Disaster tweet classification method based on improved black wing algorithm and xgboost
By combining the improved Black-winged Kirk algorithm with xgboost, the disaster tweet features are extracted using n-grams and TF-IDF technology, and the feature selection is optimized through chaos mapping and Levy flight strategy, the problem of difficult context information in disaster tweet classification is solved, and the classification accuracy and efficiency are improved.
Patent Information
- Application Number
- CN202510431777.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-08
- Publication Date
- 2025-08-01
AI Technical Summary
The prior art is difficult to effectively capture the context information of short text in disaster tweet classification, resulting in high-dimensional problems increasing computational costs and reducing classification performance.
The improved black-winged kite algorithm is combined with xgboost, and features are extracted through n-grams feature model and TF-IDF vectorization technology, combined with chaotic mapping and Levy flight strategy to optimize feature selection, and used xgboost model for classification.
It improves the accuracy and efficiency of disaster tweet classification, reduces the execution time of the feature selection process, and enhances the algorithm's global search ability and feature selection optimization effect.
Smart Images
Figure CN120407790A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of natural language processing, and particularly relates to a disaster tweet classification method based on an improved black-winged kite algorithm and xgboost. Background Technique
[0002] In the Internet era, social media platforms have become important channels for information dissemination during disaster events and play a crucial role in quickly understanding disaster-related content. However, one of the main challenges in using social media in disaster management is to reliably and quickly detect key information from a large amount of text data. Tweet classification for disaster response is a text classification task. Due to the existence of many irrelevant features in the text data, and tweet texts are usually short, informal, and multilingual, etc., it is very easy to have high-dimensional problems, ultimately increasing the computational cost and reducing the classification performance.
[0003] In the prior art, disaster tweet classification mainly relies on natural language processing (NLP) and machine learning. Traditional methods usually use the bag-of-words model (BoW) or TF-IDF for feature extraction and combine models such as naive Bayes, support vector machine (SVM), and random forest for classification. However, these methods are difficult to capture context information when dealing with short texts and may have a large number of redundant features, affecting the classification effect. Summary of the Invention
[0004] The purpose of the present invention is to overcome the deficiencies of the prior art and propose a disaster tweet classification method based on an improved black-winged kite algorithm and xgboost to improve the disaster text classification effect.
[0005] To achieve the above object of the present invention, the following technical solutions are adopted: A disaster tweet classification method based on an improved black-winged kite algorithm and xgboost, comprising the following steps:
[0006] Step S1, data preprocessing: Collect tweet data and perform cleaning, word segmentation, stop word removal, and data splitting on it;
[0007] Step S2, feature extraction and standardization: Use the n-grams feature model to capture the context semantic information of tweets, and use the TF-IDF vectorization technology to convert the split tweet data into a numerical feature matrix, and then perform data standardization processing. The specific steps are as follows:
[0008] Step S2-1, use a combination of the unigram model and the bigram model for the already cleaned data to generate n-grams features and capture the context association and lexical co-occurrence patterns inside the tweets;
[0009] Step S2-2: Use the word frequency statistics method to extract tweet features and convert each tweet into a word frequency matrix. Each row in the word frequency matrix represents a document, each column represents a feature, and the value of the matrix element represents the number of times the feature appears in the tweet.
[0010] Step S2-3: Use the TF-IDF method to extract tweet features, set the threshold parameter δ to limit the dimension of the feature space, and perform normalization processing to make the lengths of all tweet feature vectors consistent.
[0011] Step S3: Feature selection and optimization: Use the black-winged kite optimization algorithm enhanced by the chaotic mapping and Levy flight strategy to optimize and select the feature subset. The specific steps are as follows:
[0012] Step S3-1: Population initialization: Use the combined chaotic mapping and integer mapping methods to generate the initial population. Among them, the chaotic mapping is used to improve the population diversity, and the integer mapping method generates a 0-1 feature selection solution space.
