Target user mining method and device based on improved multi-layer Stacking model

Abstract

The present invention provides a target user mining method and device based on an improved multi-layer Stacking model, the method comprising: obtaining an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X types of basic classifiers; training each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first and third layers of the target user classification model only include one basic classifier with the highest classification accuracy among multiple basic classifiers, and the second layer includes all types of basic classifiers; obtaining a set of users to be classified, and inputting the user data to be classified into the target user classification model to obtain a target user set. By optimizing the traditional Stacking classification model, the present invention can optimize the classification effect of the traditional model on the data set and improve the classification accuracy of the target users.

Description

Technical Field

[0001] The present invention relates to the field of data storage technology, and in particular to a target user mining method and device based on an improved multi-layer Stacking model. Background Art

[0002] In the communications sector, product marketing is increasingly diversified and user groups are being targeted with increasing precision. This requires targeted marketing to specific groups, such as a ¥3 5G data package from midnight to 8 a.m., targeting high-data consumption and low-cost users between midnight and 8 a.m. When developing precision marketing strategies, sales personnel typically identify a group of target users and make marketing recommendations to them. However, this often results in excessive disruption to users or underperforms marketing efforts. Consequently, the process of identifying target user groups has gradually shifted from manual definition to the use of data mining models. Specifically, data mining models are incorporated into the marketing process, manually identifying a group of prioritized users for marketing push and identifying existing subscribers. Alternatively, past manual marketing data is used as training data to train a basic data mining model, generating a data mining model that can identify target users with a high probability of subscription intent. This trained data mining model is then used to identify user groups with specific discount packages, potentially improving marketing success rates.

[0003] Traditional data mining models often use a single algorithmic model for data classification and prediction, such as decision trees, support vector machines, and K-nearest neighbors. When using a single model for data classification, each model analyzes different data samples with different characteristics. For example, decision trees are more suitable for classifying discrete eigenvalues, while K-nearest neighbors are more suitable for classifying linear eigenvalues. In current data mining applications, sample data often contains multiple types of eigenvalues. Using multiple models for classification can improve classification effectiveness to a certain extent.

[0004] The Stacking classification model is an improved model that uses multiple classification models in the data mining classification process. It uses a basic classifier to analyze and predict the original sample data D, obtaining a result set R. This result set R is then used as a new feature set to form a new dataset for the original dataset D. By classifying and predicting the new dataset and using the classification results as the final model predictions, the Stacking fusion model can produce better classification results than predictions from a single model. Traditional Stacking classification algorithms use the same classification algorithm for both the original and stacked datasets. This can lead to overfitting the selected classification algorithm to the data samples, resulting in suboptimal classification results in actual applications. Summary of the Invention

[0005] In order to solve the above problems in the prior art, the present invention provides a target user mining method and device based on an improved multi-layer Stacking model, which can improve the classification accuracy of target users.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] In a first aspect, the present invention provides a target user mining method based on an improved multi-layer Stacking model, comprising:

[0008] S1. Obtain an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X basic classifiers, where X is a positive integer greater than 1;

[0009] S2. Training each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first and third layers of the target user classification model only include a base classifier with the highest classification accuracy among multiple base classifiers, and the second layer includes all types of base classifiers;

[0010] S3. Obtain a set of users to be classified, and input the user data to be classified into the target user classification model to obtain a target user set.

[0011] The beneficial effects of the present invention are: since there are large differences in the prediction results of sample data from various classification models, the present invention introduces a model with high single model analysis accuracy into the first layer of Stacking, and improves the prediction accuracy of the second layer model by adding the analysis results of the model with high accuracy. Since the classification results of a single model for different data samples fluctuate greatly, such as K-nearest neighbor is more suitable for linear distribution samples, and random forest is more suitable for discrete distribution samples; in order to avoid large differences in prediction results between different classification models in actual applications, the second layer uses a variety of different types of models for comprehensive prediction to avoid defect amplification. In the last layer of the Stacking process, the summary classification of each model is verified separately, and a model with the highest prediction accuracy is selected. By optimizing the traditional Stacking classification model, the classification effect of the traditional model on the data set can be optimized, and the classification accuracy of the target user can be improved.

