WO2021164232A1 - 用户识别方法、装置、设备及存储介质 - Google Patents
用户识别方法、装置、设备及存储介质 Download PDFInfo
- Publication number
- WO2021164232A1 WO2021164232A1 PCT/CN2020/112350 CN2020112350W WO2021164232A1 WO 2021164232 A1 WO2021164232 A1 WO 2021164232A1 CN 2020112350 W CN2020112350 W CN 2020112350W WO 2021164232 A1 WO2021164232 A1 WO 2021164232A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- user
- data
- feature
- identified
- sample
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/21—Design, administration or maintenance of databases
- G06F16/215—Improving data quality; Data cleansing, e.g. de-duplication, removing invalid entries or correcting typographical errors
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/28—Databases characterised by their database models, e.g. relational or object models
- G06F16/284—Relational databases
- G06F16/285—Clustering or classification
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/40—Document-oriented image-based pattern recognition
Definitions
- This application relates to the field of data analysis, and in particular to a user identification method, device, equipment, and storage medium.
- this application provides a user identification method, the method includes:
- the classification probability of the user to be identified is greater than the user classification threshold, it is determined that the user to be identified is the target user.
- this application also provides a user identification device, which includes:
- the data acquisition module is used to acquire offline image data and online data of the user to be identified;
- An image processing module configured to perform image processing on the offline image data to obtain offline data of the user to be identified
- the feature data module is configured to use the offline data and online data as the user data of the user to be identified, and perform data preprocessing on the user data to obtain the feature data of the user to be identified, Preprocessing includes quantification of characteristic factors, outlier processing and data cleaning;
- the classification probability module is used to input the characteristic data of the user to be identified into a pre-trained user classification model to obtain the classification probability of the user to be identified;
- the user determination module is configured to determine that the user to be identified is a target user if the classification probability of the user to be identified is greater than the user classification threshold.
- the present application also provides a computer device, the computer device includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and execute the The computer program implements the following steps:
- the classification probability of the user to be identified is greater than the user classification threshold, it is determined that the user to be identified is the target user.
- the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor implements the following steps:
- the classification probability of the user to be identified is greater than the user classification threshold, it is determined that the user to be identified is the target user.
- FIG. 1 is a schematic flowchart of a method for training a user classification model provided by an embodiment of the present application
- Figure 2 is a schematic flow chart of the steps of outputting the weight value of the feature factor by using the random forest algorithm
- FIG. 3 is a schematic flowchart of a user identification method provided by an embodiment of the present application.
- FIG. 4 is a schematic flowchart of sub-steps of the user identification method provided in FIG. 3;
- FIG. 5 is a schematic block diagram of a model training device provided by an embodiment of the present application.
- FIG. 6 is a schematic block diagram of a user identification device further provided in an embodiment of the present application.
- FIG. 7 is a schematic block diagram of the structure of a computer device according to an embodiment of the application.
- the embodiments of the present application provide a user identification method, device, computer equipment, and storage medium.
- the user identification method can be used to identify the user characteristics of the user and classify the users, so as to determine the target user based on the recognition result of the user characteristics, and improve the recognition accuracy of the target user.
- the target user is a repeat purchase user as an example for detailed description.
- FIG. 1 is a schematic flowchart of a method for training a user classification model provided by an embodiment of the present application.
- the training method of the user classification model improves the accuracy of the model for identifying target users by training sample data.
- the training method of the user classification model specifically includes:
- S101 Obtain sample data of a sample user and perform data preprocessing on the sample data to obtain characteristic data of the sample user.
- the data preprocessing includes quantification of characteristic factors, abnormal value processing, and data cleaning. Performing data preprocessing on the sample data can reduce the interference caused by the original error data to the training of the user classification model, thereby improving the recognition accuracy of the user classification model obtained by the training.
- the characteristic data includes historical purchase records, characteristic factors, and corresponding characteristic values of the sample users. After preprocessing the sample data, the characteristic data of the sample user is obtained.
- data cleaning may include performing data over-limit inspection, feature validity inspection, and data null value inspection on the data of historical purchase records, and cleaning the data.
- the null value check can be to check and delete the record with a historical purchase record of 0. The historical purchase record of 0 indicates that the sample user does not meet the characteristics.
- abnormal value processing can be to determine whether there is uncollected or abnormal data in the sample data, and then perform missing value processing, and choose different missing value processing methods according to different business rules. For example, default value replacement: For some historical purchase records of the amount/product level, the default value is set according to general business rules for calculation; if the missing value is of a non-numeric type, the mode of the attribute is used to fill in the missing value The data.
- the quantification of characteristic factors may refer to numerically representing each characteristic factor in the sample data of the sample user.
- the characteristic factor may include industry, industry sub-category, industry major category, salary level, etc. You can use 1, 2, 3, and 4 to represent the four characteristic factors of industry, industry subcategory, industry category, and salary level respectively.
- S102 Determine a user identifier corresponding to the sample user according to the historical purchase record of the sample user.
- the user identification corresponding to the sample user is determined according to the historical purchase records in the characteristic data of the sample user.
- the user identification includes a target user and a non-target user, where the target user is, for example, a repeat purchase user, and the non-target user is, for example, a non repeat purchase user.
- sample users whose users are identified as repeat-purchasing users may be regarded as positive samples, and sample users whose users are identified as non-repetitive-purchasing users may be regarded as negative samples.
- determining the user identifier corresponding to the sample user according to the historical purchase record of the sample user may be a judgment based on the product purchase situation of the sample user in a certain period of time in the past. For example, before December 2018, a user bought a product one or more times, and then checked whether the user purchased again in January-March 2019.
- the user ID of the user is determined to be a repeated purchase
- the user takes it as a positive sample, and if the user does not purchase again, the user identification of the user is determined to be a non-repetitive purchase user, and it is taken as a negative sample.
- S103 Train a classification model according to the user identification, feature factor, and feature value corresponding to the feature factor of the sample user, and use the trained classification model as a pre-trained user classification model.
- a classification algorithm is used to train the classification model according to the user identification, feature factor, and feature value corresponding to the feature factor of the sample user.
- the classification algorithm may be a logistic regression algorithm, an XGBoost algorithm, a Gaussian Naive Bayes classification algorithm, and a random forest. Algorithm and GBDT algorithm, etc.
- step S102 it may further include:
- S104 Use a random forest algorithm to classify the feature factors and corresponding feature values of the sample users and the corresponding user identifiers, so as to output the weight values of the feature factors.
- the random forest algorithm is used to output the weight value of each feature factor, and specifically, the out-of-bag data can be used to calculate the importance of each feature factor.
- the output of the weight value of the feature factor using the random forest algorithm specifically includes:
- the K-fold cross-slicing method is the StratifiedKFold method.
- the original data set refers to the collection of characteristic data of all sample users.
- the sample data in the training subset is randomly sampled using the repeated sampling method. After each sample, the sample data is replaced and the next sample is performed.
- the sample data obtained by the sample is used as a target training sub
- the data volume of the target training subset is the same as the data volume of the sample data in the training subset, the elements of different target training subsets can be repeated, and the elements of the same target training subset can also be repeated.
- the characteristic data of the sample users that are not drawn each time are taken as out-of-bag data.
- the sample data in the training subset is N
- the sample data is sampled with replacement N times to obtain N sample data as a target training subset.
- step S1042 it may further include: sampling and balancing the data in the target training subset to obtain a processed target training subset.
- the sampling balance processing may refer to using the SMOTE algorithm to construct new samples for the data in the target training subset.
- the SMOTE algorithm analyzes the minority samples and artificially synthesizes new samples based on the minority samples and adds them to the data set.
- the algorithm flow can be as follows:
- a sampling ratio is set according to the sample imbalance ratio to determine the sampling magnification N. For each minority sample x, several samples are randomly selected from its K neighbors, assuming the selected neighbor x n .
- a random forest algorithm is used to generate a decision tree based on the feature factor, a random forest model is formed according to the generated multiple decision trees, and the feature is determined according to the average of the predicted values of the multiple decision trees The final prediction result of the factor, and use the prediction result as the output result of the random forest model.
- S1044 Based on the output result of the random forest model, calculate the weight value of the feature factor by using the out-of-bag data.
- the weight value assignment method the ability to select and discriminate data during the model training process can be further ensured, and the rapid and effective convergence of the model calculation results can be achieved.
- the out-of-bag data is used to calculate the weight value of the feature factor.
- the factor classification threshold can be freely set by the user.
- the multiple feature factors are sorted from high to low according to the weight values of the multiple feature factors, and feature factors with a weight value greater than the factor classification threshold are used as target feature factors, and the target feature factors are used to identify the target user The characteristic factor.
