Credit control processing method and device based on artificial intelligence, computer equipment and medium

Abstract

The invention relates to the technical field of communication, and provides a credit control processing method and device based on artificial intelligence, computer equipment and a medium, and the method carries out the prediction based on the first historical consumption data of a target object, and can accurately predict the consumption data of the next day. After the risk assessment result set of the target object is obtained, the credit control risk level of the target object is obtained in combination with the next-day predicted consumption data and the risk assessment result set, so that the risk management is more flexible and timely; and performing credit control processing on the account of the target object according to the credit control risk level. Through accurate risk assessment, a differentiated credit control strategy based on risk levels is realized, so that the system can process high-risk users more efficiently, and meanwhile, excessive intervention on low-risk users is avoided. The effective hit rate of the credit control message is improved by performing key monitoring on the high-risk user, and the waste of software and hardware resources can be greatly reduced.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a signal control processing method, device, computer equipment and medium based on artificial intelligence. Background Art

[0002] When users make phone calls, use data to surf the Internet, etc., the calculation of real-time consumption data of users will be triggered. Based on the packages subscribed by users, real-time consumption data, balance, arrears information, etc., the system needs to determine whether the users meet the requirements of sending reminder SMS, suspending and resuming the service, and other information control actions.

[0003] The existing communication system will perform indiscriminate calculations for each user, but the number of users who actually trigger credit control actions such as payment reminders and shutdowns is less than 20% of the average number of users who trigger them. This indiscriminate credit control strategy leads to a large waste of software and hardware resources. Summary of the invention

[0004] Based on this, a signal control processing method, device, computer equipment and medium based on artificial intelligence are proposed, aiming to solve the problem of waste of software and hardware resources caused by indiscriminate calculation in the prior art.

[0005] The first aspect of the present application provides a signal control processing method based on artificial intelligence, the method comprising:

[0006] Make a prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data;

[0007] Obtaining a risk assessment result set of the target object;

[0008] Obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set;

[0009] Perform credit control processing on the target object's account according to the credit control risk level.

[0010] Optionally, performing prediction based on the first historical consumption data of the target object to obtain the predicted consumption data for the next day includes:

[0011] Classifying the change trend of the first historical consumption data;

[0012] When the first historical consumption data is of the first type, calling a pre-trained first consumption data prediction model, performing prediction based on the first historical consumption data, and obtaining next-day predicted consumption data;

[0013] When the first historical consumption data is of the second type, a pre-trained second consumption data prediction model is called to perform prediction based on the first historical consumption data to obtain predicted consumption data for the next day.

[0014] Optionally, the training process of the first consumption data prediction model includes:

[0015] Traverse the preset smoothing coefficient sequence;

[0016] Each time a smoothing coefficient is traversed, a prediction is performed based on the second historical consumption data using a cubic exponential smoothing method to obtain a historical prediction result, and a historical prediction error is obtained based on the historical prediction result and the corresponding historical true result;

[0017] Comparing the historical prediction errors corresponding to all the smoothing coefficients in the preset smoothing coefficient sequence;

[0018] Select the smoothing coefficient with the smallest historical forecast error as the target smoothing coefficient;

[0019] The first consumption data prediction model is obtained according to the target smoothing coefficient.

[0020] Optionally, the training process of the second consumption data prediction model includes:

[0021] Arrange the second historical consumption data in chronological order to obtain a data time series;

[0022] Calculate the autocorrelation function value and the partial autocorrelation function value according to the data time series;

[0023] Determine the initial parameters of the model according to the autocorrelation function value and the partial autocorrelation function value;

[0024] The preset network structure is trained based on the initial parameters of the model to obtain a second consumption data prediction model.

[0025] Optionally, obtaining a risk assessment result set of the target object includes:

[0026] Get the preset risk rule set;

[0027] configuring a risk factor table for the target object according to the risk rule set;

[0028] Generate a factor processing script set according to the risk factor table;

[0029] configuring a data extraction task according to the factor processing script set, and executing the data extraction task to extract target data from a data source corresponding to the target object;

[0030] A risk assessment result set of the target object is obtained according to the target data and the factor processing script set.

[0031] Optionally, obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set includes:

[0032] Obtaining the risk assessment level and risk assessment reason in the risk assessment result set;

[0033] The risk assessment level is adjusted according to the next day's predicted consumption data and the risk assessment reasons to obtain a credit control risk level.

[0034] Optionally, the risk assessment level is adjusted according to the next day's predicted consumption data and the risk assessment reason to obtain the credit control risk level, including:

[0035] Acquire multiple historical consumption data of the target object;

[0036] Obtaining historical average consumption data according to the plurality of historical consumption data;

[0037] Determine the change amplitude of the predicted consumption data according to the historical average consumption data and the predicted consumption data for the next day;

[0038] quantifying the risk assessment reasons to obtain a quantified value of the risk assessment reasons;

[0039] The risk assessment level is adjusted according to the predicted consumption data change amplitude and the risk assessment reason quantification value to obtain the credit control risk level.

[0040] A second aspect of the present application provides a signal control processing device based on artificial intelligence, the device comprising:

[0041] A prediction module, used to make a prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data;

[0042] An acquisition module, used to acquire a risk assessment result set of the target object;

[0043] A determination module, configured to obtain the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set;

[0044] A processing module is used to perform credit control processing on the account of the target object according to the credit control risk level.

[0045] The third aspect of the present application provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the artificial intelligence-based signal control processing method when executing the computer program.

[0046] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the artificial intelligence-based signal control processing method are implemented.

