Risk quantification regulation and control method and device based on user behaviors, equipment and medium

Abstract

The invention relates to the technical field of intelligent decision making, can be applied to business system platforms of financial science and technology, medical health and the like, and discloses a risk quantification regulation and control method, device and equipment based on user behaviors and a medium, and the method comprises the steps of obtaining multi-source data of the user behaviors, and performing timestamp synchronization on the multi-source data to obtain multi-source synchronous data; performing fusion feature extraction on the multi-source synchronous data to obtain multi-source fusion features; constructing an evaluation model according to the multi-source fusion features, and performing risk score analysis on the multi-source synchronous data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronous data; performing weighted fusion double-model scoring on the risk score according to a preset historical claim settlement data weight to obtain a user behavior risk total score; and dividing risk levels according to the user behavior risk total score, and mapping the risk levels to a control value floating coefficient. Through risk analysis of the multi-source data, the target control value is dynamically adjusted, and the accuracy of the result is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent decision-making, and in particular to a risk quantification regulation method and device based on user behavior, equipment and medium. BACKGROUND

[0002] User behavior-based insurance refers to an innovative insurance mode that takes actual behavior data of users as the core risk assessment basis. User behavior-based insurance covers multiple fields, such as in the medical health field, the HRV of a smart bracelet and voice tone analysis are used to assess the psychological state of a user, such as the voiceprint characteristics of depression, for the pricing of depression insurance; in the field of financial technology, driving systems combined with OBD devices collect data such as sudden acceleration and brake frequency to classify and price users of vehicle insurance, and safe drivers can get a 30% discount on premiums.

[0003] Currently, traditional insurance control values mainly rely on static historical data (such as age, vehicle type, and loss record) and coarse-grained group division, but this approach has problems such as single data dimension, lagging risk assessment, and fuzzy user portrait. SUMMARY

[0004] The present application provides a risk quantification regulation method and device based on user behavior, which dynamically adjusts the target control value through risk analysis of multi-source data, and improves the accuracy of the results.

[0005] In a first aspect, a risk quantification regulation method based on user behavior is provided, comprising: Obtaining multi-source data of user behavior, and synchronizing the time stamps of the multi-source data to obtain multi-source synchronized data; Performing fusion feature extraction on the multi-source synchronized data to obtain multi-source fusion features; Constructing an evaluation model according to the multi-source fusion features, and performing risk score analysis on the multi-source synchronized data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronized data; Weighted fusion double-model scoring of the risk score according to a preset historical claim data weight to obtain a user behavior risk total score; Dividing a risk level according to the user behavior risk total score, and mapping the risk level to a control value floating coefficient; Obtaining a dynamic discount factor, and adjusting the control value floating coefficient according to the dynamic discount factor, and generating a target control value based on the adjusted control value floating coefficient and a preset base value.

[0006] In a second aspect, a risk quantification regulation device based on user behavior is provided, comprising: An acquisition synchronization module is configured to acquire multi-source data of user behavior, and synchronize timestamps of the multi-source data to obtain multi-source synchronized data; An extraction module is configured to extract fused features from the multi-source synchronized data. A construction analysis module is configured to construct an evaluation model according to the multi-source fused features, and perform risk score analysis on the multi-source synchronized data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronized data. A scoring module is configured to perform weighted fused double-model scoring on the risk score according to a preset historical claim data weight to obtain a user behavior risk total score. A division mapping module is configured to divide a risk level according to the user behavior risk total score, and map the risk level to a control value floating coefficient. An acquisition adjustment module is configured to acquire a dynamic discount factor, and adjust the control value floating coefficient according to the dynamic discount factor.

[0007] A generation module is configured to generate a target control value based on the adjusted control value floating coefficient and a preset base value.

[0008] In a third aspect, a computer device is provided, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the above-mentioned risk quantification and regulation method based on user behavior when executing the computer program.

[0009] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program, and the computer program implements the steps of the above-mentioned risk quantification and regulation method based on user behavior when executed by a processor.

[0010] In the scheme implemented by the above risk quantification regulation method and device based on user behavior, multi-source data can comprehensively depict user behavior, time stamp synchronization ensures that different source data is aligned in time sequence, providing a unified time reference for subsequent analysis and avoiding feature deviation caused by time misalignment; by fusing features of different dimension data (such as behavior features, environment features, etc.), deep correlation information that cannot be reflected by a single data source can be mined, the completeness and representativeness of the features are improved, and the evaluation model based on the fused features can quantitatively quantify user behavior risk, abstract behavior is converted into comparable numerical values through risk scoring, providing an objective basis for risk assessment and reducing subjective judgment errors; the historical claim data weight is introduced to improve the reliability of the risk total score, making the score more suitable for the real risk scenario; the continuous risk total score is converted into discrete risk levels, which is convenient for intuitive understanding of the risk degree; mapping to the floating coefficient provides a quantitative basis for subsequent control value calculation, realizing the correlation between risk and control strategy; the dynamic discount factor can adapt to the real-time changing environment, such as market fluctuations and user behavior mutations, making the control value more flexible and time-effective; the target control value generated in combination with the basic value can realize personalized risk control in a standardized framework. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor under the premise of the drawings.

[0012] Figure 1 is an application environment schematic diagram of a risk quantification regulation method based on user behavior in an embodiment of the present application; Figure 2 is a flowchart of a risk quantification regulation method based on user behavior in an embodiment of the present application; Figure 3 is a structure schematic diagram of a risk quantification regulation device in an embodiment of the present application; Figure 4 is a structure schematic diagram of a computer device in an embodiment of the present application; Figure 5 is another structure schematic diagram of a computer device in an embodiment of the present application. DETAILED DESCRIPTION

[0013] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0014] The embodiment of the present invention provides a risk quantification and control method based on user behavior, which can be applied in the following situations: Figure 1 In an application environment, the client communicates with the server through a network. The server can obtain multi-source data of user behavior, and synchronize the timestamps of the multi-source data to obtain multi-source synchronized data; extract fusion features from the multi-source synchronized data to obtain multi-source fusion features; construct an evaluation model based on the multi-source fusion features, and perform risk score analysis on the multi-source synchronized data based on the evaluation model to obtain the risk score corresponding to the multi-source synchronized data; perform weighted fusion dual-model scoring on the risk score based on the preset historical claims data weight to obtain the total risk score of the user behavior; divide the risk level according to the total risk score of the user behavior, and map the risk level to the control value floating coefficient; obtain a dynamic discount factor, and adjust the control value floating coefficient according to the dynamic discount factor, generate a target control value based on the adjusted control value floating coefficient and the preset basic value, and feed the target control value back to the client. The present invention provides a risk quantification and control device based on user behavior. For the target control value business, the target control value is dynamically adjusted through risk analysis of multi-source data to improve the accuracy of the result. Among them, the client can be but is not limited to various personal computers, laptops, smart phones, tablet computers and portable wearable devices. The server side can be implemented by an independent server or a server cluster composed of multiple servers. The present invention will be described in detail below through specific embodiments.

