User portrait generation method based on characteristic variable scoring, device, vehicle, and storage medium

Abstract

Provided in the present application is a user portrait construction method based on characteristic variable scoring, comprising: collecting and acquiring driving travel and charging travel data of a vehicle, performing preprocessing, and, according to the attributes of the travel data, dividing same into four characteristic data sets of different dimensions; adding labels to the data in the data sets, and using binning operation to discretize characteristic variables; calculating the weight of evidence of each of the bins and a related prediction capability IV, and, according to the IV, picking out characteristic variables having high degrees of contribution to user portrait labels, so as to construct a multi-dimensional characteristic variable database; separately training a logistic regression model to generate models of different dimensions; testing the model of each dimension, and, by means of an ROC curve and scorecard conversion based on AUC values, performing evaluation and verification to obtain an optimal model; and, according to driving behavior habits of a user, classifying driving behavior characteristics of the user by using the optimal model, and, according to the classification result, scoring the driving behavior habits of the user, so as to generate a user portrait. The present application can effectively improve driving safety.

Description

User portrait generation method based on feature variable scoring, equipment, automobile and storage medium Technical Field

[0001] The present application relates to the field of computer information processing technology, and in particular to a method for constructing a user profile based on driving behavior preferences. Background Art

[0002] With the rapid development of internet and information technology, massive amounts of data are continuously generated. This ample data can directly present a comprehensive picture of users. Based on their social attributes, consumption behaviors, and lifestyle habits, we can outline their basic characteristics and framework, establish corresponding symbolic labels, and accurately locate customers.

[0003] By creating a user vehicle usage profile, we can truly reflect user behavior during vehicle use, understand user preferences, and provide personalized services. Analyzing vehicle usage profiles can reveal user characteristics during driving, including speed preferences, driving style, preferred time of day, and driving proficiency. Leveraging these characteristics, business personnel can quickly gain insights into user information, identify significant vehicle usage patterns, and gain insights for business development. User profiles can also be used to refine operations, providing fine-grained recommendations and other services tailored to users with specific characteristics. In terms of data product applications, user profiles are the foundation of many data products, such as recommendation systems, advertising systems, and intelligent push notification platforms.

[0004] Existing methods for constructing user profiles collect static and dynamic user data, namely, user attribute and behavior data, to analyze user groups' interests, preferences, and needs. Alternatively, methods such as clustering are used to model user data and divide user groups and analyze the characteristics of each group. However, these methods are unable to accurately measure each user's preferences and needs.

[0005] Publication No. CN111552926A, titled "Driving Behavior Evaluation Method, System, and Storage Medium Based on the Internet of Vehicles," discloses a driving behavior evaluation method, system, and storage medium based on the Internet of Vehicles. The method comprises the following steps: 1. extracting user behavior data from an Internet of Vehicles database; 2. repeatedly screening feature variables using stepwise regression until the explanatory variable with the highest significance is selected; 3. binning the feature variables and performing Word of Equation (WOE) transformation on them to construct independent variables; training the feature variables using a logistic regression model to obtain a trained model and output the model's variable parameters; 4. outputting scores for four dimensional variables (driving habits, charging habits, safety, and energy consumption) using a scorecard conversion method, with higher scores indicating better driving behavior; and 5. weighting the scores of the four dimensional variables to obtain a final driving behavior score, performing a graded conversion, and outputting a driving behavior rating. The present invention comprehensively rates driving behavior based on driving habits, charging habits, safety, and energy consumption.

[0006] Qualitative or statistical user portraits are used. Qualitative user portraits may ignore some users with unique characteristics, or over-express outliers with unusual opinions. It is also difficult to use these analyses to state the criteria used to classify users.

[0007] Summary of the Invention

[0008] In view of this, the present application addresses the problem that existing technologies cannot accurately measure the preferences and needs of each user and mark preference portraits.

