A user portrait construction method, system, device and medium based on AI

Abstract

The present application relates to a method, system, device and medium for constructing a user portrait based on AI, which includes collecting user data and preprocessing it to form a main data set, inputting the main data set into a prediction model to obtain a basic prediction label; extracting user data whose basic prediction label accuracy does not meet the standard from the main data set to form a data subset; performing correlation analysis on the user basic information in the data subset and classifying the data subset to obtain several classification subsets; filtering the user behavior information in the classification subset according to preset screening rules to obtain potential features; inputting the potential features into the prediction model to obtain an optimized prediction label; fusing the optimized prediction label with the basic prediction label to obtain a target prediction label. The present application has the effect of making the model label more accurate.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to an AI-based user portrait construction method, system, device and medium. Background Art

[0002] With the continuous development of artificial intelligence technology, user portrait construction is increasingly widely used in various fields. User portrait construction refers to dividing users into different groups by analyzing their behavioral events and basic attributes, and labeling each group accordingly. These labels can reflect information such as users' interests, preferences, and spending power, thereby providing companies with more accurate marketing strategies.

[0003] The specific user portrait construction process is as follows: the construction system collects the user's original data, and then performs data cleaning, analysis, statistics and other processing to obtain derived features. The derived features are then input into the prediction model and the model label is output. For example, a credit score is obtained, which is a score or grade used to represent the borrower's credit risk.

[0004] However, due to the complexity of some user behaviors, in the subsequent monitoring process, it was found that the key business indicators of some users declined. This means that the model labels of some users are inaccurate, which in turn affects the business application effect. Further improvement is needed in this regard. Summary of the invention

[0005] The purpose of this application is to provide an AI-based user portrait construction method, system, device and medium, which has the characteristic of making model labels more accurate.

[0006] In the first aspect, the present application provides a method for constructing a user portrait based on AI, which adopts the following technical solution:

[0007] A user portrait construction method based on AI, comprising:

[0008] Collect user data and preprocess it to form a master data set, which is then input into the prediction model to obtain basic prediction labels.

[0009] User data whose basic prediction label accuracy does not meet the standard is extracted from the main data set to form a data subset;

[0010] Perform correlation analysis on user basic information in the data subset and classify the data subset to obtain several classification subsets;

[0011] Filter the user behavior information in the classification subset according to the preset filtering rules to obtain potential features;

[0012] According to the potential features input into the prediction model, the optimized prediction label is obtained;

[0013] The optimized prediction label is fused with the basic prediction label to obtain the target prediction label.

[0014] By adopting the above technical solution, after obtaining the preliminary basic prediction label results according to the traditional prediction model, the user data whose basic prediction label accuracy does not meet the standard is extracted, and then classified and processed into different user groups. For each different user group, the corresponding potential features are further obtained, and the optimized prediction label is fused with the basic prediction label to obtain a more accurate target prediction label. This method classifies different user groups, further increases the potential features for corresponding adjustments, and refines the model output results.

[0015] Optionally, the step of performing correlation analysis on the user basic information in the data subset and classifying the data subset to obtain a plurality of classified subsets includes:

[0016] The FP-Growth algorithm is used to identify new basic features related to user data whose basic prediction label accuracy does not meet the standard for the basic user information in the data subset;

[0017] The data subsets are classified according to the newly added basic features to obtain several classification subsets.

[0018] By adopting the above technical solution and using the FP-Growth algorithm for correlation analysis, specific user groups can be effectively located.

[0019] Optionally, the step of filtering the user behavior information in the classification subset according to a preset filtering rule to obtain potential features includes:

[0020] Process the user behavior information in the classification subset according to the preset candidate rules to obtain candidate features;

[0021] According to the newly added basic features, the correlation degree is matched with the candidate features respectively to obtain the matching value;

[0022] The candidate features whose matching values ​​exceed the preset threshold are set as potential features for adding new basic features.

[0023] By adopting the above technical solution, on the basis of classification, the correlation between the latent features and the newly added basic features of the classification can be further emphasized, which can effectively reduce errors. For example, there is a situation where college students go to a certain industry to work at will during their internship. This setting can effectively reduce such errors.