[0013] Step S3-2: Fitness evaluation: Define the fitness function, use the xgboost model with the accuracy as the performance evaluation index, and combine the scale of the feature subset. Through the weight parameter α, a balance is achieved between the two. The fitness function is expressed as:
[0014]
[0015] where a is the accuracy; R is the size of the feature subset; C is the total number of features; α ∈ [0,1], β = 1 - α;
[0016] Step S3-3: Population position update: Update the population through the attack behavior and migration behavior. The update formula for the attack behavior of the black-winged kite in each generation is expressed as:
[0017]
[0018] where and represent the positions of the i-th individual in the j-th dimension at the k-th and (k + 1)-th iteration steps respectively; r is a random number ranging from 0 to 1; p is a constant value of 0.9; K is the total number of iterations; k is the number of iterations completed so far;
[0019] The update formula for the migration behavior of the black-winged kite in each generation is expressed as:
[0020]
[0021] where represents the leader among the black-winged kite individuals in the j-th dimension at the k-th iteration; and respectively represent the position of the \(i\)-th black-winged kite in the \(j\)-th dimension in the \(k\)-th and \((k + 1)\)-th iterations; \(F\) i represents the fitness value of any black-winged kite in the \(j\)-th dimension in the \(k\)-th iteration; \(F\) ri represents the fitness value of the random position of any black-winged kite in the \(j\)-th dimension in the \(k\)-th iteration; \(C(0, 1)\) represents Cauchy mutation; \(m\) represents the update step size;
[0022] Step S3-4: Iteratively optimize until the stop condition is reached, output the optimal solution, and obtain the optimal feature subset best_set.
[0023] Step S4: Classifier training and evaluation: Use the optimized feature subset best_set to train the xgboost classifier, and comprehensively evaluate the performance of the model through accuracy, classification report, ROC curve, and AUC value
[0024] As a preferred technical solution of the present invention, in the population initialization of step S2-2, a combined chaotic mapping is used to initialize each generation of the population, and the specific steps are as follows:
[0025] Step S2-2-1: Adopt three relatively common chaotic mappings;
[0026] Step S2-2-2: Create a two-dimensional array of all zeros, with a total of \(e\) individuals. The number of features is determined by the training set processed by TF-IDF. For each position of each individual, a random initial value is generated, ranging from [0, 1];
[0027] Step S2-2-3: Randomly select a chaotic mapping and apply it to the initial value \(t\) times to obtain the initial population after chaotic mapping.
[0028] As a preferred technical solution of the present invention, the integer mapping in step S3-1 specifically includes: mapping the continuous numerical values obtained by the chaotic initialization method into a 0-1 integer matrix, where 0 indicates that the corresponding feature is not selected, and 1 indicates that the corresponding feature is selected. The specific implementation process is as follows:
[0029] S3-3-1: Use combined chaotic initialization to randomly select each individual in the population, and its form is a continuous numerical array \(X\) of dimension equal to the number of features i = [x1, x2, …, x n , where x j ∈[0, 1], j = 1, 2, 3....n;
[0030] S3-3-2: For each position \(x\) in each individual j ;
[0031] If \(x\) jIf it is ≥ 0.5, then set this feature to 1, indicating that this feature is selected; if x j < 0.5, then set this feature to 0, indicating that this feature is not selected; the individual after integer mapping is represented as an integer array B i =[b1,b2,…,b n , where b j ∈{0,1}; the individual after integer mapping will be used in the feature selection process, and a fitness function is used to evaluate the quality of each individual (solution).
[0032] As a preferred technical solution of the present invention, the Levy flight strategy is incorporated into the black-winged kite algorithm, that is, the Levy flight strategy with probability θ is added to the attack behavior to enhance the global search ability of the algorithm. The updated formula for the improved attack behavior is as follows:
[0033]
[0034] Among them, L(λ) is the description formula of the Levy distribution; u and v are random numbers obeying the normal distribution; λ is the control parameter of the Levy distribution.
[0035] For a disaster tweet classification method based on an improved black-winged kite algorithm and xgboost according to the present invention, compared with the prior art by adopting the above technical solutions, it has the following technical effects:
[0036] 1. The present invention adopts a black-winged kite algorithm combined with combined chaotic initialization and Levy flight strategy, reduces the data dimension, enhances the global search ability of the algorithm, speeds up the convergence speed, and optimizes the feature selection process. Moreover, in order to capture the context information in the text during the feature extraction process, n-grams features are further generated before feature extraction, retaining the information at the lexical level and overcoming the problem of difficulty in capturing the context information at the phrase level.
[0037] 2. The present invention selects the xgboost model for training, reduces the overall execution time in the feature selection process, and improves the classification efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] Figure 1 is the flowchart of the method of the present invention;
[0039] Figure 2 is an integer mapping example;
[0040] Figure 3 is a schematic diagram of the ROC curve and AUC value of Embodiment 1 of the present invention;
[0041] Figure 4 is a schematic diagram of the ROC curve and AUC value of Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE INVENTION
[0042] To enable those skilled in the art to better understand the technical solutions in the embodiments of this application, some symbols in the embodiments of this application are first explained to facilitate the understanding of those skilled in the art.
