Information screening method, device, equipment and storage medium based on decision tree

By extracting keywords to generate user portraits, classification and purity value calculations, and building a decision tree model, it solves the problem that the decision tree cannot accurately screen high-quality customers in customer information analysis, and realizes efficient screening of customer information.

CN111444944BActive Publication Date: 2024-11-19CHINA PING AN LIFE INSURANCE CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010182272.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-03-16
Publication Date
2024-11-19
Estimated Expiration
2040-03-16

AI Technical Summary

Technical Problem

The existing decision trees cannot accurately obtain user characteristics when analyzing customer information, resulting in the inability to screen out high-quality customers from the huge customer information data in a timely and effective manner.

Method used

By obtaining the original data, extracting keywords to generate user portraits, classifying user portraits according to preset classification conditions, counting the number of users in the user group to determine the purity value of the category label, building a decision tree model, and filtering user information to The smallest leaf node to summarize the filtered user information.

Benefits of technology

It realizes accurate screening of customer information, can screen out high-quality customers in a timely and effective manner from a large amount of data, and improves the accuracy and efficiency of decision tree analysis.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN111444944B_ABST
    Figure CN111444944B_ABST
Patent Text Reader

Abstract

The present application relates to the field of data analysis technology, and in particular to a method, device, equipment and storage medium for information screening based on a decision tree, including: obtaining raw data, extracting keywords from the raw data to generate user portraits; classifying user portraits to generate user groups with category labels; counting the number of users in each user group and determining the purity value of each category label; determining the position of each category label in the decision tree according to the purity value of each category label, and obtaining a decision tree model; inputting user information in the raw data into the root node of the decision tree model, and screening the raw data from the root node of the decision tree model to the smallest leaf node according to the restriction conditions of each level of nodes in the decision tree model, and summarizing the screened user information. This solves the problem that the decision tree cannot accurately obtain the characteristics of the user when analyzing customer information, and cannot timely and effectively screen out high-quality customers from customer information data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of data analysis, and in particular to a decision tree-based information screening method, device, equipment and storage medium. Background Art

[0002] The decision tree is a comprehensive evaluation and classification analysis tool. It mainly uses a series of complex calculations such as multi-factor regression to quickly screen out the main influencing factors from a large number of influencing factors with complex interactions, and effectively conducts hierarchical analysis, thereby completing the role of accurately predicting or classifying the research object.

[0003] At present, the existing decision trees cannot accurately obtain the characteristics of users when analyzing customer information, and thus cannot timely and effectively screen out high-quality customers from the huge amount of customer information data. Summary of the invention

[0004] Based on this, in view of the problem that the current decision tree cannot accurately obtain user characteristics when analyzing customer information, and thus cannot timely and effectively screen out high-quality customers from huge customer information data, a decision tree-based information screening method, device, equipment and storage medium are provided.

[0005] A decision tree-based information screening method comprises the following steps:

[0006] Acquire raw data, extract keywords from the raw data, and generate user portraits based on the keywords;

[0007] Classify the user portraits according to preset classification conditions to generate user groups with category labels;

[0008] Counting the number of users in each of the user groups, and determining the purity value of each category label according to the number of users;

[0009] According to the purity value of each category label, determine the position of each category label in the decision tree to obtain a decision tree model;

[0010] The user information in the original data is input into the root node of the decision tree model, and according to the restriction conditions of each level of nodes in the decision tree model, the original data is screened level by level from the root node of the decision tree model to the smallest leaf node, and the screened user information is summarized.

[0011] In one possible embodiment, the obtaining of raw data, extracting keywords from the raw data, and generating a user profile according to the keywords includes:

[0012] Acquire original data, and divide the original data into dynamic data and static data according to a preset classification rule;

[0013] Obtaining a property keyword list pre-stored in a database, and extracting property keywords from the static data according to the property keyword list;

[0014] The dynamic data is traversed to obtain user behavior information corresponding to the attribute keywords, and a user profile is created based on the attribute keywords and the user behavior information.

