Bond project online review method and system

Abstract

The invention provides a bond project online review method and system. The method comprises the following steps: acquiring bond basic data in a bond project to be reviewed; performing information extraction on the bond basic data by using a natural language processing technology to obtain a preliminary review demand list; based on the preliminary review demand list of the bond project, performing risk assessment on the bond basic data by using a project risk scoring model, and outputting a risk score of the bond project; according to the risk score, auditing the bond basic data by using a rule engine, and generating an online review report as an online review result of the bond project; through combination of natural language processing, deep learning and the rule engine, bond project basic data is automatically extracted, the preliminary review demand is generated, risk assessment is performed by using the risk scoring model, then the data is audited through the rule engine, and the online review report is generated, so that the online review efficiency is improved, manual intervention is reduced, and the online review efficiency is improved. And the accuracy and consistency of review results are ensured.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to an online review method and system for a bond project. Background Art

[0002] Since special bond projects usually involve large amounts of funds and complex project evaluations, the traditional review method for special bond projects has several shortcomings. First, the traditional review process has low transparency, often relying on paper materials and face-to-face meetings, lacks an open and transparent mechanism, and is prone to conflicts of interest. Second, the inconsistent review quality and lack of standardized processes lead to different experts’ judgments on the same project, which in turn affects the stability of the review results. Third, traditional reviews are highly subjective, and the personal experience and opinions of experts often affect the review results. The lack of data-based decision support may lead to review bias. Finally, due to the lack of online system integration, project data and review information are difficult to share, forming information islands, which greatly affects the accuracy and timeliness of decision-making.

[0003] In summary, in the existing technology, the review process for special bond projects has low efficiency and accuracy due to reliance on manual processing and manual review, and lacks standardized review models and data-driven decision support, which makes the review results of special bond projects unstable, highly subjective, and lacks information sharing. Summary of the invention

[0004] In order to solve the problems in the prior art of the review process of special bond projects, which relies on manual processing and manual review, has low efficiency and accuracy, lacks standardized review models and data-driven decision support, and makes the review results of special bond projects unstable, highly subjective, and lacks information sharing, the present invention proposes an online review method for bond projects, including:

[0005] Obtain the basic bond data in the bond projects to be reviewed;

[0006] Using natural language processing technology to extract information from the bond basic data, and obtaining a preliminary review requirements list for the bond project;

[0007] Based on the preliminary review requirement list of the bond project, a risk assessment is performed on the bond basic data using a pre-built project risk scoring model, and a risk score of the bond project is output;

[0008] According to the risk score, the bond basic data is reviewed using a rule engine to generate an online review report as an online review result of the bond project;

[0009] The project risk scoring model is obtained by training a deep neural network using historical bond basic data as training data.

[0010] Optionally, the project risk scoring model includes the following construction process:

[0011] Use historical bond basic data as input for training data;

[0012] Using the risk score of the historical bond basic data as the output of training data;

[0013] According to the input and output of the training data, the deep neural network is trained using the gradient descent method and the Adam optimizer to obtain a project risk scoring model.

[0014] Optionally, the use of natural language processing technology to extract information from the bond basic data to obtain a preliminary review requirements list for the bond project includes:

[0015] Use the TextRank algorithm to extract attributes from the bond basic data to obtain key attribute information in the bond project;

[0016] Using a correlation analysis method to perform data analysis on the key attribute information, and obtaining the review requirements of the bond project;

[0017] Prioritize the review requirements using a reinforcement learning algorithm to obtain a preliminary review requirement list for the bond project;

[0018] The key attribute information includes one or more of the following: issue date, annual interest rate and debt default risk.

[0019] Optionally, the step of reviewing the bond basic data using a rule engine according to the risk score and generating an online review report as the online review result of the bond project includes:

[0020] According to the risk score, the basic data of the bond is reviewed using a rule engine to obtain a review opinion of the bond project;

[0021] Generate an online review report based on the review and audit opinions;

[0022] A digital signature is generated using public key encryption technology, and the online review report is signed using the digital signature, and the signed online review report is used as the online review result of the bond project.

[0023] Optionally, generating an online review report based on the review and audit opinions includes:

[0024] Based on the review and audit opinions, using data analysis methods, generate audit standard requirements corresponding to the bond project;

[0025] Generate a risk analysis report based on the audit standard requirements and the risk score;

[0026] The risk analysis report is visualized on the Web, and an online review report corresponding to the bond project is output.

[0027] Optionally, the bond basic data includes one or more of the following: credit rating, macroeconomic data, industry data, debt repayment data, default records and market data.

[0028] Optionally, the macroeconomic data includes one or more of the following: interest rate level, economic growth rate and inflation rate;

[0029] The industry data includes one or more of the following: industry prosperity, industry historical default rate and risk data;

[0030] The debt repayment data includes one or more of the following: debt repayment capacity index, debt-equity ratio and interest coverage ratio;

[0031] The market data includes one or more of the following: stock market fluctuation data and commodity price fluctuation data.

