Social risk early warning method and system based on multi-source data

Abstract

The invention discloses a social risk early warning method and system based on multi-source data, and the method comprises the steps: collecting multi-source heterogeneous data, and constructing a risk index library; identifying key risk indexes from the evaluation values by using a clustering algorithm; finding an alternative index for an unquantized key risk index, and establishing a quantitative model; updating the risk index library, performing hierarchical division, and calculating the weight of each risk index; constructing an index comprehensive weight matrix; quantized values of the key risk indexes are obtained through a quantitative model, and a comprehensive score is obtained in combination with the weight matrix and used for social risk early warning. According to the method, for social risk indexes which are difficult to quantify, index substitutes and a measurement and calculation model are constructed through expert experience and data analysis, and index quantification is achieved. Meanwhile, an analytic hierarchy process is adopted to construct a risk index system and give a weight, and a risk comprehensive scoring mechanism is formed. According to the method, the index system is dynamically adjusted by continuously collecting risk data and applying a time sequence analysis method and the like, so that the effectiveness of early warning is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of information technology, and in particular relates to a social risk early warning method and system based on multi-source data. Background Art

[0002] When establishing a social risk early warning system, we are faced with the technical difficulties of how to select appropriate risk indicators, quantitative indicators, and determine the weights of indicators. First, different types of social risks have different characteristics and influencing factors. Therefore, it is necessary to conduct in-depth analysis of the causes and evolution laws of various risks in order to select key indicators that can sensitively reflect the risk status and trends. Secondly, many risk indicators are difficult to quantify directly. It is necessary to explore suitable indicator substitutes or establish a scientific indicator measurement model to truly portray the risk level. Furthermore, the impact of each indicator on the risk is different, and there may be interactive effects between indicators. How to reasonably determine the weight of the indicator is a major difficulty. The determination of weights requires comprehensive qualitative and quantitative analysis, which requires both expert experience and judgment, and mathematical and statistical methods to balance subjectivity and objectivity. In addition, due to the dynamic evolution of social risks, the risk indicator system also needs to be dynamically adjusted and optimized, and new influencing factors should be incorporated in a timely manner, and the indicator quantitative model and weights should be dynamically calibrated to ensure the effectiveness and accuracy of risk early warning. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a social risk early warning method and system based on multi-source data to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above object, the present invention provides a social risk early warning method based on multi-source data, comprising:

[0005] Acquire multi-source heterogeneous data related to social risks and build a risk indicator library; obtain an evaluation value of each indicator in the risk indicator library based on the multi-source heterogeneous data; obtain key risk indicators based on clustering algorithms and evaluation value analysis;

[0006] Screening out unquantified indicators from the key risk indicators, matching substitute indicators of the unquantified risk indicators, and constructing a risk indicator quantitative measurement model based on the unquantified risk indicators and the substitute indicators;

[0007] Update the risk indicator library according to the key risk indicators and divide them into levels, calculate the importance weight of each risk indicator in each level; obtain the indicator comprehensive weight matrix based on the importance weight of each risk indicator and the relationship between each risk indicator in each level;

[0008] Based on the risk indicator quantitative measurement model, the indicator quantitative values ​​of key risk indicators are obtained, based on the indicator quantitative values ​​and the indicator comprehensive weight matrix, a comprehensive score is obtained, and social risk warning is carried out based on the comprehensive score.

[0009] Optionally, the risk indicator library includes but is not limited to social stability dimension, economic development dimension, environmental protection dimension and public safety dimension; a number of indicators are set under each dimension.

[0010] Optionally, the process of constructing a risk indicator quantitative measurement model includes: constructing an initial indicator quantitative measurement model based on the correlation between unquantified risk indicators and alternative indicators; obtaining historical data of unquantified risk indicators and alternative indicators; preprocessing the historical data to obtain a risk indicator quantitative analysis data set; based on the risk indicator quantitative data set and a machine learning algorithm, training and optimizing the initial indicator quantitative measurement model to obtain a risk indicator quantitative measurement model.

[0011] Optionally, the process of obtaining importance weights includes

[0012] The updated risk indicator library is divided into multiple levels using the analytic hierarchy process (AHP). The importance score of each indicator in each level is obtained based on expert opinions and domain knowledge. The judgment matrix corresponding to the level is obtained based on the importance score. The judgment matrix is ​​tested for consistency. If the test is passed, the eigenvalue method is used to calculate the eigenvector corresponding to the maximum eigenvalue of the judgment matrix to obtain the weight vector of the risk indicators in each level. The indicator interaction matrix is ​​obtained based on the relationship between the risk indicators in each level. The weight vector of the indicators at each level is corrected based on the indicator interaction matrix. The comprehensive weight matrix is ​​obtained based on the corrected weight vector.

[0013] Optionally, multiple levels include but are not limited to a target layer, a criterion layer, and a solution layer; a comprehensive weight matrix is ​​obtained by multiplying the corrected weight vector with the indicator weight of the solution layer.