[0012] Optionally, step S2 includes:

[0013] S21, using the training set D1 in the original data set D to train the X basic classifiers of the first layer respectively, and using the test set D TVerify the classification accuracy of various basic classifiers, use the first classifier with the highest classification accuracy as the first-layer classifier, and predict the original data set D based on the first-layer classifier to obtain the first-layer meta-data set D';

[0014] S22. Use the first-layer metadata set D' to train the X basic classifiers of the second layer to obtain trained second classifiers. Use the X second classifiers as second-layer classifiers and use the second-layer classifiers to predict the first-layer metadata set D' to obtain a second-layer metadata set D".

[0015] S23. Use the second-layer metadata set D″ to train the X basic classifiers in the third layer, and use the third classifier with the highest classification accuracy after training as the third-layer classifier to obtain a target user classification model.

[0016] Optionally, after step S23, the following steps are further included:

[0017] S24. Use the original data set D to test the target user classification model with different numbers of iterations to obtain the classification accuracy of each basic classifier at different numbers of iterations, and assign the iteration number with the highest classification accuracy in each basic classifier to the corresponding basic classifier in the target user classification model.

[0018] According to the above description, in the traditional Stacking model, only a single model is used to predict samples and iterate once. If the accuracy of the first layer iteration is too low, it will affect the result prediction of the second layer. After determining the choice of the basic model, the accuracy of the classification results can be further improved by selecting the number of iterations.

[0019] Optionally, after step S24, the following steps are further included:

[0020] S25. Construct n first classifiers using different initialization parameters. When n is a different positive integer, multiple candidate first-layer classifiers are obtained. Each candidate first-layer classifier is tested using the original data set D to obtain the classification accuracy of each candidate first-layer classifier. The candidate first-layer classifier with the highest classification accuracy is selected as the final first-layer classifier.

[0021] S26. Obtain X types of second classifiers, construct n' classifiers for each type of second classifier, and obtain a total of X*n' classifiers. When n' is a different positive integer, multiple alternative second-layer classifiers are obtained. Each alternative second-layer classifier is then tested to obtain the classification accuracy of each alternative second-layer classifier. The one with the highest classification accuracy among the alternative second-layer classifiers is used as the final second-layer classifier to obtain a trained target user classification model.

[0022] According to the above description, after determining the basic model and the number of iterations, the accuracy of the classification results can be further improved by selecting the number of iteration results added to the original data set.

[0023] Optionally, the basic classifier includes at least two of RandomForest, DecisionTree, SVC, and KNeighbors.

[0024] In a second aspect, the present invention provides a target user mining device based on an improved multi-layer Stacking model, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the device achieves the following:

[0025] S1. Obtain an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X basic classifiers, where X is a positive integer greater than 1;

[0026] S2. Training each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first and third layers of the target user classification model only include a base classifier with the highest classification accuracy among multiple base classifiers, and the second layer includes all types of base classifiers;

[0027] S3. Obtain a set of users to be classified, and input the user data to be classified into the target user classification model to obtain a target user set.

[0028] Optionally, when the processor executes step S2 of the computer program, the steps include:

[0029] S21, using the training set D1 in the original data set D to train the X basic classifiers of the first layer respectively, and using the test set D T Verify the classification accuracy of various basic classifiers, use the first classifier with the highest classification accuracy as the first-layer classifier, and predict the original data set D based on the first-layer classifier to obtain the first-layer meta-data set D';

[0030] S22. Use the first-layer metadata set D' to train the X basic classifiers of the second layer to obtain trained second classifiers. Use the X second classifiers as second-layer classifiers and use the second-layer classifiers to predict the first-layer metadata set D' to obtain a second-layer metadata set D".

[0031] S23. Use the second-layer metadata set D″ to train the X basic classifiers in the third layer, and use the third classifier with the highest classification accuracy after training as the third-layer classifier to obtain a target user classification model.

[0032] Optionally, after the processor executes step S23 of the computer program, the process further includes:

[0033] S24. Use the original data set D to test the target user classification model with different numbers of iterations to obtain the classification accuracy of each basic classifier at different numbers of iterations, and assign the iteration number with the highest classification accuracy in each basic classifier to the corresponding basic classifier in the target user classification model.

[0034] Optionally, after the processor executes step S24 of the computer program, the process further includes:

[0035] S25. Construct n first classifiers using different initialization parameters. When n is a different positive integer, multiple candidate first-layer classifiers are obtained. Each candidate first-layer classifier is tested using the original data set D to obtain the classification accuracy of each candidate first-layer classifier. The candidate first-layer classifier with the highest classification accuracy is selected as the final first-layer classifier.