[0032] Based on the same inventive concept, the present invention also provides an online bond project review system, including:

[0033] The data acquisition module is used to obtain the basic bond data in the bond project to be reviewed;

[0034] An information extraction module, used to extract information from the bond basic data using natural language processing technology to obtain a preliminary review requirements list for the bond project;

[0035] A risk scoring module is used to perform risk assessment on the bond basic data based on the preliminary review requirement list of the bond project using a pre-built project risk scoring model, and output a risk score for the bond project;

[0036] An online review module, used to review the bond basic data using a rule engine according to the risk score and generate an online review report as an online review result of the bond project;

[0037] The project risk scoring model is obtained by training a deep neural network using historical bond basic data as training data.

[0038] Optionally, the review system further includes: a model building module, which is used to:

[0039] Use historical bond basic data as input for training data;

[0040] Using the risk score of the historical bond basic data as the output of training data;

[0041] According to the input and output of the training data, the deep neural network is trained using the gradient descent method and the Adam optimizer to obtain a project risk scoring model.

[0042] Optionally, the information extraction module includes:

[0043] The attribute extraction submodule is used to extract attributes from the bond basic data using the TextRank algorithm to obtain key attribute information in the bond project;

[0044] A correlation analysis submodule, for performing data analysis on the key attribute information using a correlation analysis method to obtain the review requirements of the bond project;

[0045] A requirement sorting submodule is used to prioritize the review requirements using a reinforcement learning algorithm to obtain a preliminary review requirement list for the bond project;

[0046] The key attribute information includes one or more of the following: issue date, annual interest rate and debt default risk.

[0047] Optionally, the online audit module includes:

[0048] An opinion output submodule is used to review the bond basic data using a rule engine according to the risk score to obtain the review opinion of the bond project;

[0049] A report generation submodule is used to generate an online review report based on the review and audit opinions;

[0050] The result output submodule is used to generate a digital signature using public key encryption technology, and to sign the online review report using the digital signature, and to use the signed online review report as the online review result of the bond project.

[0051] Optionally, the report generation submodule includes:

[0052] A data analysis unit, used to generate the audit standard requirements corresponding to the bond project based on the review and audit opinions using a data analysis method;

[0053] A risk scoring unit, used to generate a risk analysis report according to the audit standard requirements and the risk score;

[0054] The visualization unit is used to perform Web visualization on the risk analysis report and output an online review report corresponding to the bond project.

[0055] For example, the bond basic data includes one or more of the following: credit rating, macroeconomic data, industry data, debt repayment data, default records and market data.

[0056] For example, the macroeconomic data includes one or more of the following: interest rate level, economic growth rate and inflation rate;

[0057] For example, the industry data includes one or more of the following: industry prosperity, industry historical default rate and risk data;

[0058] Illustratively, the debt repayment data includes one or more of the following: debt repayment capacity index, debt-equity ratio, and interest coverage ratio;

[0059] Illustratively, the market data includes one or more of the following: stock market fluctuation data and commodity price fluctuation data.

[0060] In yet another aspect, the present invention further provides a computer device, comprising: one or more processors;

[0061] A memory for storing one or more programs;

[0062] When the one or more programs are executed by the one or more processors, an online bond project review method as described above is implemented.

[0063] On the other hand, the present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed, the online bond project review method as described above is implemented.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] The present invention provides a bond project online review method and system, comprising: obtaining bond basic data in a bond project to be reviewed; extracting information from the bond basic data using natural language processing technology to obtain a preliminary review requirement list for the bond project; based on the preliminary review requirement list for the bond project, using a pre-constructed project risk scoring model to perform risk assessment on the bond basic data, and outputting a risk score for the bond project; based on the risk score, using a rule engine to review the bond basic data, and generating an online review report as an online review result for the bond project; wherein the project risk scoring model is obtained by training a deep neural network using historical bond basic data as training data; the present application automatically extracts key information from the bond basic data through natural language processing technology to generate preliminary review requirements, which is conducive to improving data processing efficiency and accuracy; performing accurate risk assessment on the project based on a deep neural network model trained with historical data to generate a risk score; the rule engine automatically reviews data and generates an online review report based on the risk score, which is conducive to reducing manual intervention and errors, ensuring the accuracy and consistency of the review results, and thus significantly improving the review efficiency and quality of the bond project. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A schematic diagram of a flow chart of an online bond project review method provided by the present invention;

[0067] Figure 2 A project review flow chart of a special bond project expert online review method provided by the present invention;

[0068] Figure 3 A schematic diagram of the expert avoidance and automatic allocation process in an online expert review method for a special bond project provided by the present invention;

[0069] Figure 4 A schematic diagram of project allocation in an online expert review method for a special bond project provided by the present invention;

[0070] Figure 5 A schematic diagram of rule allocation in an online expert review method for a special bond project provided by the present invention;

[0071] Figure 6 A schematic diagram of expert parallel review in an online expert review method for a special bond project provided by the present invention;

[0072] Figure 7 A schematic diagram of submitting review and generating a report in an online review method for a special bond project expert provided by the present invention;

[0073] Figure 8A schematic diagram of expert avoidance and automatic allocation in a specific embodiment of a method for online expert review of special bond projects provided by the present invention;

[0074] Fig. 9 A schematic diagram of the structural composition of an online bond project review system provided by the present invention. DETAILED DESCRIPTION

[0075] The present invention proposes a bond project online review method, system, device and medium. The specific implementation methods of the present invention are further described in detail below in conjunction with the accompanying drawings.