[0047] This application can accurately predict the next day's consumption data by predicting based on the first historical consumption data of the target object, providing strong support for credit control risk assessment; after obtaining the risk assessment result set of the target object, the credit control risk level of the target object is obtained by combining the next day's predicted consumption data and the risk assessment result set, making risk management more flexible and timely; credit control processing is performed on the target object's account according to the credit control risk level. Through accurate risk assessment, a differentiated credit control strategy based on risk level is implemented, enabling the system to handle high-risk users more efficiently while avoiding excessive intervention in low-risk users. By focusing on monitoring high-risk users, the effective hit rate of credit control messages is improved, which can greatly reduce the waste of software and hardware resources. Automated and intelligent credit control processing methods reduce manual intervention and decision-making time, thereby reducing operating costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 It is a flow chart of the signal control processing method based on artificial intelligence provided in the embodiment of the present application.

[0050] Figure 2 It is a data diagram based on the quadratic smoothing index provided in an embodiment of the present application.

[0051] Figure 3 This is another data schematic diagram based on the quadratic smoothing index provided in an embodiment of the present application.

[0052] Figure 4 It is a data diagram based on the cubic smoothing index provided in an embodiment of the present application.

[0053] Figure 5 This is another data schematic diagram based on the cubic smoothing index provided in an embodiment of the present application.

[0054] Figure 6 It is a functional module diagram of the artificial intelligence-based signal control processing device provided in an embodiment of the present application.

[0055] Figure 7It is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0057] With the rapid development of operators' data services, the network has evolved from 3G and 4G to the current 5G, and the number of billing call records and messages has increased exponentially. For example, the daily computing volume in a certain province is as high as 270 million.

[0058] When users make phone calls, use data to surf the Internet, etc., the calculation of real-time consumption data of users will be triggered. Based on the packages subscribed by users, real-time consumption data, balance, arrears information, etc., the system needs to determine whether the users meet the requirements of sending reminder SMS, suspending and resuming the service, and other information control actions.

[0059] It can be seen that the shortcomings of the existing technology are: indiscriminate calculations are performed for each user, and the number of users who actually trigger credit control behaviors such as payment reminders and shutdowns is less than 20% of the average number of triggers. Although the existing technical solutions can meet the real-time computing requirements to a certain extent, in high-concurrency scenarios, this indiscriminate calculation leads to a large waste of software and hardware resources.

[0060] Figure 1 A flow chart of an artificial intelligence-based signal control processing method provided in an embodiment of the present application, the artificial intelligence-based signal control processing method includes the following steps.

[0061] S11, making a prediction based on the first historical consumption data of the target object to obtain the predicted consumption data for the next day.

[0062] Among them, the target object refers to the user or entity that needs to be monitored for credit control (credit control). The target object can be either an individual user or a corporate customer. The first historical consumption data refers to the consumption record of the target object in the past period of time. For example, in the communication business scenario, the first historical consumption data may include cost data generated by call duration, traffic usage, text message sending, etc. The first historical consumption data helps to predict the future consumption trend of the target object in communication services. In the banking business scenario, the first historical consumption data may include various transaction records generated in the target object's bank account, such as deposits, withdrawals, transfers, payments, etc. The first historical consumption data helps the bank evaluate the financial status and credit rating of the target object. Depending on the specific business scenario, the first historical consumption data will also contain different information, such as purchase records of insurance products, shopping records on e-commerce platforms, etc.

[0063] Since consumption data usually has certain regularities and trends, such as periodic fluctuations, seasonal changes, etc. By analyzing and mining these regularities and trends, it is possible to predict the consumption data that the target object may generate on the next day based on the target object's first historical consumption data.

[0064] In order to more clearly explain the technical principles of the present application and its technical effects in practical applications, the embodiment of the present application takes the communication business scenario as an example, collects and analyzes the user's first historical consumption data, uses artificial intelligence technology to establish a prediction model, and achieves more accurate and efficient credit control processing, thereby providing strong support for credit control and risk management in the communication industry.

[0065] In an optional embodiment, performing prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data includes:

[0066] Classifying the change trend of the first historical consumption data;

[0067] When the first historical consumption data is of the first type, calling a pre-trained first consumption data prediction model, performing prediction based on the first historical consumption data, and obtaining next-day predicted consumption data;

[0068] When the first historical consumption data is of the second type, a pre-trained second consumption data prediction model is called to perform prediction based on the first historical consumption data to obtain predicted consumption data for the next day.

[0069] According to the characteristics of consumption data, it is divided into different types or categories, which reflect the change trend of the first historical consumption data. The first historical consumption data of the target object is preprocessed, such as cleaning, deduplication, normalization, etc., and the classification algorithm (such as decision tree, support vector machine, clustering algorithm, etc.) is applied to classify the change trend of the first historical consumption data, such as stable type, growth type, fluctuation type, etc.

[0070] After obtaining the change trend type of the first historical consumption data, it is determined whether the change trend of the first historical consumption data is the first type or the second type, thereby calling different prediction models according to the change trend type of the first historical consumption data to improve the accuracy of the prediction.

[0071] If the first historical consumption data belongs to the first type (time series data with obvious trends and seasonality), the pre-trained first consumption data prediction model, such as a cubic exponential smoothing model, is called. The first historical consumption data is input into the first consumption data prediction model, which predicts the next-day consumption data of the target object based on the first historical consumption data and outputs prediction data (next-day prediction consumption data).