- the feature data of the sample user is screened according to the target feature factor, so as to obtain the feature value corresponding to the sample user and the target feature factor.
- the feature data of the sample users is filtered according to the target feature factors, which reduces the interference of irrelevant feature factors and feature values when training the user classification model, and reduces the amount of calculation during training.
- S107 Train a classification model according to the user identification, target feature factor, and feature value corresponding to the target feature factor of the sample user, and use the trained classification model as a pre-trained user classification model.
- the user classification model may be trained by using a logistic regression algorithm, an XGBoost algorithm, a Gaussian Naive Bayes classification algorithm, a random forest algorithm, and a GBDT algorithm.
- the above-mentioned embodiment provides a user classification model training method, which performs data preprocessing on the acquired sample data, removes dirty data in the sample data and fills it with data, improves the quality of the sample data, and reduces the original error data pair
- the interference of the trained user classification model The random forest algorithm is used to output the weight value of the feature factor.
- the feature factor is used as the target feature factor, and the feature data of the sample user is screened based on the target feature factor, and the sample obtained according to the screening
- the data is used to train the user classification model, which reduces the interference of irrelevant feature factors and feature values when training the user classification model, and reduces the amount of calculation during training.
- FIG. 3 is a schematic flowchart of a user identification method provided by an embodiment of the present application.
- the user identification method can perform image processing on the offline image data of the user to be identified, and use a pre-trained user classification model to identify the user based on the user data of the user to be identified after processing, so as to identify the target user from the user.
- the user identification method specifically includes: step S201 to step S205.
- the offline image data of the user to be identified may include the basic information form image that the user to be identified fills out offline, the visit record image of the user to be identified by the business personnel, the offline activity check-in form image of the user to be identified, and the like.
- the basic information form, visit record, and offline activity check-in form of the user to be identified may be scanned or photographed to obtain offline image data.
- the online data includes online purchase data of the user to be identified, user portrait data, and the like.
- S202 Perform image processing on the offline image data to obtain offline data of the user to be identified.
- performing image processing on the offline image data refers to identifying from the image data of the user to be identified to obtain offline data of the user to be identified.
- step S202 specifically includes:
- the preprocessing includes binarization, noise removal and tilt correction. Preprocess the image of the basic information form filled in offline by the user to be identified, the visit record image of the user to be identified by the business personnel, and the offline activity check-in form image of the user to be identified, so as to obtain the preprocessed offline fill in the user to be identified Basic information form image, visit record image and offline activity check-in form image.
- binarization refers to processing the color image so that the image only includes foreground information and background information, and defines the foreground information as black and the background information as white. Due to the large amount of information in color images, in order to improve the accuracy and speed of character recognition, the color image can be binarized first.
- layout analysis refers to dividing the preprocessed offline image into paragraphs and lines.
- Character recognition refers to the feature extraction of offline image data, so as to extract the text data in the offline image data, and use the extracted text data as the recognition result.
- the recognition result of the basic information table image includes basic information such as the user's name, gender, and age
- the recognition result of the visit record image includes the user's evaluation of the currently purchased product, whether he intends to continue to purchase the product, and other information.
- the recognition result is restored and post-processed, so that the recognition result can be arranged according to the format in the offline image data, and the recognition result is corrected according to the relationship of the specific language context, so as to determine the offline data.
- the form data corresponding to the basic information form image filled by the user to be identified is determined, and the visit data and line corresponding to the visit record image Download the check-in data corresponding to the activity check-in form image, and use the form data, visit data, and check-in data together as offline data.
- S203 Use the offline data and online data as user data of the user to be identified, and perform data preprocessing on the user data to obtain characteristic data of the user to be identified.
- using the offline data and online data as the user data of the user to be identified refers to entering the offline data identified through image recognition, so as to be the user of the user to be identified together with the online data data. Improve the speed and efficiency of offline data entry, and also improve the accuracy of offline data entry.
- the preprocessing includes feature factor quantification, abnormal value processing and data cleaning.
- the process of data cleaning needs to be processed in conjunction with the characteristic factors of each user. For some interactive behavior feature factors, zero value can be filled if it cannot be collected or missing; for user income and assets, the average value can be filled; for gender, education, occupation, etc., the blank value can be left as unknown; and for the most recent purchase Data such as time should be filled with a larger value to indicate that there is no recent purchase.
- the characteristic data of the user to be identified includes historical purchase records, characteristic factors, and corresponding characteristic values of the user to be identified.
- the historical purchase record is the record of the user's purchase of the company's products, including the time of purchase of the product and the detailed information of the product, such as life insurance products, including the name of the insurance product, the method of purchase, the number of purchases, and who is insured.
- entering the offline data specifically includes:
- visit data and sign-in data of the user to be identified the offline data of the same user to be identified is associated; according to the form data of the user to be identified and the online data are the same
- the form data of the user to be identified is associated with online data; based on the association relationship between the form data of the user to be identified and the online data, the visit data and sign-in data of the user to be identified are entered.
- the offline data of the same user to be identified is correlated according to the form data, visit data, and check-in data of the user to be identified based on the name, gender, and other information, and then the name, gender, and gender in the form data of the user to be identified are correlated.
- Basic information such as age, retrieve online data matching the user to be identified, and establish an association relationship between the form data of the user to be identified and the online data, and finally based on the form of the user to be identified
- the relationship between the data and the online data, the visit data and the sign-in data associated with the form data are entered, so that the form data, the visit data and the sign-in data and the online data are used as the user data of the user to be identified .
- the characteristic data of the user to be identified is input into a pre-trained user classification model, and the user classification model predicts the user to be identified, and outputs the classification probability of the user to be identified.
- the classification probability is the probability that the user to be identified is a target user or the probability that the user to be identified is a non-target user.
- S205 If the classification probability of the user to be identified is greater than the user classification threshold, determine that the user to be identified is the target user.
- the user classification threshold may be preset by business personnel, or may be selected after classification and prediction of sample data. It should be noted that the user classification threshold can be adjusted according to application scenarios and business requirements.
- the user classification threshold when the user classification threshold is selected after classifying and predicting the sample data, it may specifically include:
- the classification probability of the pre-trained user classification model for the sample user is obtained, and multiple test thresholds are set.
- the sample user is defined as Target users, on the contrary, define the sample users as non-target users, and accordingly calculate the confusion matrix corresponding to multiple test thresholds.
- the confusion matrix is shown in Table 1:
- TP True Positive
- FP False Positive
- FN False Negative
- TN True Negative
- the Kolmogorov-Smolov test values corresponding to multiple test thresholds are calculated respectively, that is, the KS value.
- the calculation formula is as follows:
- TPR represents the real interest rate
- FPR represents the interest rate anyway.
- the test threshold corresponding to the maximum KS value can be selected from them, and the test threshold can be used as the user classification threshold.
- the user identification method provided in the above embodiment obtains offline image data and online data of the user to be identified, and then performs data processing on the offline image data to obtain the offline data of the user to be identified, and then converts the offline data
- online data is used as the user data of the user to be identified, and data preprocessing is performed on the user data to obtain the characteristic data of the user to be identified, and the characteristic data of the user to be identified is input into the pre-trained user
- the classification model is used to obtain the classification probability of the user to be identified, and if the classification probability of the user to be identified is greater than the user classification threshold, the user to be identified is determined to be the target user.
- the offline data and online data are used as user data of the user to be identified, and the speed and accuracy of offline data entry are improved. Then input the characteristic data of the user to be identified into a pre-trained user classification model, and determine whether the user to be identified is a target user according to the classification probability output by the user classification model, so as to improve the identification accuracy of the target user.
- FIG. 5 is a schematic block diagram of a model training device provided by an embodiment of the present application.
- the model training device may be configured in a server to execute the aforementioned user classification model training method.
- the model training device 300 includes: a data processing module 301, a user identification module 302, a model training module 303, a user classification module 304, a target feature module 305, a data screening module 306, and a target model training module 307.
- the data processing module 301 is configured to obtain sample data of a sample user and perform data preprocessing on the sample data to obtain characteristic data of the sample user.
- the characteristic data includes historical purchase records and characteristic factors of the sample user And the corresponding characteristic value.
- the user identification module 302 is configured to determine a user identification corresponding to the sample user according to the historical purchase record of the sample user, and the user identification includes a target user and a non-target user.
- the model training module 303 is configured to train a classification model according to the user identification, feature factors, and feature values corresponding to the feature factors of the sample users, and use the trained classification model as a pre-trained user classification model.
- the user classification module 304 is configured to use a random forest algorithm to classify the feature factors and corresponding feature values of the sample users and the corresponding user identifiers to output the weight values of the feature factors.