[0015] See also Figure 2 As shown, Figure 2 A flowchart of a method for quantifying and regulating risk based on user behavior provided by an embodiment of the present invention includes the following steps: S1. Acquire multi-source data of user behavior, and synchronize timestamps of the multi-source data to obtain multi-source synchronized data.

[0016] In an embodiment of the present invention, the acquisition refers to collecting user behavior information from multiple different data sources, and the timestamp synchronization refers to adding a unified time tag (timestamp) to data from different sources during the data collection and processing process to ensure the consistency of these data in the time dimension.

[0017] Specifically, various behavior information of users is collected from multiple different channels, such as device records, platform interactions, sensor data and the like, and then the data from different sources is aligned according to a unified time standard, such as a time stamp accurate to seconds, to finally form a set of integrated data indexed by time and containing multi-dimensional behavior records.

[0018] Further, the multi-source data of user behavior includes user health data, driving behavior data and other additional data, real-time physiological indicators such as heart rate, sleep quality, step count and blood oxygen saturation are collected through wearable devices such as smart watches and health monitoring bracelets, and user medical records such as physical examination reports, chronic disease history and medication records are integrated; state data during driving of a vehicle, such as acceleration, sudden braking, overspeed, night driving and fatigue driving (such as continuous driving time) are collected through vehicle-mounted sensors, and high-precision map data is used to assist in analyzing road complexity, and the complexity level is divided into high, medium and low (such as mountainous areas and urban congested roads are considered to have high complexity); user basic attributes (age, occupation, residence), historical insurance claim records and other data.

[0019] In the medical health specific scenario, multi-source information of a patient, such as wearable device motion data, hospital visit records, medication reminders of a medication APP and diet records, is integrated, and through time stamp synchronization, time sequence relationships such as "blood sugar change after exercise in a certain period" and "association between medication time and sleep quality" can be clearly presented, to help doctors more accurately analyze the influence of patient behavior on health and develop personalized diagnosis and treatment plans.

[0020] In the financial technology scenario, user data such as consumption records, transfer behavior, login device location and credit repayment records are integrated, and after time stamp synchronization, time-related risk behaviors such as "large consumption after logging in from a different place" and "abnormal transfer before continuous overdue" can be captured in time, to provide more comprehensive time sequence evidence for anti-fraud monitoring and credit evaluation, and to improve the accuracy of risk identification.

[0021] In the embodiment of the application, the time stamp synchronization of the multi-source data obtains multi-source synchronized data, including: Data cleaning is performed on the multi-source data, and time stamp standardization is performed on the cleaned data to obtain standard data; The standard data is time stamped and aligned, and the alignment result is filled with missing values to obtain filled data; The filled data is redundantly de-duplicated and merged to obtain multi-source synchronized data.

[0022] In the embodiment of the present application, the data cleaning refers to pre-processing the collected multi-source data to remove noise, errors, repetitions and inconsistent information, the timestamp standardization refers to converting the timestamps in different data sources into a unified time format and time zone to ensure the consistency and comparability of time data, the timestamp alignment refers to matching and integrating data records with the same or similar timestamps in different data sources, and the redundancy removal and merging refers to further processing the data after timestamp alignment to remove redundant information, merge similar records, and obtain a refined and consistent data set.

[0023] Specifically, the multi-source data is processed to remove abnormal values, complete missing values, remove noise, errors, repetitions and inconsistent information, etc., and then the cleaned data is standardized in time stamp, unified time format, time precision, and unified time zone. The specific operation is to convert the timestamps of all data into the same format, for example, the original record of the smart bracelet "2025 / 7 / 3 8:30" is converted to "2025-07-03 08:30:00"; the low-precision timestamp (e.g. accurate to minute) is supplemented to second level (e.g. "09:00" is supplemented to "09:00:00"), and the high-precision timestamp (e.g. accurate to millisecond) is truncated to second level (e.g. "09:00:00.123" is retained as "09:00:00"), to ensure consistent time granularity; the timestamps in different time zones are converted to the target time zone (e.g. unified to Beijing time), for example, the UTC time "01:00:00" (8 hours later than Beijing time) is converted to "09:00:00".

[0024] Further, when the standard data is timestamp aligned, the timestamp of the earliest data is taken as the starting point and the timestamp of the latest data is taken as the ending point to generate a continuous second-level time axis, and each record of each data source is matched to the corresponding time point on the time axis, for example, the "500 steps" recorded by the smart bracelet at "08:00:00" corresponds to "08:00:00" on the time axis.

[0025] Further, the aligned results are filled with missing values, specifically, for a data source that does not record data at a certain time point, the mean value of the previous 10 seconds and the next 10 seconds is used for filling, to ensure that all data sources have corresponding data at each time point; finally, the repeated records of the same data source at the same timestamp are deleted, for example, the smart bracelet repeatedly uploads two "500 steps" data at "08:00:00" due to failure, and one is retained; the records of different data sources at the same timestamp are combined into one data to form a structured entry of "timestamp + multi-source features".

[0026] In the embodiment of the present application, the abnormal and error data are removed, the time format is unified, the data accuracy and consistency are ensured, and a reliable foundation is laid for subsequent processing; the multi-source data is matched in the time dimension, the missing information is supplemented, and the analysis deviation caused by time misplacement and data gap is avoided; the data amount is simplified, the repeated information is eliminated, the effective content of multi-source data is integrated, and the data utilization rate and subsequent processing efficiency are improved.

