Method and device for excavating high-diving cable in real time

Abstract

The invention relates to the field of fusion of financial science and technology and information technology, and particularly provides a method and device for mining high-potential clues in real time, and the method comprises the following steps: S1, data collection and convergence; s2, managing and processing multi-source data; s3, clue analysis and mining; and S4, the clues are pushed in real time. Compared with the prior art, the method can ensure that the user can obtain the latest risk dynamic state of the debtor in the first time so as to make a quick response and decision.

Description

Technical Field

[0001] The present invention relates to the field of integration of financial technology and information technology, and specifically provides a method and device for real-time mining of high-potential clues. Background Art

[0002] In the financial industry, the management and collection of non-performing assets is an important part of maintaining financial stability and protecting the rights and interests of creditors. Traditional collection methods are limited by inherent drawbacks such as low efficiency, high cost and difficulty in accurately predicting judicial risks. In particular, in the process of judicial collection of insufficient capital contribution, false capital contribution, capital reduction and one-person companies, the collection work is full of uncertainty due to the complex legal procedures and changeable debtor behaviors.

[0003] In view of this, the requirement for real-time data has become particularly urgent. An efficient and intelligent device and system must be able to capture and process multi-source data from corporate business information databases, honorary qualifications, public projects, judicial litigation information databases, social media, industry reports, etc. in real time, and timely discover the risk attributes of corporate debtors by deeply mining multi-source data, and predict the probability of enterprises becoming high-potential clues. With the rapid growth of Internet data of financial industry enterprises, real-time data processing has become increasingly important. Traditional batch processing calculation methods can no longer meet real-time requirements.

[0004] Stream computing technology has become the future trend of real-time data processing. Stream computing is a method for processing large-scale, high-speed inflow data streams. It can process and analyze data in real time, thereby providing real-time decision support. Applying stream computing technology to the bad asset recovery scenario, timely responding to the real-time updates of multiple data sources of the enterprise and collecting, acquiring, analyzing and mining the updated enterprise data is the key to ensuring that the system can accurately reflect the latest status of the debtor, timely discover potential risks and respond quickly. It can guide lawyers and other users to flexibly adjust strategies during the recovery process to achieve the maximum debt recovery benefits at the lowest cost, thereby effectively promoting the healthy, stable and sustainable development of the financial market.

[0005] Therefore, how to capture data in real time, optimize the collection process through technological empowerment, reduce operating costs, improve collection efficiency and success rate, and provide an intelligent path is an urgent problem that technical personnel in this field need to solve. Summary of the invention

[0006] The present invention aims at the above-mentioned deficiencies of the prior art and provides a method for mining high-potential clues in real time with strong practicality.

[0007] A further technical task of the present invention is to provide a method and device for real-time mining of high-potential clues that is reasonably designed, safe and applicable.

[0008] The technical solution adopted by the present invention to solve the technical problem is:

[0009] A method for mining high potential clues in real time, comprising the following steps:

[0010] S1, data collection and aggregation;

[0011] S2, multi-source data management and processing;

[0012] S3, clue analysis and mining;

[0013] S4. Real-time push of clues.

[0014] Furthermore, in step S1, the aggregated enterprise multi-source data includes three forms: public data collection, API interface docking, and batch push data;

[0015] Use automated collection technology to collect enterprise-related information from various public channels. By presetting a list of various public authoritative channels that need to be collected, automated collection technology automatically parses web pages from various channels, analyzes the web page data structure, and extracts the required web page content;

[0016] The API interface form is to obtain third-party data by actively calling the third-party interface, and the batch pushed data is to obtain batch data pushed by a third party at a scheduled time by batch receiving.

[0017] Furthermore, in step S2, the enterprise multi-source data gathered in various ways will be temporarily stored in Redis, and then the SparkSQL component will be used to perform distributed reading of large batches of data, perform data cleaning, remove inconsistent formats and process noise data, and form a structured and standardized data set.