[0014] Optionally, the process for obtaining a composite score includes:

[0015] The quantitative values ​​of the indicators are standardized, and the comprehensive weight matrix and the quantitative values ​​of the indicators after the standardization are multiplied by the matrix operation library to obtain the comprehensive score of each risk indicator.

[0016] Optionally, the process of conducting social risk early warning based on comprehensive scores includes:

[0017] A risk level threshold is preset, and the current risk level is judged based on the decision tree algorithm, the risk level threshold and the comprehensive score, and a warning signal of a corresponding type is generated and a warning is issued according to the current risk level.

[0018] Optionally, the method further includes:

[0019] Obtain relevant data on key risk indicators and construct a time series data set; perform time series analysis on key risk indicators to obtain the changing trend of risk indicators; use feature engineering methods to extract new features of risk evolution from the dynamic changing trend of risk indicators and construct a risk evolution feature set; based on the risk evolution feature set, use a clustering algorithm to cluster risks and obtain different risk categories; analyze the characteristic attributes of each type of risk, construct a risk assessment model, and calculate the probability of risk occurrence and the degree of impact in real time; based on the risk assessment results, update the weights and thresholds of key risk indicators.

[0020] The present invention also provides a social risk early warning system based on multi-source data, comprising:

[0021] The risk indicator system construction module is used to analyze the causes of risks, establish an indicator library, and screen key indicators;

[0022] The risk indicator quantification module is used to explore indicator substitutes, establish measurement models and perform quantitative processing;

[0023] The indicator weight calculation module is used to divide the levels, calculate the weights and construct the weight matrix; the risk assessment and early warning module is used to calculate the comprehensive score, determine the risk level and generate early warning signals;

[0024] Dynamic optimization and adjustment module, used to analyze changing trends, capture new features and optimize the indicator system.

[0025] Compared with the prior art, the present invention has the following advantages and technical effects:

[0026] The present invention discloses a social risk early warning method based on multi-source data. First, multi-source heterogeneous data related to social risks are obtained to construct a risk indicator library; based on the multi-source heterogeneous data, an evaluation value of each indicator in the risk indicator library is obtained; based on the clustering algorithm and the evaluation value analysis, key risk indicators are obtained; secondly, unquantified indicators are screened out from the key risk indicators, alternative indicators of the unquantified risk indicators are matched, and a risk indicator quantitative measurement model is constructed based on the unquantified risk indicators and the alternative indicators; then, the risk indicator library is updated according to the key risk indicators and hierarchical division is performed, and the importance weight of each risk indicator is calculated in each hierarchy; based on the importance weight of each risk indicator and the relationship between each risk indicator in each hierarchy, an indicator comprehensive weight matrix is ​​obtained; finally, based on the risk indicator quantitative measurement model, the indicator quantitative value of the key risk indicator is obtained, based on the indicator quantitative value and the indicator comprehensive weight matrix, a comprehensive score is obtained, and social risk early warning is performed based on the comprehensive score.

[0027] The present invention aims at social risk indicators that are difficult to quantify, constructs indicator substitutes and measurement models through expert experience and data analysis, and realizes indicator quantification. At the same time, the analytic hierarchy process is used to construct a risk indicator system and assign weights to form a comprehensive risk scoring mechanism. The present invention also continuously collects risk data and uses time series analysis and other methods to dynamically adjust the indicator system to ensure the effectiveness of early warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0029] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0030] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0031] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0032] Embodiment 1

[0033] like Figure 1 As shown, this embodiment provides a social risk early warning method based on multi-source data, including:

[0034] Obtain multi-source heterogeneous data related to social risks and build a risk indicator library; obtain the evaluation value of each indicator in the risk indicator library based on multi-source heterogeneous data; obtain key risk indicators based on clustering algorithms and evaluation value analysis;

[0035] In some specific implementations, the risk indicator library includes, but is not limited to, social stability dimension, economic development dimension, environmental protection dimension, and public safety dimension; a number of indicators are set under each dimension.

[0036] Furthermore, multi-source heterogeneous data related to social risks are obtained, including text data, statistical data, and event data; a risk indicator library is constructed according to the types and characteristics of social risks, and the risk indicator library includes multiple dimensions and multiple indicators; the multi-source heterogeneous data is preprocessed to obtain preprocessed data that meets the requirements of the risk indicator library; the preprocessed data is used to calculate the index value of each indicator according to the definition and calculation method of each indicator in the risk indicator library; based on the calculated index value, a clustering algorithm or a classification algorithm is used to evaluate the current social risk status, and by comparing the differences in index values ​​between different clusters, indicators that have an important impact on the distinction of risk status, namely key risk indicators, can be identified. For example, if the values ​​of an indicator in high-risk and low-risk clusters are significantly different, then this indicator is likely to be a key risk indicator reflecting the social risk status.