[0027] In the embodiment of the present application, the time stamp synchronization breaks the time barrier of different source data, ensures the alignment of data in the time dimension, avoids the analysis deviation caused by time misplacement, provides accurate and consistent basic data for subsequent fusion analysis, and ensures the timeliness and relevance of risk assessment.

[0028] S2, performing fusion feature extraction on the multi-source synchronous data to obtain multi-source fusion features.

[0029] In the embodiment of the present application, the fusion feature extraction refers to the process of organically integrating and refining different source and type data information through specific methods and technologies from the multi-source synchronous data, to generate more representative, more information-rich and comprehensive user behavior or system state feature.

[0030] Specifically, the key information with relevance and representativeness is extracted from different source and type behavior data based on the multi-source data synchronized by the time stamp, and the information is integrated into a feature set that can comprehensively reflect the essence of user behavior.

[0031] Further, for example, health features such as BMI, chronic disease label, family medical history, heart rate, sleep quality score, and exercise frequency are extracted from user behavior health data; and driving features such as risk operation frequency (such as the number of emergency braking per 100 kilometers), average vehicle speed standard deviation, night driving proportion, and road risk coefficient (combined with map data to mark high-risk road sections) are extracted from user driving behavior data.

[0032] In the medical health scene, from the multi-source data such as synchronized wearable device exercise data, hospital examination data, medication records, and diet logs, fusion features such as “matching degree of exercise intensity and medication time” and “correlation features of blood glucose fluctuation and diet structure” are extracted, which help doctors more comprehensively judge the health status of patients and predict disease risks, and provide a basis for personalized health management solutions.

[0033] In the financial scene, from the multi-source data such as synchronized user consumption records, transfer data, login information, and credit records, fusion features such as “consumption features of abnormal login period” and “matching features of repayment ability and consumption habits” are extracted, which improve the accuracy of user credit rating and fraud behavior identification, and provide support for risk control and financial service optimization.

[0034] In the embodiment of the present application, the fusion feature extraction on the multi-source synchronous data obtains multi-source fusion features, which comprises: The multi-source synchronous data is subjected to data type layering to obtain layered data; The layered data is subjected to single-source feature preliminary extraction to obtain layered single-source features; The layered single-source features are subjected to cross-source feature association fusion to obtain associated fusion features; The associated fusion features are subjected to feature dimension reduction screening to obtain multi-source fusion features.

[0035] In the embodiment of the present application, the data type layering refers to dividing the multi-source synchronous data into multiple levels or categories according to different dimensions of data internal attributes, generation methods, forms of expression or business logics, the single-source feature preliminary extraction refers to using corresponding methods and technologies to mine initial features capable of reflecting data internal rules and characteristics from each layer of data after data type layering, the cross-source feature association fusion refers to associating and integrating single-source features extracted from different data sources (i.e. different layers) to mine their internal relations and interactions, thereby generating more comprehensive and representative fusion features, and the feature dimension reduction screening refers to selecting features most representative and having greatest impact on business targets from a large number of features obtained after cross-source feature association fusion, while reducing the dimension of features and reducing the risk of overfitting and computational complexity.

[0036] Specifically, the multi-source synchronous data is classified into structured data, unstructured data and semi-structured data according to data types, and basic features are extracted for each type of data source, for example, for structured data: statistical features such as daily average movement steps, weekly fluctuation amplitude of consumption amount, frequency of driving sudden braking, trend features such as growth / drop slope of steps for 7 consecutive days; for unstructured data: keywords such as “high blood pressure” and “insomnia” in medical records are extracted through natural language processing (NLP), and sentiment inclinations such as positive / negative emotion proportion in social dynamics; for semi-structured data: time sequence features such as time consumption from “shipping” to “signature” of logistics orders, and daily start / close times of equipment.

[0037] Further, the association of multi-source features is established based on timestamps, and through time sequence association or scene association, for example, the health state has an impact on driving behavior features are extracted from the fact that the smart bracelet records poor night sleep quality (structured) → the next day car insurance driving data shows that the proportion of low-speed driving increases (structured); single-source features are combined into multi-dimensional feature vectors in time sequence, for example, “step number + consumption amount + driving sudden braking times” are spliced into [8000 steps, 200 yuan, 3 times], and finally feature interaction is performed to generate new features through cross calculation.

[0038] Further, when performing feature dimension reduction screening on the associated fusion features, redundant features are removed first, then dimension reduction processing is performed, more representative features are reserved through correlation analysis (such as the correlation coefficient of two features > 0.9), for example, “daily average steps” and “weekly average steps” are highly correlated, and the former is reserved; for high-dimensional features, such as a vector containing 100 atomic features, the dimension is compressed through principal component analysis (PCA) or the like, and the core information is reserved, for example, 100 dimensions are compressed to 20 dimensions, and the calculation amount of the subsequent model is reduced.

[0039] In the embodiment of the application, the data is classified according to the data type, the data structure is clearer, targeted processing is facilitated, and the subsequent feature extraction efficiency is improved; the basic features are mined from the single-source data of each layer, the core information of the single-source data is reserved, and original feature materials are provided for cross-source fusion; the correlation of features from different sources is mined, multi-dimensional information is integrated, and the completeness and risk recognition of the features are enhanced.

[0040] In the embodiment of the application, more representative key features are extracted from scattered multi-source data, the data redundancy is reduced, the complementary value of multi-dimensional information is integrated, the features are more in line with the actual risk law, and the evaluation accuracy of the subsequent model is improved.

[0041] S3, constructing an evaluation model according to the multi-source fusion features, and performing risk score analysis on the multi-source synchronous data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronous data.

[0042] In the embodiment of the application, the construction refers to establishing a mathematical structure or calculation framework capable of mapping input features to risk evaluation results based on multi-source fusion features, using appropriate algorithms and mathematical methods, and the risk score analysis refers to performing risk evaluation on multi-source synchronous data using the constructed evaluation model, and calculating a risk score for each data sample, which reflects the risk degree corresponding to the data sample.

[0043] Specifically, based on the extracted multi-source fusion features (i.e., key features integrating multi-channel and multi-dimensional information), a model for evaluating risk is built, and the multi-source data synchronized by the time stamp is input into the model, the hidden risk in the data is quantitatively analyzed through the model, and finally a score reflecting the risk degree corresponding to the data is output.