[0009] According to one aspect of the present application, the present invention proposes a scoring-based user profile construction method, including: collecting and obtaining vehicle driving and charging travel data, preprocessing the obtained travel data, and dividing it into four different dimensional feature variable data sets according to the attributes of the travel data; marking positive and negative sample labels for the data in the data set, and discretizing the feature variables using a binning operation; calculating the evidence weight value of each bin, and the predictive ability IV value related to each feature variable, sorting according to the IV value, screening features that contribute greatly to the user portrait label to construct a feature variable database; screening feature variables in the feature variable data set, dividing the training set and the test set, and using the feature variables of each dimension in the training set to train the logistic regression model separately, and generating different dimensional models after the training is completed; using the test set data to test the corresponding dimensional model, and using the ROC curve combined with the AUC value to use the scoring card conversion to score the user portrait score, and evaluate and test to obtain the optimal model; using the optimal model to classify the user's driving behavior characteristics, score the user's driving behavior habits according to the classification results, and generate a user profile.

[0010] Further preferably, the feature variable data set divided into four different dimensions according to the attributes of the travel data includes: constructing a multidimensional feature data table including four dimensions of driving habits, charging habits, safety and energy consumption according to the attributes of the travel data, the multidimensional feature data table including multiple features, and the feature variables and variable descriptions included in the features of each dimension.

[0011] Further preferably, preprocessing the acquired travel data includes: filling missing null values, filtering abnormal data, filling missing values ​​NAN in the data set with 0; and deleting abnormal values ​​and extreme value data using the 3σ criterion.

[0012] Further preferably, the 3σ criterion is to extract data within the interval (μ-σ, μ+σ) as feature data based on the mean μ and standard deviation σ of the collected travel data, and filter out data outside the interval.

[0013] Further preferably, the feature variables in the multi-dimensional feature data table of the preprocessed data set are marked with 1 and 0 in a certain ratio to obtain the positive and negative sample labels of the feature variables in each dimensional data set, and generate a feature sample data set with positive and negative sample labels in multiple dimensions.

[0014] Further preferably, the characteristic variable values ​​in each dimensional feature are sorted, and a predetermined portion of data at the front of the ranking is used as a negative sample, and a predetermined portion of data at the back of the ranking is used as a positive sample, the positive sample is marked as 1, and the negative sample is marked as 0.

[0015] Further preferably, the calculation of the weight of evidence value for each bin and the predictive ability IV value associated with each feature variable specifically includes: calling the formula according to the positive sample ratio p and the negative sample ratio n of each feature in the feature data set of each dimension:

[0016] WOE = ln(p / n)*100%, calculate the weight of evidence WOE; according to the formula: Calculate the information content IV of the i-th characteristic variable i , construct the relationship table between information IV value and prediction ability, where N is the number of characteristic variables, WOE i is the weight of evidence for the i-th feature variable.

[0017] Further preferably, the screening of features that contribute greatly to the user portrait label to construct a feature variable database includes: calling the IV value and prediction ability relationship table, screening feature variables with IV>=0.1 as feature samples in the multi-dimensional feature variable database, and dividing the feature samples into training sets and test sets according to a certain ratio.

[0018] Further optimization is to calculate the AUC value of the ROC curve of the evaluation model based on the prediction results of the feature model of each dimension, and use the score card conversion to perform score conversion based on the AUC value to obtain the ratio of good users to bad users and the weights of different bins corresponding to the feature variables, and obtain the optimal model of each dimension. Combined with the evidence weight of each bin in the feature variable and the probability value of good users and bad users, the coefficients A and B values ​​corresponding to the specific proportion point score are determined, and the scoring result of each bin is calculated.

[0019] Further preferably, the test set samples are input into the corresponding dimension feature models for testing, the number of positive classes and negative classes is determined according to the prediction accuracy, the true positive rate TPR is calculated according to the formula: TPR = TP / (TP+FN), the false positive rate FPR is calculated according to FPR = FP / (TN+FP), the true positive rate TPR is used as the vertical axis, and the false positive rate FPR is used as the horizontal axis, the corresponding coordinate points are obtained under different thresholds, and the coordinate points are connected to obtain the ROC curve, and the user portrait score is scored using the score card conversion according to the area AUC value under the ROC curve, where TP is the true positive class, FN is the false negative class, FP is the false positive class, and TN is the true negative class.

