Feature data extraction method and device

Abstract

The invention relates to a feature data extraction method and device, electronic equipment, a computer readable medium and a computer program product. The method comprises the following steps: acquiring user information on a plurality of time nodes; the user information on the multiple time nodes is input into the multiple tree models, multiple sets of calculation result sets are generated, and the calculation result sets comprise multiple calculation sub-results corresponding to the multiple tree models; arranging the plurality of calculation sub-results in each group of calculation result set according to a preset sequence; generating a result vector according to the arrangement sequence and the plurality of calculation sub-results; generating a sequence vector through the result vectors corresponding to the multiple groups of calculation result sets; and extracting feature data of the user information through the sequence vector. According to the method, the characteristics of the tree model and the neural network can be fully combined, and the time sequence characteristics of the user information and the incidence relation among the multiple characteristics are deeply mined, so that characteristic data extraction is adaptively carried out.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer information processing, and in particular, to a feature data extraction method and device, electronic equipment, computer readable medium and computer program product. BACKGROUND

[0002] With the development of big data and artificial intelligence technology, the extraction of user feature data in network user monitoring is crucial in various application scenarios. Through in-depth analysis of user behavior, preferences and historical data, accurate user portraits can be provided for enterprises, and personalized recommendations, precision marketing, behavior prediction and other functions can be supported. The extraction of feature data not only helps enterprises better understand user needs, but also optimizes decision-making and improves business efficiency through data-driven methods.

[0003] Early user feature extraction methods usually rely on simple statistics and rule-based techniques, such as generating features based on user behavior frequency, amount, duration, etc. This method is simple to implement and easy to understand, but it has limited effectiveness in handling complex multi-dimensional data or non-linear relationships, making it difficult to capture deep associations in the data. In existing technology, decision trees and their variants (such as Random Forest, XGBoost, etc.) are common choices for feature extraction. By dividing user data into detailed tree structures, high-dimensional data can be effectively processed and feature interactions can be generated. In particular, integrated learning algorithms such as XGBoost improve the accuracy of feature extraction and the generalization ability of the model. However, tree models are good at handling static user features and cannot actively extract dynamic time-varying features. Therefore, when dealing with time series related information, experts and technical personnel need to conduct multi-faceted investigations and set time parameters and sequence extraction features. However, this method is limited by the cognition and experience of technical personnel and cannot dig deeper feature data.

[0004] Therefore, a new feature data extraction method, device, electronic equipment, computer readable medium and computer program product are needed.

[0005] The above information disclosed in the background section is only intended to strengthen the understanding of the background of the present application, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0006] Therefore, the present application provides a feature data extraction method and device, electronic equipment, computer readable medium and computer program product, which ingeniously combines tree models and neural networks, fully utilizes the strengths of these two models, and thus deeply mines the time series features of user information and the correlation between multiple features, thereby adaptively extracting feature data.

[0007] Other features and advantages of the present application will become apparent from the following detailed description, taken in conjunction with the accompanying drawings, or can be learned by practice of the application.

[0008] According to an aspect of the present application, a feature data extraction method is provided, which comprises: obtaining user information at multiple time nodes; inputting the user information at the multiple time nodes into multiple tree models respectively to generate multiple sets of calculation results, wherein the calculation results set comprises multiple calculation sub-results corresponding to the multiple tree models; arranging the multiple calculation sub-results in each set of calculation results in a preset order; generating a result vector through the arrangement order and the multiple calculation sub-results; generating a sequence vector through the result vectors corresponding to the multiple sets of calculation results; and extracting feature data of the user information through the sequence vector.

[0009] Optionally, inputting the user information at the multiple time nodes into multiple tree models respectively to generate multiple sets of calculation results comprises: inputting the user information at each time node into an XGB model that has been trained, wherein the XGB model comprises multiple tree models; performing calculation on the user information through the multiple tree models; and extracting calculation sub-results of the multiple tree models after the calculation is completed to generate a calculation result set.

[0010] Optionally, extracting calculation sub-results of the multiple tree models after the calculation is completed to generate a calculation result set comprises: extracting a hit leaf node in each tree model in the multiple tree models after the calculation is completed; and generating the calculation sub-result through the leaf node.