[0072] If the first historical consumption data belongs to the second type (relatively stable time series data), the pre-trained second consumption data prediction model is called, for example, the AutoRegressive Integrated Moving Average (ARIMA) model. The first historical consumption data is input into the second consumption data prediction model, and the second consumption data prediction model predicts the next day's consumption data of the target object based on the first historical consumption data, and outputs the prediction data (next day prediction consumption data).

[0073] In the above optional embodiment, by classifying the change trend of the first historical consumption data and calling different prediction models for prediction based on the classification results, not only the accuracy and reliability of the prediction are improved, but also the prediction results are more in line with the actual consumption behavior of the target object.

[0074] In an optional embodiment, the training process of the first consumption data prediction model includes:

[0075] Traverse the preset smoothing coefficient sequence;

[0076] Each time a smoothing coefficient is traversed, a prediction is performed based on the second historical consumption data using a cubic exponential smoothing method to obtain a historical prediction result, and a historical prediction error is obtained based on the historical prediction result and the corresponding historical true result;

[0077] Comparing the historical prediction errors corresponding to all the smoothing coefficients in the preset smoothing coefficient sequence;

[0078] Select the smoothing coefficient with the smallest historical forecast error as the target smoothing coefficient;

[0079] The first consumption data prediction model is obtained according to the target smoothing coefficient.

[0080] The preset smoothing coefficient sequence is a set of pre-set smoothing coefficient values. In time series analysis, the smoothing coefficient is used to control the degree of smoothing process and determines the degree of influence of historical data on the current forecast value. A suitable smoothing coefficient can help the model better fit the historical data while reducing the influence of noise.

[0081] The second historical consumption data is a set of historical consumption data used to train the first consumption data prediction model, which comes from the Hadoop big data platform. The historical consumption data of all objects can be collected from multiple data sources and different database types (such as Oracle, MySQL). The collected historical consumption data is cleaned, corrected, formatted, etc. to improve data quality and ensure data consistency and accuracy. The processed data is stored in the Hadoop big data platform.

[0082] Traverse each smoothing coefficient in the preset smoothing coefficient sequence. For each traversed smoothing coefficient, use the cubic exponential smoothing method to predict the second historical consumption data to obtain a set of historical prediction results. Each historical prediction result corresponds to a historical true result. Calculate the difference between the historical prediction result and the corresponding historical true result to obtain the historical prediction error corresponding to each smoothing coefficient. Compare the historical prediction errors corresponding to all smoothing coefficients in the preset smoothing coefficient sequence to find the minimum historical prediction error. Select the smoothing coefficient with the minimum historical prediction error as the target smoothing coefficient. The target smoothing coefficient is the optimal smoothing coefficient that can minimize the prediction error of the model. Thus, the most accurate consumption data prediction model is constructed based on the selected target smoothing coefficient.

[0083] Next, the exponential smoothing method is introduced: The exponential smoothing method is a special weighted moving average method. It takes the actual number of occurrences in this period and the predicted value of the previous period as the base, gives different relative numbers to each of them, calculates the exponential smoothing value, and uses this to determine the prediction result.

[0084] Single exponential smoothing forecasting model: S t =aX t +(1-a)S t-1 , S t is the smoothed value of period t (the current period forecast value), X t is the observed value at the current time point, St-1 is the smoothed value of period t-1 (previous forecast value), and a is the smoothing coefficient. One limitation of the first-order exponential smoothing method is that the forecasts made for the rising (or falling) trend series are often too low (or too high), that is, they lag behind the actual observed values.

[0085] Secondary exponential smoothing method: Based on the difference between the first exponential smoothing and the second exponential smoothing, this difference is added to the first exponential smoothing and then the trend change value, a forecast value that is closer to the actual value can be obtained.

[0086] Prediction model of quadratic exponential smoothing method: F t+T =a t +b t T, F t+T is the predicted value of t+T, T is the interval between period t and the predicted period, a t 、b t as a parameter.

[0087] a t =S t +(S t -S″ t )=2S t -S″ t

[0088]

[0089] S t and S″ t They are first exponential smoothing value and second exponential smoothing value respectively.

[0090] S t =aX t +(1-a)S t-1

[0091] S″ t =aS t +(1-a)S″ t-1

[0092] To predict the t+T period, just t Add T b to the base t Therefore, the general expression of the quadratic exponential smoothing forecasting model can be obtained: F t+T =a t +b t T.

[0093] See also Figure 2 and Figure 3 As shown, Figure 2 is the prediction result when the smoothing coefficient a=0.4, Figure 3 This is the prediction result when the smoothing coefficient a=0.6.

[0094] For example, to predict the phone bill on the 19th, the first smoothing value calculated using the single exponential smoothing method = a * phone bill on the 19th + (1 - a) * smoothing value on the 18th = 0.4 * 3.48 + 0.6 * 3.4 = 3.43, and the second smoothing value calculated using the double exponential smoothing method = a * first smoothing value + (1 - a) * second smoothing value on the 18th = 0.4 * 3.43 + 0.6 * 3.4 = 3.41.

[0095] According to the formula a t = S t +(S t - S″ t ) = 2S t - S″ t , the calculated at = 2 * 3.43 - 3.41 = 3.45.

[0096] According to the formula , the calculated Bt = (0.4 / 0.6)(3.43 - 3.41) = 0.01.

[0097] According to the formula F t+T = a t + b t T, the calculated predicted value: 3.45 + 0.01 * 1 = 3.46.