[0024] Optionally, the step of processing the user behavior information in the classification subset according to a preset candidate rule to obtain candidate features includes:

[0025] Perform statistical analysis on the user behavior information in the classification subset, calculate the correlation coefficient between each user behavior information and the target variable, and select user behavior information with a correlation coefficient exceeding a preset threshold as a candidate feature.

[0026] By adopting the above technical solutions, statistical analysis can objectively identify the user behavior information most relevant to the target variable and reduce the bias of subjective judgment. At the same time, selecting features that are highly correlated with the target variable can improve the accuracy and generalization ability of the prediction model.

[0027] Optionally, the step of fusing the optimized prediction label with the basic prediction label to obtain the target prediction label includes:

[0028] The optimized prediction label and the basic prediction label are multiplied by the relevant weight coefficients and superimposed to obtain the target prediction label. The specific formula is as follows:

[0029] ;

[0030] Among them, F is the target prediction label value, G is the basic prediction label value, and L is the optimized prediction label value. is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient.

[0031] By adopting the above technical solution and combining two different prediction labels, the respective advantages and disadvantages can be balanced to improve the accuracy of the overall prediction.

[0032] Optionally, the step of obtaining the basic prediction label weight coefficient and the optimized prediction label weight coefficient includes:

[0033] The correlation coefficients exceeding the preset threshold in each classification subset are counted, and the optimized prediction label weight coefficient and the basic prediction label weight coefficient are determined according to the number. The specific formula is as follows;

[0034] ;

[0035] ;

[0036] in, is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient, is the number of correlations that exceed the preset threshold.

[0037] By adopting the above technical solution, the weights are adjusted according to the number of correlation coefficients of different classification subsets, and a more personalized prediction model can be achieved, and each classification subset can obtain the weight configuration that best suits its characteristics. Specifically, the more potential features there are in the classification subset, the greater the influence of the optimized prediction label value on the target prediction label value.

[0038] In the second aspect, the present application provides an AI-based user portrait construction system, which adopts the following technical solutions:

[0039] An AI-based user portrait construction system, including:

[0040] Preliminary prediction module: used to collect and preprocess user data to form a main data set, and input the main data set into the prediction model to obtain basic prediction labels;

[0041] Data extraction module: used to extract user data whose basic prediction label accuracy does not meet the standard from the main data set to form a data subset;

[0042] Subset classification module: used to perform correlation analysis on user basic information in the data subset and classify the data subset to obtain several classification subsets;

[0043] Feature screening module: used to screen the user behavior information in the classification subset according to the preset screening rules to obtain potential features;

[0044] Optimized prediction module: used to obtain optimized prediction labels based on potential features input into the prediction model;

[0045] Label fusion module: used to fuse the optimized prediction label with the basic prediction label to obtain the target prediction label.

[0046] In a third aspect, the present application provides an electronic device, which adopts the following technical solution:

[0047] An electronic device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the above-mentioned method for constructing a user portrait based on AI when executing the computer program.

[0048] In a fourth aspect, the present application provides a computer storage medium, which adopts the following technical solution:

[0049] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned AI-based user portrait construction method.

[0050] In summary, the beneficial technical effects of this application include:

[0051] After obtaining preliminary basic prediction label results according to the traditional prediction model, user data whose basic prediction label accuracy does not meet the standard is extracted, and then classified and processed into different user groups. For each different user group, its corresponding potential features are further obtained, and then the optimized prediction label is fused with the basic prediction label to obtain a more accurate target prediction label. This method classifies different user groups and further increases the potential features for corresponding adjustments, making the model output results more refined. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 It is a flowchart of an AI-based user portrait construction method in one of the embodiments of the present application.

[0053] Figure 2 It is a schematic diagram of the specific step flow of step S3 in another embodiment of the present application.

[0054] Figure 3 It is a schematic diagram of the specific step flow of step S4 in another embodiment of the present application.

[0055] Figure 4 It is a structural diagram of an AI-based user portrait construction system of one of the embodiments of the present application.

[0056] Figure 5 It is a principle block diagram of an electronic device in one embodiment of the present application.

[0057] In the figure, 1. Preliminary prediction module; 2. Data extraction module; 3. Subset classification module; 4. Feature screening module; 5. Optimization prediction module; 6. Label fusion module. DETAILED DESCRIPTION

[0058] All actions of obtaining data or information or data in this application are carried out in compliance with the relevant data protection laws and policies of the country where the application is located and with the authorization of the corresponding user.