[0011] Optionally, generating the calculation sub-result through the leaf node comprises: generating the calculation sub-result through a serial number of the leaf node; and / or generating the calculation sub-result through a corresponding feature parameter of the leaf node.

[0012] Optionally, arranging the multiple calculation sub-results in each set of calculation results in a preset order comprises: arranging the multiple calculation sub-results in each set of calculation results in an order corresponding to the multiple tree models in the XGB model that has been trained.

[0013] Optionally, generating a result vector through the arrangement order and the multiple calculation sub-results comprises: merging the arrangement order and the multiple calculation results as parameters to generate a result vector.

[0014] Optionally, merging the arrangement order and the multiple calculation results as parameters to generate a result vector comprises: determining an incremental index value corresponding to the arrangement order; and merging the index value and its corresponding calculation sub-result according to the arrangement order to generate the result vector.

[0015] Optionally, the sequence vector is generated by the result vectors corresponding to the multiple sets of calculation results, including: arranging the result vectors corresponding to the multiple sets of calculation results in the order of the time nodes corresponding to the result vectors to generate the sequence vector with time characteristics.

[0016] Optionally, the feature data of the user information is extracted by the sequence vector, including: inputting the sequence vector into a neural network model to extract the feature data of the user information.

[0017] Optionally, the feature data of the user information is extracted by inputting the sequence vector into a neural network model, including: determining analysis object data; inputting the analysis object data and the sequence vector into the neural network model; and extracting the feature data of the user information by the neural network model taking the analysis object as a focus point.

[0018] According to an aspect of the present application, a feature data extraction device is provided, which includes: an information module configured to obtain user information at multiple time nodes; a model module configured to input the user information at the multiple time nodes into multiple tree models respectively to generate multiple sets of calculation results, wherein the calculation results set includes multiple calculation sub-results corresponding to the multiple tree models; a sorting module configured to arrange the multiple calculation sub-results in each set of calculation results in a predetermined order; a vector module configured to generate a result vector by the arrangement order and the multiple calculation sub-results; a sequence module configured to generate a sequence vector by the result vectors corresponding to the multiple sets of calculation results; and a feature module configured to extract feature data of the user information by the sequence vector.

[0019] According to an aspect of the present application, an electronic device is provided, which includes: one or more processors; a storage device configured to store one or more programs; and when the one or more programs are executed by the one or more processors, the one or more processors implement the method as described above.

[0020] According to an aspect of the present application, a computer readable medium is provided, which stores a computer program, and the program is executed by a processor to implement the method as described above.

[0021] According to an aspect of the present application, a computer program product is provided, which includes: computer programs / instructions, and the computer programs / instructions are executed by a processor to implement the method as described above.

[0022] According to the feature data extraction method, device, electronic equipment, computer readable medium and computer program product, the user information at the plurality of time nodes is obtained; the user information at the plurality of time nodes is input into a plurality of tree models respectively to generate a plurality of sets of calculation results, wherein the plurality of sets of calculation results include a plurality of calculation sub-results corresponding to the plurality of tree models; the plurality of calculation sub-results in each set of calculation results are arranged in a preset order; a result vector is generated through the arrangement order and the plurality of calculation sub-results; a sequence vector is generated through the result vectors corresponding to the plurality of sets of calculation results; and the feature data of the user information is extracted through the sequence vector. In this way, the tree model and the neural network are ingeniously combined, the advantages of the two models are fully utilized, the time sequence features of the user information and the correlation between the multiple features are deeply mined, the feature data is adaptively extracted, and the efficiency and accuracy in the real-time processing of the dynamic user data in network monitoring are effectively improved.

[0023] It should be understood that the foregoing general description and the following detailed description are only examples and are not limiting the present application. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description when taken in conjunction with the accompanying drawings in which:

[0025] Figure 1 is a flowchart of a feature data extraction method according to an example embodiment.

[0026] Figure 2 is a flowchart of a feature data extraction method according to another example embodiment.

[0027] Figure 3 is a schematic diagram of a feature data extraction method according to another example embodiment.

[0028] Figure 4 is a flowchart of a feature data extraction method according to another example embodiment.

[0029] Figure 5 is a block diagram of a feature data extraction device according to another example embodiment.