[0020] More preferably, the score of a specific point with a ratio of θ0 is set to P0, and the score of a point with a ratio of 2θ0 is set to P0+PDO, according to the formula: Calculate the coefficients A and B corresponding to the specific point score, according to the formula: score 总 =A+B*ln(odds), calculate the score scale, odds is the ratio of the probability of bad users to the probability of good users.

[0021] More preferably, (B*w n )*WOE n1 +(B*w n )*WOE n2 +Λ+(B*w n )*WOE ni Calculate the user's driving behavior score in a certain feature dimension, obtain the user's driving behavior habit score in each dimension, rate the user's driving behavior habit based on the score and the set threshold range, and generate a user portrait label, where w n Represents the weight of the nth feature variable, WOE ni Represents the evidence weight value of the i-th bin of the n-th feature variable.

[0022] According to a second aspect of the present invention, there is further provided an electronic device, comprising:

[0023] A processor; and a memory for storing a program, wherein the program includes instructions that, when executed by the processor, cause the processor to execute the user profile construction method based on scoring according to any one of the above items.

[0024] According to a third aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is further provided, wherein the computer instructions are used to enable the computer to execute any of the above-described methods for constructing a user profile based on scoring.

[0025] According to a fourth aspect of the present invention, there is also provided an automobile comprising the electronic device according to the present invention and a central control device.

[0026] One or more technical solutions provided in the embodiments of this application can achieve the technical effect of constructing user profiles based on scorecards. Using a scorecard approach can accurately measure user preferences without ignoring their unique characteristics. By controlling the weights of variables, the interpretability of user classification criteria is enhanced, allowing for precise quantitative characterization of user driving behavior, thereby improving driving safety and comfort. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] FIG1 shows a schematic diagram of a method for generating a user portrait based on driving behavior preferences according to an exemplary embodiment of the present application;

[0028] FIG2 is a block diagram showing the principle structure of generating a user portrait based on driving behavior preferences according to the present application. DETAILED DESCRIPTION

[0029] Further details, features and advantages of the present application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings.

[0030] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are for illustrative purposes only and are not intended to limit the scope of protection of the present application.

[0031] It should be understood that the various steps described in the method embodiments of the present application can be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present application is not limited in this respect.

[0032] The term "including" and its variations used in this document are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". The relevant definitions of other terms will be given in the description below. It should be noted that the concepts of "first", "second", etc. mentioned in this application are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0033] It should be noted that the modifications of "one" and "multiple" mentioned in this application are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly indicated in the context, they should be understood as "one or more".

[0034] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0035] The solution of the present application is further described below with reference to the accompanying drawings. As shown in Figure 1, it is a flow chart of a user portrait construction method according to one embodiment of the present invention, including: obtaining driving trip data and charging trip data; data preprocessing; constructing features and generating a data set containing labels; feature discretization, WOE encoding, and IV ranking; training and evaluating models; scorecard conversion, result verification, and label generation.

[0036] Figure 2 shows a block diagram of the principles for generating a user profile based on driving behavior preferences, according to one embodiment of the present invention. The process includes: acquiring data, inspecting data, preprocessing data, adding data labels, discretizing feature variables, encoding word of eye (WOE), sorting IV values, partitioning data, training models, testing models, evaluating models, converting to a scorecard, and finally generating user profiles.

[0037] The vehicle's driving and charging mileage data is collected through the vehicle's central control instrument data collection. The data is checked for null values ​​and abnormal data outside the normal range. If null values ​​exist, they are filled with 0. If abnormal values ​​exist, they are deleted using the 3σ principle. The preprocessed data is labeled with positive and negative samples to generate a labeled data set. The feature variables are discretized using binning operations. The word of essence (WOE) value is calculated and the independent variable construction is completed through WOE transformation. The IV value of each feature variable is calculated and sorted according to the IV value to complete the feature variable screening. The data set is divided into a training set and a test set. The model is trained using the training set data using a logistic regression model in a cross-validation manner. The trained model is tested on the test set. The model accuracy is evaluated and tested using the ROC curve and AUC value to obtain the optimal model. The scoring card conversion method is used to score each user based on the classification results of the user's driving behavior characteristics. The user's driving and charging behavior habits are evaluated based on the score to generate a user profile.