[0059] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below in conjunction with the drawings in the embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this application, not all of the embodiments.

[0060] In the description of the embodiments of the present application, words such as "for example" or "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "for example" or "for example" is intended to present related concepts in a specific way.

[0061] In the description of the embodiments of the present application, the meaning of the term "multiple" refers to two or more. For example, multiple systems refer to two or more systems, and multiple screen terminals refer to two or more screen terminals. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. The terms "include", "comprise", "have" and their variations all mean "including but not limited to", unless otherwise specifically emphasized.

[0062] In financial services, credit score labels can be used to build a credit risk profile of borrowers. Traditional models may predict the default probability of borrowers based on broad demographic data and historical credit behavior. However, traditional models may be biased against certain specific groups, such as college students, etc. College students may be incorrectly rated as high risk due to lack of sufficient credit history.

[0063] Therefore, the present application embodiment provides a method for constructing a user portrait based on AI. Figure 1 , Figure 1 This is a flowchart of a method for building a user portrait based on AI provided by an embodiment of the present application. The method can be implemented by a computer program, which can be integrated into an application or run as an independent tool application. The method can also be implemented by a single-chip microcomputer or run in an AI-based user portrait building system. Specifically, the method can include the following steps:

[0064] S1. Collect user data and preprocess it to form a master data set, and input the master data set into the prediction model to obtain basic prediction labels.

[0065] Among them, user data includes basic user information and user behavior information. User data is preprocessed, that is, data cleaning is performed to deal with missing values, outliers, duplicate records, etc. The master data set includes historical credit behavior. The master data set is input into the prediction model to obtain the basic prediction label.

[0066] S2. User data whose basic prediction label accuracy does not meet the standard is extracted from the main data set to form a data subset.

[0067] Specifically, whether the accuracy of the basic prediction labels meets the standards is determined by whether the key business indicators reach the preset thresholds. User data whose basic prediction labels do not meet the standards are extracted and integrated from the main data set to form a data subset.

[0068] S3. Perform correlation analysis on the user basic information in the data subset and classify the data subset to obtain several classification subsets.

[0069] Specifically, by classifying the basic information of users in the data subset, a more detailed division of user groups can be obtained, such as college students, etc.

[0070] S4. Filter the user behavior information in the classification subset according to the preset filtering rules to obtain potential features.

[0071] Specifically, the user behavior information of college students and the users involved is screened to identify potential features related to credit scores, such as "educational background", "residential stability", "work industry", "purchase of insurance products", etc. These features help to more comprehensively assess the user's credit risk.

[0072] S5. Input the potential features into the prediction model to obtain optimized prediction labels.

[0073] Specifically, the potential features may be input into the prediction model used in step S1, or other prediction models may be input. The prediction model selection here may be selected according to the characteristics of the potential features and business requirements.

[0074] S6. Fusion the optimized prediction label with the basic prediction label to obtain the target prediction label.

[0075] Specifically, by fusing the optimized prediction label with the basic prediction label to obtain the target prediction label, the contribution of the two labels to the target prediction label can be more reasonably integrated, which is more accurate than a single label.

[0076] Through the above method, user data whose basic prediction labels do not meet the accuracy standards are classified and processed, and the corresponding potential features are further obtained for each different user group, and then the basic prediction labels are integrated to obtain more accurate target prediction labels. This method can make corresponding adjustments for different user groups to refine the model output results.

[0077] Reference Figure 2 Based on the above embodiment, as an optional embodiment, step S3: performing correlation analysis on the user basic information in the data subset and classifying the data subset to obtain several classification subsets, including:

[0078] S30. Using the FP-Growth algorithm, the user basic information in the data subset is used to identify new basic features related to user data whose basic prediction label accuracy does not meet the standard.

[0079] S31. Classify the data subsets according to the newly added basic features to obtain several classification subsets.