[0076] Embodiment 1:

[0077] The present invention provides a bond project online review method, the flow chart is as follows Figure 1 As shown, including:

[0078] Step 1: Obtain the basic bond data of the bond project to be reviewed;

[0079] Step 2: Use natural language processing technology to extract information from bond basic data and obtain a preliminary review requirements list for bond projects;

[0080] Step 3: Based on the preliminary review requirements list of the bond project, use the pre-built project risk scoring model to conduct risk assessment on the bond basic data and output the risk score of the bond project;

[0081] Step 4: Based on the risk score, the rule engine is used to review the basic data of the bond and generate an online review report as the online review result of the bond project;

[0082] Among them, the project risk scoring model is obtained by training the deep neural network with historical bond basic data as training data.

[0083] For example, the above-mentioned bond basic data may include one or more of the following: credit rating, macroeconomic data, industry data, debt repayment data, default records and market data.

[0084] For example, the above-mentioned macroeconomic data may include one or more of the following: interest rate level, economic growth rate and inflation rate;

[0085] For example, the above-mentioned industry data may include one or more of the following: industry prosperity, industry historical default rate and risk data;

[0086] For example, the above-mentioned debt repayment data may include one or more of the following: debt repayment capacity index, debt-equity ratio and interest coverage ratio;

[0087] By way of example, the above-mentioned market data may include one or more of the following: stock market fluctuation data and commodity price fluctuation data;

[0088] In one implementation, the above-mentioned project risk scoring model may include the following construction process:

[0089] Use historical bond basic data as input for training data;

[0090] The risk scores of historical bond basic data are used as the output of training data;

[0091] According to the input and output of the training data, the deep neural network is trained using the gradient descent method and the Adam optimizer to obtain the project risk scoring model;

[0092] In this implementation, a high-precision project risk scoring model was successfully constructed by training the deep neural network using the gradient descent method and the Adam optimizer with historical bond basic data as input and risk scores as output. This method uses deep learning technology and can effectively extract complex feature relationships in bond basic data through training on large-scale historical data, providing strong predictive and generalization capabilities for project risk scoring. Compared with traditional risk assessment methods based on rules or linear models, the creativity of this implementation lies in the use of deep neural networks to mine nonlinear associations between data, which can more comprehensively capture the comprehensive impact of different bond project characteristics on risk. The combination of the gradient descent method and the Adam optimizer ensures the rapid convergence and efficiency of model training, and can maintain accuracy and stability even on high-dimensional and complex data sets. Ultimately, this implementation can significantly improve the accuracy and intelligence of risk scoring, laying the foundation for scientific evaluation and decision support of bond projects. At the same time, the automated training and optimization process reduces human intervention and errors, further improves the reliability and practicality of the model, and promotes the development of the bond review field towards intelligence and data-driven.

[0093] In one implementation, the process of extracting information from bond basic data using natural language processing technology to obtain a preliminary review requirements list for bond projects may include:

[0094] The TextRank algorithm is used to extract the attributes of the bond basic data and obtain the key attribute information in the bond project;

[0095] Use correlation analysis methods to analyze key attribute information and obtain bond project review requirements;

[0096] Use reinforcement learning algorithms to prioritize review requirements and obtain a preliminary review requirement list for bond projects;

[0097] Among them, the key attribute information may include one or more of the following: issue date, annual interest rate and debt default risk;

[0098] In this implementation, by combining natural language processing technology, correlation analysis method and reinforcement learning algorithm, the key information of bond projects is intelligently extracted and a preliminary review requirements list is generated. First, the TextRank algorithm can efficiently identify the key attribute information in the project, such as the issue date, annual interest rate and debt default risk, by extracting the attributes of the basic bond data, laying a solid foundation for subsequent analysis. Next, the correlation analysis method is used to analyze the data of these key attribute information, automatically revealing the internal relationship between the attributes, thereby generating review requirements related to the bond project to ensure the comprehensiveness and logic of the review content. In addition, by using the reinforcement learning algorithm to prioritize the review requirements, the importance of the requirements is dynamically adjusted according to the characteristics of the project and historical data, ensuring that priority is given to the review elements that have the greatest impact on the risk. Through this intelligent processing method, not only the accurate extraction of review requirements is achieved, but also the review process can be flexibly adjusted according to the actual situation, which significantly improves the efficiency and accuracy of bond project review, reduces manual intervention, and improves the level of automation.

[0099] In one implementation, the process of reviewing the bond basic data using a rule engine based on risk scoring and generating an online review report as the online review result of the bond project may include:

[0100] Based on the risk score, the rule engine is used to review the basic data of the bond and obtain the review and audit opinions of the bond project;

[0101] Generate online review report based on review and audit opinions;

[0102] Generate a digital signature using public key encryption technology, and sign the online review report with the digital signature, and use the signed online review report as the online review result of the bond project;

[0103] In this implementation, by combining the rule engine, encryption technology and digital signature, the online review report of the bond project is intelligently generated, and the security and non-tamperability of the report are ensured. First, based on the risk score of the bond project, the basic bond data is automatically reviewed by the rule engine to generate review opinions. The rule engine can quickly execute the preset review standards and automatically evaluate the risks and compliance of the bond project, avoiding deviations and inefficiencies in the manual review process. Then, an online review report is generated based on the review opinions to ensure the comprehensiveness and accuracy of the report content. What is innovative is that the implementation also uses public key encryption technology to generate digital signatures and sign the generated online review report to ensure the integrity, authenticity and non-tamperability of the report. This signature mechanism not only improves the security of the report, but also provides reliable evidence for subsequent audits and compliance checks. Through this implementation, the intelligent review method combining the rule engine and digital signature realizes the automatic generation and secure transmission of bond project review results, greatly improves work efficiency, reduces human errors, improves the transparency and credibility of reports, and promotes the development of bond project management in a more intelligent, secure and standardized direction.