[0038] The vehicle's central control device's data acquisition unit collects all driving and charging trip data. Based on the data's attributes, a multidimensional feature data table is constructed, encompassing four dimensions: driving habits, charging habits, safety, and energy consumption. This multidimensional feature data table includes multiple features. The feature variables and variable descriptions for each dimension are shown in Tables 1-4 below.

[0039] Table 1: Driving habit dimension characteristics table

[0040] Table 2: Charging habit dimension characteristics

[0041] Table 3 Security dimension characteristics

[0042] Table 4 Energy consumption dimension characteristics

[0043] The collected feature data of each dimension are preprocessed. Data preprocessing includes checking whether the data contains null values, abnormal data outside the normal range, etc.

[0044] Collected driving and charging trip data often contains noise, which can cause model prediction errors. First, we examine the collected charging and discharging behavior data for noise, such as null values ​​and abnormal data outside the normal range. We then perform data cleaning by filling in missing null values ​​and filtering out abnormal data.

[0045] The existence of missing values ​​will affect the effectiveness of data analysis and mining. Missing value NAN is an indicator value that indicates that some observations in the data set are missing.

[0046] Common methods for filling missing values ​​include mean and mode filling, linear regression filling based on the data without missing values, which can lead to collinearity if filled well and be worthless if filled poorly, and removing or setting variables. This paper uses 0 to fill the missing values ​​NAN in the dataset.

[0047] The 3σ criterion is used to delete outliers and extreme value data. All features of each dimension conform to the normal distribution. According to the mean μ and standard deviation σ of each feature, the distribution of feature data conforms to the 3σ principle:

[0048] The probability of the value being distributed in (μ-2σ, μ+2σ) is 0.6827, the probability of the value being distributed in (μ-3σ, μ+3σ) is 0.9545, and the probability of the value being distributed in (μ-3σ, μ+3σ) is 0.9973. Therefore, it can be assumed that the values ​​of the feature data are almost concentrated in the interval (μ-3σ, μ+3σ). Most values ​​outside this range are noise data. Therefore, the feature data within the interval (μ-σ, μ+σ) are extracted, and the data outside this interval is filtered out.

[0049] To build a feature dataset, we use preprocessed driving and charging trip data, call a multi-dimensional feature data table, and annotate the feature attributes into four different dimensions: driving habits, charging habits, safety, and energy consumption. We then label the feature dataset with positive and negative sample labels, generating a multi-dimensional feature sample dataset containing both positive and negative sample labels. These positive and negative sample feature data serve as input variables for the training, validation, and prediction models.

[0050] The feature data of each dimension in the dataset is divided into positive and negative samples, and the feature variables in each dimension dataset are labeled with 0 and 1 to generate a dataset containing positive and negative sample labels. In this embodiment, the positive and negative samples can be divided by sorting the feature variable values ​​in each dimension feature, and the data at the front of the ranking is used as negative samples, and the data at the back of the ranking is used as positive samples, with positive samples labeled as 1 and negative samples labeled as 0.

[0051] For example, driving habit features include the standard deviation of vehicle speed and the standard deviation of accelerator pedal depth. For example, the top 20% of single trip data with the largest standard deviations of speed and accelerator pedal depth can be selected as negative samples and marked as 0, while the bottom 20% can be selected as positive samples and marked as 1.

[0052] Safety features include driving duration and peak speed. Driving trip data is sorted by driving duration and peak speed, with the first 20% of the data being negative samples and labeled 0, and the last 20% being positive samples and labeled 1.

[0053] Energy consumption features include the vehicle's energy consumption per 100 kilometers during driving. The vehicle's energy consumption per 100 kilometers during driving is sorted, with the top 20% with the highest energy consumption as negative samples, marked as 0, and the top 20% with the lowest energy consumption as positive samples, marked as 1.

[0054] Charging habit characteristics include: power battery level, power charged in fast charging mode, etc. Charging trip data is grouped and counted by week. The data is then sorted by power battery maximum level (smallest to largest) and power charged in fast charging mode (highest to lowest). The first 20% of the data is used as negative samples and marked as 0, and the last 20% is used as positive samples and marked as 1.