[0080] Specifically, the user basic information is converted into a transaction set format, the minimum support threshold is set, and the FP-Growth algorithm is run to find all frequent item sets that meet the minimum support. Then, based on the frequent item sets, association rules are generated, and the confidence and lift of each rule are calculated. Then, association rules with high confidence and lift are screened out. From the screened association rules, new basic features related to user data whose basic prediction label accuracy does not meet the standard are identified. Then, the data subsets are classified according to the new basic features to obtain several classification subsets. Using the FP-Growth algorithm for correlation analysis can effectively locate specific user groups. Furthermore, the FP-Growth algorithm is different from the Apriori algorithm. The FP-Growth algorithm is more efficient than the traditional Apriori algorithm because it avoids multiple scans of the database, but uses a data structure called FP-tree (Frequent Pattern Tree) to compress data, and FP-Growth does not need to generate candidate item sets, which reduces the amount of calculation and improves the running speed of the algorithm.

[0081] Reference Figure 3 Based on the above embodiment, as an optional embodiment, step S4: filtering the user behavior information in the classification subset according to the preset filtering rules to obtain potential features, including:

[0082] S40: Process the user behavior information in the classification subset according to preset candidate rules to obtain candidate features.

[0083] S41. Perform correlation matching on the newly added basic features and the candidate features to obtain matching values.

[0084] S42: setting candidate features whose matching values ​​exceed a preset threshold as potential features for adding new basic features.

[0085] Specifically, the set candidate features can be preset manually or obtained by statistical analysis of user behavior information in the classification subset, and then the newly added basic features are matched with the candidate features respectively, and the candidate features with matching values ​​exceeding the preset threshold are set as potential features of the newly added basic features. For example, college students are matched with candidate features such as "educational background", "work industry", and "purchase of insurance products" to obtain matching values. When the matching value exceeds the preset threshold, it is selected as a potential feature. The above can further emphasize the correlation between the potential features and the newly added basic features of the classification on the basis of classification, which can effectively reduce errors. For example, there is a situation where college students arbitrarily go to a certain work industry to work during their internship. This setting can effectively reduce such errors.

[0086] Based on the above embodiment, as an optional embodiment, step S40: processing the user behavior information in the classification subset according to the preset candidate rule to obtain the candidate features includes:

[0087] Perform statistical analysis on the user behavior information in the classification subset, calculate the correlation coefficient between each user behavior information and the target variable, and select user behavior information with a correlation coefficient exceeding a preset threshold as a candidate feature.

[0088] Specifically, statistical analysis is performed on user behavior information such as "education background", "residence stability", "work industry", and "purchase of insurance products", and the correlation coefficient between each user behavior information and the target variable is calculated. The target variable is the credit score. For example, a higher education level may mean that an individual has better employment opportunities and higher income potential, and also a better credit score. Therefore, the credit score is correlated with it. Through statistical analysis, the user behavior information most relevant to the target variable can be objectively identified, reducing the bias of subjective judgment. At the same time, selecting features that are highly correlated with the target variable can improve the accuracy and generalization ability of the prediction model.

[0089] Based on the above embodiment, as an optional embodiment, step S6: fusing the optimized prediction label with the basic prediction label to obtain the target prediction label includes:

[0090] The optimized prediction label and the basic prediction label are multiplied by the relevant weight coefficients and superimposed to obtain the target prediction label. The specific formula is as follows:

[0091] ;

[0092] Among them, F is the target prediction label value, To predict the label value based on To optimize the predicted label value, is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient.

[0093] A weight is determined by the basic prediction label value and the optimized prediction label value respectively, and the basic prediction label value and the optimized prediction label value are multiplied by their corresponding weights. The weighted scores are summed and then divided by the sum of the weights to obtain the final credit score, i.e., the target prediction label value. By combining two different prediction labels, the advantages and disadvantages of each can be balanced to improve the accuracy of the overall prediction.

[0094] Furthermore, the steps for obtaining the basic prediction label weight coefficient and the optimized prediction label weight coefficient include:

[0095] The correlation coefficients exceeding the preset threshold in each classification subset are counted, and the optimized prediction label weight coefficient and the basic prediction label weight coefficient are determined according to the number. The specific formula is as follows;

[0096] ;

[0097] ;

[0098] in, is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient, is the number of correlations that exceed the preset threshold.