[0104] In this implementation, the process of generating an online review report based on the review and audit opinions may include:

[0105] Based on the review and audit opinions, use data analysis methods to generate audit standard requirements corresponding to the bond project;

[0106] Generate risk analysis report according to audit standard requirements and risk scores;

[0107] Perform web visualization of the risk analysis report and output the online review report corresponding to the bond project.

[0108] In this implementation, the intelligent generation and visualization of bond project review reports are realized by combining data analysis methods with Web visualization technology. First, based on the review and audit opinions, the review standard requirements are automatically generated using data analysis methods to ensure the accuracy and systematicness of the review standards. Through this automated standard generation process, the key risk factors of bond projects can be accurately identified, and in-depth analysis can be conducted based on these standards. Then, combined with the risk score of the bond project, a comprehensive risk analysis report is generated to ensure that all risk factors that may affect the bond project are clearly identified and evaluated. Innovatively, this implementation visualizes the generated risk analysis report on the Web, so that complex risk data and review results can be presented to users intuitively. This visualization processing method not only improves the readability of the report, but also facilitates all parties to understand the risk situation of the project in a timely manner and make more accurate decisions. Through this intelligent and automated report generation and visualization technology, the efficiency and quality of bond project review are greatly improved, the deviation of manual operation is reduced, and the comprehensiveness, transparency and efficiency of the report are ensured, providing more scientific and clear data support for the management and decision-making of bond projects.

[0109] In summary, the present invention aims at the problem that the review process of special bond projects in the prior art is low in efficiency and accuracy due to reliance on manual processing and manual review, and lacks standardized review models and data-driven decision support, which makes the review results of special bond projects unstable, subjective, and insufficient information sharing. A bond project review method based on natural language processing, deep learning and rule engine is proposed, which extracts key information in the project in an automated way, and uses the trained risk scoring model to perform accurate risk assessment, and automatically generates a review report, which can not only eliminate errors in manual review, but also ensure information transparency, unified review standards, and support online sharing, solving the problem of information islands in the traditional review process. In addition, the integrity and legitimacy of the data are ensured by digital signature technology, which improves the security and traceability of bond project review. The present invention can significantly improve the efficiency of bond project review, reduce manual intervention, reduce subjectivity, improve review quality, and promote bond project management to develop in a more intelligent and standardized direction. Based on differentiated project review standards, intelligent automatic allocation strategies, and efficient expert review processes, the feasibility of the project, the balance of income and expenditure, and the risk situation are comprehensively evaluated, and the digital, intelligent, and standardized management of project review work is realized. Improve the review efficiency and quality of special bond projects, reduce project risks, and assist in effective supervision by finance and government departments.

[0110] Embodiment 2:

[0111] The present invention also provides a method for online expert review of special bond projects, which automatically allocates review experts by project and field based on the avoidance principle, conducts expert parallel review, and automatically generates review reports by citing digital signatures and other technologies, calculates the scoring results according to the rules through a scoring algorithm, and performs final review and storage. To achieve the above purpose, Figure 2 As shown, the solution used to define the project review workflow is as follows:

[0112] Step S1 Review project preparation: mainly includes basic information maintenance and review project import, as follows:

[0113] Step 1.1 Basic information maintenance: including review rule configuration, institutional information and expert information maintenance; review rules are divided into judgment and scoring categories, expert information includes expert type, institution and review rule configuration of the expert's field of expertise, and projects can be assigned by field when assigning tasks; institutional information can include banks, accounting firms, law firms, etc.

[0114] Step 1.2 Import special bond projects: Import special bond project information, upload review attachments and complete basic project information, such as project name, applicant, region, funding direction, project type and related agency information. After project preparation is completed, the workflow enters the task allocation node.

[0115] Step S2 Review Task Management: Review tasks are the main carrier of the expert review process, mainly including task creation, project allocation, and task start and stop, as follows:

[0116] Step 2.1 Review task creation: Review tasks mainly include task name, task type, whether it is time-limited, time limit, review instructions and other information. When creating a review task, you can choose to save the review items, review experts and review rules included in the task to be reviewed;

[0117] Step 2.2 Avoidance principle and project allocation: The avoidance principle is followed when allocating tasks, which is divided into automatic allocation by project and by field, as follows:

[0118] Step 2.2.1 Avoidance Principles: When reviewing task assignment projects, follow these principles:

[0119] 1) Avoid experts from banks, clubs, and law firms who are involved in preparing implementation plans, financial evaluation reports, and legal opinions for the same project.

[0120] 2) Avoid financial experts in the same jurisdiction.

[0121] 3) The number of experts from the same institution participating in the review of the same project shall not exceed 1.

[0122] 4) Ensure that the number of review projects assigned to experts is generally balanced.