[0018] Furthermore, multi-source enterprise data will be aligned and standardized according to the enterprise's unique identifier, and stored in different dimensions according to unified storage standards; the real-time processed data will be sent to Kafka in real time. Kafka is a message component that receives real-time data streams. After passing through the Kafka pipeline component, the real-time data will be sent to the real-time data receiving component for clue analysis and mining. The real-time processed data will be sent to the big data storage media HDFS and tiDB cluster at the same time for batch data storage.

[0019] Furthermore, in step S3, it further includes:

[0020] S3-1. Conduct in-depth analysis and pattern recognition on large-scale offline stored batch data to establish an enterprise high-potential clue indicator system;

[0021] S3-2. Use big data intelligent analysis technology to adopt supervised machine learning methods to establish a high-potential clue model for the enterprise, and use the random forest algorithm built into the SparkML component to establish the high-potential clue model;

[0022] S3-3. Streaming computing technology predicts the probability of high-potential leads for enterprises in real time.

[0023] Further, in step S3-1,

[0024] (1) For the risk of insufficient capital contribution, for the selected enterprises, the company's industrial and commercial registration information and financial statements are checked, and the shareholders' subscribed capital contribution and paid-in capital contribution are compared to identify the insufficient capital contribution situation;

[0025] Conduct shareholder credit review and use the credit information system to confirm whether current and past shareholders have been included in the list of dishonest persons subject to enforcement or the list of persons subject to restrictions on high consumption, so as to assess their debt repayment ability;

[0026] (2) For the risk of false capital contribution, we will conduct an in-depth analysis of the company's annual report, check the subscribed and paid-in capital contribution methods, especially the non-monetary capital contribution, and verify its true value through a third-party evaluation agency to identify false capital contributions;

[0027] Conduct shareholder credit checks to ensure that all relevant shareholders are in good credit standing;

[0028] (3) For capital reduction risks, track the changes in the company's registered capital, focus on capital reduction activities, analyze the reasons for capital reduction, whether the procedures are legal and compliant, and their impact on the interests of creditors;

[0029] Confirm the list of shareholders before and after the capital reduction, conduct another credit review of the shareholders, and assess the legal risks brought about by the capital reduction.

[0030] (4) For one-person company risks, check the current type of the enterprise and its previous change records to identify the one-person company, including sole proprietorship by a natural person and sole proprietorship by a legal person; conduct a credit review of shareholders and simultaneously implement step (3) to identify capital reduction risks. If necessary, initiate a financial independence assessment mechanism to assess the degree of financial independence between the shareholders of the one-person company and the company by reviewing the financial audit reports and bank statements for the past three years.

[0031] Furthermore, in step S3-2, first, the positive and negative sample screening logic and sampling method of the high potential clue model are determined. Through the experience of lawyer experts, debtor companies with the same characteristics as the target companies for asset liquidation are screened, and companies that have not been involved in bankruptcy cases and have been enforced for breach of trust are screened as positive samples. The training sample and verification sample data are sampled from the corporate data of recent years in the stock data;

[0032] Secondly, based on the positive and negative samples and high potential clue model indicator system formed by sampling, feature processing, feature extraction, and indicator importance evaluation of the training set data are carried out, and iterative tuning of feature screening is carried out according to the feature importance and feature screening results;

[0033] Finally, the random forest algorithm built into the SparkML component is used for model training, the hyperopt parameter optimization method is used to optimize the model parameters, the AUC and KS metrics are used to verify the model effectiveness, and the model is iteratively tuned according to the model measurement results. Finally, a high-potential clue model for the enterprise is established. The model inputs various clue feature indicators of the enterprise and outputs the probability that the enterprise is predicted to be a high-potential debtor clue.