[0036] S26. Obtain X types of second classifiers, construct n' classifiers for each type of second classifier, and obtain a total of X*n' classifiers. When n' is a different positive integer, multiple alternative second-layer classifiers are obtained. Each alternative second-layer classifier is then tested to obtain the classification accuracy of each alternative second-layer classifier. The one with the highest classification accuracy among the alternative second-layer classifiers is used as the final second-layer classifier to obtain a trained target user classification model.

[0037] Optionally, the basic classifier includes at least two of RandomForest, DecisionTree, SVC, and KNeighbors.

[0038] Among them, the technical effects corresponding to the target user mining device based on the improved multi-layer Stacking model provided by the second aspect refer to the relevant description of the target user mining method based on the improved multi-layer Stacking model provided by the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 This is a schematic diagram of the main process of the target user mining method based on the improved multi-layer Stacking model according to an embodiment of the present invention;

[0040] Figure 2A schematic diagram of the accuracy of different iteration times involved in an embodiment of the present invention;

[0041] Figure 3 This is a schematic diagram of the accuracy of different numbers of first classifiers involved in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the algorithm flow of the trained target user classification model involved in an embodiment of the present invention;

[0043] Figure 5 Schematic diagram of the structure of a target user mining device based on an improved multi-layer Stacking model according to an embodiment of the present invention.

[0044] [Description of Reference Numerals]

[0045] 1: Target user mining device based on improved multi-layer Stacking model;

[0046] 2: Processor;

[0047] 3: Memory. DETAILED DESCRIPTION

[0048] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments described herein. Instead, these embodiments are provided to enable a clearer and more thorough understanding of the present invention and to fully convey the scope of the present invention to those skilled in the art.

[0049] Example 1

[0050] This embodiment is applicable to application scenarios for classifying target user groups. The following describes the specific method in conjunction with the process of defining a user group for a specific discount package.

[0051] When marketing and promoting products in the communications field, it is usually necessary to identify a small range of target users and conduct precision marketing for the target user group. The process of identifying the target user group has gradually changed from manual definition to the introduction of data mining model analysis. This embodiment improves the data mining model for target group identification. By optimizing the traditional Stacking classification model, the traditional Stacking classification model is improved in the classification process by selecting the basic model, the number of iterations, and the number of iteration results added to the original data set, and a new Stacking model is obtained to optimize the classification effect of the traditional model on the data set. The optimized model is used in the process of identifying user groups for specific discount packages, which improves the marketing success rate of the target user group to a certain extent and reduces the phenomenon of user over-interference.

[0052] Please refer to Figures 1 to 4 , a target user mining method based on an improved multi-layer Stacking model includes the following steps:

[0053] S1. Obtain the original dataset and the original Stacking classification model. The original dataset is pre-labeled user data. Each layer of the original Stacking classification model is set with X basic classifiers, where X is a positive integer greater than 1.

[0054] Among them, the original data set is obtained by manually identifying a group of priority experience users for marketing push, and obtaining the subscribed users to obtain the original data set.

[0055] That is, prepare the original data set D = {s i |i=1,…,m}={x i ,y i |i=1,…,m},y∈{c1,…,c l}, where m is the number of samples, c is the classification result, the sample contains l categories, s is the feature and classification data of each sample, x is the original feature value, and y is the classification result. In this embodiment, the original data set D is divided into a training set D1 and a test set D2 according to the ratio of 7:3. T .

[0056] In this embodiment, the basic classifiers include four types: RandomForest (random forest classifier), DecisionTree (decision tree classifier), SVC (support vector machine classifier), and KNeighbors (K nearest neighbor classifier).

[0057] S2. Train each layer of the original Stacking classification model in sequence based on the original dataset to obtain a trained target user classification model. The first and third layers of the target user classification model only include the base classifier with the highest classification accuracy among multiple base classifiers, and the second layer includes all types of base classifiers.