[0055] In the above examples, the top or bottom 20% of a certain indicator are considered positive or negative samples, and other proportions can be set based on experience.

[0056] All features in each dimension feature dataset are divided into positive and negative sample sets as described above.

[0057] Among the features in the multidimensional feature data table, some fully reflect the driver's habits and preferences and contribute significantly to the user profile. Other features are less influenced by the user's behavior and preferences and contribute less to the user profile. Therefore, it is necessary to identify features that significantly impact the results, filter out those with the highest contribution, discretize the features in the database, and calculate the weight of evidence (WOE) coding and IV information value ranking.

[0058] This example discretizes the feature variables through binning and then performs Word of Equation (WOE) encoding conversion to complete the independent variable construction. WOE represents the difference between the proportion of negative samples in the current group and the proportion of positive samples in the current group. A larger WOE indicates a greater difference and a greater likelihood of negative samples in that group. A smaller WOE indicates a smaller difference and a lower likelihood of negative samples in that group.

[0059] According to the proportion of positive samples and negative samples in the feature data set of each dimension, the formula is called:

[0060] WOE = ln(p / n)*100%, where p represents the percentage of positive samples for each feature and n represents the percentage of negative samples for each feature. Calculate the weight of evidence (WOE).

[0061] Calculate the IV value of each characteristic variable, sort according to the IV value, and screen the characteristic variables. IV measures the amount of information of a characteristic variable, according to the formula:

[0062] Calculate the information volume IVi of the i-th feature variable, which represents the predictive ability of the various variables, where p represents the proportion of positive samples of each feature, n represents the proportion of negative samples of each feature, N is the number of feature variables, and WOE i is the weight of evidence for the i-th feature variable.

[0063] This embodiment constructs a table of the relationship between information volume IV value and prediction ability based on a large amount of experimental data analysis and empirical values. As shown in Table 5:

[0064] Table 5: Relationship between IV value and predictive ability:

[0065] The IV value and prediction ability relationship table is called to determine the characteristic variables in the training sample set of each dimension database, that is, the elements in the characteristic database.

[0066] In this embodiment, feature variables with prediction capabilities greater than or equal to "medium" in the prediction table may be selected, that is, feature variables with IV>=0.1 may be selected as feature variables in the multi-dimensional database sample set.

[0067] For example, the IV value of each characteristic variable is calculated based on the weight of evidence, and characteristic variables with a predictive ability greater than or equal to "medium" are selected based on the IV value. The characteristic variables of each dimension of the sample set of this embodiment include:

[0068] Driving habit characteristics: average speed, average accelerator pedal stroke, average motor speed, average angular velocity, average motor torque, average driving energy consumption, idle speed ratio, low speed ratio, medium speed ratio, and high speed ratio.

[0069] Safety features: average speed, proportion of high speed, number of acceleration pedal strokes per 100 kilometers, number of brake pedal strokes per 100 kilometers and peak speed.

[0070] Energy consumption characteristics: air conditioning on / off ratio, medium speed ratio, average air conditioning set temperature, average accelerator pedal stroke count, number of acceleration pedal strokes per 100 kilometers, and number of brake pedal strokes per 100 kilometers.

[0071] Charging habit characteristics: number / percentage of fast charges, fast charge amount, number / percentage of overcharges, number / percentage of deep use, and charging duration.

[0072] The samples in the data set composed of the screened feature variables of each dimension are divided into a training set and a test set according to a certain ratio (such as the ratio of 8:2 in this embodiment). The training set is used to train the model, and the test set is used to test the model.

[0073] The training set samples are input into the logistic regression model for training. The four different dimension feature variables are trained separately to generate the feature model of each dimension.

[0074] The logistic regression model is a binary classification model. This example uses the ROC curve combined with the AUC value method for evaluation. Using cross-validation, the test set is used to test the feature model of each dimension generated after training. The ROC sensitivity curve and the area under the ROC curve (AUC) value are used to evaluate and verify the model accuracy.

[0075] Repeat the above training and testing process until the optimal parameters and model are obtained.