[0123] When assigning tasks, the project will automatically avoid the review of the materials that the institution experts participated in compiling (such as implementation plans, financial evaluation reports, legal opinions, etc.). According to the regional jurisdiction rules, the financial experts of the region to which the project belongs cannot review the project of that region, and it must be reviewed by financial experts from other regions or the same level. In addition, the same institution can only assign one expert to the same project. The system automatically assigns the project to the expert with the least tasks based on the number of tasks assigned to each expert, ensuring balanced project distribution.

[0124] Based on the above principles, the implementation method is described as follows:

[0125] First, initialize the expert task set N1 (recording the number of projects assigned to each expert) and the organization member group T1 (recording the expert set of each organization).

[0126] Divide the tasks according to the institution category (1: bank, 2: club, 3: law firm, 4: finance), use multi-threaded parallel processing, start multiple threads (TASK_WORKER_G1, TASK_WORKER_G2, TASK_WORKER_G3, TASK_WORKER_G4) according to the institution classification number to allocate experts in parallel. The execution process is as follows:

[0127] 1) Start traversing the project, initialize the blacklist organization set N2, and add the organization ID of the current project to N2.

[0128] 2) According to the difference set formula: Perform a difference operation on the expert task set N1 and the expert set corresponding to the institution ID in the blacklist set N2 to obtain the candidate expert set N3.

[0129] 3) Sort the candidate set N3 in ascending order by the number of expert tasks, select the expert with the first index and add it to the selected expert set N4 and update the number of tasks in the expert task set N1. Find the ID of the institution to which the expert belongs and add it to the blacklist institution N2 to ensure that only one expert is assigned to the same project in the same institution.

[0130] 4) Determine whether the number of people in N4 meets the requirement. If not, continue searching and repeat steps 2 and 3 until the number of people in N4 meets the requirement. If the number requirement is still not met after repeated execution and the upper limit of the number of searches is reached (the default is the total number of people required, and the maximum value can be configured), it will be included in the manual allocation queue.

[0131] 5) After the number of selected experts in set N4 meets the requirements, the blacklist N2 data is cleared, and the result set N4 is merged and saved with other thread data, such as Figure 3 shown.

[0132] Step 2.2.2 Assign by project: The same project can be reviewed by multiple experts, that is, each expert needs to review all the rules of a project, and the final review result of the project is determined by the expert review results. Figure 4 shown.

[0133] Step 2.2.3 Allocation by field: Since special bond project funds are invested in various fields, there are also differences in project scoring standards. For example, highway projects focus more on capacity, route selection, safety design, etc., while water conservancy projects focus on hydrological analysis, ecological impact, flood control and drought relief, etc. Therefore, for special bond projects in different fields, targeted review rules should be applied according to the purpose of funds, project nature and risk characteristics, and relevant industry experts should be invited to conduct professional evaluations. The projects are split and allocated according to the review rules. The same review rule can be reviewed by multiple review experts at the same time and generate scoring results. The total project score is generated by summing the scores of all review rules for the project, such as Figure 5 As shown:

[0134] Step S3 Expert Review: After the project is assigned, the review task is started, and the workflow instance flows to the expert review node, enters the expert parallel review and calculates the project scoring results, as follows:

[0135] Step 3.1 Expert Parallel Review: According to the project allocation results, the main task of each project in the workflow instance is broken down into multiple subtasks for expert review. Each subtask is executed in parallel and independently of each other. After all the subtasks of each project are completed, the project score is automatically calculated. The main task node of the project workflow instance automatically flows to the next node, and the flow record of each subtask and the main task is recorded, such as Figure 6 shown.

[0136] Project review can be divided into review by project mode and review by task mode.

[0137] 1) Review by project: Experts can view the project list and select projects for review.

[0138] 2) Review by task: Experts review projects one by one in order, and they must complete the review of the current project before moving on to the next one. The review can be divided into two situations: time-limited and unlimited. Time-limited review: After the expert starts the review, the system automatically counts down, and automatically submits the review results when the time is up, and prompts whether to move on to the next project.

[0139] The project review view is divided into two areas, left and right. The left area displays the review rules related to the experts, and the experts can enter the review results and opinions item by item. When assigned by project, each expert displays the same review rules; when assigned by field, the experts only display the scoring rules in their field of expertise. The right area loads the project review attachments, and the experts can browse the contents of the attachments online, and quickly locate the details of the review criteria through automatic retrieval and keyword segmentation functions, thereby improving the efficiency and quality of the review. After the review results are filled in, the experts can preview the review report in real time and submit the review results. When the expert signs online, the system generates a picture signature and digitally signs it through the CA certificate USB KEY. The signature data is submitted to the server together with the timestamp for verification. After the verification is passed, the review results, timestamp and digital signature will be saved, and a watermarked PDF review report will be generated through the back-end service, and a digital signature will be attached to ensure the security and validity of the document. After the expert confirms that the review results are correct, the review is deemed completed. Figure 7 shown.