[0037] For example, taking the social risk of "mass incidents" as an example, how to conduct risk assessment and early warning is elaborated in detail. First, it is necessary to obtain multi-source heterogeneous data related to mass incidents. Text data can include news reports, social media posts, online forum comments, etc., from which information such as public sentiment, event causes, and development trends can be extracted. Statistical data can include demographic data, economic development indicators, unemployment rates, etc. These data can reflect the social structure and development status and provide background information for the occurrence of mass incidents. Event data can include records of historical mass incidents, such as time, location, number of participants, causes, results, etc. These data can be used to establish event patterns and predict the probability of future events. At the same time, collect news reports, online public opinion data, and statistical data such as economic development indicators, unemployment rates, and population density in the region during the same period. Secondly, it is necessary to build a risk indicator library. For mass incidents, the indicator library can include multiple dimensions such as social stability, economic development, environmental protection, and public safety. Multiple specific indicators can be set under each dimension. For example, the social stability dimension can include indicators such as public satisfaction, social tension index, and the incidence of mass incidents. The economic development dimension can include indicators such as GDP growth rate, unemployment rate, and income gap among residents. The environmental protection dimension can include indicators such as air quality index, water pollution level, and environmental complaint rate. The public safety dimension can include indicators such as crime rate, traffic accident rate, and fire rate. Then, the collected multi-source heterogeneous data needs to be preprocessed to meet the requirements of the risk indicator library. For example, for text data, it is necessary to perform word segmentation, stop word removal, sentiment analysis, and other processing to extract key information and public sentiment. For statistical data, it is necessary to perform standardization processing to facilitate comparison between different indicators. For event data, it is necessary to perform classification, statistics, and aggregation processing to extract event characteristics and laws. For example, the collected news reports are segmented, keywords are extracted, and sentiment analysis is performed to determine whether the public sentiment is positive, negative, or neutral. The collected economic indicators are standardized, such as converting the GDP growth rate into a value between 0 and 1. Next, the indicator value of each indicator needs to be calculated according to the definition and calculation method of each indicator in the risk indicator library. For example, public satisfaction can be obtained through questionnaire surveys, the social tension index can be calculated by analyzing online public opinion data, and the incidence of mass incidents can be calculated by statistical historical event data. By calculating these index values, we can quantify various aspects of social risk. For example, by analyzing and processing news reports and social media data, we can calculate that public satisfaction is 0.6 (assuming the full score is 1), the social tension index is 0.8 (assuming the full score is 1), and the incidence of mass incidents is 0.1 (per 10,000 people per year). After obtaining the index values, clustering algorithms or classification algorithms can be used to evaluate the current social risk status.For example, the K-Means clustering algorithm can be used to cluster the social risk status of different regions or time periods to identify high-risk areas or time periods.

[0038] Screen out unquantified indicators from key risk indicators, match alternative indicators for unquantified risk indicators, and build a quantitative measurement model for risk indicators based on unquantified risk indicators and alternative indicators;

[0039] In some specific embodiments, the process of constructing a risk indicator quantitative measurement model includes: constructing an initial indicator quantitative measurement model based on the correlation between unquantified risk indicators and alternative indicators; obtaining historical data of unquantified risk indicators and alternative indicators; preprocessing the historical data to obtain a risk indicator quantitative analysis data set; based on the risk indicator quantitative data set and a machine learning algorithm, training and optimizing the initial indicator quantitative measurement model to obtain a risk indicator quantitative measurement model.

[0040] Specifically, according to the business characteristics of risk indicators, a method combining expert experience judgment and data analysis is used to find suitable indicator substitutes for risk indicators that are difficult to quantify directly. By analyzing the correlation between risk indicators and indicator substitutes, a quantitative measurement model for risk indicators is constructed, and expert experience judgment is converted into a quantitative mathematical model. Historical data related to risk indicators is obtained, and the data of indicator substitutes are collected and preprocessed to form a data set for quantitative analysis of risk indicators. Machine learning algorithms are used to train and optimize the quantitative measurement model for risk indicators. Through model verification and evaluation, the quantitative accuracy of the model is continuously optimized through iteration.