[0044] For example, according to the user health data and the driving behavior data, a health state evaluation model and a driving behavior evaluation model are constructed respectively, when the user health risk is evaluated, the scale of the user data involved is large, the LightGBM model is selected to construct the health state evaluation model, the input of which is the user health data feature, and the output is the health risk score; the driving behavior feature contains an important feature of night driving, which is a time sequence feature, so the LSTM model is used to train the risk evaluation model; the input is the driving behavior feature, and the output is the driving risk score (such as 0-100 points, the higher the score, the lower the risk).

[0045] In the specific scene of medical health, a health risk evaluation model is constructed by using the multi-source fusion features of the patient, such as the exercise and blood sugar correlation feature, the medication compliance and symptom change feature, etc., the synchronized patient data such as continuous physiological indicators, diagnosis and treatment records, etc. are analyzed to obtain the corresponding health risk score, which helps medical staff to quickly judge the potential disease risk of the patient and provides a basis for early intervention.

[0046] In the financial scene, a financial risk evaluation model is constructed based on the multi-source fusion features of the user, such as the consumption and repayment ability matching feature, the abnormal behavior correlation feature, etc., the synchronized user data such as transaction records, login information, etc. are analyzed to obtain the credit risk or fraud risk score, which assists the financial institutions to accurately evaluate the user credit level, identify fraudulent transactions, and improve the risk control efficiency and decision accuracy.

[0047] In the embodiment of the application, the risk score analysis of the multi-source synchronous data according to the evaluation model obtains the risk score corresponding to the multi-source synchronous data, comprising: Each record in the multi-source synchronous data is matched with the feature dimension in the evaluation model to obtain the risk dimension of each record; The risk score of the risk dimension is calculated according to the evaluation model, and the multi-dimensional risk score integration of the risk score is performed through the preset dimension weight to obtain the comprehensive risk score of each record; The trend analysis adjustment of the comprehensive risk score is performed according to the time sequence rule in the evaluation model to obtain the risk score corresponding to the multi-source synchronous data.

[0048] In the embodiment of the present application, the matching refers to finding the corresponding data field or information in each data record according to the characteristics defined by the evaluation model, ensuring that each record can provide complete characteristic data according to the model requirements, the calculation refers to using the pre-set algorithm and rules in the evaluation model to calculate the risk score corresponding to each dimension for each record matched with the risk dimension characteristic data, the multi-dimensional risk score integration refers to assigning a corresponding weight to each dimension according to the influence degree of different risk dimensions on the overall risk, and then weighting and summing the risk scores of each dimension according to the weight, thereby obtaining the comprehensive risk score of each record, and the trend analysis adjustment refers to considering the change trend of the data record in the time sequence, combining the time sequence rules set in the evaluation model, and correcting and adjusting the comprehensive risk score.

[0049] Specifically, each record (corresponding to the user behavior at a certain moment) in the multi-source synchronous data is matched with the characteristic dimension in the evaluation model, and the risk evaluation index corresponding to each data is determined, for example, the driving behavior data (such as the number of emergency braking, the length of overspeed) is matched with the “driving risk dimension” in the model, the health behavior data (such as medication adherence, exercise frequency) is matched with the “health risk dimension” in the model, and the consumption behavior data (such as abnormal consumption frequency, number of delayed repayments) is matched with the “credit risk dimension” in the model, and the implicit behavior not directly matched is mapped to the corresponding risk dimension through the association rules pre-set in the model.

[0050] Further, for each risk dimension after matching, the single-dimensional risk score is calculated according to the judgment standard (such as threshold, grade division rule) in the evaluation model, for example, the driving risk dimension: the model stipulates that “the number of emergency braking > 5 times / hour” corresponds to a risk score of 80 points (a full score of 100 points, the higher the score, the higher the risk), and the number of emergency braking in a certain data is 6 times / hour, so the dimension gets 80 points.

[0051] Further, according to the dimension weight set in the evaluation model (such as driving risk proportion 40%, health risk proportion 30%, credit risk proportion 30%), the single-dimensional risk score is weighted and integrated to obtain the comprehensive risk score of each multi-source synchronous data; the trend analysis is performed on the risk score of the continuous time stamp, and the score is adjusted in combination with the time sequence rules in the evaluation model, for example, if the driving risk scores of 3 consecutive data of a certain user are all rising, the model determines that there is a “risk deterioration trend”, and the latest score is increased by 5 points; finally, according to the abnormal detection rule in the evaluation model, the score obviously deviating from the normal range is identified, and compared with the historical data of the same period, if the difference is too large, the historical mean value is adjusted.

[0052] In the embodiment of the present application, each record corresponds to the model feature dimension, the specific direction of the explicit evaluation is ensured, the key dimension is focused on in the risk analysis, and the core risk point is avoided to be missed; the multi-dimensional score is integrated through the dimension weight, the influence of different risk dimensions is balanced, the more comprehensive single record risk quantization result is obtained, and the single dimension deviation is reduced; the risk change trend is analyzed in combination with the time sequence rule, the scoring not only reflects the current state, but also embodies the dynamic evolution of the risk, and the timeliness and foresight of the scoring are improved.

[0053] In the embodiment of the present application, the abstract fusion feature is converted into a quantifiable risk score through the model, the preliminary quantization of the user behavior risk is realized, the subjectivity of the artificial evaluation is avoided, and the model based on the fusion feature can more comprehensively capture the potential risk, thereby providing standardized basic data for subsequent risk total score calculation.

[0054] S4, the risk score is weighted and fused according to the preset historical claim data weight to obtain a user behavior risk total score.

[0055] In the embodiment of the present application, the process of combining the risk score results of the two different models and comprehensively calculating according to the preset rule.

[0056] Specifically, the influence proportion (i.e., the preset weight) of each risk factor in the historical claim data is combined, the risk scores output by the two different evaluation models are calculated according to the weight, and finally a total score reflecting the overall behavior risk level of the user is obtained.

[0057] In the embodiment of the present application, the risk score is weighted and fused according to the preset historical claim data weight to obtain a user behavior risk total score, including: According to the preset historical claim data weight, a weight is configured for each model in the double model to obtain the weight of each model; The risk score is weighted and calculated according to the weight to obtain a comprehensive weighted score; The comprehensive weighted score is adjusted according to the preset time dimension weight to obtain a user behavior risk total score.