[0015] In one possible embodiment, classifying the user portraits according to preset classification conditions to generate user groups with category labels includes:

[0016] Input the sample data that meets the preset classification conditions as a training set into a preset support vector machine model for training to obtain a trained classification model;

[0017] Inputting the user portrait into the trained classification model for classification to obtain multiple initial user groups;

[0018] Calculating the correlation between the initial user groups, if the correlation between any two initial user groups is greater than a preset correlation threshold, merging the corresponding two initial user groups into one user group;

[0019] The category names corresponding to the user groups are used as category labels to obtain multiple user groups with category labels.

[0020] In one possible embodiment, counting the number of users in each of the user groups and determining the purity value of each category label according to the number of users includes:

[0021] Obtaining entity names of users in the user group with category labels, packaging users with the same entity name into a user group, and counting the number of users in the user group;

[0022] Calculate the ratio of the number of users in each user group to the total number of users in the user group, and input the ratio into a preset purity calculation formula to obtain the purity value of the user group, wherein the purity value calculation formula is:

[0023]

[0024] In the formula, H represents purity, and P(i) represents the ratio of the number of users in the i-th user group to the total number of users in the user group.

[0025] In one possible embodiment, determining the position of each category label in the decision tree according to the purity value of each category label to obtain the decision tree model includes:

[0026] Obtaining a first category label corresponding to the maximum value of the purity value, and using the first category label as the root node of the decision tree;

[0027] Obtaining a second category label corresponding to the minimum value in the purity value, and using the second category label as the minimum leaf node of the decision tree;

[0028] After arranging the other category labels from large to small according to the purity value, they are sequentially filled between the root node and the smallest leaf node to obtain the decision tree model.

[0029] In one possible embodiment, after inputting the user information in the original data into the root node of the decision tree model, filtering the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and aggregating the filtered user information, the method further includes:

[0030] Summarize pending user information that has not passed the minimum leaf node, and send the pending user information to the review terminal for review;

[0031] Receive the audit result of the audit terminal, re-input the correction information in the audit result into the decision tree model for re-screening, and delete the user information output by the minimum leaf node of the decision tree model.

[0032] An information screening device based on a decision tree comprises the following modules:

[0033] A portrait generation module is used to obtain raw data, extract keywords from the raw data, and generate a user portrait based on the keywords;

[0034] A label assignment module, used to classify the user portraits according to preset classification conditions and generate user groups with category labels;

[0035] A purity value generating module, used for counting the number of users in each of the user groups, and determining the purity value of each category label according to the number of users;

[0036] A decision tree generation module, used to determine the position of each category label in the decision tree according to the purity value of each category label, and obtain a decision tree model;

[0037] The user screening module is used to input the user information in the original data into the root node of the decision tree model, screen the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and summarize the screened user information.

[0038] In one possible embodiment, the portrait generation module is further used to:

[0039] Acquire original data, and divide the original data into dynamic data and static data according to a preset classification rule;

[0040] Obtaining a property keyword list pre-stored in a database, and extracting property keywords from the static data according to the property keyword list;

[0041] The dynamic data is traversed to obtain user behavior information corresponding to the attribute keywords, and a user profile is created based on the attribute keywords and the user behavior information.

[0042] A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the above-mentioned information screening method based on the decision tree.

[0043] A storage medium storing computer-readable instructions, when the computer-readable instructions are executed by one or more processors, enables the one or more processors to perform the steps of the above-mentioned decision tree-based information screening method.

[0044] Compared with the existing mechanism, the present application obtains the original data, extracts the keywords in the original data, generates the user portrait according to the keywords; classifies the user portrait according to the preset classification conditions, generates the user group with the category label; counts the number of users in each user group, and determines the purity value of each category label according to the number of users; determines the position of each category label in the decision tree according to the purity value of each category label, and obtains the decision tree model; inputs the user information in the original data into the root node of the decision tree model, and according to the restriction conditions of each level of nodes in the decision tree model, filters the original data step by step from the root node of the decision tree model to the smallest leaf node, and summarizes the filtered user information. This solves the problem that the intelligent decision tree cannot accurately obtain the user's characteristics when analyzing customer information, and thus cannot timely and effectively filter out high-quality customers from the huge customer information data. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the following detailed description of the preferred embodiment.The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application.