- the user classification module 304 includes a subset division submodule 3041, a target subset submodule 3042, a random forest model submodule 3043, and a weight calculation submodule 3044.
- the subset dividing sub-module 3041 is used to use the K-fold cross-slicing method to divide the feature factors and corresponding feature values of the sample users and the corresponding user identifiers to obtain multiple disjoint training subsets .
- the target subset sub-module 3042 is used to perform multiple random replacement sampling on multiple training subsets by repeated sampling to obtain multiple target training subsets, and remove the samples that have not been drawn.
- the user's feature factor and corresponding feature value and the corresponding user ID are used as out-of-bag data.
- the random forest model sub-module 3043 is used to obtain a random forest model by using a random forest algorithm based on the feature factor and a plurality of the target training subsets.
- the weight calculation sub-module 3044 is configured to calculate the weight value of the feature factor by using the out-of-bag data based on the output result of the random forest model.
- the target feature module 305 is configured to determine that the feature factor is a target feature factor when the weight value of the feature factor is greater than the factor classification threshold.
- the data screening module 306 is configured to screen the characteristic data of the sample user according to the target characteristic factor to obtain a characteristic value corresponding to the sample user and the target characteristic factor.
- the target model training module 307 is configured to train a classification model according to the user identification of the sample user, the target feature factor, and the feature value corresponding to the target feature factor, and use the trained classification model as a pre-trained user classification model.
- FIG. 6 is a schematic block diagram of a user identification device provided in an embodiment of the present application, and the user identification device is used to execute the aforementioned user identification method.
- the user identification device can be configured in a server or a terminal.
- the server can be an independent server or a server cluster.
- the terminal can be an electronic device such as a mobile phone, a tablet computer, a notebook computer, a desktop computer, a personal digital assistant, and a wearable device.
- the user identification device 400 includes: a data acquisition module 401, an image processing module 402, a characteristic data module 403, a classification probability module 404, and a user determination module 405.
- the data acquisition module 401 is used to acquire offline image data and online data of the user to be identified.
- the image processing module 402 is configured to perform image processing on the offline image data to obtain offline data of the user to be identified.
- the image processing module 402 includes a preprocessing submodule 4021, a recognition result submodule 4022, and a data determination submodule 4023.
- the preprocessing submodule 4021 is used to preprocess the offline image data, and the preprocessing includes binarization, noise removal and tilt correction;
- the recognition result submodule 4022 is used to perform preprocessing on the preprocessed image data.
- the offline image data is subjected to layout analysis and character recognition to obtain a recognition result;
- a data determination sub-module 4023 is used to determine the offline data according to the recognition result.
- the feature data module 403 is configured to use the offline data and online data as the user data of the user to be identified, and perform data preprocessing on the user data to obtain the feature data of the user to be identified.
- the preprocessing includes quantification of characteristic factors, processing of outliers and data cleaning.
- the classification probability module 404 is configured to input the characteristic data of the user to be identified into a pre-trained user classification model to obtain the classification probability of the user to be identified.
- the user determination module 405 is configured to determine that the user to be identified is a target user if the classification probability of the user to be identified is greater than the user classification threshold.
- the above-mentioned user identification device may be implemented in the form of a computer program, and the computer program may be run on a computer device as shown in FIG. 7.
- FIG. 7 is a schematic block diagram of the structure of a computer device provided by an embodiment of the present application.
- the computer equipment can be a server or a terminal.
- the computer device includes a processor, a memory, and a network interface connected through a system bus, where the memory may include a non-volatile or volatile storage medium and an internal memory.
- Non-volatile or volatile storage media can store operating systems and computer programs.
- the computer program includes program instructions, and when the program instructions are executed, the processor can execute any user identification method.
- the processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
- the internal memory provides an environment for the operation of a computer program in a non-volatile or volatile storage medium.
- the processor can execute any user identification method.
- the network interface is used for network communication, such as sending assigned tasks.
- the network interface is used for network communication, such as sending assigned tasks.
- FIG. 7 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied.
- the specific computer device may Including more or fewer parts than shown in the figure, or combining some parts, or having a different arrangement of parts.
- the processor may be a central processing unit (Central Processing Unit, CPU), and the processor may also be other general-purpose processors, digital signal processors (Digital Signal Processor, DSP), and application specific integrated circuits (Application Specific Integrated Circuits). Circuit, ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc.
- the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.
- the processor is used to run a computer program stored in a memory to implement the following steps:
- the preprocessing includes quantification of characteristic factors, abnormal value processing, and data cleaning;
- the characteristic data of the user is input into a pre-trained user classification model to obtain the classification probability of the user to be identified; if the classification probability of the user to be identified is greater than the user classification threshold, the user to be identified is determined to be the target user.
- the processor when the processor implements the image processing on the offline image data to obtain the offline data of the user to be identified, the processor is used to implement:
- the recognition result determines offline data.
- the processor is further configured to implement:
- sample data of a sample user Acquiring sample data of a sample user and performing data preprocessing on the sample data to obtain characteristic data of the sample user, the characteristic data including historical purchase records, characteristic factors, and corresponding characteristic values of the sample user;
- the historical purchase record of the sample user determines the user ID corresponding to the sample user, and the user ID includes target users and non-target users; training according to the user ID, feature factor, and feature value corresponding to the feature factor of the sample user Classification model, and use the trained classification model as a pre-trained user classification model.
- the processor before implementing the training of the classification model based on the user identification, feature factor, and feature value corresponding to the feature factor of the sample user, the processor is further configured to implement:
- the random forest algorithm is used to classify the characteristic factors and corresponding characteristic values of the sample users and the corresponding user identifiers to output the weight values of the characteristic factors; when the weight values of the characteristic factors are greater than the factor classification threshold, it is determined that the The characteristic factor is the target characteristic factor; the characteristic data of the sample user is screened according to the target characteristic factor to obtain the characteristic value corresponding to the sample user and the target characteristic factor; the processor is implementing the basis When training a classification model with the user identification, feature factor, and feature value corresponding to the feature factor of the sample user, it is used to implement: training the classification according to the user identification of the sample user, the target feature factor, and the feature value corresponding to the target feature factor Model.
- the processor is further configured to implement:
- the processor is configured to use the random forest algorithm to classify the feature factors and corresponding feature values and corresponding user identifiers of the sample users to output the weight values of the feature factors. accomplish:
- the K-fold cross-segmentation method is used to segment the characteristic factors and corresponding characteristic values of the sample users and the corresponding user identifications to obtain multiple disjoint training subsets; repeated sampling is used for multiple training subsets
- the method performs multiple random replacement sampling to obtain multiple target training subsets, and uses the feature factors and corresponding feature values and corresponding user identifications of the sample users that have not been selected as out-of-bag data; based on The characteristic factor and the plurality of target training subsets are obtained by using a random forest algorithm to obtain a random forest model; based on the output result of the random forest model, the weight value of the characteristic factor is calculated by using the out-of-bag data.
- the processor before the processor implements the random forest algorithm to obtain a random forest model based on the feature factor and a plurality of the target training subsets, it is further configured to implement:
- the data in the target training subset is sampled and balanced to obtain a processed target training subset.
- the embodiments of the present application also provide a computer-readable storage medium.
- the computer-readable storage medium may be volatile or non-volatile.
- the computer-readable storage medium stores a computer program.
- the computer program includes program instructions, and the processor executes the program instructions to implement the following steps:
- the classification probability of the user to be identified is greater than the user classification threshold, it is determined that the user to be identified is the target user.
- the computer-readable storage medium may be the internal storage unit of the computer device described in the foregoing embodiment, for example, the hard disk or memory of the computer device.
- the computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk equipped on the computer device, a smart memory card (Smart Media Card, SMC), and a Secure Digital (SD) ) Card, Flash Card, etc.
- a plug-in hard disk equipped on the computer device such as a smart memory card (Smart Media Card, SMC), and a Secure Digital (SD) ) Card, Flash Card, etc.