[0099] By adjusting the weights according to the number of correlation coefficients of different classification subsets, a more personalized prediction model can be achieved, and each classification subset can get the weight configuration that best suits its characteristics. Specifically, the more potential features there are in a classification subset, the greater the impact of the optimized prediction label value on the target prediction label value.

[0100] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0101] The embodiment of the present application also provides an AI-based user portrait construction system, which corresponds one-to-one to an AI-based user portrait construction method in the embodiment. Figure 4 , the user portrait construction system based on AI includes: preliminary prediction module 1, data extraction module 2, subset classification module 3, feature screening module 4, optimization prediction module 5 and label fusion module 6. The detailed description of each functional module is as follows:

[0102] Preliminary prediction module 1: used to collect user data and preprocess it to form a main data set, and input the main data set into the prediction model to obtain basic prediction labels.

[0103] Data extraction module 2: used to extract user data whose basic prediction label accuracy does not meet the standard from the main data set to form a data subset.

[0104] Subset classification module 3: used to perform correlation analysis on user basic information in the data subset and classify the data subset to obtain several classification subsets.

[0105] Feature screening module 4: used to screen the user behavior information in the classification subset according to preset screening rules to obtain potential features.

[0106] Optimized prediction module 5: used to obtain optimized prediction labels according to the potential features input into the prediction model.

[0107] Label fusion module 6: used to fuse the optimized prediction label with the basic prediction label to obtain the target prediction label.

[0108] Among them, the system can collect user data and preprocess it through the preliminary prediction module 1 to form a main data set, input the main data set into the prediction model to obtain a basic prediction label, and then use the data extraction module 2 to extract user data whose basic prediction label accuracy does not meet the standard from the main data set to form a data subset, and then use the subset classification module 3 to perform correlation analysis on the user basic information in the data subset and classify the data subset to obtain several classification subsets, and then use the feature screening module 4 to screen the user behavior information in the classification subset according to the preset screening rules to obtain potential features, and then use the optimization prediction module 5 to input the potential features into the prediction model to obtain optimized prediction labels, and finally use the label fusion module 6 to fuse the optimized prediction labels with the basic prediction labels to obtain the target prediction labels.

[0109] For the specific definition of the AI-based user portrait construction system, please refer to the definition of the AI-based user portrait construction method in the context, which will not be repeated here. Each module in the above-mentioned AI-based user portrait construction system can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the electronic device in the form of hardware, or can be stored in the memory of the electronic device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0110] In this embodiment, an electronic device is provided, and the electronic device is a computer. Figure 5, the electronic device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the electronic device is used to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the electronic device is used to store a detection data table. The network interface of the electronic device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a user portrait construction method based on AI is implemented.

[0111] In one embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:

[0112] S1. Collect user data and preprocess it to form a master data set, and input the master data set into the prediction model to obtain basic prediction labels.

[0113] S2. User data whose basic prediction label accuracy does not meet the standard is extracted from the main data set to form a data subset.

[0114] S3. Perform correlation analysis on the user basic information in the data subset and classify the data subset to obtain several classification subsets.

[0115] S4. Filter the user behavior information in the classification subset according to the preset filtering rules to obtain potential features.

[0116] S5. Input the potential features into the prediction model to obtain optimized prediction labels.

[0117] S6. Fusion the optimized prediction label with the basic prediction label to obtain the target prediction label.

[0118] Step S3 specifically includes the following steps:

[0119] S30. Using the FP-Growth algorithm, the user basic information in the data subset is used to identify new basic features related to user data whose basic prediction label accuracy does not meet the standard.

[0120] S31. Classify the data subsets according to the newly added basic features to obtain several classification subsets.

[0121] Step S4 specifically includes the following steps:

[0122] S40: Process the user behavior information in the classification subset according to preset candidate rules to obtain candidate features.

[0123] S41. Perform correlation matching on the newly added basic features and the candidate features to obtain matching values.

[0124] S42: setting candidate features whose matching values ​​exceed a preset threshold as potential features for adding new basic features.

[0125] An embodiment of the present application also discloses a computer-readable storage medium storing a computer program that can be loaded and executed by a processor. When the computer program is executed by the processor, the steps of any of the above-mentioned AI-based user portrait construction methods are implemented and the same effect can be achieved.