[0046] Figure 1 This is an overall flow chart of an information screening method based on a decision tree in one embodiment of the present application;

[0047] Figure 2 A schematic diagram of a portrait generation process in a decision tree-based information screening method in one embodiment of the present application;

[0048] Figure 3 This is a structural diagram of a decision tree-based information screening device in one embodiment of the present application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0050] Those skilled in the art will appreciate that, unless otherwise stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0051] Figure 1 This is an overall flow chart of an information screening method based on a decision tree in one embodiment of the present application. The information screening method based on a decision tree includes the following steps:

[0052] S1. Obtaining original data, extracting keywords from the original data, and generating user portraits based on the keywords;

[0053] Specifically, the raw data in this step can be the user's behavior log data, purchase history data, favorite products and other data generated by the client when the user purchases an insurance policy or browses the Golden Butler app. When performing a keyword query on the raw data, it can be the user's birthday, age, neighborhood, constellation, gender, and terminal model used. After combining these keywords, you can get a user portrait, such as Zhang San, a user, has purchased flight delay insurance 10 times in the past 30 days; browsed the Golden Butler app e-commerce website for a total of 10 hours; Li Si user purchased 30 orders within 1 month, and the payment amount for each order was more than 200. User portraits mainly include three elements: time, place, and person. Each user behavior is essentially a random event, which can be described in detail as: what user, at what time, where, and what did he do.

[0054] S2. Classify the user portraits according to preset classification conditions to generate user groups with category labels;

[0055] Specifically, the preset classification conditions are generated by the user's label model, which includes inherent attribute labels, user preference labels, model labels, etc. Since the user's model label (that is, the label is based on the data collection results of the user in a certain period of time, it is possible that Zhang San likes high-end brands during this period, and it is also possible that Zhang San likes other high-end brands after a period of time) changes dynamically over time, based on this, the label dynamic maintenance, that is, the label management function is added, which can update the model label of a certain user as the user data changes dynamically. Zhang San's labels are as follows: A label (5 times), B label (2 times), C label (1 time). Zhang San's A behavior type weight = a / (a+b+c) = 5 / (5+2+1) = 5 / 8 time decay; it is believed that the earlier the user browses or buys a certain product, the less impact it has on the existing label. For example, Zhang San bought a black and white TV 10 years ago, and he should consider buying a color TV 10 years later. Time decay coefficient formula: 1 / [100+(current time - past purchase time)]; User tag weight calculation formula: number of behaviors * time decay coefficient * behavior type weight

[0056] Example: Name Number of purchases Time Purchased Product Name

[0057] Zhang San bought accident insurance 5 times on 2009-01-01 (think of it as label A)

[0058] Zhang San bought accident insurance 2 times on 2019-01-01 (think of label B)

[0059] According to the label weight calculation formula:

[0060] Zhang San A label weight: 5*[1 / 100+(2019-2009)]*5 / 8=5*(1 / 110)*5 / 8=0.02840

[0061] Zhang San B's label weight: 2*[1 / 100+(2019-2019)]*5 / 8=5*(1 / 100)*5 / 8=0.03125. Zhang San can be divided into different user groups according to the label weight.

[0062] S3, counting the number of users in each of the user groups, and determining the purity value of each category label according to the number of users;

[0063] The purity value refers to the ratio of the number of users in a user group that meet the preset conditions to the total number of users.

[0064] For example, the data of user group A is as follows:

[0065]

[0066] According to whether the business trip insurance is purchased (yes, no), annual income (less than 1 million, greater than or equal to 1 million), a data classification table is obtained;

[0067] Annual income 50 102 110 120 Premium Customers no yes yes yes Split Point 50 100 110 120

[0068] According to the purity calculation formula:

[0069] In the formula, H represents purity, and P(i) represents the ratio of the number of users in the i-th user group to the total number of users in the user group.