- SD Secure Digital
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Databases & Information Systems (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
一种用户识别方法、装置、设备及存储介质,涉及数据分析领域,具体使用了用户分类模型识别目标用户,所述方法包括:获取待识别用户的线下图像数据和线上数据;对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据;将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。提高了线下数据录入的准确度和基于用户特征的目标用户的识别准确度。
Description
本申请要求于2020年2月17日提交中国专利局、申请号为CN202010097654.6、名称为“用户识别方法、装置、设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。
本申请涉及数据分析领域,尤其涉及一种用户识别方法、装置、设备及存储介质。
目前,在对用户进行目标用户的识别时,现有的方法大多是基于业务规则,对用户进行特征分析,从而基于分析出的用户特征进行目标用户的识别。发明人意识到该方法会受到个人主观的影响,容易产生数据上的遗漏,无法完整分析到所有特征,导致识别出的用户数据特征不完整。并且,业务规则的建立是基于人工观察和数据比较,因此用户数据特征不能完全准确地区别开,所以按照这样的规则进行目标用户的识别时,识别准确率较低。
因此,如何基于用户的特征进行识别,以提高识别目标用户的准确率成为亟待解决的问题。
发明内容
第一方面,本申请提供了一种用户识别方法,所述方法包括:
获取待识别用户的线下图像数据和线上数据;
对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;
将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;
将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;
若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
第二方面,本申请还提供了一种用户识别装置,所述装置包括:
数据获取模块,用于获取待识别用户的线下图像数据和线上数据;
图像处理模块,用于对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;
特征数据模块,用于将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;
分类概率模块,用于将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;
用户确定模块,用于若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
第三方面,本申请还提供了一种计算机设备,所述计算机设备包括存储器和处理器;所述存储器用于存储计算机程序;所述处理器,用于执行所述计算机程序并在执行所述计算机程序时实现如下步骤:
获取待识别用户的线下图像数据和线上数据;
对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;
将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行 数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;
将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;
若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
第四方面,本申请还提供了一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序被处理器执行时使所述处理器实现如下步骤:
获取待识别用户的线下图像数据和线上数据;
对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;
将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;
将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;
若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
为了更清楚地说明本申请实施例技术方案,下面将对实施例描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图是本申请的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动的前提下,还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的一种用户分类模型的训练方法的示意流程图;
图2是采用随机森林算法输出特征因子的权重值的步骤示意流程图;
图3是本申请实施例提供的一种用户识别方法的示意流程图;
图4是图3中提供的用户识别方法的子步骤示意流程图;
图5是本申请一实施例提供的一种模型训练装置的示意性框图;
图6是本申请的实施例还提供一种用户识别装置的示意性框图;
图7为本申请一实施例提供的一种计算机设备的结构示意性框图。
下面将结合本申请实施例中的附图,对本申请实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例是本申请一部分实施例,而不是全部的实施例。基于本申请中的实施例,本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例,都属于本申请保护的范围。
附图中所示的流程图仅是示例说明,不是必须包括所有的内容和操作/步骤,也不是必须按所描述的顺序执行。例如,有的操作/步骤还可以分解、组合或部分合并,因此实际执行的顺序有可能根据实际情况改变。
应当理解,在此本申请说明书中所使用的术语仅仅是出于描述特定实施例的目的而并不意在限制本申请。如在本申请说明书和所附权利要求书中所使用的那样,除非上下文清楚地指明其它情况,否则单数形式的“一”、“一个”及“该”意在包括复数形式。
还应当理解,在本申请说明书和所附权利要求书中使用的术语“和/或”是指相关联列出的项中的一个或多个的任何组合以及所有可能组合,并且包括这些组合。
本申请的实施例提供了一种用户识别方法、装置、计算机设备及存储介质。用户识别方法可用于对用户进行用户特征的识别,并对用户进行分类,从而基于用户特征的识别结果确定目标用户,提高目标用户的识别准确率。在本申请中,为了便于描述,以目标用户为重复购买用户为例进行详细说明。
下面结合附图,对本申请的一些实施方式作详细说明。在不冲突的情况下,下述的实 施例及实施例中的特征可以相互组合。
请参阅图1,图1是本申请实施例提供的一种用户分类模型的训练方法的示意流程图。该用户分类模型的训练方法通过对样本数据进行训练,从而提高模型对于目标用户的识别准确率。
如图1所示,该用户分类模型的训练方法,具体包括:
S101、获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据。
具体地,所述数据预处理包括特征因素量化、异常值处理和数据清洗。对所述样本数据进行数据预处理,能够减少原始错误数据对用户分类模型的训练造成的干扰,从而提高训练得到的用户分类模型的识别准确度。所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值。对样本数据进行数据预处理后,得到样本用户的特征数据。
其中,数据清洗可以包括对历史购买记录的数据进行数据超限值检验、特征有效性检验、数据空值检验,对数据进行清洗。在具体实施过程中,空值检查可以是检查历史购买记录为0的记录并予以删除,历史购买记录为0表示样本用户不符合特征性。
其中,异常值处理可以是判断样本数据中是否存在未采集或异常数据的情况,然后进行缺失值处理,根据不同业务规则选用不同的缺失值处理方法。比如,默认值替换:对于某些历史购买记录的金额/产品等级的情况,按普遍的业务规则设置默认值进行计算;如果缺失值是非数值类型的,则用该属性的众数来补齐缺失的数据。
其中,特征因素量化可以是指将样本用户的样本数据中的各个特征因子进行数值化表示。具体地,特征因子可以包括行业、行业小类、行业大类、薪酬等级等。可以采用1、2、3、4分别表示行业、行业小类、行业大类、薪酬等级四个特征因子。
S102、根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识。
具体地,根据样本用户的特征数据中的历史购买记录确定样本用户对应的用户标识。其中,所述用户标识包括目标用户和非目标用户,其中,目标用户例如为重复购买用户,非目标用户例如为非重复购买用户。
在一些实施例中,可以将用户标识为重复购买用户的样本用户作为正样本,将用户标识为非重复购买用户的样本用户作为负样本。其中,根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识可以是根据过去某一时段样本用户的产品购买情况进行判断。例如,2018年12月以前,某用户买过一次或多次某产品,然后查看该用户在2019年1-3月份是否再次购买,如果该用户再次购买,则确定该用户的用户标识为重复购买用户,将其作为正样本,如果该用户没有再次购买,则确定该用户的用户标识为非重复购买用户,将其作为负样本。
S103、根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
具体地,根据样本用户的用户标识、特征因子和与特征因子对应的特征值采用分类算法训练分类模型,所述分类算法可以是逻辑回归算法、XGBoost算法、高斯朴素贝叶斯分类算法、随机森林算法和GBDT算法等。
在另一实施例中,在步骤S102之后,还可以包括:
S104、采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值。
具体地,利用随机森林算法输出每个特征因子的权重值,具体的可以利用袋外数据来计算每个特征因子的重要性。
在一些实施例中,请参考图2,采用随机森林算法输出特征因子的权重值具体包括:
S1041、采用K折交叉切分方法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分割,以得到多个不相交的训练子集。
具体地,K折交叉切分方法为StratifiedKFold方法,通过分层采样,可以确保各个训练子集中各个用户标识的样本用户的比例与原始数据集中相同。其中,原始数据集是指所有样本用户的特征数据的集合。
S1042、对多个所述训练子集采用重复抽样的方法进行多次随机有放回采样,以得到多个目标训练子集,并将未被抽到的所述样本用户的特征因子和对应的特征值以及对应的用户标识作为袋外数据。
具体地,具体地,对所述训练子集中的样本数据使用重复抽样的方法进行随机的采样,每一次采样后都有放回的再进行下一次采样,采样得到的样本数据作为一个目标训练子集,目标训练子集的数据量和所述训练子集中的样本数据的数据量相同,不同目标训练子集的元素可以重复,同一个目标训练子集的元素也可以重复。并且将每次未被抽到的所述样本用户的特征数据作为袋外数据。
例如,训练子集中的样本数据为N,对该样本数据进行N次有放回的采样,得到N个样本数据作为一个目标训练子集。
在一些实施例中,在步骤S1042之后,还可以包括:对所述目标训练子集中的数据进行抽样平衡化处理,以得到处理后的目标训练子集。