[0034] Furthermore, in step S3-3, the managed multi-source data of various types of enterprises will be transmitted to the clue analysis and mining module in real time through the Kafka message component for real-time streaming data processing. The SparkSreaming component receives the processed multi-source data of the enterprise and extracts the various dimensional indicators on which the enterprise's high-potential clue model depends. It determines whether the enterprise triggers various clue models based on the clue judgment rules and rule trigger thresholds, and loads the trained high-potential clue classification model at the same time. It predicts the probability of high-potential clues of the enterprise through the processed clue indicators. The prediction results of each clue and the probability of high-potential clues will be transmitted to the system application end through the Kafka component, and distributed and pushed in real time according to the system application end configuration rules.

[0035] Furthermore, in step S4, real-time clue push is used to connect the clue analysis module and the system application. The real-time push mechanism is configured through the system end. The system will generate instant clue prompts according to the preset push rules and mechanisms for enterprises that trigger various clue models and have high potential clue probabilities. These prompts will be pushed to the user end through diversified message push methods. The information of various debtors will be displayed vividly and intuitively through a visual interface, which is convenient for guiding users to make quick responses and decisions.

[0036] A method and device for mining high-potential clues in real time, comprising: at least one memory and at least one processor;

[0037] The at least one memory is used to store a machine-readable program;

[0038] The at least one processor is used to call the machine-readable program to execute a method for mining high-potential clues in real time.

[0039] Compared with the prior art, the method and device for real-time mining of high-potential clues of the present invention have the following outstanding beneficial effects:

[0040] First, enhance the real-time and dynamic monitoring capabilities of clues: In the field of asset preservation and debt collection, the timeliness of information is crucial. The present invention ensures the instant capture and dynamic tracking of debtor risk clues by real-time collection, processing and analysis of various authoritative data sources. The system can capture relevant clues at the first time and automatically trigger risk assessment and early warning mechanisms. This real-time clue monitoring capability enables lawyers and financial institutions to respond quickly to market changes, adjust strategies in a timely manner, and effectively avoid decision-making delays caused by information asymmetry. At the same time, the system's continuous monitoring function also ensures the continuous updating and iteration of risk information, providing strong support for long-term and stable asset preservation and collection work.

[0041] Second, improve case handling efficiency: In the process of handling asset preservation and debt collection cases, lawyers often need to invest a lot of time to perform tedious tasks such as in-depth research on corporate credit, verification of capital contribution status, and multi-dimensional risk assessment. This system integrates advanced data processing and intelligent analysis technologies to achieve instant identification and in-depth analysis of corporate judicial risk points, including but not limited to capital contribution defects (such as insufficient or untrue), capital reduction behavior, and risks unique to single-shareholder companies. This move significantly shortens the time period for lawyers to obtain key case information and reduces the burden of manual sorting and analysis of massive raw data, thereby greatly improving the timeliness and accuracy of case handling.

[0042] Third, strengthen the rigor and effectiveness of legal argumentation: The detailed corporate portraits and judicial risk assessment data generated by the system provide lawyers with solid data support and decision-making basis. This information not only helps lawyers to more accurately judge the possibility of winning a case, but also provides a scientific basis for formulating legal strategies. Lawyers can adjust litigation strategies, preservation measures or collection plans in a targeted manner based on the evaluation results to improve the success rate and effectiveness of case handling.

[0043] Fourth, reduce professional risks and enhance professional safety: In the face of complex legal risks in the field of asset preservation and collection, including but not limited to litigation failures caused by information blind spots and improper implementation of preservation measures, this system plays an indispensable role in risk warning and prevention. By fully revealing the true situation of the enterprise and potential legal traps, the system effectively helps lawyers avoid the risks of negligence and misjudgment in the practice process. In addition, the credit review function provided by the system can instantly screen and warn of transaction risks with dishonest entities or restricted high-consumption personnel, providing lawyers with pre-emptive risk avoidance strategies, further consolidating the safety line of practice. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the embodiments of the present invention 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 some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Attached Figure 1 is a flowchart of a method for mining high-potential leads in real time. DETAILED DESCRIPTION

[0046] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with specific implementation methods. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] A best embodiment is given below:

[0048] like Figure 1 As shown, a method for real-time mining of high-potential clues in this embodiment has the following steps:

[0049] S1, data collection and aggregation;

[0050] The aggregated enterprise multi-source data mainly includes three forms: public data collection, API interface docking, and batch data push.