[0070] H(D)=-0.75*log2(0.75)-0.25*log2(0.25)

[0071] =0.75*0.415+0.25*2

[0072] =0.311+0.5=0.811

[0073] According to the above example, high-quality customers account for 3 / 4; non-high-quality customers account for 1 / 4, and the annual income of these customers is greater than 1 million.

[0074] S4. Determine the position of each category label in the decision tree according to the purity value of each category label, and obtain a decision tree model;

[0075] Specifically, the larger the purity value, the higher the position in the decision tree model. For example, the purity value of annual income is 0.999, the purity value of manager is 0.92, and the purity value of age is 0.56. Then the annual income is at the root node of the decision tree model, the manager is the first-level leaf node, and the age is the second-level leaf node.

[0076] S5. Input the user information in the original data into the root node of the decision tree model, filter the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and summarize the filtered user information.

[0077] Specifically, when inputting user information into the decision tree model, the root node label of the decision tree model is first used as a keyword to filter the user information. When filtering, the word vector conversion method can be used, that is, each word in the user information is converted into a word vector, and then each node label of the decision tree model is converted into a word vector. Among them, the word vector conversion can use a word vector conversion tool such as Word2vec. If there is a word in the user information that is consistent with the root node word vector of the decision tree model, it means that the user information meets the root node condition of the decision tree model. At this time, there is no need to filter the user through the child nodes of the decision tree model. For words that are not consistent with the root node word vector in the user information, the word vectors of the leaf nodes at each level are used for filtering step by step.

[0078] In this embodiment, the purity value is used to set the position of each category label on the decision tree model, thereby achieving the technical effect of timely and effectively screening out high-quality customers from huge customer information data.

[0079] Figure 2 This is a schematic diagram of a portrait generation process in a decision tree-based information screening method in one embodiment of the present application. As shown in the figure, the S1, obtaining original data, extracting keywords from the original data, and generating a user portrait based on the keywords, includes:

[0080] S11, obtaining original data, and dividing the original data into dynamic data and static data according to a preset classification rule;

[0081] Among them, the original data can be divided into two categories: static data and dynamic data. Static data: such as the user's demographic attributes, business attributes, consumption characteristics, life style and other aspects of the data are static data, static data is relatively fixed, and rarely changes with the changes in user behavior; and there are many ways to obtain it, the most common one is data mining, data mining can be used in many forms, such as group discussions, user interviews, logs, questionnaires, etc., mainly through open questions to explore the real psychological needs of users. The key to the questionnaire method is how to model and analyze the collected data in the later stage. Its purpose is mainly to verify the hypothesis to obtain the rules of user attributes;

[0082] Dynamic data: user's constantly changing behavior information. A user opens a web page and buys a personal accident insurance; the user's phone number changes, he buys a book about insurance knowledge on website A, and books a flight to Shanghai on website B, and at the same time buys personal accident insurance, flight delay insurance, etc. These are all user behaviors, and various dynamic behavior data can be recorded.

[0083] S12, obtaining an attribute keyword list pre-stored in a database, and extracting attribute keywords from the static data according to the attribute keyword list;

[0084] Specifically, a keyword may represent content, and the user may be interested in, prefer, or demand the content, etc. When extracting keywords, a stroke similarity comparison method may be used, that is, the attribute keyword in the static data and the words in the attribute keyword list are subjected to stroke similarity calculations. If the calculation result is less than or equal to a similarity threshold, the two are similar, and the words in the attribute keyword list are used as attribute keywords extracted from the static data.

[0085] S13. Traverse the dynamic data to obtain user behavior information corresponding to the attribute keywords, and create a user profile based on the attribute keywords and the user behavior information.

[0086] Specifically, the verb next to the attribute keyword in the dynamic data is the user behavior. For example, Manager Zhang (keyword) purchased 30 orders. The user behavior is "purchase".