[0030] Figure 6 is a block diagram of an electronic device according to an example embodiment. DETAILED DESCRIPTION

[0031] Example embodiments now will be described more fully hereinafter with reference to the accompanying drawings. Example embodiments, however, can be implemented in many different forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of example embodiments to those skilled in the art. Like reference numerals refer to like elements throughout the several views and the description of the figures.

[0032] Figure 1 is a flowchart of a feature data extraction method according to an example embodiment. The feature data extraction method 10 includes at least steps S102-S112.

[0033] As shown in S102, user information at multiple time nodes is obtained. More specifically, relevant data of a user can be collected at different time points. For example, for a user of a consumption platform, financial information of the user such as income, expenditure, credit score, total consumption, number of returns, etc. can be collected every month. Figure 1

[0034] Collecting user information at multiple time nodes can reflect the trend of user behavior changes, such as whether the number of orders of the user is increasing or decreasing, whether the purchase categories of the user are increasing or decreasing. More specifically, the selection of time nodes can depend on business needs, such as collecting data weekly, monthly, quarterly, or annually.

[0035] In S104, the user information at the multiple time nodes is input into multiple tree models to generate multiple sets of calculation results, wherein the set of calculation results includes multiple calculation sub-results corresponding to the multiple tree models.

[0036] In an embodiment, user information at each time node can be input into a trained XGB model, which includes multiple tree models. The user information is calculated by the multiple tree models, and calculation sub-results of the multiple tree models are extracted after the calculation to generate a set of calculation results.

[0037] In actual application scenarios, XGB (eXtreme Gradient Boosting) model can be used to process user data. XGB model is a machine learning algorithm based on decision trees, which can effectively process structured data and generate models with high prediction accuracy. XGB model contains multiple decision trees, and each tree makes a prediction on the input data and outputs a “calculation sub-result”. When the input is user information from multiple time nodes, the model generates multiple sets of prediction results based on the data at each time point.

[0038] ​For example, assuming there is a user's monthly product purchase data (such as the purchase amount and payment information in January, February, and March), each month of data will be processed by all tree models in the trained XGB model, and each tree outputs a corresponding calculation result. These calculation results constitute a data set for each month.

[0039] In S106, the plurality of calculation sub-results in each set of calculation results is arranged in a predetermined order. For example, the plurality of calculation sub-results in each set of calculation results can be arranged in the order corresponding to the plurality of tree models in the trained XGB model.

[0040] More specifically, the data result set for each time node can be sorted. Since the XGB model includes a plurality of tree models, each tree corresponds to a calculation sub-result. Therefore, the calculation results can be arranged in the order of the trees in the XGB model to maintain consistency. Of course, the order of arrangement and combination can also be adjusted according to actual business needs or different focuses, and the application is not limited in this regard.

[0041] For example, assuming there are three trees in the XGB model, and the results of the three trees for a certain time node (such as data in January) are 0.4, 0.6, and 0.7. Then, according to the order of the trees, the final calculation result set will be arranged as [0.4, 0.6, 0.7].

[0042] In S108, a result vector is generated by the arrangement order and the plurality of calculation sub-results. For example, the arrangement order and the plurality of calculation results can be combined to generate a result vector as parameters.

[0043] More specifically, for example, an increasing index value corresponding to the arrangement order can be determined; the index value and its corresponding calculation sub-result can be combined according to the arrangement order to generate the result vector. The arranged calculation sub-results are combined to generate a "result vector". The result vector can be understood as a list containing a plurality of elements, each element corresponding to the calculation result of a certain tree for the data of a time node.

[0044] More specifically, the increasing index value can be determined according to the arrangement order, for example, a weight or identifier can be assigned to each sub-result. Assuming the three sub-results are 0.4, 0.6, and 0.7, they can be assigned corresponding identifiers (such as tree 1, tree 2, and tree 3), and then combined into a result vector [tree 1:0.4, tree 2:0.6, tree 3:0.7].

[0045] In S110, a sequence vector is generated by the result vectors corresponding to the plurality of sets of calculation results. The result vectors corresponding to the plurality of sets of calculation results are arranged in the order of their corresponding time nodes to generate a sequence vector with time characteristics.