其中,抽样平衡化处理可以是指对目标训练子集中的数据采用SMOTE算法构建新样本。其中,SMOTE算法是对少数类样本进行分析并根据少数类样本人工合成新样本添加到数据集中,算法流程可以如下:
对于少数类中每一个样本x,以欧氏距离为标准计算它到少数类样本集S
min中所有样本的距离,得到其K近邻。
根据样本不平衡比例设置一个采样比例以确定采样倍率N,对于每一个少数类样本x,从其K近邻中随机选择若干个样本,假设选择的近邻x
n。
对于每一个随机选出的近邻x
n,分别与原样本按照如下的公式构建新的样本:
X
new=x+rand(0,1)*|x-x
n|
S1043、基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型。
具体地,对于每一个目标训练子集,采用随机森林算法基于所述特征因子生成一颗决策树,根据生成的多颗决策树组成随机森林模型,按照多颗决策树的预测值的均值决定特征因子的最终预测结果,并将该预测结果作为随机森林模型的输出结果。
S1044、基于所述随机森林模型的输出结果,利用所述袋外数据计算所述特征因子的权重值。
具体地,通过权重值的赋予方式,可以进一步确保模型训练过程中对数据的选取甄别的能力,实现模型计算结果的快速有效收敛。根据随机森林模型的输出结果,采用袋外数据计算特征因子的权重值,具体的计算方法如下:对于随机森林中的每一颗决策树,使用相应的袋外数据来计算它的袋外数据误差,记为errOOB1;随机地对袋外数据所有样本的特征因子X加入噪声干扰,再次计算它的袋外数据误差,记为errOOB2;假设随机森林中有K棵树,那么对于特征因子X的重要性=∑(errOOB2-errOOB1)/K。
之所以可以用这个表达式作为相应特征因子的重要性的度量值是因为若给某个特征因子随机加入噪声之后,袋外的准确率大幅度降低,则说明这个特征因子对于样本的分类结果影响很大,也就是说它的重要程度比较高。
S105、当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子。
具体地,因子分类阈值可以是用户自由设定的。根据所述多个特征因子的权重值对多个特征因子按照从高到低进行排序,将权重值大于因子分类阈值的特征因子作为目标特征因子,所述目标特征因子是指用于识别目标用户的特征因子。
S106、根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值。
具体地,在确定目标特征因子后,根据目标特征因子对样本用户的特征数据进行筛选,从而得到样本用户与目标特征因子对应的特征值。根据目标特征因子对样本用户的特征数据进行筛选,降低训练用户分类模型时无关特征因子和特征值的干扰,并且减少训练时的运算量。
S107、根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
具体地,所述用户分类模型可以采用逻辑回归算法、XGBoost算法、高斯朴素贝叶斯分类算法、随机森林算法和GBDT算法进行训练得到。
上述实施例提供的一种用户分类模型的训练方法,对获取到的样本数据进行数据预处理,清除样本数据中的脏数据并且对其进行数据填充,提高样本数据的质量,减少原始错误数据对训练的用户分类模型的干扰。采用随机森林算法输出特征因子的权重值,当该权重值大于因子分类阈值时,将该特征因子作为目标特征因子,并基于目标特征因子对样本用户的特征数据进行筛选,从而根据筛选得到的样本数据进行用户分类模型的训练,降低了训练用户分类模型时无关特征因子和特征值的干扰,并且减少训练时的运算量。
请参阅图3,图3是本申请实施例提供的一种用户识别方法的示意流程图。该用户识别方法可以对待识别用户的线下图像数据进行图像处理,根据处理后得到的待识别用户的用户数据利用预先训练的用户分类模型对用户进行识别,以从用户中识别出目标用户。
如图3所示,该用户识别方法,具体包括:步骤S201至步骤S205。
S201、获取待识别用户的线下图像数据和线上数据。
具体地,所述待识别用户的线下图像数据可以包括将待识别用户线下填写的基本信息表格图像、业务人员对待识别用户的拜访记录图像、待识别用户的线下活动签到表格图像等。在一些实施例中,可以对待识别用户的基本信息表格、拜访记录和线下活动签到表格进行扫描或拍照,以得到线下图像数据。所述线上数据包括待识别用户的线上购买数据、用户画像数据等。
S202、对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据。
具体地,对所述线下图像数据进行图像处理是指从待识别用户的图像数据中进行识别,以得到所述待识别用户的线下数据。
其中,在一些实施例中,请参考图4,步骤S202具体包括:
S2021、对所述线下图像数据进行预处理。
具体地,所述预处理包括二值化、噪声去除和倾斜矫正。分别对待识别用户线下填写的基本信息表格图像、业务人员对待识别用户的拜访记录图像、待识别用户的线下活动签到表格图像进行预处理,从而得到预处理后的待识别用户线下填写的基本信息表格图像、拜访记录图像和线下活动签到表格图像。
其中,二值化是指对彩色图像进行处理,使图片只包括前景信息与背景信息,并将前景信息定义为黑色,背景信息为白色。由于彩色图像的信息量较大,为了提高文字识别的准确度和识别速度,可以先对彩色图像进行二值化处理。
S2022、对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果。
具体地,版面分析是指对预处理后的线下图像进行分段落和分行。字符识别是指对线下图像数据进行特征提取,从而提取出线下图像数据中的文字数据,并将提取出的文字数据作为识别结果。
分别对预处理后的待识别用户线下填写的基本信息表格图像、拜访记录图像和线下活动签到表格图像进行版面分析和字符识别,以分别得到基本信息表格图像、拜访记录图像和线下活动签到表格图像对应的识别结果。其中,所述基本信息表格图像的识别结果包括 用户姓名、性别、年龄等基本信息,拜访记录图像的识别结果包括用户对于当前所购买产品的评价、是否有意向继续购买该产品等信息。
S2023、根据所述识别结果确定线下数据。
具体地,对识别结果进行版面恢复和后处理,使识别结果能够按照线下图像数据中的格式进行排列,并且根据特定的语言上下文的关系,对识别结果进行矫正,从而确定线下数据。
根据待识别用户填写的基本信息表格图像、拜访记录图像和线下活动签到表格图像对应的识别结果分别确定待识别用户填写的基本信息表格图像对应的表格数据,拜访记录图像对应的拜访数据和线下活动签到表格图像对应的签到数据,并将所述表格数据、拜访数据和签到数据共同作为线下数据。
S203、将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据。
具体地,将所述线下数据和线上数据作为所述待识别用户的用户数据,是指将经过图像识别识别出的线下数据进行录入,从而与线上数据共同作为待识别用户的用户数据。提高了线下数据录入的速度和效率,并且也提高了线下数据录入的准确性。
其中,所述预处理包括特征因素量化、异常值处理和数据清洗。数据清洗的过程需要结合每个用户的特征因子进行处理。对于一些互动行为特征因子,无法采集或者缺少的情况下可以进行零值填充;对于用户的收入、资产可以填充均值;对于性别、学历、职业等可以保留空值为未知;而对于最近一次的购买时间这类数据,应该填充为较大的数值表示最近没有购买。
所述待识别用户的特征数据包括所述待识别用户的历史购买记录、特征因子和对应的特征值。历史购买记录为用户购买公司产品的记录,包括购买产品的时间及产品的详细信息,如人身保险类产品,则包括该保险产品的名称、购买方式、够买件数、为谁投保等。
在一些实施例中,将所述线下数据进行录入,具体包括:
根据所述待识别用户的所述表格数据、拜访数据和签到数据将同一所述待识别用户的线下数据进行关联;根据所述待识别用户的表格数据和所述线上数据将同一所述待识别用户的所述表格数据与线上数据进行关联;基于待识别用户的所述表格数据与线上数据之间的关联关系,将所述待识别用户的拜访数据和签到数据进行录入。
具体地,根据待识别用户的表格数据、拜访数据和签到数据中的姓名、性别等信息将同一待识别用户的线下数据进行关联,然后再根据待识别用户的表格数据中的姓名、性别、年龄等基本信息,调取出与该待识别用户匹配的线上数据,并建立该待识别用户的所述表格数据与线上数据之间的关联关系,最后基于该待识别用户的所述表格数据与线上数据之间的关联关系,将与所述表格数据相关联的拜访数据和签到数据进行录入,从而使表格数据、拜访数据和签到数据与线上数据共同作为待识别用户的用户数据。
通过表格数据调取待识别用户的线上数据,提高了数据录入的准确率和匹配程度,并且基于表格数据与拜访数据和签到数据之间的关联关系以及表格数据与线上数据之间的关联关系,对线下数据进行录入,提高了数据录入的效率。
S204、将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率。
具体地,将待识别用户的特征数据输入预先训练的用户分类模型,由用户分类模型对待识别用户进行预测,输出所述待识别用户的分类概率。其中,该分类概率为所述待识别用户为目标用户的概率或者所述待识别用户为非目标用户的概率。
S205、若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
具体地,若用户分类模型输出的该待识别用户的分类概率大于用户分类阈值,则确定 所述待识别用户为目标用户。其中,该用户分类阈值可以是业务人员预先设置的,也可以是通过对样本数据进行分类预测后选出的。需要说明的是,该用户分类阈值可以根据应用场景以及业务要求进行调整。
在一些实施例中,在通过对样本数据进行分类预测后选出用户分类阈值时,具体可以包括:
获取所述预先训练的用户分类模型对于所述样本用户的分类概率;基于多个测试阈值分别计算多个与所述测试阈值对应的混淆矩阵;根据多个所述混淆矩阵计算多个柯尔莫哥洛夫-斯摩洛夫检验值;将多个所述柯尔莫哥洛夫-斯摩洛夫检验值中的最大值对应的测试阈值作为用户分类阈值。
具体地,获取所述预先训练的用户分类模型对于样本用户的分类概率,并且设置多个测试阈值,当用户分类模型输出的某样本用户的分类概率大于该测试阈值时,将该样本用户定义为目标用户,反之,则将该样本用户定义为非目标用户,据此,分别计算出多个测试阈值对应的混淆矩阵。其中,混淆矩阵具体如表1所示:
表1
其中,True Positive(TP)表示真实值是positive,模型也认为是positive的数量;False Positive(FP)表示真实值是negative,模型认为是positive的数量;False Negative(FN)表示真实值是positive,模型认为是negative的数量;True Negative(TN)表示真实值是negative,模型也认为是negative的数量。
根据得到的混淆矩阵,分别计算多个测试阈值对应的柯尔莫哥洛夫-斯摩洛夫检验值,也即KS值。计算公式如下所示:
KS=max(TPR-FPR)
TPR=TP/(TP+FN)
FPR=FP/(FP+TN)
其中,TPR表示真正利率,FPR表示反正利率。
由于KS值可以反应模型的最优区分效果,因此,KS值越大,则说明此时的预测效果越好。因此对于多个测试阈值计算出的KS值,可以从中选取KS最大值所对应的测试阈值,并将该测试阈值作为用户分类阈值。
上述实施例提供的用户识别方法,通过获取待识别用户的线下图像数据和线上数据,然后对线下图像数据进行数据处理,得到待识别用户的线下数据,再将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率,若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。通过对线下图像数据进行图像处理,以得到线下数据,从而将线下数据和线上数据作为待识别用户的用户数据,提高线下数据录入的速度和准确率。然后将待识别用户的特征数据输入预先训练的用户分类模型,根据所述用户分类模型输出的分类概率判断该待识别用户是否为目标用户,提高目标用户的识别准确度。
请参阅图5,图5是本申请一实施例提供的一种模型训练装置的示意性框图,该模型训练装置可以配置于服务器中,用于执行前述的用户分类模型的训练方法。