[0087] In this embodiment, a user portrait is accurately established through static data and dynamic data, thereby ensuring the accuracy of screening the original data.

[0088] In one embodiment, classifying the user portraits according to preset classification conditions to generate user groups with category labels includes:

[0089] Input the sample data that meets the preset classification conditions as a training set into a preset support vector machine model for training to obtain a trained classification model;

[0090] Specifically, a support vector machine (SVM) is a generalized linear classifier that performs binary classification on data using supervised learning. Its decision boundary is the maximum-margin hyperplane found for learning samples. SVM can be used to effectively train and classify classification conditions.

[0091] Inputting the user portrait into the trained classification model for classification to obtain multiple initial user groups;

[0092] Specifically, the initial user groups, such as product managers, sales managers, senior engineers, junior engineers, etc.

[0093] Calculating the correlation between the initial user groups, if the correlation between any two initial user groups is greater than a preset correlation threshold, merging the corresponding two initial user groups into one user group;

[0094] Among them, the correlation degree is the degree of correlation between two things. In mathematics, it refers to the degree of similarity between two functions. It is a term in grey system analysis. By calculating the correlation degree, the product manager and sales manager in the previous step can be merged into a business manager.

[0095] The category names corresponding to the user groups are used as category labels to obtain multiple user groups with category labels.

[0096] In this embodiment, user portraits are effectively grouped, thereby facilitating the placement of user tags into corresponding nodes of the decision tree.

[0097] In one embodiment, counting the number of users in each of the user groups and determining the purity value of each category label according to the number of users includes:

[0098] Obtaining entity names of users in the user group with category labels, packaging users with the same entity name into a user group, and counting the number of users in the user group;

[0099] Among them, the entity names can be extracted from the user group using knowledge graph technology.

[0100] Calculate the ratio of the number of users in each user group to the total number of users in the user group, and input the ratio into a preset purity calculation formula to obtain the purity value of the user group, wherein the purity value calculation formula is:

[0101]

[0102] In the formula, H represents purity, and P(i) represents the ratio of the number of users in the i-th user group to the total number of users in the user group.

[0103] In this embodiment, by calculating the purity value, the user attributes of each user group can be effectively obtained, thereby simplifying the dimension of raw data screening.

[0104] In one embodiment, determining the position of each category label in the decision tree according to the purity value of each category label to obtain the decision tree model includes:

[0105] Obtaining a first category label corresponding to the maximum value of the purity value, and using the first category label as the root node of the decision tree;

[0106] Obtaining a second category label corresponding to the minimum value in the purity value, and using the second category label as the minimum leaf node of the decision tree;

[0107] After arranging the other category labels from large to small according to the purity value, they are sequentially filled between the root node and the smallest leaf node to obtain the decision tree model.

[0108] Specifically, if the purity values ​​corresponding to any two of the category labels are consistent, the logical relationship between the category labels is obtained;

[0109] If the logical relationship is "and", the category labels are written into multiple leaf nodes at the same level;

[0110] If the logical relationship is "not", the purity value of the category label is recalculated.

[0111] In one embodiment, after inputting the user information in the original data into the root node of the decision tree model, filtering the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and aggregating the filtered user information, the method further includes:

[0112] Summarize pending user information that has not passed the minimum leaf node, and send the pending user information to the review terminal for review;

[0113] Specifically, the parameter values ​​that failed in the user information that failed the minimum leaf node are sent to the audit terminal. For example, if the passing condition is: the premium is greater than 100,000, and the premium of customer A is 30,000, then customer A's information will be sent to the audit terminal for audit. If it is found after the audit that user A's premium is 30,000 per year for 5 years, and the 100,000 in the minimum leaf node is the total amount, then this user will be marked as a problem user and his premium will be modified to 150,000.

[0114] Receive the audit result of the audit terminal, re-input the correction information in the audit result into the decision tree model for re-screening, and delete the user information output by the minimum leaf node of the decision tree model.

[0115] Among them, the user information output by the smallest leaf node is the user information that does not meet the high-quality customer conditions in this solution.