[0140] Step 3.2 Calculation of scoring results: After multiple expert review subtasks for the same project are completed, the project review results are calculated. First, the score of each review rule of the project is calculated according to the granularity of the review rules. The score of each review rule of the project is calculated and generated by multiple expert review results. Scoring rules are divided into judgment and scoring categories. Judgment rules are divided into "one is yes" and "one is no", that is, one vote for approval and one vote for rejection, and do not participate in the scoring calculation. The scoring rule calculation method is to first apply the average score formula to each review rule to calculate the rule score μ, the formula is as follows:

[0141]

[0142] Among them, m represents the total number of experts, x i represents the score of the i-th expert;

[0143] Then the variance of each review rule is calculated based on the average score, and the formula is as follows:

[0144]

[0145] Calculate the standard deviation σ of the rule score, the formula is as follows:

[0146]

[0147] The relative standard deviation CV is calculated by the standard deviation σ and the average score μ. The formula is as follows:

[0148]

[0149] According to the mean μ, calculate the sample standard deviation s, the formula is as follows:

[0150]

[0151] According to the sample standard deviation s, calculate the standard error SE, the calculation formula is:

[0152]

[0153] The relative standard deviation (CV) reflects the volatility or dispersion of expert scores. A small value indicates consistent scores, while a large value indicates a large difference in scores. The sample standard error (SE) evaluates the reliability and accuracy of the scores. The background records the score, relative standard deviation, and sample standard error of each rule, and summarizes and calculates the final score of the project. The review differences are marked according to the relative standard deviation and threshold, and the reliability of the score is evaluated with reference to the sample standard error, which is reflected in the project review results. The preliminary review conclusion of the project is obtained by combining the project score, relative standard deviation, and sample standard error.

[0154] Step S4 Financial Final Review: The project review status is displayed by experts and by rules respectively. By expert, the review status of each expert on each rule of the project is displayed; by rule, the review status of multiple experts on each rule is displayed. Based on the preliminary review conclusions and review status of the project, the approved projects are selected for storage. For rejected projects, the reasons for rejection are filled in and the affiliated units are notified.

[0155] The effect of this embodiment is: through the online expert review method of special bond projects, the avoidance principle is followed, and experts are automatically allocated by project and field to ensure the fairness, balance and transparency of project allocation, realize parallel expert review and generate review reports by citing digital signatures and scoring calculation algorithms, improve review efficiency, improve the consistency and scientificity of review quality, drive decision-making with data, reduce subjectivity, and improve the comprehensiveness of decision-making.

[0156] Embodiment 3:

[0157] The specific implementation process of the above-mentioned method for online expert review of special bond projects is described with a specific embodiment, and the steps are as follows:

[0158] Step A1 Review Project Preparation:

[0159] First, maintain the basic information of the system, including zoning information, institutional information, funding areas, expert information, rule information, etc., as shown in the following table:

[0160] Table 1 is the institutional information, Table 2 is the expert information, and the financial expert division codes are represented as 9900, 9901, and 9902 respectively.

[0161] Table 1 Institutional information

[0162]

[0163]

[0164] Table 2 Expert information

[0165]

[0166] Table 3 is a scoring standard table, with R1 to R10, a total of 10 rules.

[0167] Table 3 Scoring criteria

[0168]

[0169]

[0170] Tables 4 and 5 show the areas where project funds are invested, and Table 6 is a scoring configuration table that records the correspondence between the areas where funds are invested and the scoring rules:

[0171] Table 4 Project funding investment areas Table 1

[0172] ID coding name Parent ID GUID CODE NAME PARENT_ID TXLY1 001 Investment field 001 TXLY2 001001 Investment field 001001 TXLY1 TXLY3 002 Investment field 002 TXLY4 002001 Investment field 002001 TXLY3

[0173] Table 5 Project funding investment areas Table 2

[0174] Rule ID ID of the field to which funds are invested RULE_ID TXLY_ID R1 TXLY2 R2 TXLY2 R3 TXLY2 R4 TXLY2 R5 TXLY2 R6 TXLY2 R7 TXLY2 R8 TXLY2 R9 TXLY2 R10 TXLY2

[0175] The review projects are imported according to the template, mainly including: project name, project code, financial division, applicant unit, application year, application amount, project type and other information.

[0176] Table 6 Scoring configuration table

[0177]

[0178]

[0179] After the project information is imported into the model, the applicant unit user will supplement the project's bank information, accounting firm, law firm and other information, and upload the implementation plan, legal opinion, financial evaluation report and other attachments involved in the review. After the project information is complete, submit the workflow task to enter the review project allocation process.

[0180] Step A2 Task creation and project allocation:

[0181] Create an unlimited time review task, select 100 projects with project numbers XM1, XM2, .... XM100, etc., include the 22 experts in the expert information table in this review task, and select all the scoring rules in the example table as the review rules for this review. The review task allocation method is to allocate by project, and the allocation ratio is 2:1:2:1, that is, each project consists of 2 banking experts, 1 accounting expert, 2 lawyer experts and 1 financial expert. The set of experts corresponding to each institution is initialized as shown in Table 8-1 to Table 8-4:

[0182] Table 8-1 Membership composition of institution 1

[0183]

[0184] Table 8-2 Membership composition of institution 2

[0185] Clubs expert JG5 K1 JG6 K2 JG7 K3 JG8 K4

[0186] Table 8-3 Composition of members of Institution 3

[0187]

[0188] Table 8-4 Composition of members of Institution 4

[0189]

[0190]

[0191] Assume that the review task ID is T0, all data sets are cached using Redis, initialize the review task number set N1 of all experts, and use ordered sets to cache key-value pairs. The key is "TASK+task ID", here is TASK_T0, and the value is the task number set of each expert (<B1,0> ,<B2,0> ...<C6,0> ), and initialize the member set T1 of all organizations and write it into the cache, with the organization ID as the key and the value as the corresponding organization member set. As shown below: (JG1,<B1,B2> ),(JG2,<B3,B4> ), (JG5, <k1>), (JG6, <k2>)...(9902,<C5,C6> ). Divide tasks by institution type (1: bank, 2: club, 3: law firm, 4: finance), use multi-threading for parallel processing, and name the threads with "TASK_WORKER_G+category number".