[0041] Furthermore, risk indicators that are difficult to quantify directly often need to rely on expert experience for judgment. For example, it is difficult to measure the indicator of social stability with an exact value. At this time, some quantifiable indicator substitutes can be found. For example, indicators such as the number of mass incidents, the negative sentiment index of online public opinion, and the satisfaction of key social groups can be used as substitutes for social stability. Although these indicators are not completely equivalent to social stability, they are closely related to it and can reflect the changing trend of social stability to a certain extent. By analyzing the correlation between these indicators and social stability, a quantitative measurement model can be constructed. Constructing a quantitative measurement model for risk indicators is essentially to convert expert experience judgment into a quantitative mathematical model. For example, experts believe that when the number of mass incidents exceeds a certain threshold, social stability will significantly decrease. This expert experience can be converted into a mathematical model through machine learning algorithms such as logistic regression. For example, the model can set a threshold. When this threshold is exceeded, the value of the social stability indicator output by the model will be significantly reduced. Obtain historical data related to risk indicators. For example, data such as the number of mass incidents, the negative sentiment index of online public opinion, and the satisfaction of key social groups in the past ten years can be collected. At the same time, these data need to be preprocessed, such as removing outliers, filling missing values, and performing data standardization to ensure data quality and consistency. In this way, a data set for quantitative analysis of risk indicators is formed. Machine learning algorithms are used to train and optimize the risk indicator quantitative measurement model. For example, the collected historical data is used to train the logistic regression model, and the model parameters are adjusted to make the model output as close as possible to the expert's judgment on social stability. After the model training is completed, the model needs to be verified and evaluated, for example, the model is tested using a part of the data that is not used for training to evaluate the model's prediction accuracy. If the model's accuracy is not high enough, the model needs to be further optimized, such as adjusting the model parameters, replacing the machine learning algorithm, etc. This process needs to be iterated continuously until the quantitative accuracy of the model meets the requirements. The optimized risk indicator quantitative measurement model is applied to actual business scenarios. For example, when a new mass incident occurs, the relevant data is input into the model, and the model will output a quantitative social stability index value. The quantified risk indicator value is analyzed and evaluated. For example, if the social stability index value output by the model is lower than the preset threshold, it is considered that there is a risk of social stability and corresponding control measures need to be taken. For example, increase police deployment and strengthen public opinion monitoring. Continue to track changes in risk indicators, for example, regularly collect data such as the number of mass incidents, negative sentiment index of online public opinion, and satisfaction of key social groups, and input them into the model to monitor the changing trend of social stability. Regularly retrain and optimize the quantitative measurement model. For example, over time, the social environment may change, and the original model may no longer be applicable.At this time, the model needs to be retrained with new data to ensure the accuracy and timeliness of the risk indicator quantification. For example, if it is found that the weight of the impact of the satisfaction of key social groups on social stability has changed, the corresponding parameters in the model need to be adjusted.

[0042] Update the risk indicator library according to the key risk indicators and divide them into levels, calculate the importance weight of each risk indicator in each level; obtain the indicator comprehensive weight matrix based on the importance weight of each risk indicator and the relationship between each risk indicator in each level;

[0043] In some specific implementations, the process of obtaining the importance weight includes:

[0044] The updated risk indicator library is divided into multiple levels using the analytic hierarchy process (AHP). The importance score of each indicator in each level is obtained based on expert opinions and domain knowledge. The judgment matrix corresponding to the level is obtained based on the importance score. The judgment matrix is ​​tested for consistency. If the test is passed, the eigenvalue method is used to calculate the eigenvector corresponding to the maximum eigenvalue of the judgment matrix to obtain the weight vector of the risk indicators in each level. The indicator interaction matrix is ​​obtained based on the relationship between the risk indicators in each level. The weight vector of the indicators at each level is corrected based on the indicator interaction matrix. The comprehensive weight matrix is ​​obtained based on the corrected weight vector.

[0045] In some specific implementations, multiple levels include but are not limited to a target level, a criterion level, and a solution level; a comprehensive weight matrix is ​​obtained by multiplying the corrected weight vector with the indicator weight of the solution level.

[0046] Specifically, according to the risk indicator system, the hierarchical analysis method is used to divide the indicators into multiple levels such as the target layer, the criterion layer and the scheme layer to form a hierarchical structure model. In each level, by consulting experts and using questionnaires, etc., the subjective scores of experts on the importance of each indicator are obtained to obtain the judgment matrix. The judgment matrix is ​​tested for consistency. If it passes the test, the eigenvalue method is used to calculate the eigenvector corresponding to the maximum eigenvalue of the judgment matrix to obtain the weight vector of the indicator at this level. At the criterion level, by analyzing the causal relationship and correlation between indicators, the indicator interaction matrix is ​​constructed, and the fuzzy mathematical method is used to operate the matrix to obtain the modified weight after considering the interaction of indicators. At the scheme level, the weight of the upper criterion is multiplied by the weight of the indicator at this level to obtain the comprehensive weight vector.