[0116] In this embodiment, the erroneous information is corrected through the audit terminal, thereby improving the accuracy of the decision tree model in screening user information.

[0117] The technical features mentioned in any of the above corresponding embodiments or implementations are also applicable to the present application. Figure 3 The corresponding embodiments will not be described in detail later.

[0118] The above is an explanation of a decision tree-based information screening method in the present application, and the following is a description of an apparatus for executing the decision tree-based information screening method.

[0119] like Figure 3 The structure diagram of a decision tree-based information screening device is shown in FIG. 1 , which can be applied to decision tree-based information screening. The decision tree-based information screening device in the embodiment of the present application can achieve the above Figure 1 The steps of the information screening method based on the decision tree executed in the corresponding embodiment. The functions implemented by the information screening device based on the decision tree can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware.

[0120] In one embodiment, a decision tree-based information screening device is proposed, such as Figure 3 As shown, it includes the following modules:

[0121] A portrait generation module 10 is used to obtain raw data, extract keywords from the raw data, and generate a user portrait based on the keywords;

[0122] A label assigning module 20, used to classify the user portraits according to preset classification conditions and generate user groups with category labels;

[0123] The purity value generating module 30 is used to count the number of users in each of the user groups and determine the purity value of each category label according to the number of users;

[0124] A decision tree generation module 40 is used to determine the position of each category label in the decision tree according to the purity value of each category label to obtain a decision tree model;

[0125] The user screening module 50 is used to input the user information in the original data into the root node of the decision tree model, screen the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and summarize the screened user information.

[0126] In one embodiment, the portrait generation module is further used to:

[0127] Acquire original data, and divide the original data into dynamic data and static data according to a preset classification rule;

[0128] Obtaining a property keyword list pre-stored in a database, and extracting property keywords from the static data according to the property keyword list;

[0129] The dynamic data is traversed to obtain user behavior information corresponding to the attribute keywords, and a user profile is created based on the attribute keywords and the user behavior information.

[0130] In one embodiment, a computer device is proposed, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the decision tree-based information screening method in the above-mentioned embodiments.

[0131] In one embodiment, a storage medium storing computer-readable instructions is proposed, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the information screening method based on the decision tree in the above embodiments. The storage medium can be a non-volatile storage medium or a volatile storage medium, which is not specifically limited in this application.

[0132] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium, and the storage medium may include: a read-only memory (ROM), a random access memory (RAM), a disk or an optical disk, etc.

[0133] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] The above-described embodiments only express some exemplary embodiments of the present application, in which the description is relatively specific and detailed, but it cannot be understood as limiting the scope of the present application. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present application, which all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the attached claims.

Claims

1. A decision tree-based information screening method, characterized in that: The information screening method based on the decision tree comprises: Acquire raw data, extract keywords from the raw data, and generate user portraits based on the keywords; Classify the user portraits according to preset classification conditions to generate user groups with category labels; Counting the number of users in each of the user groups, and determining the purity value of each category label according to the number of users; According to the purity value of each category label, determine the position of each category label in the decision tree to obtain a decision tree model; Inputting the user information in the original data into the root node of the decision tree model, filtering the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and summarizing the filtered user information; The counting of the number of users in each of the user groups and determining the purity value of each category label according to the number of users includes: Obtaining entity names of users in the user group with category labels, packaging users with the same entity name into a user group, and counting the number of users in the user group; Calculate the ratio of the number of users in each user group to the total number of users in the user group, and input the ratio into a preset purity calculation formula to obtain the purity value of the user group, wherein the purity value calculation formula is: ; In the formula, H represents purity, P(i) represents the ratio of the number of users in the i-th user group to the total number of users in the user group; Determining the position of each category label in the decision tree according to the purity value of each category label to obtain the decision tree model includes: Obtaining a first category label corresponding to the maximum value of the purity value, and using the first category label as the root node of the decision tree; Obtaining a second category label corresponding to the minimum value in the purity value, and using the second category label as the minimum leaf node of the decision tree; After arranging the other category labels from large to small according to the purity value, they are sequentially filled between the root node and the smallest leaf node to obtain the decision tree model.