[0058] In the embodiment of the present application, the weight configuration refers to assigning a relatively important value to each model in the double model according to the preset rule and actual demand, and the value is the weight; the weighted calculation refers to the process of multiplying the risk scores of the two models according to the weight configured for the double model and then adding them to obtain a comprehensive weighted score.

[0059] Specifically, the key indicators related to risks in historical claim data (such as disease claim rate in the medical field, default claim amount proportion in the financial field, etc.) are combed, the contribution of the two models to risk prediction in historical data is analyzed, and then the two models are assigned corresponding weights based on these contribution degrees, and the influence proportion of each model in the final score is determined; for example, according to the analysis of historical car insurance claim data, it is found that driving behavior is the main cause of car insurance claim cases, accounting for about 60%.

[0060] Further, the respective risk scores output by the two models, for example, 80 points for the user health assessment model and 60 points for the driving behavior assessment model, are mathematically calculated according to the weights determined in the previous step (such as 60% and 40%) (80 x 60% + 60 x 40% = 72 points), to obtain a comprehensive score integrating the information of the two models, that is, a comprehensive weighted score.

[0061] Further, combined with the preset time dimension weight (such as higher weight for recent data and lower weight for long-term data, for example, 70% weight for the last 3 months and 30% weight for 3-6 months), the comprehensive weighted score in different time intervals is adjusted again (such as recent comprehensive score 80 points x 70% + long-term comprehensive score 70 points x 30% = 77 points), and finally the user behavior risk total score is obtained.

[0062] In the embodiments of the present application, the model weight is calibrated through historical claim data, which can make the weight configuration more consistent with the risk occurrence rules in actual business, and avoid the deviation of subjective assignment; the scores of the two models are fused through weighted calculation, which can not only retain the advantages of the two models, but also balance their respective limitations through weight adjustment, avoid the one-sidedness of a single model, and make the score more comprehensively reflect the risk characteristics.

[0063] In the embodiments of the present application, reasonable weights are assigned to the two models in combination with the actual influence of historical claim data, so that the model advantages are matched with the actual risk rules of the business, and the reliability of the score is improved; the scores of the two models are fused through weights, the output deviation of different models is balanced, the advantages of multiple models are integrated, and a more comprehensive risk quantification result is obtained; the time dimension weight is introduced, the total score is more consistent with the timeliness change of risks, and the misjudgment of historical data on current risks is avoided.

[0064] S5, dividing a risk level according to the user behavior risk total score, and mapping the risk level to a control value floating coefficient.

[0065] In the embodiments of the present application, the division refers to dividing the user into different risk categories according to the user behavior risk total score according to pre-set rules and standards, and the mapping refers to establishing a corresponding relationship between the divided risk level and the control value floating coefficient, that is, each risk level corresponds to a specific control value floating coefficient.

[0066] Specifically, according to the high or low of the user behavior risk total score, the user behavior risk total score is divided into different risk levels (such as low risk, medium risk, high risk) according to preset standards, and then the divided risk levels are converted into specific coefficient values according to the corresponding rules between the risk levels and the control value floating coefficient, so as to reflect the adjustment range of different risk levels on the control value (such as premium, quota, etc.).

[0067] In the embodiment of the application, the risk level is divided according to the user behavior risk total score, and the risk level is mapped to the control value floating coefficient, including: obtaining a risk level division standard; mapping the user behavior risk total score to the risk level division standard to obtain a risk level; obtaining the risk level and actual loss data corresponding to the risk level of the historical user, and performing loss rate statistics according to the risk level and the actual loss data to obtain a loss rate corresponding to the risk level; determining the corresponding relationship between the risk level and the preset floating coefficient according to the loss rate; mapping the risk level according to the corresponding relationship to obtain a control value floating coefficient.

[0068] In the embodiment of the application, the obtaining refers to extracting criteria for dividing the user behavior risk total score into different risk levels from the given data source, rule base or expert experience, the mapping refers to corresponding the specific user behavior risk total score to the corresponding risk level according to the obtained risk level division standard, the second obtaining refers to collecting the risk level information of the user in the past period of time and the actual loss situation data of the user under each risk level from the database, business system or other related data sources of the enterprise, and the loss rate statistics refers to calculating the loss rate corresponding to each risk level according to the collected risk level and actual loss data.

[0069] Specifically, according to the preset risk level division standard, such as 0-30 points for “low risk”, 31-60 points for “medium risk”, and 61-100 points for “high risk”, the user behavior risk total score is classified into the corresponding level; for example, if the user behavior risk total score is 25 points, it matches “low risk”; if the total score is 50 points, it matches “medium risk”; and if the total score is 75 points, it matches “high risk”.

[0070] Further, a large number of historical user risk levels (such as low / medium / high risk) and corresponding actual loss data (such as car insurance claim amount, frequency) are collected, the average loss rate of different risk levels is counted, and the loss rate difference is converted into a coefficient adjustment rule according to the business profit target; based on the above analysis, a corresponding relationship is preliminarily set, such as low risk 0.8, medium risk 1.0, and high risk 1.5, and the historical data is tried, if the high-risk user is calculated according to the coefficient of 1.5, the claim rate meets the expectation, then it is retained; otherwise, adjust the coefficient (such as increase to 1.6); verify the rationality of the coefficient through new user data (such as whether the actual claim of the high-risk user matches the coefficient of 1.5), and continuously fine-tune combined with feedback (such as adding a "medium-high risk" level corresponding to a coefficient of 1.2), and finally form a stable corresponding relationship, thereby establishing a corresponding table.

[0071] Further, the risk level-floating coefficient corresponding table is called, such as low risk corresponding to floating coefficient 0.8, medium risk corresponding to 1.0, and high risk corresponding to 1.5, and the output risk level is converted into a specific control value floating coefficient; for example, "low risk" is mapped to 0.8, "medium risk" is mapped to 1.0, and "high risk" is mapped to 1.5.

[0072] In the embodiment of the application, the continuous risk total score is converted into an intuitive risk level, which is convenient for quickly understanding the risk degree; at the same time, the floating coefficient is mapped to establish the association between risk and actual control indicators (such as premium, limit), which provides an operable adjustment basis for the generation of subsequent target control values, and realizes the connection between risk assessment and business decision.