[0051] Use automated collection technology to collect enterprise-related information from various public and authoritative channels, including but not limited to enterprise business information, honorary qualifications, public projects, bankruptcy reorganization, industry reports, and judicial litigation. By presetting a list of various public and authoritative channels that need to be collected, automated collection technology automatically parses web pages from various channels, analyzes the data structure of the web pages, and extracts the required web page content;

[0052] The API interface form is to obtain third-party data by actively calling the third-party interface, and the batch pushed data is to obtain batch data pushed by a third party at a scheduled time by batch receiving.

[0053] S2, multi-source data management and processing;

[0054] The enterprise data gathered through public data collection and different data docking forms are cleaned, integrated and pre-processed by using big data processing technology to eliminate redundant, erroneous and inconsistent information to form a structured and standardized data set.

[0055] The multi-source data of enterprises gathered through various means will first be temporarily stored in Redis, a database that supports fast, large-scale reading and writing, supports multiple data types, and is subscribed; then the SparkSQL component will be used for distributed reading of large-scale data, data cleaning, removal of inconsistent formats, processing of noise data, etc., such as inconsistent registered capital units, empty field data, etc., and structured processing of semi-structured text data such as judicial documents, announcement content, etc., to form a structured and standardized data set. Multi-source enterprise data will align and standardize various types of data according to the unique identification of the enterprise, and store them in different dimensions according to unified storage standards; the data after real-time processing will be sent in real time to Kafka, a message component that receives real-time data streams. After passing through the pipeline component such as Kafka, the real-time data will be sent to the real-time data receiving component in the clue mining module for clue analysis and mining. The data after real-time processing will be sent to the big data storage medium HDFS and tiDB cluster at the same time for batch data storage.

[0056] S3, clue analysis and mining;

[0057] include:

[0058] S3-1. Determine the enterprise's high-potential lead indicator system;

[0059] Conduct in-depth analysis and pattern recognition on large-scale offline stored batch data to establish an enterprise high-potential clue indicator system, including but not limited to enterprise capital changes, external investment changes, newly acquired qualifications and honors, separations and mergers, judicial risks, etc. The enterprise high-potential clue indicator system includes but is not limited to insufficient capital contribution, false capital contribution, capital reduction, one-person company and other dimensions.

[0060] (1) Risk of insufficient capital contribution;

[0061] For the screened enterprises, by querying the business registration information, financial statements, etc., comparing the shareholders' subscribed capital with the paid-in capital, we can identify the situation of insufficient capital contribution; implement shareholder credit review, and use the credit information system to confirm whether current and historical shareholders are included in the list of dishonest debtors or the list of persons restricted from high consumption, in order to assess their debt repayment ability.

[0062] (2) Risk of false capital contribution;

[0063] We conduct in-depth analysis of the company's annual report, check the subscribed and paid-in capital contribution methods, especially the non-monetary capital contribution, verify its true value through a third-party appraisal agency, and identify false capital contributions; conduct shareholder credit review to ensure that all relevant shareholders are in good credit status.

[0064] (3) Risk of capital reduction;

[0065] Track all changes in the company's registered capital, focus on capital reduction, analyze the reasons for capital reduction, whether the procedures are legal and compliant, and its impact on the interests of creditors; confirm the list of shareholders before and after the capital reduction, and conduct another credit review of the shareholders to assess the possible legal risks brought about by the capital reduction.