[0076] The ROC curve is a visual tool for comparing two classification models. It shows the trade-off between a given model's true positive rate (TPR) and false positive rate (FPR). The ROC curve is formed by connecting the corresponding coordinate points at different thresholds, with the true positive rate (TPR) on the vertical axis and the false positive rate (FPR) on the horizontal axis.

[0077] The test set samples are input into the corresponding feature model of each dimension for testing. The test result classification is compared with the sample label, and the number of positive and negative classes is determined based on the prediction accuracy. One instance is positive and is predicted to be positive as a true positive class TP, one instance is positive but predicted to be negative as a false negative class FN, one instance is negative but predicted to be positive as a false positive class FP, and one instance is negative and is predicted to be negative as a true negative class TN. According to the above-mentioned number of categories output by the model prediction, according to the formula:

[0078] TPR=TP / (TP+FN), calculate the true positive rate TPR,

[0079] FPR=FP / (TN+FP), calculate the false positive rate FPR.

[0080] AUC represents the area under the ROC curve. According to industry standards, AUC>0.7 is generally required. The larger the AUC value, the better the model.

[0081] The score conversion was performed using the score card conversion according to the AUC value, and the score was controlled within 100 points.

[0082] According to the formula:

[0083] Calculate the ratio of the user probability (1-p) to the bad user probability (p), odds.

[0084] The score scale set by the scorecard can be defined by expressing the score as the logarithm of the ratio.

[0085] According to the formula: score 总=A+B*ln(odds), calculate the total score, and determine the score scale θ. Set the score of a specific point with a ratio of θ to P0, and the score of a point with a ratio of 2θ to P0+PDO. Where P0 and PDO are constants, substituting them into the above formula yields: P0=A+Bln(θ0)P0+PDO=A+Bln(2θ0)

[0086] Calculate the coefficients A and B corresponding to the specific point score: A = P0-Bln(θ0).

[0087] According to the ratio of the probability of good users to the probability of bad users, the feature variable x after WOE binning, the weight wi of the i-th feature variable, i = 1, 2, ... n, n represents the number of feature variables.

[0088] Calling formula:

[0089] score 总 =A+B+(θ Τ x)=A+B*(w0+w1x1+Λ+w n x n )=(A+B*w0)+B*w1x1+Λ+B*w n x n Calculate the total score of the output feature variable.

[0090] By repeatedly training the logistic regression model with training set samples, the AUC value of the ROC curve of the evaluation model is calculated based on the classification results predicted by the model. Based on the AUC value, the score card conversion is used to convert the score to a score, and the probability of good and bad users is obtained. The weights of the feature variables corresponding to different bins during model training can also be obtained. The optimal model for each dimension is obtained.

[0091] Combine the weight of evidence WOE of each bin in the feature variable and the probability value of good users and bad users to determine the coefficients A and B corresponding to the score of the specific proportion point, and calculate the score result of each bin. For example, in this embodiment, the score corresponding to each variable is multiplied by the WOE of each bin in the variable to obtain the score of each bin, that is, according to the formula: (B*w n )*WOE ni Calculate the score of the i-th bin of the n-th feature variable.

[0092] The following table shows the calculation table of the weight scores of feature variables corresponding to different bins.

[0093] [Article 91, 26.11.2024] Table 6: Score calculation table for feature variables corresponding to different bin weights

[0094] Add up all the scores of all feature variables to get the score of user behavior habits in a certain feature dimension. That is, according to the formula:

[0095] (B*w n )*WOE n1 +(B*w n )*WOE n2 +Λ+(B*w n )*WOE ni Calculate the score of user behavior habits in a certain feature dimension. n Represents the weight of the nth feature variable, WOE ni Represents the weight of evidence WOE value of the i-th bin of the n-th feature variable.

[0096] The vehicle's central control system collects characteristic variables of various dimensions filtered from a user's driving and charging trip data. The user's driving behavior score is calculated for each dimension according to the above formula to obtain the user's driving behavior habits in different dimensions. The driving behavior habits are then rated based on the set threshold range.