[0073] S6, acquiring a dynamic discount factor, and adjusting the control value floating coefficient according to the dynamic discount factor, generating a target control value based on the adjusted control value floating coefficient and a preset base value.

[0074] In the embodiment of the application, the acquisition refers to collecting, extracting or generating a value that changes with time, business status or other related factors from a specific data source, a calculation model or an external system, which is used to dynamically adjust the control value floating coefficient, and the adjustment refers to modifying and optimizing the original control value floating coefficient according to the acquired dynamic discount factor according to a preset rule or algorithm, so that it can better adapt to the current dynamic changes.

[0075] Specifically, the dynamic discount factor (such as a reward discount or a punitive premium ratio) is determined according to the real-time changes of user behavior (such as risk improvement or deterioration), and then the control value floating coefficient (adjustment ratio based on risk level) is adjusted again using the factor, and finally the adjusted floating coefficient is combined with the preset base value (such as the benchmark premium, the basic limit, etc.) to calculate the final target control value (such as the actual premium, the actual credit limit, etc.) for execution.

[0076] In the embodiment of the present application, the obtaining of the dynamic discount factor and the adjusting of the control value floating coefficient according to the dynamic discount factor comprises: obtaining behavior data of a user in a preset time period, performing risk score change analysis on the behavior data to obtain a dynamic discount factor; determining a dynamic reward and punishment mechanism according to the dynamic discount factor; adjusting the control value floating coefficient based on the dynamic reward and punishment mechanism.

[0077] In the embodiment of the present application, the risk score change analysis refers to using a specific risk assessment model and algorithm to calculate corresponding risk scores according to the behavior data of the user at different time points, and comparing the risk scores in different time periods to analyze the change trend, and the determining refers to formulating a set of corresponding reward and punishment rules according to the change of the risk score of the user.

[0078] Specifically, the behavior data of the user in a period of time (such as 3 months) is collected, the change trend of the risk score is compared, if the risk score is improved for 3 consecutive months, such as from high risk to medium risk, or the medium risk is continuously optimized, it is determined as “behavior improvement”, the incentive mechanism is triggered, and the dynamic discount factor is set as “-5%” (additional 5% reduction based on the original floating coefficient); if the high-risk behavior is not improved (such as maintaining high risk for 2 consecutive months, or new high-risk behavior occurs), it is determined as “behavior deterioration”, the punishment mechanism is triggered, and the dynamic discount factor is set as “+5%” (additional 5% increase based on the original floating coefficient); if the behavior has no obvious change (such as stable in medium risk), the dynamic discount factor is set as “0” (no adjustment).

[0079] Further, the determined “control value floating coefficient” is called, such as 10% for low risk, 0 for medium risk, and -10% for high risk; the control value floating coefficient is adjusted according to the dynamic discount factor, if the dynamic discount factor is -5% (incentive mechanism), the original coefficient is additionally reduced by 5% (such as low risk original coefficient 10% to 5%); if the dynamic discount factor is +5% (punishment mechanism), the original coefficient is additionally increased by 5%, such as high risk original coefficient -10% to -15%, but it is ensured that it does not exceed the total interval of ±10%, and the threshold is limited to -10%; if the dynamic discount factor is 0, the control value floating coefficient remains unchanged, and finally the adjusted control value floating coefficient is output.

[0080] Finally, according to the formula the final result is the target control value, wherein, is the adjusted control value floating coefficient, and the basic premium is a preset basic value.

[0081] In the embodiment of the present invention, the real-time fluctuations of user behavior risks are captured, the discount factor is adapted to the recent behavior trend, and the lag of static evaluation is avoided; the discount factor is converted into clear reward and punishment rules, so that the risk changes are associated with the actual incentives / constraints, and the operability of the mechanism is enhanced; through the dynamic correction coefficient of the reward and punishment mechanism, the control value can more accurately reflect the user's current risk, thereby improving the flexibility and pertinence of risk management.

[0082] It can be seen that in the above scheme, for the target control value business, multi-source data of user behavior is obtained, and the multi-source data is timestamped to obtain multi-source synchronized data; fusion feature extraction is performed on the multi-source synchronized data to obtain multi-source fusion features; an evaluation model is constructed based on the multi-source fusion features, and a risk score analysis is performed on the multi-source synchronized data based on the evaluation model to obtain the risk score corresponding to the multi-source synchronized data; the risk score is weighted and fused with a dual model score based on the preset historical claims data weight to obtain the total risk score of user behavior; the risk level is divided according to the total risk score of user behavior, and the risk level is mapped to the control value floating coefficient; a dynamic discount factor is obtained, and the control value floating coefficient is adjusted according to the dynamic discount factor, and the target control value is generated based on the adjusted control value floating coefficient and the preset basic value. Through risk analysis of multi-source data, the target control value is dynamically adjusted to improve the accuracy of the result.

[0083] It should be understood that the order of execution of the steps in the above embodiments does not necessarily mean the order of execution. The order of execution of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0084] In one embodiment, a risk quantification and control device based on user behavior is provided, and the risk quantification and control device based on user behavior corresponds one to one with the risk quantification and control method based on user behavior in the above embodiment. Figure 3 As shown, the risk quantification and control device based on user behavior includes an acquisition and synchronization module 101, an extraction module 102, a construction and analysis module 103, a scoring module 104, a division and mapping module 105, an acquisition and adjustment module 106, and a generation module 107. The functional modules are described in detail as follows: An acquisition and synchronization module 101 is used to acquire multi-source data of user behavior and synchronize timestamps of the multi-source data to obtain multi-source synchronized data; An extraction module 102 is configured to extract fusion features from the multi-source synchronized data to obtain multi-source fusion features; Constructing an analysis module 103, configured to construct an evaluation model based on the multi-source fusion features, and perform risk score analysis on the multi-source synchronized data based on the evaluation model to obtain a risk score corresponding to the multi-source synchronized data; The scoring module 104 is configured to weight and fuse the double model scores according to preset historical claim data weights to obtain a user behavior risk total score. The division and mapping module 105 is configured to divide a risk level according to the user behavior risk total score and map the risk level to a control value floating coefficient. The acquisition and adjustment module 106 is configured to acquire a dynamic discount factor and adjust the control value floating coefficient according to the dynamic discount factor.