[0046] The result vectors of multiple time nodes are arranged in chronological order, thereby generating a sequence vector. The sequence vector reflects the characteristics of the user over time, and can capture the dynamic changes in user behavior.

[0047] For example, assume there are three time nodes (January, February, and March), and the result vectors of each time node are [Tree 1: 0.4, Tree 2: 0.6, Tree 3: 0.7], [Tree 1: 0.5, Tree 2: 0.7, Tree 3: 0.8], and [Tree 1: 0.3, Tree 2: 0.5, Tree 3: 0.6], respectively. Combining these vectors in chronological order, the generated sequence vector will reflect the behavior changes of the user's purchase amount in these three months.

[0048] In S112, feature data of the user information is extracted through the sequence vector. For example, the sequence vector can be input into a neural network model to extract the feature data of the user information. The generated sequence vector is further processed by the neural network model. Neural networks are a type of deep learning model that is good at extracting high-level features from complex, non-linear data. After inputting the sequence vector into the neural network, the neural network can learn the user's behavior patterns and extract deeper feature data. For example, in the authorized authentication scenario, when credit assessment is performed on the monitored user, these features can help determine whether the user has credit risk (i.e., whether there is a risk of loss of credit, low credit, etc. that may compromise data / resource / platform security), thereby supporting business decisions.

[0049] In one specific embodiment, assume that a user's purchase of product A in January, February, and March is 1000, 1200, and 1500, respectively, and the purchase of product B is 500, 400, and 300, respectively. Input the data of these three months into the trained XGB model, and generate a calculation result for each tree, such as:

[0050] January: [0.4, 0.5, 0.6]

[0051] February: [0.45, 0.55, 0.65]

[0052] March: [0.35, 0.4, 0.5]

[0053] Arrange these result vectors in chronological order to obtain the sequence vector:

[0054] [0.4, 0.5, 0.6]

[0055] [0.45, 0.55, 0.65]

[0056] [0.35, 0.4, 0.5]

[0057] By processing these sequence vectors through a neural network, the long-term purchase habit characteristics of the user can be extracted, helping the platform to predict future behavior trends.

[0058] According to the feature data extraction method of the present application, user information at multiple time nodes is obtained; the user information at the multiple time nodes is respectively input into multiple tree models to generate multiple sets of calculation results, wherein the calculation result set includes multiple calculation sub-results corresponding to the multiple tree models; the multiple calculation sub-results in each set of calculation results are arranged in a preset order; a result vector is generated through the arrangement order and the multiple calculation sub-results; a sequence vector is generated through the result vectors corresponding to the multiple sets of calculation results; and the feature data of the user information is extracted through the sequence vector. This method ingeniously combines tree models and neural networks, fully utilizes the strengths of the two models, and thus deeply mines the time sequence characteristics of the user information and the correlation between multiple features, thereby adaptively extracting feature data.

[0059] It should be clearly understood that the present application describes how to form and use specific examples, but the principles of the present application are not limited to any details of these examples. On the contrary, based on the teachings of the disclosure of the present application, these principles can be applied to many other embodiments.

[0060] Figure 2 is a flowchart of a feature data extraction method according to another exemplary embodiment. Figure 2 The flowchart 20 shown is for Figure 1 The detailed description of S104 "inputting the user information at the multiple time nodes into the multiple tree models respectively to generate multiple sets of calculation results" in the flowchart shown.

[0061] As shown in Figure 2 In S202, the user information at each time node is input into the trained XGB model, which includes multiple tree models.

[0062] Figure 3 The structure diagram of an XGB model is shown in Figure 3 The model is composed of multiple sub-tree models (decision trees), and in general, the number of sub-tree models is between 100 and 1000. Integrated learning is performed through multiple sub-tree models, and each tree is adjusted and optimized according to the error of the previous tree. Each tree has a different structure, splits according to the input user features, and filters out the nodes that can best distinguish the input features layer by layer until the leaf nodes are reached. The leaf node is the final output point of the decision tree, representing the final prediction result of the current input data.

[0063] More specifically, after inputting user data into the XGB model, the model splits layer by layer according to these features. The root node of each tree will make the first split according to a certain feature (such as income level). For example, assume that the root node of the tree model makes a decision according to whether the income is greater than 5000 yuan. If the income is greater than 5000 yuan, the data will flow to the left node, otherwise it will flow to the right node. Next, the internal node continues to split according to other features (such as age, debt ratio, etc.) until it reaches the leaf node.