[0066] (4) Risks of a one-person company;

[0067] Check the current type of the enterprise and its previous change records to identify the status of one-person companies, including sole proprietorship by a natural person and sole proprietorship by a legal person; conduct a credit review of shareholders and simultaneously implement step (3) to identify the risk of capital reduction. If necessary, initiate a financial independence assessment mechanism to assess the degree of financial independence between the shareholders of a one-person company and the company by reviewing the financial audit reports and bank statements for the past three years, in order to prevent shareholders from abusing the company's independent personality and limited liability to evade debts.

[0068] S3-2. Establish a high potential lead model for the enterprise;

[0069] Use SparkML and other big data intelligent analysis technologies and adopt supervised machine learning methods to establish a high-potential clue model for enterprises. The built-in random forest algorithm of the SparkML component is mainly used to establish the high-potential clue model. First, determine the positive and negative sample screening logic and sampling method of the high-potential clue model. Through the experience of lawyer experts, screen debtor companies with the same characteristics as the target companies of asset recovery, such as those who have been dishonest debtors, bankruptcy cases, foreign investment reductions, and registered capital reductions in the past year as negative samples. Screen companies with good credit status, tax credit status A+ and above, companies that have never had asset changes, bankruptcy cases, dishonest executions, etc. as positive samples. The training sample and verification sample data are mainly sampled from the corporate data of the past three years in the stock data; secondly, based on the positive and negative samples formed by sampling And the high-potential clue model indicator system is used to perform feature processing, feature extraction, and indicator importance evaluation of the training set data, and iterative tuning of feature screening is performed according to the feature importance and feature screening results; finally, the random forest algorithm built into the SparkML component is used for model training, the hyperopt parameter optimization method is used to optimize the model parameters, the AUC and KS measurement indicators are used to verify the effectiveness of the model, and the model is iteratively tuned according to the model measurement results. Finally, a high-potential clue model for the enterprise is established. The model inputs various clue feature indicators of the enterprise and outputs the probability that the enterprise is predicted to be a high-potential debtor clue.

[0070] S3-3, streaming computing technology predicts the probability of high-potential leads for enterprises in real time;

[0071] After governance, various types of enterprise multi-source data will be transmitted to the clue analysis and mining module in real time through the Kafka message component for real-time streaming data processing. The SparkSreaming component receives the processed enterprise multi-source data and extracts the various dimensional indicators that the enterprise's high-potential clue model relies on. It determines whether the enterprise triggers various clue models based on the clue judgment rules and rule trigger thresholds. At the same time, it loads the trained high-potential clue classification model, and predicts the probability of high-potential clues of the enterprise through the processed clue indicators. The prediction results of each clue and the probability of high-potential clues will be transmitted to the system application end through the Kafka component, and distributed and pushed in real time according to the system application end configuration rules.

[0072] S4. Real-time push of clues;

[0073] The real-time clue push module is used to connect the clue analysis module with the system application, and configure the real-time push mechanism through the system end, including the prediction mechanism of the real-time data stream high-potential clue model, the prediction cycle, the time range for processing data, and the system push clues, high-potential clue probability, the content of various risks of high-potential clue enterprises, the push cycle, and the push method; the system will generate instant clue prompts according to the preset push rules and mechanisms for enterprises that trigger various clue models and have high-potential clue probabilities, and push them to the user end through diversified message push methods such as SMS and email, to ensure that lawyers and other users can obtain debtor information at the first time, and know at the first moment which enterprises have various risks that have triggered which clue models, what is the probability of these enterprises being predicted as high-potential clues, what are the specific contents of various risks occurring in these enterprises, etc. All kinds of debtor information will be displayed vividly and intuitively through a visual interface, which is convenient for guiding users to make quick responses and decisions.

[0074] A real-time high-potential clue mining system includes a data collection and aggregation module, a multi-source data management and processing module, a clue analysis and mining module, and a clue real-time push module;

[0075] The data collection and aggregation module is responsible for collecting enterprise-related information from various public and authoritative channels, and completing the integration and aggregation of data in various forms such as data collection, API interface docking, and batch push.