[0192] For example, the processing flow corresponding to thread TASK_WORKER_G3 (law firm) is as follows:

[0193] 1) Start traversing the project, initialize the blacklist organization set N2, and add the organization ID JG10 to which the current project belongs to N2. N2 caches key-value pairs, with the key being "DENY_G+organization category+underscore+task ID+underscore+project ID" and the value being the set of disabled organization IDs; T0 is the review task ID, XM1 is the project ID, and the N2 set is initialized to (DENY_G3_T0_XM1,[JG10]).

[0194] 2) Calculate the difference between the sets corresponding to the organization IDs in N1 and N2. Here, the REDIS ZDIFF command is used to implement the difference operation. Substitute the required parameters into the command: ZDIFF (TASK_T0, JG10, N3), and the set of experts to be selected N3 is obtained.

[0195] 3) Sort the experts in the N3 set in ascending order. Here, use the REDIS ZRANGE command to obtain the first expert L3 in the index and add it to the N4 set for caching. At the same time, update the value of the expert's review task in the N1 cache by 1. Add the expert's institution ID JG11 to the blacklist institution set N2. At this time, N2 is (DENY_G3_T0_XM1, [JG10, JG11]).

[0196] 4) Determine whether the number of people in N4 meets the limit number. If not, repeat steps A2 and A3 until the condition is met or the search limit is reached. If the search limit is reached, it is included in the manual allocation queue and manually allocated later.

[0197] 5) After the conditions are met, clear the N2 blacklist cache and get the result set N4. When allocating by project, N4 uses "TASK_G+institution number+underline+task ID+project ID" as the key and the expert ID set as the value for cache; when allocating by field, use "TASK_G+institution number+underline+task ID+underline+project ID+underline+rule ID" as the value for cache. The example here is allocation by project, and the N4 result set is represented as (TASK_G3_T0_XM1, [L3, L5]), where ID L5 is the expert ID added in subsequent execution.

[0198] 6) After multiple threads are processed in parallel, the allocation results are grouped by project and the data is summarized. Finally, the summarized results are batch inserted into the database for storage. After the project allocation is completed, the review task is opened to enter the expert review, such as Figure 8 shown.

[0199] Step A3 Expert parallel review and result calculation:

[0200] After the review task is started, the main task of the project workflow is split into expert subtasks for parallel review according to the distribution of the review rules. For example, the review rules R1, R2, R3, and R4 of the project XM1 are assigned to four experts B1, B3, B5, and B6, R5 and R6 are assigned to K1 and K2, R7 and R8 are assigned to L1 and L2, and R9 and R10 are assigned to C3 and C6. According to the above distribution, 10 workflow subtasks are split. Each subtask is independent and processed in parallel. Experts only need to review the assigned review rules. After the review is completed, they sign online and submit to generate the review report. When all subtasks of a project are completed, the review score, relative standard deviation, and sample standard error of each rule are automatically calculated. Taking the full score of R1 rule as 20 as an example, the expert scoring samples are: 14, 12, 10, and 18.

[0201] Calculate the average score using the mean formula:

[0202] Calculate the variance based on the mean μ:

[0203] The standard deviation is calculated as:

[0204] According to the relative standard deviation formula: Substituting the values:

[0205] Calculate the sample standard deviation based on the mean μ:

[0206] Therefore, the sample standard error is:

[0207] The review score, relative standard deviation and sample standard error of each rule are calculated in turn, and the scoring scores are summarized and summed to obtain the final score of the project.

[0208] The relative standard deviation (CV) is compared with the preset threshold rules, the degree of dispersion or fluctuation of each rule is marked, and a prompt is given during the final review and storage.

[0209] The relative standard deviation can refer to the following rules:

[0210] 1) CV < 10%; low dispersion. The data are generally considered to be relatively stable and reliable.

[0211] 2) 10% ≤ CV < 20%; medium dispersion. Used when data fluctuates, suitable for relatively loose scenarios.

[0212] 3) 20% ≤ CV < 30%; high degree of dispersion. The data fluctuates greatly, and there may be more uncertainties or unstable factors.

[0213] 4) CV ≥ 30%; high dispersion. The data is very unstable and there is a high risk or uncertainty.

[0214] Comprehensively evaluate key indicators such as the total project score, relative standard deviation (CV), sample standard error (SE), analyze the overall situation of the project, and draw a preliminary conclusion on the project.

[0215] Step A4 Final review and storage:

[0216] Combining the preliminary review conclusions and project evaluation results, the financial user comprehensively evaluates the project feasibility, balance of income and expenditure, and potential risks, and chooses to approve the project for inclusion in the database. If the project is not approved, the financial user will fill in the reasons for rejection and notify the affiliated unit.