如图5所示,该模型训练装置300,包括:数据处理模块301、用户标识模块302、模型训练模块303、用户分类模块304、目标特征模块305、数据筛选模块306和目标模型训练模块307。
数据处理模块301,用于获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据,所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值。
用户标识模块302,用于根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识,所述用户标识包括目标用户和非目标用户。
模型训练模块303,用于根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
用户分类模块304,用于采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值。
其中,用户分类模块304包括子集划分子模块3041、目标子集子模块3042、随机森林模型子模块3043和权重计算子模块3044。
具体地,子集划分子模块3041,用于采用K折交叉切分方法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分割,以得到多个不相交的训练子集。
目标子集子模块3042,用于对多个所述训练子集采用重复抽样的方法进行多次随机有放回采样,以得到多个目标训练子集,并将未被抽到的所述样本用户的特征因子和对应的特征值以及对应的用户标识作为袋外数据。
随机森林模型子模块3043,用于基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型。
权重计算子模块3044,用于基于所述随机森林模型的输出结果,利用所述袋外数据计算所述特征因子的权重值。
目标特征模块305,用于当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子。
数据筛选模块306,用于根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值。
目标模型训练模块307,用于根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
请参阅图6,图6是本申请的实施例还提供一种用户识别装置的示意性框图,该用户识别装置用于执行前述的用户识别方法。其中,该用户识别装置可以配置于服务器或终端中。
其中,服务器可以为独立的服务器,也可以为服务器集群。该终端可以是手机、平板电脑、笔记本电脑、台式电脑、个人数字助理和穿戴式设备等电子设备。
如图6所示,用户识别装置400包括:数据获取模块401、图像处理模块402、特征数据模块403、分类概率模块404和用户确定模块405。
数据获取模块401,用于获取待识别用户的线下图像数据和线上数据。
图像处理模块402,用于对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据。
其中,图像处理模块402包括预处理子模块4021、识别结果子模块4022和数据确定子模块4023。
具体地,预处理子模块4021,用于对所述线下图像数据进行预处理,所述预处理包括二值化、噪声去除和倾斜矫正;识别结果子模块4022,用于对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果;数据确定子模块4023,用于根据所述识别 结果确定线下数据。
特征数据模块403,用于将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗。
分类概率模块404,用于将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率。
用户确定模块405,用于若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
需要说明的是,所属领域的技术人员可以清楚地了解到,为了描述的方便和简洁,上述描述的模型训练装置和各模块的具体工作过程,和上述描述的用户识别装置和各模块的具体工作过程,可以参考前述用户分类模型的训练方法和用户识别方法实施例中的对应过程,在此不再赘述。
上述的用户识别装置可以实现为一种计算机程序的形式,该计算机程序可以在如图7所示的计算机设备上运行。
请参阅图7,图7是本申请实施例提供的一种计算机设备的结构示意性框图。该计算机设备可以是服务器或终端。
参阅图7,该计算机设备包括通过系统总线连接的处理器、存储器和网络接口,其中,存储器可以包括非易失性或易失性存储介质和内存储器。
非易失性或易失性存储介质可存储操作系统和计算机程序。该计算机程序包括程序指令,该程序指令被执行时,可使得处理器执行任意一种用户识别方法。
处理器用于提供计算和控制能力,支撑整个计算机设备的运行。
内存储器为非易失性或易失性存储介质中的计算机程序的运行提供环境,该计算机程序被处理器执行时,可使得处理器执行任意一种用户识别方法。
该网络接口用于进行网络通信,如发送分配的任务等。本领域技术人员可以理解,图7中示出的结构,仅仅是与本申请方案相关的部分结构的框图,并不构成对本申请方案所应用于其上的计算机设备的限定,具体的计算机设备可以包括比图中所示更多或更少的部件,或者组合某些部件,或者具有不同的部件布置。
应当理解的是,处理器可以是中央处理单元(Central Processing Unit,CPU),该处理器还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现场可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。其中,通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
其中,在一个实施例中,所述处理器用于运行存储在存储器中的计算机程序,以实现如下步骤:
获取待识别用户的线下图像数据和线上数据;对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
在一个实施例中,所述处理器在实现所述对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据时,用于实现:
对所述线下图像数据进行预处理,所述预处理包括二值化、噪声去除和倾斜矫正;对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果;根据所述识别 结果确定线下数据。
在一个实施例中,所述处理器还用于实现:
获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据,所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值;根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识,所述用户标识包括目标用户和非目标用户;根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
在一个实施例中,所述处理器在实现所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型之前,还用于实现:
采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值;当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子;根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值;所述处理器在实现所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型时,用于实现:根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型。
在一个实施例中,所述处理器还用于实现:
获取所述预先训练的用户分类模型对于所述样本用户的分类概率;基于多个测试阈值分别计算多个与所述测试阈值对应的混淆矩阵;根据多个所述混淆矩阵计算多个柯尔莫哥洛夫-斯摩洛夫检验值;将多个所述柯尔莫哥洛夫-斯摩洛夫检验值中的最大值对应的测试阈值作为用户分类阈值。
在一个实施例中,所述处理器在实现所述采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值时,用于实现:
采用K折交叉切分方法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分割,以得到多个不相交的训练子集;对多个所述训练子集采用重复抽样的方法进行多次随机有放回采样,以得到多个目标训练子集,并将未被抽到的所述样本用户的特征因子和对应的特征值以及对应的用户标识作为袋外数据;基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型;基于所述随机森林模型的输出结果,利用所述袋外数据计算所述特征因子的权重值。
在一个实施例中,所述处理器在实现所述基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型之前,还用于实现:
对所述目标训练子集中的数据进行抽样平衡化处理,以得到处理后的目标训练子集。
本申请的实施例中还提供一种计算机可读存储介质,所述计算机可读存储介质可以是易失性,也可以是非易失性,所述计算机可读存储介质存储有计算机程序,所述计算机程序中包括程序指令,所述处理器执行所述程序指令,实现如下步骤:
获取待识别用户的线下图像数据和线上数据;
对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;
将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;
将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;
若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
其中,所述计算机可读存储介质可以是前述实施例所述的计算机设备的内部存储单元,例如所述计算机设备的硬盘或内存。所述计算机可读存储介质也可以是所述计算机设备的 外部存储设备,例如所述计算机设备上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。
以上所述,仅为本申请的具体实施方式,但本申请的保护范围并不局限于此,任何熟悉本技术领域的技术人员在本申请揭露的技术范围内,可轻易想到各种等效的修改或替换,这些修改或替换都应涵盖在本申请的保护范围之内。因此,本申请的保护范围应以权利要求的保护范围为准。
Claims (20)
- 一种用户识别方法,其中,包括:获取待识别用户的线下图像数据和线上数据;对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
- 根据权利要求1所述的用户识别方法,其中,所述对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据,包括:对所述线下图像数据进行预处理,所述预处理包括二值化、噪声去除和倾斜矫正;对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果;根据所述识别结果确定线下数据。