[0085] The generation module 107 is configured to generate a target control value based on the adjusted control value floating coefficient and a preset base value.

[0086] In an embodiment, the acquisition synchronization module 101 is configured to, when synchronizing time stamps of the multi-source data to obtain multi-source synchronized data: perform data cleaning on the multi-source data and perform time stamp standardization on the cleaned data to obtain standard data; perform time stamp alignment on the standard data and perform missing value filling on the alignment result to obtain filled data; perform redundancy de-duplication and merging on the filled data to obtain the multi-source synchronized data.

[0087] In an embodiment, the extraction module 102 is configured to, when performing fusion feature extraction on the multi-source synchronized data to obtain multi-source fusion features: perform data type layering on the multi-source synchronized data to obtain layered data; perform single-source feature preliminary extraction on the layered data to obtain layered single-source features; perform cross-source feature association fusion on the layered single-source features to obtain associated fusion features; perform feature dimension reduction screening on the associated fusion features to obtain the multi-source fusion features.

[0088] In an embodiment, the construction and analysis module 103 is configured to, when performing risk score analysis on the multi-source synchronized data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronized data: match each record in the multi-source synchronized data with a feature dimension in the evaluation model to obtain a risk dimension of each record; calculate a risk score of the risk dimension according to the evaluation model, and integrate the risk scores in multiple dimensions by a preset dimension weight to obtain a comprehensive risk score of each record; perform trend analysis and adjustment on the comprehensive risk score according to a time sequence rule in the evaluation model to obtain the risk score corresponding to the multi-source synchronized data.

[0089] In an embodiment, the scoring module 104 is configured to, when weighting and fusing the risk scores according to preset historical claim data weights to obtain a user behavior risk total score, perform the following operations: configuring weights for each model in the dual model according to preset historical claim data weights to obtain the weights of each model; weighting and calculating the risk scores according to the weights to obtain a comprehensive weighted score; adjusting the comprehensive weighted score according to a preset time dimension weight to obtain the user behavior risk total score.

[0090] In an embodiment, the division and mapping module 105 is configured to, when dividing risk levels according to the user behavior risk total score and mapping the risk levels to a control value floating coefficient, perform the following operations: obtaining a risk level division standard; mapping the user behavior risk total score to the risk level division standard to obtain a risk level; obtaining risk levels of historical users and actual loss data corresponding to the risk levels, performing loss rate statistics according to the risk levels and the actual loss data to obtain a loss rate corresponding to the risk level; determining a corresponding relationship between the risk level and a preset floating coefficient according to the loss rate; mapping the risk level according to the corresponding relationship to obtain a control value floating coefficient.

[0091] In an embodiment, the obtaining and adjusting module 106 is configured to, when obtaining a dynamic discount factor and adjusting the control value floating coefficient according to the dynamic discount factor, perform the following operations: obtaining behavior data of a user in a preset time period, performing risk score change analysis on the behavior data to obtain a dynamic discount factor; determining a dynamic reward and punishment mechanism according to the dynamic discount factor; adjusting the control value floating coefficient based on the dynamic reward and punishment mechanism.

[0092] The application provides a risk quantification regulation device based on user behavior, acquires multi-source data of user behavior for target control value business, and synchronizes time stamps of the multi-source data to obtain multi-source synchronized data; multi-source fusion feature extraction is performed on the multi-source synchronized data to obtain multi-source fusion features; an evaluation model is constructed according to the multi-source fusion features, risk score analysis is performed on the multi-source synchronized data according to the evaluation model, and a risk score corresponding to the multi-source synchronized data is obtained; the risk score is weighted and fused according to a preset historical claim data weight to obtain a user behavior risk total score; a risk level is divided according to the user behavior risk total score, and the risk level is mapped to a control value floating coefficient; a dynamic discount factor is acquired, and the control value floating coefficient is adjusted according to the dynamic discount factor; a target control value is generated based on the adjusted control value floating coefficient and a preset basic value; through risk analysis on multi-source data, the target control value is dynamically adjusted, and the accuracy of the result is improved.

[0093] The specific limitation of the risk quantification regulation device based on user behavior can be referred to the limitation of the risk quantification regulation method based on user behavior in the above, which will not be repeated here. Each module in the risk quantification regulation device based on user behavior can be realized by software, hardware and their combination in whole or in part. Each module can be embedded in or independent of the processor in the computer device in hardware form, or stored in the memory in the computer device in software form, so as to call and execute the operation corresponding to each module by the processor.

[0094] In one embodiment, a computer device can be a server, and its internal structure diagram can be as shown in Figure 4 The computer device includes a processor, a memory, a network interface and a database connected by a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with the external client through the network connection. The computer program is executed by the processor to realize the function or step of the server side of the risk quantification regulation method based on user behavior.

[0095] In one embodiment, a computer device can be a client, and its internal structure diagram can be as shown in Figure 5As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium, an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The network interface of the computer device is used to communicate with the external server through the network connection. The computer program is executed by the processor to realize the function or step of the client side of the risk quantification regulation method based on user behavior.

[0096] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor executing the computer program to implement the following steps: Obtain multi-source data of user behavior, and synchronize the timestamps of the multi-source data to obtain multi-source synchronized data; Perform fusion feature extraction on the multi-source synchronized data to obtain multi-source fusion features; Construct an evaluation model according to the multi-source fusion features, and perform risk score analysis on the multi-source synchronized data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronized data; Weighted fusion double-model scoring of the risk score according to a preset historical claim data weight to obtain a user behavior risk total score; According to the user behavior risk total score, divide the risk level, and map the risk level to a control value floating coefficient; Obtain a dynamic discount factor, and adjust the control value floating coefficient according to the dynamic discount factor, and generate a target control value based on the adjusted control value floating coefficient and a preset base value.