[0047] Furthermore, the analytic hierarchy process (AHP) is a multi-criteria decision-making method that combines qualitative analysis with quantitative analysis. It decomposes complex decision-making problems into multiple levels, judges the relative importance of each factor through pairwise comparison, and finally obtains the comprehensive weight of each solution. In this example, the analytic hierarchy process will be used to build an enterprise information system security risk assessment model. First, the goal of risk assessment is set as "ensuring the security of enterprise information systems". Then, according to the common risks of information system security, the indicators are divided into three levels: target layer, criterion layer and solution layer. The target layer has only one indicator, namely "ensuring the security of enterprise information systems". The criterion layer includes four aspects: "physical security", "network security", "data security" and "management security". The solution layer is refined to specific risk indicators, such as "computer room environment security" and "equipment security" under physical security; "firewall configuration" and "intrusion detection" under network security; "data backup" and "data encryption" under data security; "security policy" and "security audit" under management security. In this way, a hierarchical model is formed. Next, it is necessary to compare the indicators of each level in pairs to determine their relative importance. For example, at the criterion level, it is necessary to compare the relative importance of physical security, network security, data security, and management security. The judgment matrix can be obtained through expert scoring. Assume that the expert scoring results for the four aspects are as follows: the importance of physical security relative to network security is 3, the importance relative to data security is 2, and the importance relative to management security is 1; the importance of network security relative to data security is 2, and the importance relative to management security is 1; the importance of data security relative to management security is 2. Based on these scoring results, a judgment matrix can be constructed. Then, the judgment matrix needs to be checked for consistency to ensure the logical consistency of the expert scoring. If the consistency check passes, the eigenvalue method can be used to calculate the eigenvector corresponding to the maximum eigenvalue of the judgment matrix to obtain the weight vector of the indicator at this level. For example, after calculation, the weight vectors of the criterion layer are: physical security 0.37, network security 0.28, data security 0.21, and management security 0.14. This shows that in the process of ensuring the security of enterprise information systems, physical security is the most important, followed by network security, then data security, and finally management security. At the criterion level, the interactive effects between indicators also need to be considered. For example, there is a certain correlation between network security and data security, and the degree of perfection of network security measures will directly affect the level of data security. By constructing an indicator interaction matrix and using fuzzy mathematics to calculate, the modified weight after considering the interaction of indicators can be obtained. At the solution level, the weight of the upper-level criterion needs to be multiplied by the weight of the indicator at this level to obtain a comprehensive weight vector. For example, the "firewall configuration" indicator belongs to the network security criterion. Assuming that its weight in the solution layer is 0.6, its comprehensive weight is 0.28*0.6=0.168.Finally, the indicator weight vectors at each level are combined according to the hierarchical model to construct a comprehensive weight matrix of the multi-level risk indicator system, which can quantify the contribution of risk indicators to the overall risk.

[0048] Based on the risk indicator quantitative measurement model, the quantitative values ​​of key risk indicators are obtained, and based on the indicator quantitative values ​​and the indicator comprehensive weight matrix, a comprehensive score is obtained, and social risk warning is carried out based on the comprehensive score.

[0049] In some specific embodiments, the process of obtaining the comprehensive score includes:

[0050] The quantitative values ​​of indicators are standardized, and the comprehensive weight matrix and the quantitative values ​​of indicators after standardization are multiplied by the matrix operation library to obtain the comprehensive score of each risk indicator.

[0051] In some specific implementations, the process of performing social risk warning based on the comprehensive score includes:

[0052] A risk level threshold is preset, and the current risk level is judged based on the decision tree algorithm, the risk level threshold and the comprehensive score, and a warning signal of a corresponding type is generated and a warning is issued according to the current risk level.

[0053] Specifically, data preprocessing technology is used to standardize the quantitative values ​​of risk indicators to ensure the accuracy and consistency of data during matrix multiplication operations. The matrix operation library is used to implement the product operation of the quantitative value and the weight matrix to obtain the comprehensive score of each risk indicator. According to the preset risk level threshold, the decision tree algorithm is used to classify the comprehensive score to determine the current risk level. If the comprehensive score exceeds the high risk threshold, an emergency warning signal is generated through signal processing technology, otherwise, a conventional risk warning signal is generated. Through the real-time data monitoring system, the changes in risk indicators are continuously tracked to dynamically adjust the generation of warning signals. Machine learning models, such as support vector machines, are used to train and optimize risk assessment models to improve the accuracy of risk judgment. Finally, the generated warning signal is released to relevant departments and the public through the information release system to ensure the timely transmission of risk information.

[0054] Furthermore, data preprocessing technology is to ensure the accuracy and consistency of data during calculation. For example, suppose there are two risk indicators: server downtime (hours) and number of user complaints (number). The value range of server downtime may be 0 to hundreds, while the value range of user complaints may be 0 to thousands. Since the dimensions and orders of magnitude of the two indicators are different, direct calculation will lead to deviation in the results. Therefore, it is necessary to standardize the data, such as using the Min-Max standardization method to scale the values ​​of all indicators to between 0 and 1, so as to ensure the fairness and consistency of different indicators during calculation. The matrix operation library can perform matrix multiplication operations efficiently. For example, suppose the risk indicator weight matrix is ​​a 2x2 matrix, which represents the influence weight of the two criterion layer indicators on the target layer risk. The quantitative value is a 1x2 matrix, which represents the standardized values ​​of the two indicators. Through the matrix operation library, the two matrices can be multiplied to obtain a 1x1 matrix, which is the comprehensive score of the risk. The decision tree algorithm can classify the comprehensive score according to the preset risk level threshold. For example, assume that the risk level is divided into three levels: low, medium, and high, and the corresponding thresholds are 0.3, 0.6, and 0.9, respectively. If the calculated comprehensive score is 0.7, then according to the classification rules of the decision tree, the current risk level is judged to be medium risk. Signal processing technology can generate different types of early warning signals according to the risk level. For example, if the risk level is high risk, an emergency early warning signal is generated, such as sending an alarm message via SMS or email, and starting an emergency response mechanism. If the risk level is low risk, a conventional risk early warning signal is generated, such as displaying risk prompt information on the system interface. The real-time data monitoring system can continuously track changes in risk indicators and dynamically adjust early warning signals. For example, if the server downtime continues to increase and exceeds the preset threshold, the system will automatically increase the risk level from low risk to medium risk and update the early warning signal. Machine learning models can train and optimize risk assessment models. For example, a support vector machine model can be used to learn historical risk data, find out the relationship between risk indicators and risk levels, and continuously optimize model parameters to improve the accuracy of risk judgment. This can more accurately predict risk levels based on new data and generate more effective early warning signals. The information release system can promptly transmit the generated warning signals to relevant departments and the public. For example, the system can publish warning information to websites, apps or other information platforms to ensure that risk information can be promptly transmitted to those who need it so that appropriate measures can be taken.