[0097] After a large number of experiments and sample analysis, this embodiment sets a reasonable interval range for score rating, and the score is set in the interval [60,100]. According to the score interval range, the corresponding characteristic behavior is divided and a characteristic behavior partition table is established, such as:

[0098] Driving behavior: [0,60]: bad driving habits; [60,70]: driving habits need improvement; [70,80]: acceptable driving habits; [80,90]: good driving habits; [90,100]: excellent driving habits.

[0099] Charging habits: [0,60]: Charging habits are seriously harmful to battery health; [60,70]: Charging habits need to be improved; [70,80]: Charging habits are acceptable; [80,90]: Charging habits are good; [90,100]: Charging habits are excellent.

[0100] Generate a user profile label. Compare the obtained score with the score range in the characteristic behavior partition table, evaluate the user's driving and charging behavior habits based on the score, and generate a user profile label.

[0101] The exemplary embodiments of the present application further provide an electronic device, comprising: at least one processor; and a memory communicatively connected to the at least one processor. The memory stores a computer program executable by the at least one processor, wherein the computer program, when executed by the at least one processor, causes the electronic device to perform a method according to an embodiment of the present application.

[0102] An exemplary embodiment of the present application further provides a non-transitory computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to perform a method according to an embodiment of the present application.

[0103] An exemplary embodiment of the present application further provides a computer program product, including a computer program, wherein when the computer program is executed by a processor of a computer, it is used to cause the computer to perform the method according to the embodiment of the present application.

[0104] Electronic device is intended to refer to various forms of digital electronic computing devices, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device can also refer to various forms of mobile devices, such as personal digital assistants, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present application described and / or claimed herein.

[0105] The program code for implementing the methods of the present application can be written in any combination of one or more programming languages. Such program code can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the program code is executed by the processor or controller, the functions / operations specified in the flow charts and / or block diagrams are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0106] In the context of the present application, a machine-readable medium can be a tangible medium that can contain or store a program for use by an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0107] As used herein, the terms "machine-readable medium" and "computer-readable medium" refer to any computer program product, apparatus, and / or device (e.g., a magnetic disk, an optical disk, a memory, a programmable logic device (PLD)) for providing machine instructions and / or data to a programmable processor, including a machine-readable medium that receives machine instructions as a machine-readable signal. The term "machine-readable signal" refers to any signal for providing machine instructions and / or data to a programmable processor.

[0108] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0109] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0110] Computer systems may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The client and server relationship arises through computer programs running on the respective computers and having a client-server relationship to each other.

Claims

1. A method for constructing a user profile based on feature variable scoring, characterized in that: include: Collect and obtain the vehicle's driving and charging trip data, perform preprocessing, and divide it into four feature data sets of different dimensions according to the trip data attributes; The data in the dataset are labeled with positive and negative sample labels, and the feature variables are discretized by binning operation; the weight of evidence for each bin and the IV value of the predictive ability related to each feature variable are calculated, and the feature variables with the largest contribution to the user portrait label are selected to build a multidimensional feature variable database according to the IV value sorting; The feature variables in the feature variable data set are divided into training set samples and test set samples, and the logistic regression models are trained separately using the training set samples. After the training is completed, models of different dimensions are generated; the corresponding dimensional models are tested using the test set samples, and the ROC curve combined with the AUC value is used to convert the score card to obtain the optimal model through evaluation and verification; the optimal model is used to classify the user's driving behavior characteristics according to the user's driving behavior habits, and the user's driving behavior habits are scored according to the classification results to generate a user portrait.

2. The method according to claim 1, characterized in that The preprocessing of the acquired travel data includes: filling the missing null values ​​and filtering the abnormal data. Specifically, the missing values ​​NAN in the data set are filled with 0; and the 3σ criterion is used to delete the abnormal values ​​and extreme value data.

3. The method according to claim 2, characterized in that The 3σ criterion is to extract data in the interval (μ-σ, μ+σ) as feature data based on the mean μ and standard deviation σ of the collected travel data, and filter out data outside the interval.

4. The method according to claim 1, characterized in that: The feature variable data set is divided into four different dimensions according to the attributes of the travel data, including: a multi-dimensional feature data table including four dimensions of driving habits, charging habits, safety and energy consumption is constructed according to the attributes of the travel data, and the multi-dimensional feature data table includes multiple features, and the feature variables and variable descriptions of each dimensional feature.