[0076] The multi-source data management and processing module screens, compares and manages data according to the preset unified processing management standards. The multi-source data of the enterprise is aligned and verified according to the enterprise's unique identification and then stored in the big data cluster and relational database;

[0077] The clue analysis and mining module processes clue indicators according to the established clue indicator system, establishes supervised clue models for batch data stored offline, and uses streaming processing technology to receive real-time data, process indicators, predict clue model results, and predict comprehensive high-potential clue models for real-time transmission data;

[0078] The real-time clue push module pushes real-time clues and risk analysis result data according to the system's preset rules and mechanisms. The result output module displays the clue mining results and various decision-making suggestions in a visual form.

[0079] Based on the above, a method and device for real-time mining of high-potential clues in this embodiment includes: at least one memory and at least one processor;

[0080] The at least one memory is used to store a machine-readable program;

[0081] The at least one processor is used to call the machine-readable program to execute a method for mining high-potential clues in real time.

[0082] The above-mentioned specific implementations are only specific cases of the present invention. The patent protection scope of the present invention includes but is not limited to the above-mentioned specific implementations. Any technical solutions that conform to the above-mentioned specific implementations of the present invention and any appropriate changes or substitutions made by ordinary technicians in the relevant technical field shall fall within the patent protection scope of the present invention.

[0083] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for mining high potential clues in real time, characterized in that: The steps are as follows: S1, data collection and aggregation; S2, multi-source data management and processing; S3, clue analysis and mining; S4. Real-time push of clues.

2. The method for real-time mining of high-potential clues according to claim 1, characterized in that: In step S1, the collected enterprise multi-source data includes three forms: public data collection, API interface connection and batch push data; Use automated collection technology to collect enterprise-related information from various public channels. By presetting a list of various public authoritative channels that need to be collected, automated collection technology automatically parses web pages from various channels, analyzes the web page data structure, and extracts the required web page content; The API interface form is to obtain third-party data by actively calling the third-party interface, and the batch pushed data is to obtain batch data pushed by a third party at a scheduled time by batch receiving.

3. The method for real-time mining of high-potential clues according to claim 2, characterized in that: In step S2, the enterprise multi-source data gathered in various ways will be temporarily stored in Redis, and then the SparkSQL component will be used to perform distributed reading of large batches of data, clean the data, remove inconsistent formats and process noise data, and form a structured and standardized data set.

4. The method for real-time mining of high-potential clues according to claim 3, characterized in that: Multi-source enterprise data will be aligned and standardized according to the enterprise's unique identifier, and stored in different dimensions according to unified storage standards. The real-time processed data will be sent to Kafka in real time. Kafka is a message component that receives real-time data streams. After passing through the Kafka pipeline component, the real-time data will be sent to the real-time data receiving component for clue analysis and mining. The real-time processed data will be sent to the big data storage media HDFS and tiDB cluster at the same time for batch data storage.

5. The method for real-time mining of high-potential clues according to claim 4, characterized in that: In step S3, further comprising: S3-1. Conduct in-depth analysis and pattern recognition on large-scale offline stored batch data to establish an enterprise high-potential clue indicator system; S3-2. Use big data intelligent analysis technology to adopt supervised machine learning methods to establish a high-potential clue model for the enterprise, and use the random forest algorithm built into the SparkML component to establish the high-potential clue model; S3-3. Streaming computing technology predicts the probability of high-potential leads for enterprises in real time.