[0126] The computer-readable storage medium includes, for example, a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

Claims

1. A user portrait construction method based on AI, characterized in that: include: Collect user data and pre-process it to form the main data set; User data includes basic user information and user behavior information, and user data is pre-processed; The master dataset includes historical credit behavior, and the master dataset is input into the prediction model to obtain the basic prediction label; User data whose basic prediction label accuracy does not meet the standard is extracted from the main data set to form a data subset; Perform correlation analysis on user basic information in the data subset and classify the data subset to obtain several classification subsets; Filter the user behavior information in the classification subset according to the preset filtering rules to obtain potential features; According to the potential features input into the prediction model, the optimized prediction label is obtained; The optimized prediction label is merged with the basic prediction label to obtain the target prediction label; The step of performing correlation analysis on the user basic information in the data subset and classifying the data subset to obtain a plurality of classified subsets includes: The FP-Growth algorithm is used to identify new basic features related to user data whose basic prediction label accuracy does not meet the standard for the basic user information in the data subset; Classify the data subsets according to the newly added basic features to obtain several classification subsets; The step of filtering the user behavior information in the classification subset according to the preset filtering rules to obtain potential features includes: Process the user behavior information in the classification subset according to the preset candidate rules to obtain candidate features; According to the newly added basic features, the correlation degree is matched with the candidate features respectively to obtain the matching value; The candidate features whose matching values ​​exceed the preset threshold are set as potential features for adding new basic features; The step of processing the user behavior information in the classification subset according to the preset candidate rules to obtain candidate features includes: Perform statistical analysis on the user behavior information in the classification subset, calculate the correlation coefficient between each user behavior information and the target variable, and select user behavior information with a correlation coefficient exceeding a preset threshold as a candidate feature.

2. The method according to claim 1, characterized in that The step of fusing the optimized prediction label with the basic prediction label to obtain the target prediction label includes: The optimized prediction label and the basic prediction label are multiplied by the relevant weight coefficients and superimposed to obtain the target prediction label. The specific formula is as follows: ； Among them, F is the target prediction label value, To predict the label value based on To optimize the predicted label value, is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient.

3. The method according to claim 2, characterized in that The steps of obtaining the basic prediction label weight coefficient and the optimized prediction label weight coefficient include: The correlation coefficients exceeding the preset threshold in each classification subset are counted, and the optimized prediction label weight coefficient and the basic prediction label weight coefficient are determined according to the number. The specific formula is as follows; ； ； in, is the basic prediction label weight coefficient, To optimize the predicted label weight coefficient, is the number of correlations that exceed the preset threshold.

4. A user portrait construction system based on AI, characterized in that: include: Preliminary prediction module (1): used to collect and pre-process user data. The user data includes basic user information and user behavior information, which constitute a master data set. The master data set includes historical credit behavior. The master data set is input into the prediction model to obtain a basic prediction label. Data extraction module (2): used to extract user data whose basic prediction label accuracy does not meet the standard from the main data set to form a data subset; Subset classification module (3): used to perform correlation analysis on the basic user information in the data subset and classify the data subset, and use the FP-Growth algorithm to identify the new basic features related to the basic user information in the data subset and the user data whose basic prediction label accuracy does not meet the standard; classify the data subset according to the new basic features to obtain a number of classification subsets; Feature screening module (4): used to screen the user behavior information in the classification subset according to the preset screening rules, perform statistical analysis on the user behavior information in the classification subset according to the preset candidate rules, and calculate the correlation coefficient between each user behavior information and the target variable; Select user behavior information with a correlation coefficient exceeding a preset threshold as a candidate feature to obtain a candidate feature; match the newly added basic features with the candidate features respectively to obtain a matching value; set the candidate features with a matching value exceeding the preset threshold as potential features of the newly added basic features to obtain a potential feature; Optimized prediction module (5): used to obtain optimized prediction labels based on potential features input into the prediction model; Label fusion module (6): used to fuse the optimized prediction label with the basic prediction label to obtain the target prediction label.

5. An electronic device, characterized in that: It comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executes any one of the AI-based user portrait construction methods as claimed in claims 1 to 3.

6. A computer-readable storage medium, characterized in that: A computer program is stored which can be loaded by a processor and executes any one of the AI-based user portrait building methods as claimed in claims 1 to 3.

Images