more accurate solution for emotional metaphor related fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] The present invention is further described in detail below in conjunction with the accompanying drawings;

[0037] Figure 1 It is a flow chart of a method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0038] Figure 2 It is a flow chart of a method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0039] Figure 3 Schematic diagram of the sphere generation process of the low-quality data emotion metaphor recognition method based on fuzzy sphere modeling according to an embodiment of the present invention; wherein (a) represents the sphere list after the second iteration; (b) represents the sphere list after the eighth iteration; (c) represents the sphere list after the 21st iteration; (d) represents the sphere list after the 44th iteration; (e) represents the sphere list after the 63rd iteration; (f) represents the sphere list after the 71st iteration; (g) represents the sphere list after the 97th iteration; (h) represents the final sphere list result after the iteration ends;

[0040] Figure 4 A schematic diagram of a reduced matrix and a convolution kernel of a method for identifying low-quality data emotional metaphors based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of a single-layer convolutional neural network structure of a low-quality data emotion metaphor recognition method based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0042] Figure 6 A schematic diagram of the emotion metaphor recognition process of the low-quality data emotion metaphor recognition method based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0043] Figure 7 It is a schematic diagram of the principle structure of a low-quality data emotion metaphor recognition system based on fuzzy granular sphere modeling according to an embodiment of the present invention;

[0044] Figure 8 It is a structural block diagram of a low-quality data emotion metaphor recognition system based on fuzzy granular sphere modeling according to an embodiment of the present invention. DETAILED DESCRIPTION

[0045] The present invention is further described below through specific implementation modes.

[0046] like Figure 1 and Figure 2 As shown in the figure, the low-quality data emotional metaphor recognition method based on fuzzy granular sphere modeling is as follows:

[0047] S1, obtain low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information.

[0048] A large amount of low-quality text information with rich metaphor information and emotional tendencies is quickly obtained from some public network platforms (including Sina Weibo, Baidu Encyclopedia, Meituan Dianping, Zhihu Community and / or Douban Books) through crawlers and other methods. Among them, low-quality text information refers to low-quality text data with a lot of redundant and irrelevant information and low authenticity. Then, an emotional metaphor training dataset is constructed through manual annotation.

[0049] The word segmentation component "jieba" is used to divide text data into a sequence of several words. For example, the sentence "I love Tiananmen Square in Beijing" is divided into "I / love / Beijing / Tiananmen Square" after using the word segmentation component "jieba". After word segmentation, stop words are removed to obtain a clean word sequence.

[0050] S2, input the text data containing linguistic information into the word embedding model to obtain the emotional metaphor word vector and form a word vector matrix.

[0051] Word2Vec is used to train word vectors. Word vector training is to map words to multidimensional real number space, convert them into numerical expressions, and map words to real number vectors through distributed representation methods. Specifically: First, the word library obtained by word segmentation is traversed, and each word is mapped to a unique integer ID to generate an ID dictionary; then the Skip-Gram model in the Word2Vec framework is used for training; finally, the probability of predicting the context of the current word by the context word is used to obtain the final word vector. The word vectors are combined into a word vector matrix, and the row vector x is used. i Represents the word vector corresponding to the i-th word in the sentence, then a sentence S of length d (containing d words after word segmentation) can be represented by the word vector matrix as follows:

[0052]

[0053] Among them, x i is the word vector corresponding to the i-th word in sentence S, k represents the dimension of the word vector, and each element m in the matrix ij Represents the value of the word vector of the i-th word in the j-th dimension.

[0054] S3, using fuzzy granular ball calculation to generate a granularity list that meets the inclusion threshold, and using a variable precision dependency function based on the granularity list to perform feature simplification on the word vector matrix to obtain a simplified matrix.