[0055] In some specific implementations, risk-related data are continuously collected, and through methods such as time series analysis, the dynamic changing trends of risk indicators are characterized, new features of risk evolution are captured in a timely manner, and the risk indicator system is adjusted and optimized.

[0056] Specifically, relevant data of key risk indicators are obtained to construct a time series data set; time series analysis is performed on key risk indicators to obtain the changing trend of risk indicators; feature engineering methods are used to extract new features of risk evolution from the dynamic changing trend of risk indicators and to construct a risk evolution feature set; based on the risk evolution feature set, clustering algorithms are used to cluster risks and obtain different risk categories; characteristic attributes of each type of risk are analyzed, a risk assessment model is constructed, and the probability of risk occurrence and the degree of impact are calculated in real time; based on the risk assessment results, the weights and thresholds of key risk indicators are updated.

[0057] Exemplarily, specifically, the continuous acquisition of risk-related data is the basis for risk assessment and early warning. Various data related to social risks, such as unemployment rate, crime rate, environmental pollution index, network public opinion data, etc., can be obtained through various channels, such as government public data, news media reports, social network information, sensor data, etc. These data form a time series data set, which provides data support for subsequent analysis. After constructing the time series data set, it is necessary to perform time series analysis on risk indicators. For example, the unemployment rate change trend of a city in the past ten years can be analyzed to observe whether there are cyclical fluctuations, long-term trends or mutation points. Common time series analysis methods include moving average, exponential smoothing, ARIMA model, etc. Through these methods, the dynamic change trend of risk indicators can be understood, providing a basis for subsequent risk assessment. Suppose that through time series analysis, it is found that the unemployment rate of a city has continued to rise in the past year, and the rate of increase has gradually accelerated, which may indicate that the city has a higher economic risk. In order to more accurately characterize the characteristics of risk evolution, feature engineering methods need to be adopted. For example, some new features can be extracted from the time series data of unemployment rate, such as the growth rate, volatility, and correlation with other economic indicators of unemployment rate. Suppose that the correlation between the growth rate of unemployment rate and the consumer confidence index is gradually increasing, which may indicate that economic risks are spreading to the consumer sector. After extracting the risk evolution characteristics, clustering algorithms can be used to cluster risks. For example, the K-Means algorithm can be used to classify risk events with similar characteristics. Suppose that risk events are clustered into three categories: economic risk, social risk, and environmental risk based on indicators such as unemployment rate, inflation rate, and consumer confidence index. For each type of risk, a targeted risk assessment model needs to be built. For example, for economic risk, a model based on logistic regression can be built to predict the probability of an economic crisis in the future. Suppose the model predicts that the probability of an economic crisis in the next year is 20%, and predicts that if an economic crisis occurs, it will cause a 5% decrease in GDP. According to the risk assessment results, the weights and thresholds of risk indicators can be dynamically adjusted. For example, if the economic risk assessment model predicts a high probability of an economic crisis in the future, the weights of economic-related indicators in the risk indicator system can be increased, and the warning threshold can be lowered so that early warning signals can be issued earlier. The updated risk indicator system will be applied to risk monitoring and early warning. For example, real-time monitoring of changes in indicators such as unemployment rate and inflation rate. Once these indicators exceed the preset threshold, the warning signal will be triggered immediately. Assuming that the unemployment rate suddenly rises sharply and exceeds the preset threshold, the system will automatically generate an emergency warning signal and notify relevant departments to take countermeasures. Finally, it is necessary to continuously optimize the risk indicator system and risk assessment model based on the risk monitoring and early warning results. For example, the accuracy and effectiveness of historical early warning signals can be analyzed to find out the deficiencies of the model and make improvements.Through continuous optimization, the accuracy and effectiveness of risk management can be improved, and social stability and development can be better guaranteed. For example, by analyzing historical data, it is found that the previous model is not able to predict sudden public health events. Then it is necessary to improve the model and add relevant indicators and data to improve the model's ability to predict such risks.