[0097] In one embodiment, a computer readable storage medium is provided, having a computer program stored thereon, the computer program being executed by a processor to implement the following steps: Obtain multi-source data of user behavior, and synchronize the timestamps of the multi-source data to obtain multi-source synchronized data; Perform fusion feature extraction on the multi-source synchronized data to obtain multi-source fusion features; Construct an evaluation model according to the multi-source fusion features, and perform risk score analysis on the multi-source synchronized data according to the evaluation model to obtain a risk score corresponding to the multi-source synchronized data; Weighted fusion double-model scoring of the risk score according to a preset historical claim data weight to obtain a user behavior risk total score; According to the user behavior risk total score, a risk level is divided, and the risk level is mapped to a control value floating coefficient; A dynamic discount factor is obtained, and the control value floating coefficient is adjusted according to the dynamic discount factor. A target control value is generated based on the adjusted control value floating coefficient and a preset base value.

[0098] It should be noted that the functions or steps described above with respect to the computer readable storage medium or the computer device can correspond to the related descriptions of the server side and the client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0099] Those skilled in the art can understand that all or part of the processes in the foregoing method embodiments can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the foregoing method embodiments. In the embodiments provided in the present application, any reference to a memory, storage, database or other medium can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct RAM bus dynamic RAM (DRDRAM) and memory bus dynamic RAM (RDRAM).

[0100] Those skilled in the art can clearly understand that, for the convenience and brevity of description, only the division of the above functional units and modules is exemplified. In actual applications, the above functions can be completed by different functional units or modules according to needs, that is, the internal structure of the device is divided into different functional units or modules to complete all or part of the functions described above.

[0101] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, and not to limit them. If a software tool or component that is not from the company appears in the application examples, it is only used for example introduction and does not represent actual use. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced by equivalents. These modifications or replacements do not change the essence of the corresponding technical solutions, and should be included in the protection scope of the present application.

Claims

1. A risk quantification and control method based on user behavior, characterized in that: include: Acquire multi-source data of user behavior, and synchronize timestamps of the multi-source data to obtain multi-source synchronized data; Performing fusion feature extraction on the multi-source synchronized data to obtain multi-source fusion features; Constructing an assessment model based on the multi-source fusion features, and performing risk score analysis on the multi-source synchronized data based on the assessment model to obtain a risk score corresponding to the multi-source synchronized data; Perform weighted fusion dual-model scoring on the risk score based on the preset historical claims data weights to obtain the total user behavior risk score; Divide the risk level according to the total risk score of the user behavior, and map the risk level to a control value floating coefficient; A dynamic discount factor is obtained, and the control value floating coefficient is adjusted according to the dynamic discount factor, and a target control value is generated based on the adjusted control value floating coefficient and a preset basic value.

2. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The performing timestamp synchronization on the multi-source data to obtain multi-source synchronized data includes: Performing data cleaning on the multi-source data and performing timestamp standardization on the cleaned data to obtain standard data; Performing timestamp alignment on the standard data, and performing missing value filling on the alignment result to obtain filled data; Redundancy removal and merging are performed on the filling data to obtain multi-source synchronized data.

3. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The extracting fusion features from the multi-source synchronized data to obtain multi-source fusion features includes: Performing data type stratification on the multi-source synchronized data to obtain stratified data; Performing preliminary single-source feature extraction on the hierarchical data to obtain hierarchical single-source features; Performing cross-source feature correlation fusion on the layered single-source features to obtain correlated fusion features; Perform feature dimension reduction screening on the associated fusion features to obtain multi-source fusion features.

4. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The performing risk score analysis on the multi-source synchronized data according to the assessment model to obtain a risk score corresponding to the multi-source synchronized data includes: Matching each record in the multi-source synchronized data with the characteristic dimensions in the assessment model to obtain a risk dimension for each record; Calculating the risk scores of the risk dimensions according to the assessment model, and integrating the risk scores into multi-dimensional risk scores using preset dimension weights to obtain a comprehensive risk score for each record; The comprehensive risk score is trend-analyzed and adjusted according to the timing rules in the assessment model to obtain a risk score corresponding to the multi-source synchronized data.

5. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The risk score is weighted and integrated with the dual-model score according to the preset historical claims data weights to obtain the total user behavior risk score, including: According to the preset historical claims data weight, configure the weight for each model in the dual model to obtain the weight of each model; Performing weighted calculation on the risk scores according to the weights to obtain a comprehensive weighted score; The comprehensive weighted score is adjusted according to the preset time dimension weight to obtain the total user behavior risk score.

6. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The step of dividing the risk levels according to the total risk score of the user behavior and mapping the risk levels to control value floating coefficients includes: Obtaining a risk level classification standard; mapping the user behavior risk total score to the risk level classification standard to obtain a risk level; Obtain historical user risk levels and actual loss data corresponding to the risk levels, perform loss rate statistics based on the risk levels and the actual loss data, and obtain the loss rate corresponding to the risk levels; Determining a correspondence between the risk level and a preset floating coefficient based on the loss rate; The risk level mapping is mapped according to the corresponding relationship to obtain a control value floating coefficient.

7. The risk quantification and control method based on user behavior according to claim 1, characterized in that: The obtaining of the dynamic discount factor and adjusting the control value floating coefficient according to the dynamic discount factor includes: Obtaining user behavior data for a preset time period, analyzing risk score changes on the behavior data, and obtaining a dynamic discount factor; Determining a dynamic reward and punishment mechanism based on the dynamic discount factor; The control value floating coefficient is adjusted based on the dynamic reward and punishment mechanism.

8. A risk quantification and control device based on user behavior, characterized in that: include: An acquisition synchronization module is used to acquire multi-source data of user behavior and synchronize timestamps of the multi-source data to obtain multi-source synchronized data; An extraction module, configured to extract fusion features from the multi-source synchronized data to obtain multi-source fusion features; Constructing an analysis module, configured to construct an evaluation model based on the multi-source fusion features, and performing risk score analysis on the multi-source synchronized data based on the evaluation model to obtain a risk score corresponding to the multi-source synchronized data; A scoring module is used to perform a weighted fusion dual-model scoring on the risk score based on preset historical claims data weights to obtain a total user behavior risk score; A division and mapping module, configured to divide risk levels according to the total risk score of the user behavior, and map the risk levels to control value floating coefficients; The acquisition and adjustment module is used to acquire a dynamic discount factor and adjust the control value floating coefficient according to the dynamic discount factor. The generating module is used to generate a target control value based on the adjusted control value floating coefficient and a preset basic value.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the risk quantification and control method based on user behavior as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the risk quantification and control method based on user behavior as described in any one of claims 1 to 7 is implemented.

Images