[0055] The matrix M d×k Each row of is considered as a sample, then the matrix M d×k All row vectors of can form a sample set U = (x1, x2, ..., x d );The matrix M d×k Each column of is regarded as a feature of the sample, then the matrix M d×k All column vectors can constitute the feature set C = (a1, a2, ..., a k ); the sample decision is recorded as D = (d1, d2), where d1 and d2 represent the metaphor information and emotional category of the sample respectively. According to D, the sample set U can be divided into multiple sets {D1, D2, ..., D r}. For each feature a s ∈C, calculate the fuzzy similarity between sample x and sample y in f(x,a s ) indicates that sample x has feature a s Then calculate the fuzzy similarity based on the feature set C Furthermore, the fuzzy neighborhood membership of the sample [x] can be obtained C (y) = R C (x,y),x,y∈U.

[0056] The general decision to classify samples is to give a "yes" or "no" judgment, while the fuzzy decision allows the existence of a "maybe" state, that is, there is a [0,1] range to describe the sample's membership to the category. The fuzzy decision membership of sample x obtained by the decision set D is defined as

[0057]

[0058] Here, |·| represents the cardinality of the set, [x] C is the fuzzy neighborhood of sample x, which is a fuzzy set containing all samples y and their corresponding membership [x] C (y), x,y∈U.

[0059] So we can calculate the fuzzy decision Here is a fuzzy set, which contains all samples x and their corresponding membership

[0060] The sphere calculation is a data grouping method. Samples with small feature value differences will be divided into the same sphere. Inclusion is a quantitative indicator that describes the quality of spheres. The sphere calculation is performed according to the inclusion threshold α (0.5≤α≤1): based on the current feature subset K-means clustering is performed until the inclusion degree of each sphere reaches α. The spheres that do not meet the conditions will be deleted. After multiple iterations, a sphere list GB = {GB q ,q=1,2,...,l}, such as Figure 3 As shown, Figure 3 The inclusion threshold is 0.7. For sample x∈GB q , exist The degree of inclusion in is defined as:

[0061]

[0062] Among them, |·| represents the cardinality of the set, GB q is the qth ball, is the granular fuzzy neighborhood of sample x, which is a fuzzy set containing all samples y and their corresponding membership is the granular fuzzy neighborhood membership of the sample, which is defined as:

[0063]

[0064] The dependency function is the ratio of samples that can be accurately classified under feature subset B to the total samples. Therefore, the dependency function is an indicator for evaluating the ability of feature subset B to identify samples. The dependency function can be calculated to determine whether a candidate feature can be reduced. The variable precision dependency function of decision set D relative to feature subset B is:

[0065]

[0066] Among them, |·| represents the cardinality of the set, U is the total sample set, is the variable precision positive domain of the decision set D relative to the feature subset B, which is defined as:

[0067]

[0068] Among them, ∪ represents the union, D tThe variable precision lower approximation with respect to B is defined as:

[0069]

[0070] Among them, ∪ represents the union, That is, it satisfies the inclusion degree GB q The set of samples included is q=1,2,...,l.

[0071] Before feature selection, the initial particle sphere calculation is performed on the data set, and the initial particle sphere list GB0 and initial dependency are obtained based on the feature set C. For each feature a in the feature set C s :Based on C-{a s}Get the current ball list GB′ and dependency By comparing the dependency function, we determine whether a candidate feature can be reduced. This shows that the feature subset C-{a s} is equal to the feature set C, so feature a can be removed s , then let C=C-{a s}, otherwise, keep a s .

[0072] If all features in C have been checked, the feature selection process is terminated, the feature selection results are obtained, and the reduced matrix is ​​output.

[0073]

[0074] S4, divide the reduced matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify the metaphor information and emotion categories in the test set in parallel to obtain the emotion metaphor recognition result.

[0075] like Figure 4 and Figure 5 As shown in the figure, the convolutional neural network structure is designed, using convolution kernels of various sizes C1, C2, ..., C m , extract the semantic features of the sentence. Specifically, the steps of using the convolution kernel to generate semantic features are: define the convolution kernel This convolution kernel will extract the features of d′ consecutive words in sentence S, using c i Represents the features extracted by M′(i,j), which is calculated as follows:

[0076] c i =f(w·M′(i,j)+b)

[0077] Among them, b∈R is the bias and f is the activation function.