[0064] After a series of feature splits, the input data will eventually fall into a certain leaf node. Each leaf node has a unique number and corresponds to a specific prediction result (such as credit level, risk score, or category label, etc.).

[0065] In S204, the user information is calculated by the plurality of tree models. The plurality of tree models in the XGB model respectively calculate the input user data. Each tree makes decisions according to the input user features, divides the data layer by layer, and finally gives a prediction result at the leaf node.

[0066] For example, if a user's purchase amount in January is 1000 yuan and the number of returns is 500, the trees in the model will calculate layer by layer through these feature values, and finally classify the user's situation to a certain leaf node of the tree, and generate a specific prediction result, such as the user's business risk is 0.6 (the value range is 0 to 1, the larger the value, the higher the risk).

[0067] In S206, after the calculation is completed, the calculation sub-results of the plurality of tree models are extracted to generate a calculation result set. After the calculation is completed, for example, the hit leaf nodes in each tree model of the plurality of tree models are extracted; and the calculation sub-results are generated through the leaf nodes.

[0068] Once all the tree models in the XGB model complete the calculation, their calculation results can be extracted, and a calculation result set containing the outputs of all tree models can be generated. Each element in this set represents the calculation result of a user data by a tree. Specifically, each tree of XGB will have multiple leaf nodes, and each input data will eventually "hit" a specific leaf node, and the prediction result of the model is determined based on the feature parameters represented by this leaf node.

[0069] Among them, the leaf node in the decision tree is the final output point, and after the data passes through several layers of decisions from the root of the tree, it will finally fall into a certain leaf node. This leaf node corresponds to a certain classification or regression result.

[0070] For example, if the user's purchase amount in January is 1000 yuan and the return is 500, a certain tree in the XGB model may divide the user into a certain leaf node according to these values, which represents the user's business risk score under this data condition.

[0071] More specifically, the calculation sub-result is generated through the leaf node, which may be generated, for example, through the serial number of the leaf node; more specifically, each leaf node of a tree usually has a unique identification number, and different identification numbers represent different tree model characteristic parameters, which can be used as part of the calculation sub-result.

[0072] More specifically, the calculation sub-result is generated through the leaf node, which may be generated, for example, through the corresponding characteristic parameters (labels or scores) of the leaf node. The characteristic parameters contained in the leaf node can be used to generate the calculation sub-result. For example, a certain leaf node may represent "risk is 0.6" or "normal business probability is 80%". These characteristic parameters can be further used to generate specific calculation sub-results.

[0073] Figure 4 is a flowchart of a feature data extraction method according to another exemplary embodiment. Figure 4 The flowchart 40 shown is a detailed description of Figure 1 The flowchart S112 "extracting feature data of user information through the sequence vector" is shown in detail.

[0074] As shown in the above description, the sequence vector is generated through the analysis object data and the user information. Figure 4 As shown in S402, the analysis object data is determined. The analysis object data can be set according to business needs. For example, when the business needs require in-depth analysis of whether there is a certain correlation between the user's behavior habit characteristics and the climate, the analysis object can be set as the climate data, and the corresponding climate data can be obtained according to the time point described above.

[0075] As described above, assuming that a certain user purchases A products in January, February and March in the amount of 1000, 1200 and 1500 respectively, and purchases B products in the amount of 500, 400 and 300 respectively. The climate data of January, February and March can be obtained as the analysis object data, and the climate data can be mathematically processed to average the data of January, February and March respectively, and the climate data of the user's region can also be obtained in combination with the user's geographic location.

[0076] It is worth mentioning that the sequence vector can also be directly input into a neural network model to extract hidden features between each user sub-information in the user information, and then generate the feature data of the user information.

[0077] In S404, the analysis object data and the sequence vector are input into the neural network model. The climate data of January, February and March, and the user information sequence vector corresponding to January, February and March can be input into the neural network model.

[0078] In S406, the neural network model extracts the feature data of the user information as the analysis object.