[0098] Combined with Figure 2 and Figure 3 , it can be seen that the smoothing coefficient a is a key parameter in the double exponential smoothing method, which determines the influence degree of historical data on the current predicted value. In the single exponential smoothing method, the value of a determines whether the current predicted value depends more on the recent data points or the long-term average level of historical data. In the double exponential smoothing method, the value of a not only affects the calculation of the smoothing value, but also indirectly affects the calculation of the trend value, thus affecting the final prediction result.

[0099] For a trend time series, since the time series shows an obvious upward or downward trend, the smoothing coefficient a can take a relatively large value (0.6 < a ≤ 1) to more sensitively capture the trend changes in the time series; for a horizontal time series, since the time series remains relatively stable for a long time without an obvious trend, the smoothing coefficient a can take a relatively small value (0 < a ≤ 0.3) to reduce noise and make the predicted value more stable; for a time series that is a mixture of horizontal and trend types, the smoothing coefficient a can take a moderate value (0.3 < a ≤ 0.6) to balance the influence of historical data and trend changes on the predicted value. For a time series that is a mixture of horizontal and trend types, the smoothing coefficient a can take a moderate value (0.3 < a ≤ 0.6).

[0100] Select multiple a values ​​for testing. For each a value, use the quadratic exponential smoothing method to calculate and obtain the corresponding prediction results. Calculate the error between each prediction result and the actual result, compare the error size under different a values, and select the a value with the smallest error as the optimal value. For example, the error of a = 0.4 is significantly lower than that of a = 0.6, so 0.4 is selected as the optimal value of the smoothing coefficient a.

[0101] Based on the quadratic exponential smoothing forecasting model, the triple exponential smoothing forecasting model is obtained as follows:

[0102] F t+T =a t +b t T

[0103] a t =S t +S″ t -S″′ t

[0104]

[0105] S t =aX t +(1-a)S t-1

[0106] S″ t =aS t +(1-a)S″ t-1

[0107] S″′ t =aS″ t +(1-a)S″′ t-1

[0108] Among them, F t+T is the predicted value for the t+T period, t is the current period, T is the interval between the current period and the predicted period, a t 、b t is the parameter, S t , S″ t , S″′ t They are first exponential smoothing, second exponential smoothing, and third exponential smoothing.

[0109] See also Figure 4 and Figure 5 As shown, based on the data from April 11 to April 24, the error between the second exponential smoothing and the original value is calculated as w″=0.46, and the error between the third exponential smoothing and the original value is calculated as w″′=0.34.

[0110] It can be seen that the prediction of triple exponential smoothing is more accurate than that of quadratic exponential smoothing. As a tool for trend analysis, triple exponential smoothing has the characteristics of both moving average method and trend prediction method. Therefore, it is more accurate in determining the trend of phone charges.

[0111] In an optional embodiment, the training process of the second consumption data prediction model includes:

[0112] Arrange the second historical consumption data in chronological order to obtain a data time series;

[0113] Calculate the autocorrelation function value and the partial autocorrelation function value according to the data time series;

[0114] Determine the initial parameters of the model according to the autocorrelation function value and the partial autocorrelation function value;

[0115] The preset network structure is trained based on the initial parameters of the model to obtain a second consumption data prediction model.

[0116] The second historical consumption data is arranged in chronological order in order to capture the trend and periodicity of the data as it changes over time. The arranged data forms a data time series. You can check whether there are missing values ​​or outliers in the data time series and perform corresponding processing, such as interpolation or deletion. You can also smooth the data time series to reduce the impact of noise.

[0117] Based on the generated data time series, two key statistics are calculated: the autocorrelation function (ACF) value and the partial autocorrelation function (PACF) value. The autocorrelation function (ACF) is used to measure the correlation between data at different time points in the same time series. That is, the degree of similarity between the current value and its past values. The partial autocorrelation function (PACF) measures the correlation between data at two time points after taking into account the influence of intermediate values, identifying past values ​​that have a direct impact on predicting the current value in multiple lag periods.

[0118] According to the calculated autocorrelation function value and partial autocorrelation function value, the initial parameters of the second consumption data prediction model are determined. The initial parameters of the model may include the order p, d, and q of the model. Among them, p represents the order of the autoregressive term, d represents the order of the difference, and q represents the order of the moving average term. The initial parameters of the model may also include the variance of the constant term and the error term.

[0119] Finally, the determined model initial parameters are used to train the preset network structure. The preset network structure may be a neural network model based on time series analysis, such as a recurrent neural network (RNN) or its variants (such as LSTM, GRU, etc.), or other types of machine learning models.

[0120] The training process includes inputting time series data into the model, adjusting the model parameters through the back propagation algorithm to minimize the prediction error, and finally obtaining a second consumption data prediction model that can accurately predict future consumption data. Check whether the residual of the model is white noise, that is, whether the residual has no autocorrelation. If the residual has autocorrelation, adjust the model parameters and refit until the model performance is satisfactory.

[0121] S12, obtaining a risk assessment result set of the target object.

[0122] While predicting the next day's consumption data for the target object, the risk assessment result set of the target object can also be obtained from the distributed cache. The risk assessment result set includes the risk level and risk reason.

[0123] In an optional embodiment, obtaining the risk assessment result set of the target object includes:

[0124] Get the preset risk rule set;

[0125] configuring a risk factor table for the target object according to the risk rule set;

[0126] Generate a factor processing script set according to the risk factor table;

[0127] configuring a data extraction task according to the factor processing script set, and executing the data extraction task to extract target data from a data source corresponding to the target object;

[0128] A risk assessment result set of the target object is obtained according to the target data and the factor processing script set.