6. The method for real-time mining of high-potential clues according to claim 5, characterized in that: In step S3-1, (1) For the risk of insufficient capital contribution, for the selected enterprises, the company's industrial and commercial registration information and financial statements are checked, and the shareholders' subscribed capital contribution and paid-in capital contribution are compared to identify the insufficient capital contribution situation; Conduct shareholder credit review and use the credit information system to confirm whether current and past shareholders have been included in the list of dishonest persons subject to enforcement or the list of persons subject to restrictions on high consumption, so as to assess their debt repayment ability; (2) For the risk of false capital contribution, we will conduct an in-depth analysis of the company's annual report, check the subscribed and paid-in capital contribution methods, especially the non-monetary capital contribution, and verify its true value through a third-party evaluation agency to identify false capital contributions; Conduct shareholder credit checks to ensure that all relevant shareholders are in good credit standing; (3) For capital reduction risks, track the changes in the company's registered capital, focus on capital reduction activities, analyze the reasons for capital reduction, whether the procedures are legal and compliant, and their impact on the interests of creditors; Confirm the list of shareholders before and after the capital reduction, conduct another credit review of the shareholders, and assess the legal risks brought about by the capital reduction. (4) For one-person company risks, check the current type of the enterprise and its previous change records to identify the one-person company, including sole proprietorship by a natural person and sole proprietorship by a legal person; conduct a credit review of shareholders and simultaneously implement step (3) to identify capital reduction risks. If necessary, initiate a financial independence assessment mechanism to assess the degree of financial independence between the shareholders of the one-person company and the company by reviewing the financial audit reports and bank statements for the past three years.

7. The method for real-time mining of high-potential clues according to claim 6, characterized in that: In step S3-2, first, the positive and negative sample screening logic and sampling method of the high potential clue model are determined. Based on the experience of lawyer experts, debtor companies with the same characteristics as the target companies for asset recovery are screened, and companies that have not been involved in bankruptcy cases or have been enforced for dishonesty are screened as positive samples. The training sample and verification sample data are sampled from the corporate data of recent years in the stock data; Secondly, based on the positive and negative samples and high potential clue model indicator system formed by sampling, feature processing, feature extraction, and indicator importance evaluation of the training set data are carried out, and iterative optimization of feature screening is carried out according to the feature importance and feature screening results; Finally, the random forest algorithm built into the SparkML component is used for model training, the hyperopt parameter optimization method is used to optimize the model parameters, the AUC and KS metrics are used to verify the model effectiveness, and the model is iteratively tuned according to the model measurement results. Finally, a high-potential clue model for the enterprise is established. The model inputs various clue feature indicators of the enterprise and outputs the probability that the enterprise is predicted to be a high-potential debtor clue.

8. The method for real-time mining of high-potential clues according to claim 7, characterized in that: In step S3-3, the managed multi-source data of various types of enterprises will be transmitted to the clue analysis and mining module in real time through the Kafka message component for real-time streaming data processing. The SparkSreaming component receives the processed multi-source data of the enterprise and extracts the various dimensional indicators on which the enterprise's high-potential clue model depends. It determines whether the enterprise triggers various clue models based on the clue judgment rules and rule trigger thresholds, and loads the trained high-potential clue classification model at the same time. The probability of high-potential clues of the enterprise is predicted through the processed clue indicators. The prediction results of each clue and the probability of high-potential clues will be transmitted to the system application end through the Kafka component, and distributed and pushed in real time according to the system application end configuration rules.

9. The method for real-time mining of high-potential clues according to claim 8, characterized in that: In step S4, real-time clue push is used to connect the clue analysis module and the system application. The real-time push mechanism is configured through the system end. The system will generate instant clue prompts for enterprises that trigger various clue models and have high potential clue probabilities according to the preset push rules and mechanisms, and push them to the user end through diversified message push methods. The information of various debtors will be displayed vividly and intuitively through a visual interface, which is convenient for guiding users to make quick responses and decisions.

10. A method and device for real-time mining of high-potential clues, characterized in that: include: at least one memory and at least one processor; The at least one memory is used to store a machine-readable program; The at least one processor is configured to call the machine-readable program to execute the method according to any one of claims 1 to 9.

Images