[0078] In the current model, the column dimension of the convolution kernel is consistent with the column dimension of the input matrix, so the output matrix degenerates into a vector, that is, The pooling layer simplifies the features extracted by the convolution layer, aggregates and filters the features. Here, the maximum pooling method is used, that is, extracting The most important features:

[0079]

[0080] After the convolutional layer and pooling layer operations, it is connected to two parallel fully connected layers and Softmax layers. The first fully connected layer and Softmax layer are used to calculate the scores of whether the input sentence is a metaphorical sentence or a non-metaphorical sentence, and classify the samples into the category with high scores. The second fully connected layer and Softmax layer are used to calculate the emotional category score of the input sentence and classify the samples into the emotional categories with high scores (happy, good, angry, sad, fearful, evil, and surprised).

[0081] like Figure 6 As shown, an embodiment of the present invention provides a method for identifying low-quality data emotional metaphors based on fuzzy granular sphere modeling, which performs word embedding processing on low-quality text information to obtain semantic vector representation, combines these vectors into a matrix, and uses fuzzy granular sphere processing to obtain a granular sphere list that meets the inclusion threshold. It determines whether a candidate feature can be reduced by calculating a variable precision dependency function, and outputs the matrix after feature selection. The matrices of the training set and the test set are successively input into the convolutional neural network model for training and prediction, and the identification of whether it is a metaphor and the identification of the emotional category are performed in parallel to obtain the emotional metaphor recognition result. Through such a technical solution, the process of obtaining the granular sphere list removes noise data, and the feature selection process deletes redundant features, leaving features with strong identification ability. The efficiency of the convolutional neural network model in extracting features is improved, the problem of mixed noise information in the acquisition of a large amount of text information is solved, and a higher accuracy emotional metaphor recognition result is obtained, providing a more accurate solution for the field related to emotional metaphor.

[0082] like Figure 7 and Figure 8 As shown, the present invention also discloses a low-quality data emotional metaphor recognition system based on fuzzy granular sphere modeling, comprising:

[0083] The data acquisition and preprocessing module 801 includes an acquisition module and a preprocessing module, which are used to acquire low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information.

[0084] The word vector matrix acquisition module 802 includes a word embedding processing module, which is used to input text data containing linguistic information into the word embedding model, obtain emotional metaphor word vectors, and form a word vector matrix.

[0085] The feature reduction module 803 includes a fuzzy granular ball calculation module, which is used to generate a granularity list that meets the inclusion threshold using fuzzy granular ball calculation, and use a variable precision dependency function based on the granularity list to perform feature reduction on the word vector matrix to obtain a reduced matrix.

[0086] The model training and testing module 804 includes a training module and a testing module, which are used to divide the reduction matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify metaphor information and emotion categories in the test set in parallel to obtain an emotion metaphor recognition result.

[0087] The specific implementation of the low-quality data emotion metaphor recognition system based on fuzzy granular sphere modeling is the same as the low-quality data emotion metaphor recognition method based on fuzzy granular sphere modeling, and will not be repeated in this embodiment.

[0088] The above is only a specific implementation of the present invention, but the design concept of the present invention is not limited thereto. Any non-substantial changes to the present invention using this concept shall be deemed as an infringement of the protection scope of the present invention.

Claims

1. A method for identifying emotional metaphors in low-quality data based on fuzzy granular sphere modeling, characterized in that: The steps include: S1, obtain low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information; S2, input the text data containing linguistic information into the word embedding model to obtain the emotional metaphor word vector and form a word vector matrix; S3, using fuzzy granular ball calculation to generate a granularity list that meets the inclusion threshold, and using a variable precision dependency function based on the granularity list to perform feature simplification on the word vector matrix to obtain a simplified matrix; S4, divide the reduced matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify the metaphor information and emotion categories in the test set in parallel to obtain the emotion metaphor recognition result.

2. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1 is characterized in that: The word embedding model is the Skip-Gram model in the Word2Vec framework.

3. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1 is characterized in that: The granularity list that meets the inclusion threshold is generated by using fuzzy granular sphere calculation, as follows: Fuzzy spheres are calculated according to the inclusion threshold to obtain a sphere list: based on the current feature set, k-means clustering is performed until the inclusion of each sphere reaches the inclusion threshold, and spheres that do not meet the conditions are deleted. After multiple iterations, a sphere list is obtained; the feature set is composed of column vectors of the word vector matrix.

4. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1 is characterized in that: The variable precision dependency function based on the granularity list is used to perform feature simplification on the word vector matrix to obtain the optimal simplification matrix, as follows: The word vector matrix is ​​used as the data set to calculate the initial particle ball, and the initial particle ball list GB0 and initial dependency are obtained based on the feature set C. The dependency is the value of the variable precision dependency function; Perform feature selection and filter feature a in feature set C s :like Remove feature a from feature set C s , that is, let C = C-{a s }, otherwise, retain feature a s ; Perform feature selection on each feature in the feature set C to obtain a feature selection result, and restore a reduced matrix based on the feature selection result; Where C represents the feature set, which is composed of the column vectors of the word vector matrix; D represents the decision set, D = (d1, d2), d1 represents metaphor information, and d2 represents the emotion category; Represents the set C-{a s }dependency.

5. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1 is characterized in that: The variable precision dependency function of the decision set D with respect to the feature subset B is expressed as The details are as follows: Among them, |·| represents the cardinality of the set; feature subset C represents the feature set; U represents the sample set, which is composed of the row vectors of the word vector matrix; D represents the decision set, D = (d1, d2), d1 represents metaphor information, and d2 represents the emotion category; Represents the variable precision positive domain of the decision set D relative to the feature subset B; Indicates D t A variable precision lower approximation relative to B; Indicates that the inclusion degree is satisfied GB q The set of samples contained; x represents a sample, x∈U; ∪ represents a union; GB represents a ball list; GB q represents the qth ball, l represents the number of balls, q = 1, 2, ..., l; Indicates the degree of inclusion, represents the granular fuzzy neighborhood of sample x, represents the tth fuzzy decision, r represents the number of decisions, t=1,2,…,r; D t represents the tth sample decision; α represents the inclusion threshold.

6. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 5 is characterized in that: For a sample x∈GB q , exist The inclusion degree in It is expressed as: Among them, |·| represents the cardinality of the set; ∩ represents the intersection.

7. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 6 is characterized in that: The granular sphere fuzzy neighborhood is a fuzzy set, the granular sphere fuzzy neighborhood of sample x Contains all samples y and the corresponding membership of y It is expressed as: Where y∈U,R B (x,y) represents the fuzzy similarity between sample x and sample y, which is calculated based on the normalized value.

8. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1 is characterized in that: The convolutional neural network includes a convolutional layer, a pooling layer, two fully connected layers and two softmax layers; the convolutional layer is connected to the pooling layer, and then fully connected with two parallel fully connected layers, each fully connected layer is connected to a softmax layer, one softmax layer is used for metaphor category division, and the other softmax layer is used for emotion category division.

9. The method for identifying emotional metaphors from low-quality data based on fuzzy granular sphere modeling according to claim 1, characterized in that: In S4, the reduced matrix is ​​divided into training set and test set using the ten-fold cross validation method.

10. A low-quality data emotional metaphor recognition device based on fuzzy granular sphere modeling, comprising the following: The data acquisition and preprocessing module is used to acquire low-quality text data and perform annotation and semantic preprocessing to obtain text data containing linguistic information; The word vector matrix acquisition module is used to input text data containing linguistic information into the word embedding model to obtain the emotional metaphor word vector and form a word vector matrix; The feature reduction module is used to generate a granularity list that meets the inclusion threshold using fuzzy granular ball calculation, and to perform feature reduction on the word vector matrix using a variable precision dependency function based on the granularity list to obtain a reduced matrix; The model training and testing module is used to divide the reduced matrix into a training set and a test set, use the training set to train the convolutional neural network to obtain a trained convolutional neural network, use the test set to predict the trained convolutional neural network, and identify the metaphor information and emotion categories in the test set in parallel to obtain the emotion metaphor recognition result.

Images