- 根据权利要求1所述的用户识别方法,其中,还包括:获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据,所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值;根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识,所述用户标识包括目标用户和非目标用户;根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
- 根据权利要求3所述的用户识别方法,其中,在所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型之前,还包括:采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值;当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子;根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值;所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,包括:根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型。
- 根据权利要求3所述的用户识别方法,其中,还包括:获取所述预先训练的用户分类模型对于所述样本用户的分类概率;基于多个测试阈值分别计算多个与所述测试阈值对应的混淆矩阵;根据多个所述混淆矩阵计算多个柯尔莫哥洛夫-斯摩洛夫检验值;将多个所述柯尔莫哥洛夫-斯摩洛夫检验值中的最大值对应的测试阈值作为用户分类阈值。
- 根据权利要求4所述的用户识别方法,其中,所述采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值,包括:采用K折交叉切分方法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分割,以得到多个不相交的训练子集;对多个所述训练子集采用重复抽样的方法进行多次随机有放回采样,以得到多个目标训练子集,并将未被抽到的所述样本用户的特征因子和对应的特征值以及对应的用户标识作为袋外数据;基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型;基于所述随机森林模型的输出结果,利用所述袋外数据计算所述特征因子的权重值。
- 根据权利要求6所述的用户识别方法,其中,在所述基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型之前,还包括:对所述目标训练子集中的数据进行抽样平衡化处理,以得到处理后的目标训练子集。
- 一种用户识别装置,其中,包括:数据获取模块,用于获取待识别用户的线下图像数据和线上数据;图像处理模块,用于对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;特征数据模块,用于将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;分类概率模块,用于将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;用户确定模块,用于若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
- 一种计算机设备,其中,所述计算机设备包括存储器和处理器;所述存储器用于存储计算机程序;所述处理器,用于执行所述计算机程序并在执行所述计算机程序时实现如下步骤:获取待识别用户的线下图像数据和线上数据;对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
- 根据权利要求9所述的计算机设备,其中,所述对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据,包括:对所述线下图像数据进行预处理,所述预处理包括二值化、噪声去除和倾斜矫正;对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果;根据所述识别结果确定线下数据。
- 根据权利要求9所述的计算机设备,其中,所述处理器在执行所述计算机程序时还实现如下步骤:获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据,所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值;根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识,所述用户标识包括目标用户和非目标用户;根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
- 根据权利要求11所述的计算机设备,其中,在所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型之前,所述处理器在执行所述计算 机程序时还实现如下步骤:采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值;当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子;根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值;所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,包括:根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型。
- 根据权利要求11所述的计算机设备,其中,所述处理器在执行所述计算机程序时还实现如下步骤:获取所述预先训练的用户分类模型对于所述样本用户的分类概率;基于多个测试阈值分别计算多个与所述测试阈值对应的混淆矩阵;根据多个所述混淆矩阵计算多个柯尔莫哥洛夫-斯摩洛夫检验值;将多个所述柯尔莫哥洛夫-斯摩洛夫检验值中的最大值对应的测试阈值作为用户分类阈值。
- 根据权利要求12所述的计算机设备,其中,所述采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值,包括:采用K折交叉切分方法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分割,以得到多个不相交的训练子集;对多个所述训练子集采用重复抽样的方法进行多次随机有放回采样,以得到多个目标训练子集,并将未被抽到的所述样本用户的特征因子和对应的特征值以及对应的用户标识作为袋外数据;基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型;基于所述随机森林模型的输出结果,利用所述袋外数据计算所述特征因子的权重值。
- 根据权利要求14所述的计算机设备,其中,在所述基于所述特征因子和多个所述目标训练子集,采用随机森林算法得到随机森林模型之前,所述处理器在执行所述计算机程序时还实现如下步骤:对所述目标训练子集中的数据进行抽样平衡化处理,以得到处理后的目标训练子集。
- 一种计算机可读存储介质,其中,所述计算机可读存储介质存储有计算机程序,所述计算机程序被处理器执行时使所述处理器实现如下步骤:获取待识别用户的线下图像数据和线上数据;对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据;将所述线下数据和线上数据作为所述待识别用户的用户数据,并对所述用户数据进行数据预处理,以得到所述待识别用户的特征数据,所述预处理包括特征因素量化、异常值处理和数据清洗;将所述待识别用户的特征数据输入预先训练的用户分类模型,以得到所述待识别用户的分类概率;若所述待识别用户的分类概率大于用户分类阈值,则确定所述待识别用户为目标用户。
- 根据权利要求16所述的计算机可读存储介质,其中,所述对所述线下图像数据进行图像处理,以得到所述待识别用户的线下数据,包括:对所述线下图像数据进行预处理,所述预处理包括二值化、噪声去除和倾斜矫正;对预处理后的所述线下图像数据进行版面分析和字符识别,以得到识别结果;根据所述识别结果确定线下数据。
- 根据权利要求16所述的计算机可读存储介质,其中,所述计算机程序被处理器执行时还实现如下步骤:获取样本用户的样本数据并对所述样本数据进行数据预处理,以得到所述样本用户的特征数据,所述特征数据包括所述样本用户的历史购买记录、特征因子和对应的特征值;根据所述样本用户的历史购买记录确定所述样本用户对应的用户标识,所述用户标识包括目标用户和非目标用户;根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,并将训练得到的分类模型作为预先训练的用户分类模型。
- 根据权利要求18所述的计算机可读存储介质,其中,在所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型之前,所述计算机程序被处理器执行时还实现如下步骤:采用随机森林算法对所述样本用户的特征因子和对应的特征值以及对应的用户标识进行分类,以输出特征因子的权重值;当所述特征因子的权重值大于因子分类阈值时,确定所述特征因子为目标特征因子;根据所述目标特征因子对所述样本用户的特征数据进行筛选,以得到所述样本用户与所述目标特征因子对应的特征值;所述根据所述样本用户的用户标识、特征因子和与特征因子对应的特征值训练分类模型,包括:根据所述样本用户的用户标识、目标特征因子和与目标特征因子对应的特征值训练分类模型。
- 根据权利要求18所述的计算机可读存储介质,其中,所述计算机程序被处理器执行时还实现如下步骤:获取所述预先训练的用户分类模型对于所述样本用户的分类概率;基于多个测试阈值分别计算多个与所述测试阈值对应的混淆矩阵;根据多个所述混淆矩阵计算多个柯尔莫哥洛夫-斯摩洛夫检验值;将多个所述柯尔莫哥洛夫-斯摩洛夫检验值中的最大值对应的测试阈值作为用户分类阈值。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010097654.6A CN111459922A (zh) | 2020-02-17 | 2020-02-17 | 用户识别方法、装置、设备及存储介质 |
| CN202010097654.6 | 2020-02-17 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021164232A1 true WO2021164232A1 (zh) | 2021-08-26 |
Family
ID=71678168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2020/112350 Ceased WO2021164232A1 (zh) | 2020-02-17 | 2020-08-31 | 用户识别方法、装置、设备及存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN111459922A (zh) |
| WO (1) | WO2021164232A1 (zh) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113869321A (zh) * | 2021-09-30 | 2021-12-31 | 佛山市南海区广工大数控装备协同创新研究院 | 一种基于bm4d去噪算法和knn算法的苹果分类方法 |
| CN114444576A (zh) * | 2021-12-30 | 2022-05-06 | 北京达佳互联信息技术有限公司 | 一种数据采样方法、装置、电子设备及存储介质 |
| CN114548243A (zh) * | 2022-02-09 | 2022-05-27 | 同盾科技有限公司 | 电信诈骗受害者识别方法、系统及电子设备 |