[0058] Since there are large differences in the prediction results of sample data from various classification models, the present invention introduces a model with high single model analysis accuracy into the first layer of Stacking, and improves the prediction accuracy of the second layer model by adding the analysis results of the model with high accuracy. Since the classification results of a single model for different data samples fluctuate greatly, such as K-neighbor is more suitable for linear distribution samples, and random forest is more suitable for discrete distribution samples; in order to avoid large differences in prediction results from different classification models in actual applications, the second layer uses a variety of different types of models for comprehensive prediction to avoid defect amplification. In the last layer of the Stacking process, the summary classification of each model is verified separately, and a model with the highest prediction accuracy is selected.

[0059] Thus, step S2 includes:

[0060] S21, use the training set D1 in the original data set D to train the X basic classifiers in the first layer respectively, and use the test set D T Verify the classification accuracy of various basic classifiers, take the first classifier with the highest classification accuracy as the first-layer classifier, predict the original data set D based on the first-layer classifier, and obtain the first-layer meta-data set D';

[0061] In step S21, the training set D1 is trained by the basic classifiers RandomForest, DecisionTree, SVC, and KNeighbors classification algorithms, and the test set D T Verify the accuracy of each classification algorithm and select the classification algorithm with the highest accuracy.

[0062] In this embodiment, the prediction results of the four basic classifiers are shown in Table 1. Among them, for single model prediction, RandomForest has the highest prediction accuracy. Therefore, RandomForest is used as the first classifier to form the first layer classifier.

[0063] Table 1. Classification results of different classifiers in the first layer

[0064] RandomForest DecisionTree SVC KNeighbors Classification accuracy of a single classifier 77.27 75.45 70.91 75.54 After adding the analysis result feature 77.44 77.36 71.3 77.08

[0065] S22. Use the first-layer metadata set D' to train the X basic classifiers of the second layer to obtain trained second classifiers. Use the X second classifiers as second-layer classifiers and use the second-layer classifiers to predict the first-layer metadata set D' to obtain a second-layer metadata set D".

[0066] Among them, the analysis results of RandomForest are added to the original data set, and then the analysis accuracy of each model is verified. Through the analysis of the sample prediction results, referring to Table 1, it can be seen that after adding the analysis results with higher accuracy, the prediction accuracy of each model in the second layer is greatly improved, and is close to the prediction results of the first round of RandomForest.

[0067] S23. Use the second-layer metadata set D″ to train the X basic classifiers in the third layer, and use the third classifier with the highest classification accuracy after training as the third-layer classifier to obtain a target user classification model.

[0068] Among them, the third layer Stacking uses multiple classification models to summarize the prediction results as shown in Table 2 below:

[0069] Table 2. Classification results of different classifiers in the third layer

[0070] RandomForest DecisionTree SVC KNeighbors Classification accuracy of a single classifier 77.27 75.45 70.91 75.54 Summarize the results after stacking 78.17 77.4 74.25 77.2

[0071] Therefore, in the third layer of Stacking, for the prediction analysis results of the first two layers, RandomForest analysis still has the highest classification accuracy. Therefore, RandomForest is selected as the third classifier.

[0072] S24. Use the original data set D to test the target user classification model with different numbers of iterations to obtain the classification accuracy of each base classifier at different numbers of iterations, and assign the iteration number with the highest classification accuracy in each base classifier to the corresponding base classifier in the target user classification model.

[0073] This example verifies the effectiveness of different iterations of Stacking classification on improving the prediction effect. Multiple rounds of Stacking tests are performed on the sample data set. The test results are shown in Figure 2. Figure 2 and Table 3.

[0074] Table 3. Accuracy of different iteration times

[0075] RandomForest DecisionTree SVC KNeighbors Base classifier (one round) 77.43 75.32 71.07 75.53 Stacking two wheels 77.39 77.39 71.76 76.89 Stacking three wheels 78.44 78.03 74.3 76.95 Stacking four wheels 77.26 76.53 73.65 76.45

[0076] Verification results show that the accuracy of various classification models improves to a certain extent during the first and second rounds of stacking. Beyond a certain number of iterations, classification accuracy regresses to the prediction results of the original data. For this experimental sample, reaching three stacking iterations achieved the greatest improvement in the prediction accuracy of a single model.

[0077] It should be noted that if the number of iterations required for different classifiers to reach the optimal accuracy is different, each classifier has its own optimal number of iterations.

[0078] S25. Construct n first classifiers using different initialization parameters. When n is a different positive integer, multiple alternative first-layer classifiers are obtained. Each alternative first-layer classifier is tested using the original data set D to obtain the classification accuracy of each alternative first-layer classifier. The one with the highest classification accuracy among the alternative first-layer classifiers is used as the final first-layer classifier.