[0129] You can customize risk rules according to actual business needs. Each rule includes: conditions, priorities, risk levels, risk reasons and other elements. You can use logical operators (AND, OR, NOT) and comparison operators (=, <, >, <=, >=, !=) to combine conditions. Each rule has a priority value. The smaller the value, the higher the priority. Risk levels can include high, medium and low risk levels.

[0130] The risk factor table is the basis for risk assessment. The risk factor table lists various factors (i.e., risk factors) that may affect the target object's credit control behavior, such as real-time call charges, balances, arrears, large accounts, etc. Each risk factor has a unique name and code for identification. Assign a type to each risk factor, such as fixed risk factors (factors calculated once a day, such as creditworthiness, credit control type, etc.), variable risk factors (factors recalculated in real time, such as balances, arrears), and AI prediction factors (predicted call charges for the next day).

[0131] Then, based on the risk factor table, a script that conforms to the hive syntax format can be used to generate a factor processing script set. The factor processing script set is a collection of programs or scripts used to process and analyze the target object data. According to the requirements of the risk factor table, the required data is extracted, converted, and calculated from the data source. Each risk factor can be associated with multiple scripts, and the priority of script execution can be specified. The script can include data cleaning, format conversion, data aggregation, result output, etc. to ensure that the target data meets the needs of risk assessment.

[0132] According to the dependent data of the factor processing, the required data source type is identified to configure the data extraction task. The data extraction task defines the mapping relationship between the extracted data and the factors to ensure that the data is correctly mapped to the factors. The data extraction task also defines the rules for extracting data from the data source, including SQL queries, file reading rules, etc. The data source may include Oracle, MySQL, PostgreSQL databases, CSV files, hdfs remote files, APIs, etc., and the data source stores various data related to the target object. After executing the data extraction task, the target data that meets the requirements of the factor processing script set will be obtained. If the data value corresponding to the risk factor cannot be obtained, the default value can be used. In one embodiment, the required computing resources can also be configured according to the data extraction task, such as setting the number of concurrent extraction threads by taking the modulus of the specified field.

[0133] Finally, based on the target data and factor processing script set, the risk assessment result set of the target object is obtained. The risk assessment result set includes the specific score of the risk factor, risk level, risk probability, risk cause and other information. The risk assessment result set is stored in the distributed cache to ensure fast access and real-time update of data.

[0134] Assume that a mobile communications company wants to optimize its credit control system to reduce indiscriminate real-time consumption data calculation and resource waste. The company decides to use a risk rule customization engine to determine whether the user meets the conditions for credit control actions such as sending reminder SMS, suspending and resuming the service, etc. based on the user's real-time consumption data, balance, and arrears.

[0135] The risk rule set includes three rules:

[0136] Rule 1: If the user has a balance of 10 yuan on the day and owes 0 yuan, and it is predicted that the user will spend 20 yuan the next day, a SMS reminder for overdue payment will be sent.

[0137] Rule 2: If the user owes more than 100 yuan and the balance is less than 10 yuan, the shutdown action will be triggered.

[0138] Rule 3: If the user is a VIP user and the outstanding payment does not exceed RMB 30, no credit control action will be triggered.

[0139] Each rule contains conditions, priorities (such as Rule 2>Rule 1>Rule 3), risk levels (such as collection, shutdown, no action) and risk reasons (such as large amount of arrears, insufficient balance, VIP users, etc.).

[0140] For example, if user A owes 60 yuan for 8 days and is not a VIP user, according to rule 1, user A will be triggered to send a reminder SMS.

[0141] The risk factor configuration table includes the following 4 risk factors:

[0142] Factor 1: Amount of outstanding payments (name: outstanding payments, code: QF001, type: variable factor, data source: communication system billing system, default value: 0).

[0143] Factor 2: Duration of arrears (name: Duration of arrears, code: QFSZ001, type: variable factor, data source: communication system billing system, default value: 0 days).

[0144] Factor 3: User balance (name: balance, code: YE001, type: change factor, data source: communication system account system, default value: 0).

[0145] Factor 4: User level (name: level, code: DJ001, type: fixed factor, data source: communication system user management system, default value: ordinary user).

[0146] For example, user A owes 60 yuan, the overdue period is 8 days, the balance is 0 yuan, and the user level is ordinary user.

[0147] The factor processing scripts written include:

[0148] Script 1: Get the user's outstanding amount and duration from the communication system billing system.

[0149] Script 2: Get the user's balance from the communication system account system.

[0150] Script 3: Get the user's level from the communication system user management system.

[0151] Associate a corresponding script with each factor and specify the priority of script execution (such as script 1, script 2, and script 3 are executed in sequence).

[0152] For example, after script 1 is executed, the amount of money owed by user A is 60 yuan, and the duration of the arrears is 8 days. After script 2 is executed, the balance of user A is 0 yuan. After script 3 is executed, the level of user A is obtained as ordinary user.

[0153] According to the script set's dependency on data, identify the required data source type (such as communication system billing system, account system, user management system). The configured data extraction tasks include: data source type, connection information, extraction rules, and mapping relationships.

[0154] For example, configure data source connection information, such as the database connection string, user name, and password of the communication system billing system. Set data extraction rules, such as querying the amount of arrears and the duration of arrears of user A from the billing system. Configure data mapping relationships, such as mapping the "amount of arrears" field in the billing system to the factor "arrears".

[0155] The risk assessment results (such as risk level, risk cause, etc.) are stored in the distributed cache.