[0043] Embodiment 1
[0044] This example is a tweet dataset during the explosion of a fertilizer plant in Texas, USA from CrisisLex.org. In this dataset, 1 represents the tweet is associated with the event, and 0 represents not associated, with a total of 10,004 tweets. This dataset has characteristics such as being short, informal, high-dimensional and sparse. Part of the data is shown in Table 1.
[0045] Table 1 Partial data display in the total dataset
[0046]
[0047]
[0048] Such as Figure 1 , the specific application of a disaster tweet classification method based on an improved black-winged kite algorithm in disaster tweet classification. The obtained disaster tweet data is preprocessed, and the text data in Table 1 after preprocessing is subjected to feature extraction, extracting 10,000 features. Then, the black-winged kite algorithm fused with chaotic mapping and Levy flight strategy is used for feature selection, iterating 300 times to obtain the optimal feature subset. Finally, the xgboost model is trained using this feature subset and its performance is evaluated to better complete the disaster tweet classification task. The specific steps are as follows:
[0049] Step 1) (Data preprocessing):
[0050] Step 1-1) Text standardization: Convert all text content to lowercase to eliminate the situation where the same word is regarded as a different feature due to different cases.
[0051] Step 1-2) Remove special characters and links: Remove URL links and special symbols (such as punctuation marks and numbers) in the text, as these contents contribute less to the semantic information of the text and may increase the noise in feature extraction.
[0052] Step 1-3) Stop word filtering: After word segmentation, filter the word segmentation results to remove stop words; retain the words crucial to the text content to enhance the feature representation ability of the model.
[0053] Step 1-4) Word segmentation processing: Perform word segmentation on the text, splitting the sentence into individual words to lay the foundation for subsequent n-grams generation and feature extraction.
[0054] Step 1-5) Split the dataset: Split the preprocessed dataset into a training set and a test set according to a ratio of 8:2.
[0055] Specifically, after data preprocessing of the dataset in this example, the text data shown in Table 2 is obtained:
[0056] Table 2 Data after preprocessing
[0057]
[0058]
[0059] Step 2) (Feature extraction and standardization): Use the n-grams feature model to capture the context semantic information of tweets, and use the TF-IDF vectorization technique to convert the split tweet data into a numerical feature matrix, and then perform data standardization processing. The specific steps are as follows:
[0060] Step 2-1) Use a combination of the unigram model and the bigram model to generate n-grams features for the split training set and test set, introducing the context relationship and lexical co-occurrence pattern in the tweets into the feature representation, which not only retains the information at the word level but also captures the context information at the phrase level;
[0061] Step 2-2) Adopt the word frequency statistics method to extract tweet features and convert each tweet into a word frequency matrix. Each row in the word frequency matrix represents a document, each column represents a feature (such as a certain word or n-gram), and the value of the matrix element represents the number of times this feature appears in the tweet; word frequency statistics can intuitively reflect the distribution of vocabulary in the tweet and is a basic means of tweet featureization;
[0062] Step 2-3) To improve the discriminability of the feature representation, the TF-IDF method is further used here to extract tweet features. The TF-IDF features can take into account both the frequency of a vocabulary in the tweet and its rarity in the entire dataset. Here, it is set to retain at most 10,000 most important features.
[0063] During the TF-IDF feature extraction process, each feature is normalized, and the maximum absolute value of each feature is scaled to 1, so that the feature vectors of each tweet have the same length, eliminating the influence of tweet length and ensuring the comparability between feature vectors.
[0064] Step 3) (Feature selection and optimization): Use the black-winged kite optimization algorithm enhanced by the chaos mapping and Levy flight strategy to optimize and select the feature subset, so as to select the optimal feature subset. The specific steps are as follows:
[0065] Step 3-1) Population initialization: Generate the initial population using a combined chaotic mapping and integer mapping method; where the chaotic mapping is used to improve population diversity, and the integer mapping method generates a 0-1 feature selection solution space. The specific implementation is as follows:
[0066] Step 3-1-1) Chaotic initialization: Here, a combined chaotic mapping is used to initialize each generation of the population. The specific steps are as follows:
[0067] Step 3-1-1-1) Use three relatively common chaotic mappings. Here, the Logistic mapping, Tent mapping, and Sine mapping are selected.
[0068] Step 3-1-1-2) Create a two-dimensional array of all zeros, with a total of 100 individuals. The number of features is determined by the training set processed by TF-IDF. For each position of each individual, a random initial value is generated, ranging from [0, 1].