| CN115145904A (zh) * | 2022-07-06 | 2022-10-04 | 枣庄宏禹数字科技有限公司 | 用于ai云计算训练的大数据清洗方法及大数据采集系统 |
| CN115186749A (zh) * | 2022-07-07 | 2022-10-14 | 北京百度网讯科技有限公司 | 一种数据识别方法、装置、设备以及存储介质 |
| CN115309737A (zh) * | 2022-10-11 | 2022-11-08 | 深圳市明源云客电子商务有限公司 | 访客意向分析方法、系统、终端设备及可读存储介质 |
| CN116389647A (zh) * | 2023-06-02 | 2023-07-04 | 深圳市尚哲医健科技有限责任公司 | 急诊急救一体化平台 |
| CN116756673A (zh) * | 2023-06-21 | 2023-09-15 | 杭州鲁尔物联科技有限公司 | 多传感器异常检测方法、装置、计算机设备及存储介质 |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111459922A (zh) * | 2020-02-17 | 2020-07-28 | 平安科技(深圳)有限公司 | 用户识别方法、装置、设备及存储介质 |
| CN112035775B (zh) * | 2020-09-01 | 2023-11-14 | 中国平安财产保险股份有限公司 | 基于随机森林模型的用户识别方法、装置和计算机设备 |
| CN112085114B (zh) * | 2020-09-14 | 2024-07-12 | 杭州中奥科技有限公司 | 线上线下身份匹配方法、装置、设备及存储介质 |
| CN113988152B (zh) * | 2021-09-23 | 2024-12-27 | 北京达佳互联信息技术有限公司 | 用户类型预测模型训练方法、资源分配方法、介质及装置 |
| CN114169451B (zh) * | 2021-12-10 | 2025-12-30 | 中国建设银行股份有限公司 | 行为数据分类处理方法、装置、设备及存储介质 |
| CN114399087A (zh) * | 2021-12-22 | 2022-04-26 | 中国电信股份有限公司 | 基于Flink流处理引擎的用户数据处理方法和装置 |
| CN116244652A (zh) * | 2023-02-08 | 2023-06-09 | 中国工商银行股份有限公司 | 用户的识别方法及装置、存储介质和电子设备 |
| CN119917926B (zh) * | 2024-12-30 | 2025-09-30 | 成都微筑科技有限公司 | 基于大数据的客户行为分析方法、装置、可读存储介质 |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160217198A1 (en) * | 2015-01-26 | 2016-07-28 | Samsung Electronics Co., Ltd. | User management method and apparatus |
| CN109961296A (zh) * | 2017-12-25 | 2019-07-02 | 腾讯科技(深圳)有限公司 | 商户类型识别方法及装置 |
| CN110188796A (zh) * | 2019-04-25 | 2019-08-30 | 博彦科技股份有限公司 | 用户识别方法、装置、存储介质和处理器 |
| CN110415065A (zh) * | 2018-04-28 | 2019-11-05 | K11集团有限公司 | 用户数据收集系统和信息推送方法 |
| CN110580490A (zh) * | 2018-06-11 | 2019-12-17 | 杭州海康威视数字技术股份有限公司 | 一种确定人员行为概率的方法、装置及设备 |
| CN111459922A (zh) * | 2020-02-17 | 2020-07-28 | 平安科技(深圳)有限公司 | 用户识别方法、装置、设备及存储介质 |
| CN111506798A (zh) * | 2020-03-04 | 2020-08-07 | 平安科技(深圳)有限公司 | 用户筛选方法、装置、设备及存储介质 |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109977977B (zh) * | 2017-12-28 | 2021-04-02 | 中移动信息技术有限公司 | 一种识别潜在用户的方法及对应装置 |
| CN110070123A (zh) * | 2019-04-16 | 2019-07-30 | 北京新意互动数字技术有限公司 | 一种目标用户识别装置及服务器 |
-
2020
- 2020-02-17 CN CN202010097654.6A patent/CN111459922A/zh active Pending
- 2020-08-31 WO PCT/CN2020/112350 patent/WO2021164232A1/zh not_active Ceased
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160217198A1 (en) * | 2015-01-26 | 2016-07-28 | Samsung Electronics Co., Ltd. | User management method and apparatus |
| CN109961296A (zh) * | 2017-12-25 | 2019-07-02 | 腾讯科技(深圳)有限公司 | 商户类型识别方法及装置 |
| CN110415065A (zh) * | 2018-04-28 | 2019-11-05 | K11集团有限公司 | 用户数据收集系统和信息推送方法 |
| CN110580490A (zh) * | 2018-06-11 | 2019-12-17 | 杭州海康威视数字技术股份有限公司 | 一种确定人员行为概率的方法、装置及设备 |
| CN110188796A (zh) * | 2019-04-25 | 2019-08-30 | 博彦科技股份有限公司 | 用户识别方法、装置、存储介质和处理器 |
| CN111459922A (zh) * | 2020-02-17 | 2020-07-28 | 平安科技(深圳)有限公司 | 用户识别方法、装置、设备及存储介质 |
| CN111506798A (zh) * | 2020-03-04 | 2020-08-07 | 平安科技(深圳)有限公司 | 用户筛选方法、装置、设备及存储介质 |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113869321A (zh) * | 2021-09-30 | 2021-12-31 | 佛山市南海区广工大数控装备协同创新研究院 | 一种基于bm4d去噪算法和knn算法的苹果分类方法 |
| CN114444576A (zh) * | 2021-12-30 | 2022-05-06 | 北京达佳互联信息技术有限公司 | 一种数据采样方法、装置、电子设备及存储介质 |
| CN114548243A (zh) * | 2022-02-09 | 2022-05-27 | 同盾科技有限公司 | 电信诈骗受害者识别方法、系统及电子设备 |
| CN115145904A (zh) * | 2022-07-06 | 2022-10-04 | 枣庄宏禹数字科技有限公司 | 用于ai云计算训练的大数据清洗方法及大数据采集系统 |
| CN115186749A (zh) * | 2022-07-07 | 2022-10-14 | 北京百度网讯科技有限公司 | 一种数据识别方法、装置、设备以及存储介质 |
| CN115309737A (zh) * | 2022-10-11 | 2022-11-08 | 深圳市明源云客电子商务有限公司 | 访客意向分析方法、系统、终端设备及可读存储介质 |
| CN116389647A (zh) * | 2023-06-02 | 2023-07-04 | 深圳市尚哲医健科技有限责任公司 | 急诊急救一体化平台 |
| CN116389647B (zh) * | 2023-06-02 | 2023-08-08 | 深圳市尚哲医健科技有限责任公司 | 急诊急救一体化平台 |
| CN116756673A (zh) * | 2023-06-21 | 2023-09-15 | 杭州鲁尔物联科技有限公司 | 多传感器异常检测方法、装置、计算机设备及存储介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111459922A (zh) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021164232A1 (zh) | 用户识别方法、装置、设备及存储介质 | |
| CN107633265B (zh) | 用于优化信用评估模型的数据处理方法及装置 | |
| WO2021072876A1 (zh) | 证件图像分类方法、装置、计算机设备及可读存储介质 | |
| CN112966583A (zh) | 图像处理方法、装置、计算机设备和存储介质 | |
| CN110555372A (zh) | 数据录入方法、装置、设备及存储介质 | |
| CN113448955B (zh) | 数据集质量评估方法、装置、计算机设备及存储介质 | |
| CN111506798A (zh) | 用户筛选方法、装置、设备及存储介质 | |
| WO2018112783A1 (zh) | 图像识别方法及装置 | |
| CN108319672B (zh) | 基于云计算的移动终端不良信息过滤方法及系统 | |
| CN111159481B (zh) | 图数据的边预测方法、装置及终端设备 | |
| CN111552984A (zh) | 显示信息的加密方法、装置、设备及存储介质 | |
| WO2020056968A1 (zh) | 数据降噪方法、装置、计算机设备和存储介质 | |
| WO2019114305A1 (zh) | 对待评估样本进行分类的方法及装置 | |
| WO2022199185A1 (zh) | 用户操作检测方法及程序产品 | |
| CN114387451A (zh) | 异常图像检测模型的训练方法、装置及介质 | |
| CN111507850A (zh) | 核保方法及相关装置、设备 | |
| CN113435464B (zh) | 异常数据检测方法、装置、电子设备及计算机存储介质 | |
| CN110458024B (zh) | 活体检测方法及装置和电子设备 | |
| CN113569879A (zh) | 异常识别模型的训练方法、异常账号识别方法及相关装置 | |
| CN114743048A (zh) | 检测异常秸秆图片的方法和检测装置 | |
| CN111582722B (zh) | 风险识别方法、装置、电子设备及可读存储介质 | |
| CN113867785A (zh) | 一种数据筛选方法及相关装置 | |
| CN117975097B (zh) | 一种浮游生物分类方法、装置、电子设备及存储介质 | |
| KR20210031444A (ko) | 데이터 프로그래밍에 기반한 레이블링 모델 생성 방법 및 장치 | |
| CN112085526A (zh) | 基于用户群的信息匹配方法、装置、计算机设备及存储介质 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20920024 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20920024 Country of ref document: EP Kind code of ref document: A1 |