[0156] For example, the risk assessment result of user A is: the risk level is collection, and the risk reason is that the amount of arrears is large and the arrears period exceeds 7 days.

[0157] S13, obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set.

[0158] The credit control risk level of the target object is obtained by comprehensively considering the next day's predicted consumption data and the risk assessment results. The credit control risk level is usually used to measure the default risk that the target object (such as individuals, enterprises, etc.) may have in credit transactions or consumption activities. The credit control risk level includes low risk, medium risk, and high risk.

[0159] Compare the next day's forecast consumption data with the risk assessment results to analyze the correlation and consistency between the two. For example, if the next day's forecast consumption data shows that the target person will have a large amount of consumption, and the risk assessment results show that his credit risk is high, it may mean that the target person has a high risk of default.

[0160] In an optional embodiment, obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set includes:

[0161] Obtaining the risk assessment level and risk assessment reason in the risk assessment result set;

[0162] The risk assessment level is adjusted according to the next day's predicted consumption data and the risk assessment reasons to obtain a credit control risk level.

[0163] The next day's forecast consumption data provides the target's expected consumption in the next day, which helps to more comprehensively understand the target's financial situation and consumption habits. By analyzing this data, it is possible to assess whether the target has sufficient financial capacity to support its future consumption plans and adjust the risk assessment level accordingly.

[0164] If the next day's forecast consumption data shows that the target has sufficient financial capacity to cope with future consumption plans, and the risk factors in the risk assessment reasons are controllable or have been improved, then the risk assessment level can be appropriately lowered. On the contrary, if the next day's forecast consumption data shows that the target's financial situation is tight, or the risk factors in the risk assessment reasons still exist and may intensify, then the risk assessment level can be increased.

[0165] Finally, the credit control risk level of the target object is determined based on the adjusted risk assessment level.

[0166] In an optional embodiment, the risk assessment level is adjusted according to the next day forecast consumption data and the risk assessment reason to obtain the credit control risk level, including:

[0167] Acquire multiple historical consumption data of the target object;

[0168] Obtaining historical average consumption data according to the plurality of historical consumption data;

[0169] Determine the change amplitude of the predicted consumption data according to the historical average consumption data and the predicted consumption data for the next day;

[0170] quantifying the risk assessment reasons to obtain a quantified value of the risk assessment reasons;

[0171] The risk assessment level is adjusted according to the predicted consumption data change amplitude and the risk assessment reason quantification value to obtain the credit control risk level.

[0172] Extract multiple historical consumption records of the target object from the database, including information such as consumption amount, consumption time, and consumption type. Average the extracted historical consumption data to obtain the historical average consumption data of the target object. The historical average consumption data reflects the long-term consumption level and consumption habits of the target object. Calculate the difference or ratio between the next day's predicted consumption data and the historical average consumption data to obtain the predicted consumption data change amplitude. The predicted consumption data change amplitude is used to measure the fluctuation of the target object's consumption level.

[0173] According to the nature and impact of the risk assessment reasons, they are converted into quantifiable values ​​to more accurately assess their impact on the risk level. For example, different quantitative standards can be set based on factors such as the number of overdue payments, the number of overdue days, and the debt ratio of the target object.

[0174] The risk assessment level is adjusted by comprehensively considering the change range of the forecast consumption data and the quantitative value of the risk assessment reasons. If the change range of the forecast consumption data is large and the quantitative value of the risk assessment reasons is high, the risk assessment level may need to be increased; otherwise, the risk assessment level may be lowered.

[0175] By comprehensively considering multiple factors to assess the credit control risk level of the target object, its actual default risk level can be more accurately reflected.

[0176] Suppose there is a mobile communication user whose average consumption in the past 30 days is 200 yuan, the average consumption in the past 6 months is 220 yuan, and the average consumption in the past year is 210 yuan. The average consumption data of the past 30 days is used as the historical average consumption data, that is, 200 yuan. Based on the machine learning model, it is predicted that the user's consumption data for the next day will be 250 yuan. The risk assessment reasons for this user include: there is a recent overdue record (quantitative value: 2), and the debt ratio is high (quantitative value: 3). The predicted consumption data change amplitude is calculated as (predicted consumption data for the next day-historical average consumption data) / historical average consumption data=(250-200) / 200=0.25 (or 25%), which means that the consumption level of the target object has increased by 25% compared with the historical average level.

[0177] Assuming that the weight of each risk assessment reason is equal and the sum is 1, the total risk assessment reason quantification value = (2*0.5) + (3*0.5) = 1 + 1.5 = 2.5.

[0178] Assuming that the risk assessment level is "medium", the change amplitude weight is 0.3, the quantitative value weight is 0.7, 1 represents "low", 2 represents "medium", and 3 represents "high". Therefore, the adjusted risk assessment level = risk assessment level + (change amplitude * change amplitude weight) + (total risk assessment reason quantitative value * quantitative value weight) = 2 + (0.25 * 0.3) + (2.5 * 0.7) = 2 + 0.075 + 1.75 = 3.825.

[0179] Since the risk assessment levels are discrete (low, medium, high), the adjusted values ​​need to be mapped back to these levels. For example, below 2.5 is "low", 2.5 to 3.5 is "medium", and above 3.5 is "high"). Therefore, the adjusted risk assessment level is "high".

[0180] According to the above calculation, the credit control risk level of the target object is adjusted to "high", which means that the target object has a higher default risk than other objects and may need to take more stringent credit control measures.

[0181] S14: Perform credit control on the target object's account according to the credit control risk level.