5. The method according to claim 4, characterized in that The feature variables in the multi-dimensional feature data table are marked with 1 and 0 in a certain ratio, and the positive and negative sample labels of the feature variables in each dimensional data set are obtained, thus generating feature sample data sets with positive and negative sample labels in multiple dimensions.

6. The method according to claim 4 or 5, characterized in that: Sort the feature variable values ​​in each dimensional feature, and use the predetermined part of the data in the front of the ranking as negative samples, and the predetermined part of the data in the back of the ranking as positive samples. The positive samples are marked as 1, and the negative samples are marked as 0.

7. The method according to any one of claims 1 to 5, characterized in that: The calculation of the weight of evidence for each bin and the IV value of the predictive ability related to each feature variable specifically includes: according to the positive sample ratio p and the negative sample ratio n of each feature in the feature data set of each dimension, calling the formula: WOE=ln(p / n)*100%, calculate the weight of evidence WOE; according to the formula: Calculate the The information content of i characteristic variable IV i , construct the relationship table between information volume IV value and prediction ability, where N is the number of characteristic variables, WOE i is the evidence weight of the ith feature variable.

8. The method according to any one of claims 1 to 5, characterized in that: The AUC value of the ROC curve of the evaluation model is calculated based on the prediction results of the feature model of each dimension. The score conversion is performed based on the AUC value to obtain the ratio of good users to bad users and the weights of different bins corresponding to the feature variables. The optimal model of each dimension is obtained. The evidence weight of each bin in the feature variable and the ratio of good users to bad users are combined to determine the coefficients A and B values ​​corresponding to the specific proportion point scores, and the scoring results of each bin are calculated.

9. The method according to any one of claims 1 to 5, characterized in that: Input the test set samples into the corresponding dimensional feature models for testing, determine the number of positive and negative classes according to the prediction accuracy, calculate the true positive rate TPR according to the formula: TPR = TP / (TP+FN), and calculate the false positive rate FPR according to the formula: FPR = FP / (TN+FP). Take the true positive rate TPR as the vertical axis and the false positive rate FPR as the horizontal axis, obtain the corresponding coordinate points under different thresholds, connect the coordinate points, and get the ROC curve. Use the score card conversion to score the user portrait score according to the area AUC value under the ROC curve, where TP is the true positive class, FN is the false negative class, FP is the false positive class, and TN is the true negative class.

10. The method according to claim 7, characterized in that The method of screening features that contribute greatly to user portrait labels to construct a feature variable database includes: calling an IV value and a prediction capability relationship table, screening feature variables with IV>=0.1 as feature samples in a multi-dimensional feature variable database, and dividing the feature samples into a training set and a test set according to a certain ratio.

11. The method according to claim 8, characterized in that The score of a specific point with a ratio of θ0 is set to P0, and the score of a point with a ratio of 2θ0 is set to P0+PDO, according to the formula: Calculate the coefficients A and B corresponding to the score of a specific point according to the formula: score 总 =A+B*ln(odds), calculate the score scale, odds is the ratio of the probability of a bad user to the probability of a good user.

12. The method according to claim 7, characterized in that According to the formula: (B*w n )*WOE n1 +(B*w n )*WOE n2 +Λ+(B*w n )*WOE ni Calculate the user's driving behavior score in a certain feature dimension, obtain the user's driving behavior habit score in each dimension, rate the user's driving behavior habit based on the score and the set threshold range, and generate a user portrait label, where w n Represents the weight of the nth feature variable, WOE ni Represents the evidence weight value of the i-th bin of the n-th feature variable.

13. An electronic device, comprising: processor; And a memory for storing a program, characterized in that, wherein the program includes instructions, which, when executed by the processor, cause the processor to execute the user portrait construction method according to any one of claims 1-12.

14. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: in, The computer instructions are used to enable the computer to execute the user portrait construction method according to any one of claims 1-12.

15. A car, characterized in that: Comprising the electronic device as claimed in claim 13.

Images