[0079] Those skilled in the art can understand that all or part of the steps of the above-mentioned embodiments are implemented as a computer program executed by a CPU. When the computer program is executed by the CPU, the above-mentioned functions defined by the above-mentioned method provided by the present application are executed. The program can be stored in a computer readable storage medium, which can be a read-only memory, a magnetic disk or an optical disk, etc.

[0080] In addition, it should be noted that the above-mentioned figures are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present application, and are not for limiting purposes. It is easy to understand that the processes shown in the above-mentioned figures do not indicate or limit the time sequence of these processes. In addition, it is also easy to understand that these processes can be executed synchronously or asynchronously, for example, in multiple modules.

[0081] The following is a device embodiment of the present application, which can be used to execute the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0082] Figure 5 is a block diagram of a feature data extraction device according to another exemplary embodiment. As shown in Figure 5 The feature data extraction device 50 includes an information module 502, a model module 504, an ordering module 506, a vector module 508, a sequence module 510, and a feature module 512.

[0083] The information module 502 is used to obtain user information at multiple time nodes;

[0084] The model module 504 is used to input the user information at the multiple time nodes into multiple tree models respectively, to generate multiple sets of calculation results, wherein the calculation result set includes multiple calculation sub-results corresponding to the multiple tree models; the model module 504 is also used to input the user information at each time node into a trained XGB model, wherein the XGB model includes multiple tree models; the user information is calculated by the multiple tree models; after the calculation is completed, the calculation sub-results of the multiple tree models are extracted to generate a calculation result set.

[0085] The sorting module 506 is configured to arrange the multiple calculation sub-results in each set of calculation results in a preset order; and the sorting module 506 is further configured to arrange the multiple calculation sub-results in each set of calculation results in an order corresponding to the multiple tree models in the trained XGB model.

[0086] The vector module 508 is configured to generate a result vector by using the arrangement order and the multiple calculation sub-results; and the vector module 508 is further configured to generate the result vector by using the arrangement order and the multiple calculation results as parameters.

[0087] The sequence module 510 is configured to generate a sequence vector by using the result vectors corresponding to the multiple sets of calculation results; and the sequence module 510 is further configured to arrange the result vectors corresponding to the multiple sets of calculation results in an order according to the time nodes corresponding to the result vectors, to generate a sequence vector with a time feature.

[0088] The feature module 512 is configured to extract feature data of the user information by using the sequence vector; and the feature module 512 is further configured to input the sequence vector into a neural network model, to extract the feature data of the user information.

[0089] According to the feature data extraction apparatus, the user information at multiple time nodes is acquired; the user information at the multiple time nodes is input into multiple tree models respectively, to generate multiple sets of calculation results, wherein the calculation results set includes multiple calculation sub-results corresponding to the multiple tree models; the multiple calculation sub-results in each set of calculation results are arranged in a preset order; a result vector is generated by using the arrangement order and the multiple calculation sub-results; a sequence vector is generated by using the result vectors corresponding to the multiple sets of calculation results; and the feature data of the user information is extracted by using the sequence vector. In this way, the tree model and the neural network are combined ingeniously, the advantages of the two models are brought into full play, the time sequence feature of the user information and the correlation between multiple features are mined deeply, and the feature data is extracted adaptively.

[0090] As shown in Figure 6 FIG. 6, an embodiment of the present application provides an electronic device, which includes a processor 610, a memory 620 and a bus, wherein the processor 610 and the memory 620 complete communication with each other through the bus 640;

[0091] The memory 620 is configured to store a computer program.

[0092] The processor 610 is configured to execute the program stored in the memory 620, to implement the feature data extraction method of any of the above embodiments.

[0093] The communication interface 620 is configured to implement communication between the electronic device and other devices.

[0094] The memory 620 can include a random access memory (RAM) and can also include a non-volatile memory, such as at least one disk memory 620. Optionally, the memory 620 can also be at least one storage device located remotely from the aforementioned processor 610.

[0095] The above-described methods in the present application, if implemented in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes of the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of the above-mentioned various method embodiments can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc.