[0069] Step 3-1-2) Integer mapping: As Figure 2 , the solutions involved in feature selection are integer arrays composed of 0 and 1, where 0 represents that the feature is not selected, and 1 represents that the feature is selected; while the individuals generated by the black-winged kite algorithm are continuous value arrays. Therefore, integer mapping is required to map the continuous values to integer representations for feature selection. The specific implementation process is as follows:
[0070] Step 3-1-2-1) Use combined chaotic initialization to randomly select each individual in the population, which is in the form of a continuous numerical array X i = [x1, x2, …, x n , where x j ∈ [0, 1], j = 1, 2, 3....n.
[0071] Step 3-1-2-2) For each position x j in each individual:
[0072] If x j ≥ 0.5, then set the feature to 1, indicating that the feature is selected.
[0073] If x j < 0.5, then set the feature to 0, indicating that the feature is not selected.
[0074] The individual after integer mapping is represented as an integer array B i = [b1, b2, …, b n , where b j ∈ {0, 1}.
[0075] Specifically, by using the combined chaotic mapping method, individuals in the black-winged kite are generated as [0.34051971, 0.59905227, 0.12844472,..., 0.97485218, 0.9480087, 0.63238268], and they are mapped into an integer array [0, 1, 0,..., 1, 1, 1]. Here, 1 indicates that the corresponding feature at this position is selected for subsequent xgboost classification training; 0 indicates non-selection, and the feature at this position does not participate in the classification training.
[0076] Step 3-2) Fitness evaluation: The individuals after integer mapping will be used in the feature selection process, and the quality of each individual (solution) is evaluated through a fitness function; the xgboost model is used to train the selected features in the fitness function. Specifically, for each mapped integer solution B i , a feature subset is constructed by screening the features with a value of 1. After constructing the feature subset, the xgboost model is trained using this subset, and its accuracy a on the training set is evaluated. Finally, the fitness is determined by the model accuracy, and the population is updated according to the fitness in each generation to select the optimal individual. The fitness calculation formula is as follows:
[0077]
[0078] where a is the accuracy; R is the size of the feature subset; C is the total number of features; α ∈ [0, 1], β = 1 - α;
[0079] Step 3-3) Population position update: The attack behavior and migration behavior of each position of each individual in the population are updated. Since these updates will generate new continuous value solutions, integer mapping needs to be performed again to ensure that the individuals in each generation can be effectively used for feature selection. The specific steps are as follows:
[0080] Step 3-3-1) Perform attack behavior update: Here, a Levy flight strategy with a probability of 0.5 is included. If the randomly generated number is greater than 0.5, the attack behavior before improvement is performed, and the update formula is as follows:
[0081]
[0082] where, and respectively represent the position of the i-th individual in the j-th dimension at the k-th and (k + 1)-th iteration steps; r is a random number ranging from 0 to 1, p is a constant value of 0.9; K is the total number of iterations; k is the number of iterations completed so far.
[0083] If it is less than 0.5, the improved attack behavior, that is, the attack behavior with the Levy flight strategy added, is performed, and the update formula is as follows:
[0084]
[0085] Among them, L(λ) is the description formula of the Levy distribution; u and v are random numbers obeying the normal distribution; λ is the control parameter of the Levy distribution.
[0086] Step 3-3-2) Perform migration behavior update, and the specific update formula is as follows:
[0087]
[0088] Where: represents the leader in the black-winged kite individual in the j-th dimension at the k-th iteration; and represent the positions of the i-th black-winged kite in the j-th dimension at the k-th and (k + 1)-th iterations respectively; F i represents the fitness value of any black-winged kite in the j-th dimension at the k-th iteration; F ri represents the fitness value of the random position of any black-winged kite in the j-th dimension at the k-th iteration; C(0, 1) represents Cauchy mutation; m represents the update step size.
[0089] Step 3-3-3) Perform integer mapping again. Map the updated continuous value solution to an integer representation of 0 or 1. In this way, the improved black-winged kite algorithm can efficiently explore in the continuous value search space and map the solution to an integer after each update to find the optimal subset in the feature selection task.
[0090] Step 3-4) Iteratively optimize 300 times, output the optimal solution, and obtain the optimal feature subset best_set.