[0182] Determine the corresponding credit control strategy according to the credit control risk level. For example, when the credit control risk level is a low risk level, it is not necessary to determine whether the low risk level object needs to be urged or shut down, thereby reducing the computing pressure of the program and hardware, thereby reducing the waste of resources. When the credit control risk level is a medium risk level, the relevant data of the medium risk level object (for example, historical consumption data, next day forecast consumption data, risk factors, risk reasons, etc.) can be sent to the low-speed computing channel for slow calculation, reducing the pressure on the hardware. When the credit control risk level is a high risk level, the relevant data of the high risk level object can be sent to the high-speed computing channel for priority calculation.

[0183] According to the determined credit control risk level, the corresponding credit control processing actions are performed on the target object's account. The accounts include but are not limited to: mobile phone numbers, bank card numbers, insurance accounts, electronic wallets, shopping points accounts, membership accounts, etc. For medium and high risk users, credit control processing includes but is not limited to: sending reminder SMS, suspending service, adjusting packages, adjusting quotas, transaction restrictions, risk warnings, account freezing, etc.

[0184] It should be noted that the application can also compare the predicted consumption data for the next day with the current balance to determine whether the balance will be insufficient and arrears will occur the next day. For example, if the balance on the day is 10 yuan and the arrears are 0 yuan, and it is predicted that a fee of 20 yuan will be incurred the next day, then the user is a high-risk user and the action of arrears collection may be triggered the next day.

[0185] The artificial intelligence-based credit control processing method provided in the embodiment of the present application can accurately predict the next day's consumption data by predicting based on the first historical consumption data of the target object, thereby providing strong support for credit control risk assessment; after obtaining the risk assessment result set of the target object, the credit control risk level of the target object is obtained by combining the next day's predicted consumption data and the risk assessment result set, making risk management more flexible and timely; credit control processing is performed on the target object's account according to the credit control risk level. Through accurate risk assessment, a differentiated credit control strategy based on risk level is implemented, enabling the system to handle high-risk users more efficiently while avoiding excessive intervention in low-risk users. By focusing on monitoring high-risk users, the effective hit rate of credit control messages is improved, and the waste of software and hardware resources can be greatly reduced. The automated and intelligent credit control processing method reduces manual intervention and decision-making time, thereby reducing operating costs.

[0186] Figure 6 It is a functional module diagram of the artificial intelligence-based signal control processing device provided in an embodiment of the present application.

[0187] In some embodiments, the artificial intelligence-based signal control processing device 60 may include a plurality of functional modules composed of program code segments. The program code of each program segment in the artificial intelligence-based signal control processing device 60 may be stored in a memory of a computer device and executed by at least one processor to execute (see Figure 1 Description) Functions of signal control processing based on artificial intelligence.

[0188] In this embodiment, the signal control processing device 60 based on artificial intelligence can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a prediction module 601, an acquisition module 602, a determination module 603 and a processing module 204. The module referred to in this application refers to a series of computer-readable instruction segments that can be executed by at least one processor and can complete fixed functions, which are stored in a memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0189] The prediction module 601 is used to make a prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data;

[0190] The acquisition module 602 is used to acquire a risk assessment result set of the target object;

[0191] The determination module 603 is used to obtain the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set;

[0192] The processing module 604 is used to perform credit control processing on the target object's account according to the credit control risk level.

[0193] It should be understood that the various variations and specific embodiments of the artificial intelligence-based signal control processing method provided in the above-mentioned embodiments are also applicable to the artificial intelligence-based signal control processing device in the present embodiment. Through the detailed description of the aforementioned artificial intelligence-based signal control processing method, those skilled in the art can clearly understand the implementation process of the artificial intelligence-based signal control processing device in the present embodiment. For the sake of brevity of the specification, it will not be described in detail here.

[0194] The artificial intelligence-based credit control processing device provided in the embodiment of the present application can accurately predict the next day's consumption data by predicting based on the first historical consumption data of the target object, thereby providing strong support for credit control risk assessment; after obtaining the risk assessment result set of the target object, the credit control risk level of the target object is obtained by combining the next day's predicted consumption data and the risk assessment result set, making risk management more flexible and timely; credit control processing is performed on the target object's account according to the credit control risk level. Through accurate risk assessment, a differentiated credit control strategy based on risk level is implemented, enabling the system to handle high-risk users more efficiently while avoiding excessive intervention in low-risk users. By focusing on monitoring high-risk users, the effective hit rate of credit control messages is improved, and the waste of software and hardware resources can be greatly reduced. The automated and intelligent credit control processing method reduces manual intervention and decision-making time, thereby reducing operating costs.

[0195] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, all or part of the steps of the artificial intelligence-based signal control processing method are implemented.

[0196] See also Figure 7 FIG. 7 is a schematic diagram of the structure of a computer device provided in an embodiment of the present application. In a preferred embodiment of the present application, the computer device 7 includes a memory 701 , at least one processor 702 , and at least one communication bus 703 .

[0197] Those skilled in the art should understand that Figure 7 The structure of the computer device shown does not constitute a limitation of the embodiments of the present application. The computer device 7 may also include more or less other hardware or software than shown in the figure, or a different arrangement of components.

[0198] In some embodiments, the computer device 7 is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, and embedded devices. The computer device 7 may also include client devices, which include but are not limited to any electronic product that can interact with a client through a keyboard, mouse, remote control, touchpad, or voice-controlled device, such as a personal computer, tablet computer, smart phone, digital camera, etc.

[0199] It should be noted that the computer device 7 is only an example, and other existing or future electronic products that are suitable for the present application should also be included in the protection scope of the present application and included here by reference.