[0058] In some specific implementations, when new risk influencing factors emerge, they are promptly incorporated into the indicator system, and through data mining and other technologies, their association with existing indicators is explored, and the indicator quantification model and weight matrix are calibrated to ensure the effectiveness of risk warning.

[0059] Specifically, obtain relevant data on newly emerging risk influencing factors and incorporate them into the indicator system as new candidate indicators; use data mining technology to analyze the correlation patterns between the newly included candidate indicators and existing indicators; based on the mined correlation patterns, determine whether there is a significant logical correlation between the new indicator and the existing indicators; if there is a significant correlation between the new indicator and the existing indicators, the new indicator will be formally incorporated into the indicator system, and the weights of each indicator in the indicator quantification model will be calibrated according to the correlation patterns; if there is no significant correlation between the new indicator and the existing indicators, it will not be included in the indicator system for the time being, and continue to monitor and collect relevant data of the indicator, and then conduct correlation analysis after accumulating a certain amount; apply the calibrated indicator quantification model and weight matrix to the risk warning system to evaluate and warn new risk events; continuously track the effectiveness of risk warnings, and dynamically optimize and adjust the indicator system and quantitative model based on the feedback of warning results to ensure continuous improvement of risk warning capabilities.

[0060] Furthermore, obtain the data related to the newly emerged risk influencing factors and incorporate them into the indicator system as new candidate indicators. For example, when assessing corporate credit risk, in addition to traditional financial indicators (such as debt-to-asset ratio and current ratio), non-financial factors such as corporate social responsibility performance and environmental protection investment can also be considered as new candidate indicators. Collect data on these non-financial factors, such as the amount of corporate public welfare donations and carbon emissions, and incorporate them into the candidate indicator set of the credit risk indicator system. For the newly included candidate indicators, use data mining technology to analyze the association rules between them and existing indicators. For example, you can use association rule mining algorithms to analyze the association between corporate social responsibility performance (such as the amount of public welfare donations) and financial indicators (such as profitability). If it is found that companies with higher amounts of public welfare donations usually have stronger profitability, it indicates that there is a positive correlation between the two indicators. According to the mined association rules, determine whether there is a significant correlation between the new indicator and the existing indicators in logic. If the data mining results show that there is a significant statistical correlation between the new indicator and the existing indicators, and this correlation can also be explained in business logic, it is considered that the new indicator has a significant correlation with the existing indicators. For example, if data analysis shows that there is a positive correlation between environmental protection investment and corporate long-term value, and this relationship can also be explained from the perspective of corporate sustainable development, it can be considered that there is a significant correlation between the two indicators. If there is a significant correlation between the new indicator and the existing indicator, the new indicator will be formally included in the indicator system, and the weights of each indicator in the indicator quantification model will be calibrated according to the correlation law. For example, if it is determined that there is a significant negative correlation between the corporate social responsibility performance and credit risk, the social responsibility indicator can be formally included in the credit risk assessment model, and the weight of the indicator can be adjusted according to the strength of the correlation. The stronger the correlation, the greater the weight. Doing so can more comprehensively assess the corporate credit risk and improve the accuracy of risk assessment. If there is no significant correlation between the new indicator and the existing indicator, it will not be included in the indicator system for the time being, and the relevant data of the indicator will continue to be monitored and collected, and the correlation analysis will be carried out after a certain amount of data is accumulated. For example, assuming that a new indicator is "corporate employee satisfaction", preliminary analysis finds that there is no obvious correlation between it and corporate credit risk. At this time, the indicator will not be immediately incorporated into the credit risk assessment model, but the data of the indicator will continue to be collected and monitored. For example, the company's employee satisfaction data and the company's subsequent credit performance data will be continuously collected. After a certain amount of data is accumulated, a more in-depth analysis will be conducted to determine whether the indicator has a potential correlation with credit risk. The calibrated indicator quantification model and weight matrix are applied to the risk warning system to evaluate and warn new risk events. For example, the updated credit risk assessment model is applied to the evaluation of new loan applications, and the credit risk score of the applicant company is calculated based on the various indicator data, and risk warnings are issued based on preset thresholds.If the credit risk score exceeds the warning threshold, the system will issue a warning signal to remind relevant personnel to pay attention to the risk of the loan application. Continue to track the effectiveness of risk warnings, and dynamically optimize and adjust the indicator system and quantitative model based on the feedback of the warning results to ensure the continuous improvement of risk warning capabilities. For example, continue to track the warning results of the credit risk warning system and analyze the accuracy and false alarm rate of the warning. If it is found that the warning effect of a certain indicator is not good, it is necessary to re-evaluate the effectiveness and weight of the indicator, and even consider removing it from the indicator system. Through continuous optimization and adjustment, the accuracy and reliability of the risk warning system can be improved.