[0079] Among them, n RandomForest algorithm classifiers (RF i |i=1,…,n), that is, when the performance parameters are the same, different random number seeds can be set to construct different numbers of RandomForest algorithm classifications; using the classifier (RF i |i=1,…,n) K-fold cross validation is used to train the training set D1, and after training, the original data set D is predicted to generate the first-layer meta-training set D′={s′ i |i=1,…,m}={x i ,z i ,y i |i=1,…,m},y∈{c1,…,c l}, where z is the algorithm classifier (RF i |i=1,…,n) for each sample s, the classification result z={c rf1 ,…,c rfi |i=1,…,n}.

[0080] Therefore, when the number of Stacking iterations is fixed to 3 in step S23, the classification result C added to the original data set is increased. i {(c1,c2,…,c n |n=1,2,3,…,N}, compare and verify the impact of adding different numbers of classification result features on the accuracy of the classification results.

[0081] Among them, the first round of stacking uses random forest as the base learner. Since the model algorithm used is the same, in order to construct different random forest classifiers, the parameter random_state of each random forest classifier is set to a different value, and random forest classifiers with different structures are constructed to avoid introducing repeated classification results. The comparison of feature numbers in this round is referenced Figure 3 and Table 4.

[0082] Table 4. Accuracy of the first classifier with different numbers

[0083] i value RandomForest DecisionTree SVC KNeighbors 0 71.91 70.44 61.42 66.03 1 72.25 69.86 61.55 66.23 2 72.37 72.17 61.79 66.96 3 72.37 72.37 62.06 67.47 4 72.51 72.42 62.03 67.91 5 72.42 72.42 62.13 68.18 6 72.34 72.42 61.92 68.15 7 72.34 72.42 61.7 68.2

[0084] Depend on Figure 3 As can be seen from Table 4, when the number of the first classifier RandomForest is 4, the classification accuracy is the highest. Therefore, 4 RandomForests are selected as the first-layer classifier.

[0085] S26. Obtain X types of second classifiers, construct n' classifiers for each type of second classifier, and obtain a total of X*n' classifiers. When n' is a different positive integer, multiple alternative second-layer classifiers are obtained. Each alternative second-layer classifier is then tested to obtain the classification accuracy of each alternative second-layer classifier. The one with the highest classification accuracy among the alternative second-layer classifiers is used as the final second-layer classifier to obtain a trained target user classification model.

[0086] Among them, RandomForest, DecisionTree, SVC, and KNeighbors are used respectively. Each algorithm constructs n' classifiers, and there are T=4n' classifiers in total. Then the second-layer classifier is (C i |i=1,…,T). The first-layer metadata set D' of Stacking is divided into training set D'1 and test set D' according to the ratio of 7:3 of the number of samples. T , using the classifier (C i |i=1,…,T) trains the first layer metadata dataset D'1 of Stacking, and predicts the first layer metadata dataset D' after training. i |i=1,…,T) The classification prediction result for each sample is z′ i , then z′={z it =h t (x i ,z i )|i=1,…,m;t=1,…,T}, where h t For each classifier (C i |i=1,…,T) for sample features (x i ,z i |i=1,…,m) predicted classification probability, h t =(p1((c1|x)),…,p t ((c l |x)),t=1,…,T), where p t ((c l |x)) is the tth classifier h t Classify sample x as c l The probability value is calculated as y = {f(c,p) = c×p 2}(c is the classification model's predicted classification of the sample, and p is the probability that the model predicts that the sample belongs to classification c); the resulting stacking second-layer metadata set is

[0087] D″={s″ i |i=1,…,m}={x i ,z′ i , i |i=1,…,m},y∈{c1,…,c l}.

[0088] In this embodiment, the second round of Stacking uses four commonly used classification models as meta-learners, M2{(m RandomForest ,m DecisionTree ,m SVC ,m KNeighbors )|i=4}, four models are used to construct learners, each 4 as a group, a group in total, at this time all the learners in the second layer are recorded as C n {(c1,c2,…,c n |n=4a,a=1,2,…,N}, when the number of groups a is used as a variable, the comparison of sample feature prediction results is shown in Table 5.