[0091] Specifically, the best_set obtained after the black-winged kite iteration of this example dataset is = [1, 0, 1,......0, 0.78405807, 0.33824562]
[0092] Step 4) (Classifier training and evaluation) Use the optimized feature subset best_set to train the xgboost classifier, and comprehensively evaluate the performance of the model through accuracy, classification report, ROC curve, and AUC value.
[0093] Specifically, under the conditions of Windows10 system, Python3.9, AMD R5 4.47GHz CPU, 16GB memory, population size of 100, and 300 iteration times, optimize and classify this example dataset. Finally, the accuracy is 0.987, the classification report is shown in Table 3, and the ROC curve and AUC value are as Figure 3 .
[0094] Table 3 Classification Report
[0095] precision recall f1-score support 0 0.9793 0.9937 0.9864 952 1 0.9942 0.9809 0.9875 1049 accuracy 0.9870 2001 macroavg 0.9867 0.9873 0.9870 2001 weightedavg 0.9871 0.9870 0.9870 2001
[0096] Example 2
[0097] This example is a tweet dataset during the flood in Queensland, Australia from CrisisLex.org. In this dataset, 1 still represents the tweet is associated with the event, and 0 represents not associated, with a total of 10,029 tweets. Part of the data is shown in Table 4 as follows.
[0098] Table 4 Partial data display in this dataset
[0099]
[0100]
[0101] First, after data preprocessing of the dataset in this example, the text data shown in Table 5 is obtained:
[0102] Table 5 Dataset after preprocessing
[0103]
[0104] Next, the n-grams feature model is used to capture the context semantic information of tweets, and the TF-IDF vectorization technology is used to convert the split tweet data into a numerical feature matrix, and then data standardization processing is carried out.
[0105] Then, the black-winged kite optimization algorithm enhanced by chaotic mapping and Levy flight strategy is used to optimize and select the feature subset, so as to select the optimal feature subset.
[0106] Specifically, through the combined chaotic mapping method, individuals in the black-winged kite are generated as [0.31499848, 0.25474993, -3.56101399......3.18572156 0.28908748, 0.46533955], and they are mapped into an integer array [0, 0, 0, ……, 1, 0, 0]. Among them, 1 represents that the feature corresponding to this position is selected for subsequent xgboost classification training; 0 represents not selected, and the feature at this position does not participate in classification training.
[0107] The best_set obtained after the black-winged kite iteration of the dataset in this example is [1.22264244, 0, 1.22264244......0, 0.51195701, 1.12055823]
[0108] Finally, train the XGBoost classifier using the optimized feature subset best_set, and comprehensively evaluate the performance of the model through accuracy, classification report, ROC curve, and AUC value.
[0109] Specifically, under the conditions of Windows 10 system, Python 3.9, AMD R5 4.47GHz CPU, 16GB of memory, a population size of 100, and 300 iterations, optimize and classify the dataset in this example. The final accuracy is 0.979, the classification report is shown in Table 6, and the ROC curve and AUC value are as Figure 4 .
[0110] Table 6 Classification Report
[0111] precision recall f1-score support 0 0.9623 0.9935 0.9776 924 1 0.9943 0.9666 0.9803 1079 accuracy 0.9790 2003 macroavg 0.9783 0.9801 0.9789 2003 weightedavg 0.9795 0.9790 0.9791 2003
[0112] Through case verification, the present invention can preprocess the obtained disaster tweet data, extract features from the preprocessed text data, then use the black-winged kite algorithm that combines chaotic mapping and Levy flight strategy for feature selection to obtain the optimal feature subset, and finally use this feature subset to train the XGBoost model and evaluate its performance to better complete the disaster tweet classification task.
[0113] The specific implementation schemes described above further elaborate on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific implementation schemes of the present invention and are not intended to limit the scope of the present invention. Any equivalent changes and modifications made by those skilled in the art without departing from the concept and principles of the present invention shall fall within the scope of protection of the present invention.