[0061] This embodiment also provides a social risk early warning system based on multi-source data, including:

[0062] The risk indicator system construction module is used to analyze the causes of risks, establish an indicator library, and screen key indicators;

[0063] The risk indicator quantification module is used to explore indicator substitutes, establish measurement models and perform quantitative processing;

[0064] The indicator weight calculation module is used to divide the levels, calculate the weights and construct the weight matrix; the risk assessment and early warning module is used to calculate the comprehensive score, determine the risk level and generate early warning signals;

[0065] Dynamic optimization and adjustment module, used to analyze changing trends, capture new features and optimize the indicator system.

[0066] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A social risk early warning method based on multi-source data, characterized in that: The following steps are involved: Obtain multi-source heterogeneous data related to social risks and build a risk indicator library; Obtaining an evaluation value of each indicator in a risk indicator library based on the multi-source heterogeneous data; Obtain key risk indicators based on clustering algorithms and assessment value analysis; Screening out unquantified indicators from the key risk indicators, matching substitute indicators of the unquantified risk indicators, and constructing a risk indicator quantitative measurement model based on the unquantified risk indicators and the substitute indicators; Update the risk indicator library according to the key risk indicators and divide them into levels, and calculate the importance weight of each risk indicator in each level; Obtain the comprehensive weight matrix of indicators based on the importance weight of each risk indicator and the relationship between each risk indicator in each level; Based on the risk indicator quantitative measurement model, the indicator quantitative values ​​of key risk indicators are obtained, based on the indicator quantitative values ​​and the indicator comprehensive weight matrix, a comprehensive score is obtained, and social risk warning is carried out based on the comprehensive score.

2. The method according to claim 1, characterized in that The risk indicator library includes but is not limited to social stability dimension, economic development dimension, environmental protection dimension and public safety dimension; a number of indicators are set under each dimension.

3. The method according to claim 1, characterized in that The process of constructing a risk indicator quantitative measurement model includes: constructing an initial indicator quantitative measurement model based on the correlation between unquantified risk indicators and alternative indicators; obtaining historical data of unquantified risk indicators and alternative indicators; preprocessing the historical data to obtain a risk indicator quantitative analysis data set; based on the risk indicator quantitative data set and a machine learning algorithm, training and optimizing the initial indicator quantitative measurement model to obtain a risk indicator quantitative measurement model.

4. The method according to claim 1, characterized in that: The process of obtaining importance weights includes The updated risk indicator library is divided into multiple levels using the analytic hierarchy process. Obtain the importance score of each indicator in each level based on expert opinions and domain knowledge, obtain the judgment matrix corresponding to the level based on the importance score, and perform consistency test on the judgment matrix. If the test is passed, use the eigenvalue method to calculate the eigenvector corresponding to the maximum eigenvalue of the judgment matrix to obtain the weight vector of the risk indicators in each level; obtain the indicator interaction matrix based on the relationship between the risk indicators in each level; The weight vectors of the indicators at each level are modified based on the indicator interaction matrix; and a comprehensive weight matrix is ​​obtained based on the modified weight vectors.

5. The method according to claim 4, characterized in that Multiple levels include but are not limited to target level, criterion level, and solution level; a comprehensive weight matrix is ​​obtained by multiplying the corrected weight vector with the indicator weight of the solution level.

6. The method according to claim 1, characterized in that The process for obtaining a composite score includes: The quantitative values ​​of the indicators are standardized, and the comprehensive weight matrix and the quantitative values ​​of the indicators after the standardization are multiplied by the matrix operation library to obtain the comprehensive score of each risk indicator.

7. The method according to claim 1, characterized in that The process of social risk early warning based on comprehensive scoring includes: A risk level threshold is preset, and the current risk level is judged based on the decision tree algorithm, the risk level threshold and the comprehensive score, and a warning signal of a corresponding type is generated and a warning is issued according to the current risk level.

8. The method according to claim 1, characterized in that The method further comprises: Obtain relevant data on key risk indicators and construct a time series data set; perform time series analysis on key risk indicators to obtain the changing trend of risk indicators; use feature engineering methods to extract new features of risk evolution from the dynamic changing trend of risk indicators and construct a risk evolution feature set; based on the risk evolution feature set, use a clustering algorithm to cluster risks and obtain different risk categories; analyze the characteristic attributes of each type of risk, construct a risk assessment model, and calculate the probability of risk occurrence and the degree of impact in real time; based on the risk assessment results, update the weights and thresholds of key risk indicators.

9. A social risk early warning system based on multi-source data, characterized in that: include: The risk indicator system construction module is used to analyze the causes of risks, establish an indicator library, and screen key indicators; The risk indicator quantification module is used to explore indicator substitutes, establish measurement models and perform quantitative processing; The indicator weight calculation module is used to divide the levels, calculate the weights and construct the weight matrix; the risk assessment and early warning module is used to calculate the comprehensive score, determine the risk level and generate early warning signals; Dynamic optimization and adjustment module, used to analyze changing trends, capture new features and optimize the indicator system.

Images