[0089] Table 5. Accuracy of different groups

[0090] a value Stacking 0 77.27 1 77.36 2 78.17 3 77.54

[0091] It can be seen from Table 5 that when the number of groups a is 2, that is, after adding 8 classification result features to the original data set in the second round, the final classification effect can be maximized.

[0092] Then, the random forest algorithm is used to perform the stacking second layer metadata set D″={s″ i |i=1,…,m}={x i ,z′i,y i |i=1,…,m},y∈{c1,…,c l} is trained to obtain the final classification rules.

[0093] Combining the above, we can see that by using multiple single model classification prediction data, in the first round of stacking, four RandomForest analysis result features are added to the original data set. In the second round, multiple models are used for fusion prediction. After that, 8 classification result features are added to the original data set. Finally, the third round uses Random Forest for classification, which has the highest accuracy. The target user classification model obtained by training is referenced Figure 4 It is known.

[0094] It should be noted that Tables 1 to 5 represent the validation data for this embodiment. The sample sets used during the validation process may not be identical. Therefore, due to data quality issues with the sample sets, the accuracy rates in different tables may vary. For example, the accuracy rate in Table 4 is significantly lower than that in other tables. However, since multiple groups of experiments in the same table represent the same sample set, this does not affect the validation results of this application.

[0095] S3. Obtain a set of users to be classified, and input the data of the users to be classified into a target user classification model to obtain a target user set.

[0096] Example 2

[0097] Please refer to Figure 5 A target user mining device 1 based on an improved multi-layer Stacking model includes a memory 3, a processor 2, and a computer program stored in the memory 3 and executable on the processor 2. When the processor 2 executes the computer program, the steps in the above-mentioned embodiment 1 are implemented.

[0098] Since the apparatus / devices described in the above embodiments of the present invention are apparatus / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art will be able to understand the specific structure and variations of the apparatus / devices based on the methods described in the above embodiments of the present invention, and thus will not be described in detail here. All apparatus / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0099] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, apparatus, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0100] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (apparatus), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions.

[0101] It should be noted that, in the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims enumerating several means, several of these means may be embodied by one and the same hardware. The use of the words first, second, third etc. is for convenience only and does not indicate any order. These words may be understood as part of the component name.

[0102] In addition, it should be noted that, in the description of this specification, the description of the terms "one embodiment", "some embodiments", "embodiment", "example", "specific example" or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0103] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments after learning the basic creative concept. Therefore, the claims should be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0104] Obviously, those skilled in the art may make various modifications and variations to the present invention without departing from the spirit and scope of the present invention. Thus, if such modifications and variations fall within the scope of the claims and their equivalents, the present invention shall also include such modifications and variations.

Claims

1. The target user mining method based on the improved multi-layer Stacking model is characterized by: Including steps: S1. Obtain an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X basic classifiers, where X is a positive integer greater than 1; S2. Train each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first layer of the target user classification model includes only the first classifier with the highest classification accuracy among the multiple basic classifiers, the second layer includes all types of basic classifiers, and the third layer includes only the third classifier with the highest classification accuracy among the multiple basic classifiers; S3. Obtain a set of users to be classified, and input the user data to be classified into the target user classification model to obtain a target user set.

2. The target user mining method based on the improved multi-layer Stacking model according to claim 1 is characterized in that: The step S2 comprises: S21, using the training set D1 in the original data set D to train the X basic classifiers of the first layer respectively, and using the test set D T Verify the classification accuracy of various basic classifiers, use the first classifier with the highest classification accuracy as the first-layer classifier, and predict the original data set D based on the first-layer classifier to obtain the first-layer meta-data set D'; S22. Use the first-layer metadata set D' to train the X basic classifiers of the second layer to obtain trained second classifiers. Use the X second classifiers as second-layer classifiers and use the second-layer classifiers to predict the first-layer metadata set D' to obtain a second-layer metadata set D". S23. Use the second-layer metadata set D″ to train the X basic classifiers in the third layer, and use the third classifier with the highest classification accuracy after training as the third-layer classifier to obtain a target user classification model.

3. The target user mining method based on the improved multi-layer Stacking model according to claim 2 is characterized in that: After step S23, the following steps are further included: S24. Use the original data set D to test the target user classification model with different numbers of iterations to obtain the classification accuracy of each basic classifier at different numbers of iterations, and assign the iteration number with the highest classification accuracy in each basic classifier to the corresponding basic classifier in the target user classification model.