Claims
1. A disaster tweet classification method based on an improved black-winged kite algorithm and xgboost, characterized in that It includes the following steps: Step S1, Data preprocessing: Collect tweet data and perform cleaning, word segmentation, stop word removal, and data splitting on it; Step S2, Feature extraction and standardization: Use the n-grams feature model to capture the context semantic information of tweets, and use the TF-IDF vectorization technology to convert the split tweet data into a numerical feature matrix, and then perform data standardization processing. The specific steps are as follows: Step S2-1, Use a combination of the unigram model and the bigram model on the already cleaned data to generate n-grams features, and capture the context associations and lexical co-occurrence patterns within the tweets; Step S2-2, Adopt a word frequency statistics method to extract tweet features and convert each tweet into a word frequency matrix; each row in the word frequency matrix represents a document, each column represents a feature, and the value of the matrix element represents the number of times this feature appears in the tweet; Step S2-3, Use the TF-IDF method to extract tweet features, set the threshold parameter δ to limit the dimension of the feature space, and at the same time perform normalization processing to make the lengths of all tweet feature vectors consistent; Step S3, Feature selection and optimization: Use the black-winged kite optimization algorithm enhanced by the chaos mapping and Levy flight strategy to optimize and select the feature subset. The specific steps are as follows: Step S3-1, Population initialization: Use a combined chaos mapping and integer mapping method to generate the initial population; among them, the chaos mapping is used to improve the population diversity, and the integer mapping method generates a 0-1 feature selection solution space; Step S3-2, Fitness evaluation: Define the fitness function, use the xgboost model with the accuracy rate as the performance evaluation index, and at the same time combine the scale of the feature subset, and balance the two through the weight parameter α. The fitness function is expressed as: where a is the accuracy rate; R is the size of the feature subset; C is the total number of features; α ∈ [0,1], β = 1 - α; Step S3-3, Population position update: Update the population through the attack behavior and migration behavior. The update formula for the attack behavior of the black-winged kite in each generation is expressed as: wherein, and respectively represent the positions of the ith individual in the jth dimension at the kth and (k + 1)th iteration steps; r is a random number ranging from 0 to 1; p is a constant value of 0.9; K is the total number of iterations; k is the number of iterations completed so far; The update formula for the migration behavior of the black-winged kite in each generation is expressed as: Among them, represents the leader in the j-th dimension of the black-winged kite individual in the k-th iteration; and represent the positions of the i-th black-winged kite in the j-th dimension in the k-th and (k + 1)-th iterations respectively; F i represents the fitness value of any black-winged kite in the j-th dimension in the k-th iteration; F ri represents the fitness value of the random position of any black-winged kite in the j-th dimension in the k-th iteration; C(0, 1) represents Cauchy mutation; m represents the update step size; Step S3-4, Iteratively optimize until the stop condition is reached, output the optimal solution, and obtain the optimal feature subset best_set. Step S4, Classifier training and evaluation: Use the optimized feature subset best_set to train the xgboost classifier, and comprehensively evaluate the performance of the model through the accuracy rate, classification report, ROC curve, and AUC value.
2. The disaster tweet classification method based on the improved black-winged kite algorithm and xgboost according to claim 1, wherein, In the population initialization of Step S2-2, a combined chaos mapping is used to initialize each generation of the population. The specific steps are as follows: Step S2-2-1, Adopt three relatively common chaos mappings; Step S2-2-2, Create a two-dimensional array of all zeros, with a total of e individuals. The number of features is determined by the training set after TF-IDF processing. For each position of each individual, randomly generate an initial value within the range of [0,1]; Step S2-2-3, Randomly select a chaos mapping and apply it to this initial value t times to obtain the initial population after chaos mapping.
3. The disaster tweet classification method based on the improved black-winged kite algorithm and xgboost according to claim 2, wherein The integer mapping in step S3-1 specifically includes: mapping the continuous numerical values obtained by the chaotic initialization method into a 0-1 integer matrix, where 0 indicates that the corresponding feature is not selected and 1 indicates that the corresponding feature is selected. The specific implementation process is as follows: S3-3-1. Initialize each individual in the population by randomly selecting using combined chaos, in the form of a continuous numerical array X of dimension equal to the number of features, i = [x1, x2, …, x n , where x j ∈ [0, 1], j = 1, 2, 3....n; S3-3-2. For each position x in each individual j ; If x j ≥ 0.5, then set this feature to 1, indicating that this feature is selected; if x j < 0.5, then set this feature to 0, indicating that this feature is not selected; the individual after integer mapping is represented as an integer array B i = [b1, b2, …, b n , where b j ∈ {0, 1}; the individual after integer mapping will be used in the feature selection process, and the quality of each individual will be evaluated by a fitness function.
4. The disaster tweet classification method based on the improved black kite algorithm and xgboost according to claim 3 is characterized in that: The Levy flight strategy is integrated into the Black Kite algorithm. That is, the Levy flight strategy with probability θ is added to the attack behavior to enhance the algorithm's global search capability. The improved attack behavior update formula is as follows: Where L(λ) is the description formula of the Levy distribution; u and v are random numbers that obey the normal distribution; λ is the control parameter of the Levy distribution.