[0096] The embodiment of the present application provides a computer readable storage medium, and the computer readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the feature data extraction method of any one of the above-mentioned embodiments. For example, user information at multiple time nodes can be obtained; the user information at the multiple time nodes is respectively input into multiple tree models to generate multiple sets of calculation results, wherein the calculation result set includes multiple calculation sub-results corresponding to the multiple tree models; the multiple calculation sub-results in each set of calculation results are arranged in a preset order; a result vector is generated through the arrangement order and the multiple calculation sub-results; a sequence vector is generated through the result vectors corresponding to the multiple sets of calculation results; and feature data of the user information is extracted through the sequence vector.

[0097] The exemplary embodiments of the present application are specifically shown and described above. It should be understood that the present application is not limited to the detailed structure, arrangement or implementation method described herein; on the contrary, the present application is intended to cover various modifications and equivalent arrangements included in the spirit and scope of the appended claims.

Claims

1. A method for extracting feature data, characterized in that: include: Get user information at multiple time points; Inputting user information at multiple time nodes into multiple tree models respectively to generate multiple sets of calculation results, wherein the calculation result sets include multiple calculation sub-results corresponding to the multiple tree models; Arrange the multiple calculation sub-results in each group of calculation result sets according to a preset order; Generate a result vector by permuting the order and the plurality of calculation sub-results; Generate a sequence vector through the result vectors corresponding to multiple sets of calculation result sets; Feature data of the user information is extracted through the sequence vector.

2. The method according to claim 1, wherein Input user information at multiple time nodes into multiple tree models to generate multiple sets of calculation results, including: Input the user information at each time node into the trained XGB model, which includes multiple tree models; Calculating the user information using the multiple tree models; After the calculation is completed, the calculation sub-results of the multiple tree models are extracted to generate a calculation result set.

3. The method according to claim 2, wherein After the calculation is completed, the calculation sub-results of the multiple tree models are extracted to generate a calculation result set, including: After the calculation is completed, extracting the leaf nodes hit in each tree model of the multiple tree models; The calculation sub-result is generated through the leaf node.

4. The method according to claim 3, wherein Generating the calculation sub-result through the leaf node includes: Generate the calculation sub-result according to the sequence number of the leaf node; and / or The calculation sub-result is generated by the corresponding characteristic parameters of the leaf node.

5. The method according to claim 1, wherein Arranging the multiple calculation sub-results in each group of calculation result sets according to a preset order includes: The multiple calculation sub-results in each group of calculation result sets are arranged in the order corresponding to the multiple tree models in the trained XGB model.

6. The method according to claim 1, wherein Generating a result vector by arranging the order and the plurality of calculation sub-results includes: The arrangement order and the multiple calculation results are combined as parameters to generate a result vector.

7. The method according to claim 6, wherein The arrangement order and the plurality of calculation results are combined as parameters to generate a result vector, including: Determining the increasing index values ​​corresponding to the arrangement order; The indicator values ​​and their corresponding calculation sub-results are merged according to the arrangement order to generate the result vector.

8. The method according to claim 1, wherein Generate sequence vectors through result vectors corresponding to multiple sets of calculation result sets, including: The result vectors corresponding to the multiple groups of calculation result sets are arranged in the order of their corresponding time nodes to generate a sequence vector with time characteristics.

9. The method according to claim 1, wherein Extracting feature data of the user information through the sequence vector includes: The sequence vector is input into a neural network model to extract feature data of the user information.

10. The method according to claim 9, wherein Inputting the sequence vector into a neural network model to extract feature data of the user information includes: Determine the data to be analyzed; Inputting the analysis object data and the sequence vector into the neural network model; The neural network model takes the analysis object as a focus to extract feature data of the user information.

11. A device for extracting feature data, characterized in that: include: Information module, used to obtain user information at multiple time nodes; A model module, configured to input user information at multiple time nodes into multiple tree models respectively, and generate multiple sets of calculation result sets, wherein the calculation result sets include multiple calculation sub-results corresponding to the multiple tree models; A sorting module, configured to arrange the plurality of calculation sub-results in each set of calculation results in a preset order; A vector module, configured to generate a result vector by permuting the sequence and the plurality of calculation sub-results; A sequence module is used to generate sequence vectors from result vectors corresponding to multiple sets of calculation result sets; A feature module is used to extract feature data of the user information through the sequence vector.

12. An electronic device, characterized in that: include: one or more processors; a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 10.

13. A computer-readable medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 10 is implemented.

14. A computer program product, characterized in that The method comprises a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images