[0200] In some embodiments, a computer program is stored in the memory 701, and when the computer program is executed by the at least one processor 702, all or part of the steps in the signal control processing method based on artificial intelligence are implemented. The memory 701 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electronically erasable rewritable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, magnetic disk storage, magnetic tape storage, or any other computer-readable medium that can be used to carry or store data. Further, the computer-readable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application required for at least one function, and the like.

[0201] In some embodiments, the at least one processor 702 is the control core (Control Unit) of the computer device 7, and uses various interfaces and lines to connect the various components of the entire computer device 7, and executes various functions and processes data of the computer device 7 by running or executing the program or module stored in the memory 701, and calling the data stored in the memory 701. For example, when the at least one processor 702 executes the computer program stored in the memory, it implements all or part of the steps of the signal control processing method based on artificial intelligence described in the embodiment of the present application; or implements all or part of the functions of the signal control processing device based on artificial intelligence. The at least one processor 702 can be composed of an integrated circuit, for example, it can be composed of a single packaged integrated circuit, or it can be composed of multiple integrated circuits with the same function or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0202] In some embodiments, the at least one communication bus 703 is configured to realize connection and communication between the memory 701 and the at least one processor 702, etc. Although not shown, the computer device 7 may also include a power supply (such as a battery) for powering each component. Preferably, the power supply may be logically connected to the at least one processor 702 through a power management device, so as to realize functions such as management of charging, discharging, and power consumption management through the power management device. The power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The computer device 7 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which are not repeated here.

[0203] The above-mentioned integrated unit implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor to execute part of the method described in each embodiment of the present application.

[0204] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only illustrative, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0205] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

Claims

1. A signal control processing method based on artificial intelligence, characterized in that: The method comprises: Make a prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data; Obtaining a risk assessment result set of the target object; Obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set; Perform credit control processing on the target object's account according to the credit control risk level.

2. The signal control processing method based on artificial intelligence according to claim 1 is characterized in that: The method of performing prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data includes: Classifying the change trend of the first historical consumption data; When the first historical consumption data is of the first type, calling a pre-trained first consumption data prediction model, performing prediction based on the first historical consumption data, and obtaining next-day predicted consumption data; When the first historical consumption data is of the second type, a pre-trained second consumption data prediction model is called to perform prediction based on the first historical consumption data to obtain predicted consumption data for the next day.

3. The signal control processing method based on artificial intelligence according to claim 2 is characterized in that: The training process of the first consumption data prediction model includes: Traverse the preset smoothing coefficient sequence; Each time a smoothing coefficient is traversed, a prediction is performed based on the second historical consumption data using a cubic exponential smoothing method to obtain a historical prediction result, and a historical prediction error is obtained based on the historical prediction result and the corresponding historical true result; Comparing the historical prediction errors corresponding to all the smoothing coefficients in the preset smoothing coefficient sequence; Select the smoothing coefficient with the smallest historical forecast error as the target smoothing coefficient; The first consumption data prediction model is obtained according to the target smoothing coefficient.

4. The signal control processing method based on artificial intelligence according to claim 2 is characterized in that: The training process of the second consumption data prediction model includes: Arrange the second historical consumption data in chronological order to obtain a data time series; Calculate the autocorrelation function value and the partial autocorrelation function value according to the data time series; Determine the initial parameters of the model according to the autocorrelation function value and the partial autocorrelation function value; The preset network structure is trained based on the initial parameters of the model to obtain a second consumption data prediction model.

5. The signal control processing method based on artificial intelligence according to claim 1 is characterized in that: The step of obtaining a risk assessment result set of the target object includes: Get the preset risk rule set; configuring a risk factor table for the target object according to the risk rule set; Generate a factor processing script set according to the risk factor table; configuring a data extraction task according to the factor processing script set, and executing the data extraction task to extract target data from a data source corresponding to the target object; A risk assessment result set of the target object is obtained according to the target data and the factor processing script set.

6. The signal control processing method based on artificial intelligence according to claim 5 is characterized in that: The step of obtaining the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set includes: Obtaining the risk assessment level and risk assessment reason in the risk assessment result set; The risk assessment level is adjusted according to the next day's predicted consumption data and the risk assessment reasons to obtain a credit control risk level.

7. The signal control processing method based on artificial intelligence according to claim 6 is characterized in that: The risk assessment level is adjusted according to the next day's predicted consumption data and the risk assessment reason to obtain the credit control risk level, including: Acquire multiple historical consumption data of the target object; Obtaining historical average consumption data according to the plurality of historical consumption data; Determine the change amplitude of the predicted consumption data according to the historical average consumption data and the predicted consumption data for the next day; quantifying the risk assessment reasons to obtain a quantified value of the risk assessment reasons; The risk assessment level is adjusted according to the predicted consumption data change amplitude and the risk assessment reason quantification value to obtain the credit control risk level.

8. A signal control processing device based on artificial intelligence, characterized in that: The device comprises: A prediction module, used to make a prediction based on the first historical consumption data of the target object to obtain the next day's predicted consumption data; An acquisition module, used to acquire a risk assessment result set of the target object; A determination module, configured to obtain the credit control risk level of the target object according to the next day's predicted consumption data and the risk assessment result set; A processing module is used to perform credit control processing on the account of the target object according to the credit control risk level.

9. A computer device, characterized in that: The computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the signal control processing method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the signal control processing method based on artificial intelligence as described in any one of claims 1 to 7 are implemented.

Images