2. The information screening method based on decision tree according to claim 1 is characterized in that: The obtaining of raw data, extracting keywords from the raw data, and generating user portraits according to the keywords include: Acquire original data, and divide the original data into dynamic data and static data according to a preset classification rule; Obtaining a property keyword list pre-stored in a database, and extracting property keywords from the static data according to the property keyword list; The dynamic data is traversed to obtain user behavior information corresponding to the attribute keywords, and a user profile is created based on the attribute keywords and the user behavior information.

3. The information screening method based on decision tree according to claim 1 is characterized in that: The classifying the user portraits according to the preset classification conditions to generate user groups with category labels includes: Input the sample data that meets the preset classification conditions as a training set into a preset support vector machine model for training to obtain a trained classification model; Inputting the user portrait into the trained classification model for classification to obtain multiple initial user groups; Calculate the correlation between the initial user groups. If the correlation between any two initial user groups is greater than a preset correlation threshold, merge the corresponding two initial user groups into one user group. The category name corresponding to each user group is used as a category label to obtain multiple user groups with category labels.

4. The information screening method based on decision tree according to claim 1 is characterized in that: After inputting the user information in the original data into the root node of the decision tree model, filtering the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and aggregating the filtered user information, the method further includes: Summarize pending user information that has not passed the minimum leaf node, and send the pending user information to the review terminal for review; Receive the audit result of the audit terminal, re-input the correction information in the audit result into the decision tree model for re-screening, and delete the user information output by the minimum leaf node of the decision tree model.

5. An information screening device based on a decision tree, characterized in that: The information screening device based on the decision tree comprises: A portrait generation module is used to obtain raw data, extract keywords from the raw data, and generate a user portrait based on the keywords; A label assignment module, used to classify the user portraits according to preset classification conditions and generate user groups with category labels; A purity value generating module, used for counting the number of users in each of the user groups, and determining the purity value of each category label according to the number of users; A decision tree generation module, used to determine the position of each category label in the decision tree according to the purity value of each category label, and obtain a decision tree model; A user screening module, used to input the user information in the original data into the root node of the decision tree model, screen the original data level by level from the root node to the smallest leaf node of the decision tree model according to the restriction conditions of each level of nodes in the decision tree model, and summarize the screened user information; The purity value generating module is specifically used for: Obtaining entity names of users in the user group with category labels, packaging users with the same entity name into a user group, and counting the number of users in the user group; Calculate the ratio of the number of users in each user group to the total number of users in the user group, and input the ratio into a preset purity calculation formula to obtain the purity value of the user group, wherein the purity value calculation formula is: ; In the formula, H represents purity, P(i) represents the ratio of the number of users in the i-th user group to the total number of users in the user group; The decision tree generation module is specifically used for: Obtaining a first category label corresponding to the maximum value of the purity value, and using the first category label as the root node of the decision tree; Obtaining a second category label corresponding to the minimum value in the purity value, and using the second category label as the minimum leaf node of the decision tree; After arranging the other category labels from large to small according to the purity value, they are sequentially filled between the root node and the smallest leaf node to obtain the decision tree model.

6. The information screening device based on decision tree according to claim 5, characterized in that: The portrait generation module is also specifically used for: Acquire original data, and divide the original data into dynamic data and static data according to a preset classification rule; Obtaining a property keyword list pre-stored in a database, and extracting property keywords from the static data according to the property keyword list; The dynamic data is traversed to obtain user behavior information corresponding to the attribute keywords, and a user profile is created based on the attribute keywords and the user behavior information.

7. An information screening device based on a decision tree, comprising a memory and a processor, wherein the memory stores computer-readable instructions, characterized in that: When the computer-readable instructions are executed by the processor, the processor is caused to perform the information screening method based on a decision tree according to any one of claims 1 to 4.

8. A storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to perform the information screening method based on a decision tree according to any one of claims 1 to 4.