4. The target user mining method based on the improved multi-layer Stacking model according to claim 3 is characterized in that: After step S24, the following steps are further included: S25. Construct n first classifiers using different initialization parameters. When n is a different positive integer, multiple candidate first-layer classifiers are obtained. Each candidate first-layer classifier is tested using the original data set D to obtain the classification accuracy of each candidate first-layer classifier. The candidate first-layer classifier with the highest classification accuracy is selected as the final first-layer classifier. S26. Obtain X types of second classifiers, construct n' classifiers for each type of second classifier, and obtain a total of X*n' classifiers. When n' is a different positive integer, multiple alternative second-layer classifiers are obtained. Each alternative second-layer classifier is then tested to obtain the classification accuracy of each alternative second-layer classifier. The one with the highest classification accuracy among the alternative second-layer classifiers is used as the final second-layer classifier to obtain a trained target user classification model.

5. The target user mining method based on the improved multi-layer Stacking model according to any one of claims 1 to 4, characterized in that: The basic classifier includes at least two of RandomForest, DecisionTree, SVC, and KNeighbors.

6. A target user mining device based on an improved multi-layer Stacking model, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the following is achieved: S1. Obtain an original data set and an original Stacking classification model, wherein the original data set is pre-labeled user data, and each layer of the original Stacking classification model is provided with X basic classifiers, where X is a positive integer greater than 1; S2. Train each layer of the original Stacking classification model in sequence according to the original data set to obtain a trained target user classification model, wherein the first layer of the target user classification model includes only the first classifier with the highest classification accuracy among the multiple basic classifiers, the second layer includes all types of basic classifiers, and the third layer includes only the third classifier with the highest classification accuracy among the multiple basic classifiers; S3. Obtain a set of users to be classified, and input the user data to be classified into the target user classification model to obtain a target user set.

7. The target user mining device based on the improved multi-layer Stacking model according to claim 6 is characterized in that: When the processor executes the step S2 of the computer program, the step includes: S21, using the training set D1 in the original data set D to train the X basic classifiers of the first layer respectively, and using the test set D T Verify the classification accuracy of various basic classifiers, use the first classifier with the highest classification accuracy as the first-layer classifier, and predict the original data set D based on the first-layer classifier to obtain the first-layer meta-data set D'; S22. Use the first-layer metadata set D' to train the X basic classifiers of the second layer to obtain trained second classifiers. Use the X second classifiers as second-layer classifiers and use the second-layer classifiers to predict the first-layer metadata set D' to obtain a second-layer metadata set D". S23. Use the second-layer metadata set D″ to train the X basic classifiers in the third layer, and use the third classifier with the highest classification accuracy after training as the third-layer classifier to obtain a target user classification model.

8. The target user mining device based on the improved multi-layer Stacking model according to claim 7 is characterized in that: After the processor executes step S23 of the computer program, the method further includes: S24. Use the original data set D to test the target user classification model with different numbers of iterations to obtain the classification accuracy of each basic classifier at different numbers of iterations, and assign the iteration number with the highest classification accuracy in each basic classifier to the corresponding basic classifier in the target user classification model.

9. The target user mining device based on the improved multi-layer Stacking model according to claim 8 is characterized in that: After the processor executes step S24 of the computer program, the method further includes: S25. Construct n first classifiers using different initialization parameters. When n is a different positive integer, multiple candidate first-layer classifiers are obtained. Each candidate first-layer classifier is tested using the original data set D to obtain the classification accuracy of each candidate first-layer classifier. The candidate first-layer classifier with the highest classification accuracy is selected as the final first-layer classifier. S26. Obtain X types of second classifiers, construct n' classifiers for each type of second classifier, and obtain a total of X*n' classifiers. When n' is a different positive integer, multiple alternative second-layer classifiers are obtained. Each alternative second-layer classifier is then tested to obtain the classification accuracy of each alternative second-layer classifier. The one with the highest classification accuracy among the alternative second-layer classifiers is used as the final second-layer classifier to obtain a trained target user classification model.

10. The target user mining device based on the improved multi-layer Stacking model according to any one of claims 6 to 9, characterized in that: The basic classifier includes at least two of RandomForest, DecisionTree, SVC, and KNeighbors.

Images