[0217] This embodiment can illustrate that the present invention can achieve expert avoidance and automatic allocation, ensure the fairness, transparency and balance of project review tasks, support expert parallel review, and improve the flexibility and efficiency of the review process; automatically generate review reports by citing digital signatures to ensure the standardization, security and effectiveness of the review results, and record the entire process of review tasks to ensure the full traceability and traceability of project records.

[0218] Embodiment 4:

[0219] Based on the same inventive concept, the present invention also provides an online bond project review system, the structural composition diagram is as follows: Fig. 9 As shown, including:

[0220] The data acquisition module is used to obtain the basic bond data in the bond project to be reviewed;

[0221] The information extraction module is used to extract information from the basic bond data using natural language processing technology to obtain a preliminary review requirements list for the bond project;

[0222] The risk scoring module is used to conduct risk assessment on bond basic data based on the preliminary review requirements list of bond projects using the pre-built project risk scoring model and output the risk score of the bond project;

[0223] The online review module is used to review the basic bond data based on the risk score using the rule engine and generate an online review report as the online review result of the bond project;

[0224] Among them, the project risk scoring model is obtained by training the deep neural network with historical bond basic data as training data.

[0225] For example, the above-mentioned bond basic data may include one or more of the following: credit rating, macroeconomic data, industry data, debt repayment data, default records and market data.

[0226] For example, the above-mentioned macroeconomic data may include one or more of the following: interest rate level, economic growth rate and inflation rate;

[0227] For example, the above-mentioned industry data may include one or more of the following: industry prosperity, industry historical default rate and risk data;

[0228] For example, the above-mentioned debt repayment data may include one or more of the following: debt repayment capacity index, debt-equity ratio and interest coverage ratio;

[0229] Illustratively, the above market data may include one or more of the following: stock market fluctuation data and commodity price fluctuation data.

[0230] In one implementation, the above review system may further include: a model building module, which is used to:

[0231] Use historical bond basic data as input for training data;

[0232] The risk scores of historical bond basic data are used as the output of training data;

[0233] According to the input and output of the training data, the deep neural network is trained using the gradient descent method and Adam optimizer to obtain the project risk scoring model.

[0234] In one implementation, the information extraction module may include:

[0235] The attribute extraction submodule is used to extract attributes from bond basic data using the TextRank algorithm to obtain key attribute information in the bond project;

[0236] The correlation analysis submodule is used to analyze the key attribute information using the correlation analysis method to obtain the review requirements of the bond project;

[0237] The requirement sorting submodule is used to prioritize the review requirements using the reinforcement learning algorithm to obtain a preliminary review requirement list for the bond project;

[0238] Among them, key attribute information includes one or more of the following: issue date, annual interest rate and debt default risk.

[0239] In one implementation, the online audit module may include:

[0240] The opinion output submodule is used to review the basic data of bonds based on risk scores and use the rule engine to obtain the review opinions of the bond projects;

[0241] The report generation submodule is used to generate an online review report based on the review and audit opinions;

[0242] The result output submodule is used to generate a digital signature using public key encryption technology, and to sign the online review report with the digital signature, and to use the signed online review report as the online review result of the bond project.

[0243] In one implementation, the report generation submodule may include:

[0244] The data analysis unit is used to generate the audit standard requirements corresponding to the bond project based on the review and audit opinions using data analysis methods;

[0245] The risk scoring unit is used to generate risk analysis reports based on the audit standard requirements and risk scores;

[0246] The visualization unit is used to perform Web visualization of the risk analysis report and output the online review report corresponding to the bond project.

[0247] Embodiment 5:

[0248] Based on the same inventive concept, the present invention also provides a computer device, which includes a processor and a memory, the memory is used to store a computer program, the computer program includes program instructions, and the processor is used to execute the program instructions stored in a computer storage medium. The processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, which is suitable for implementing one or more instructions, and is specifically suitable for loading and executing one or more instructions in a computer storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of an online review method for a bond project in the above embodiment.

[0249] Embodiment 6:

[0250] Based on the same inventive concept, the present invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device for storing programs and data. It can be understood that the computer-readable storage medium here can include both built-in storage media in a computer device and, of course, extended storage media supported by the computer device. The computer-readable storage medium provides a storage space, which stores the operating system of the terminal. In addition, one or more instructions suitable for being loaded and executed by a processor are also stored in the storage space, and these instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of an online review method for a bond project in the above embodiment.

[0251] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0252] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0253] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0254] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0255] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit its protection scope. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that after reading the present invention, those skilled in the art can still make various changes, modifications or equivalent substitutions to the specific implementation methods of the application, but these changes, modifications or equivalent substitutions are all within the protection scope of the claims to be approved.

Claims

1. A bond project online review method, characterized in that: include: Obtain the basic bond data in the bond projects to be reviewed; Using natural language processing technology to extract information from the bond basic data, and obtaining a preliminary review requirements list for the bond project; Based on the preliminary review requirement list of the bond project, a risk assessment is performed on the bond basic data using a pre-built project risk scoring model, and a risk score of the bond project is output; According to the risk score, the bond basic data is reviewed using a rule engine to generate an online review report as an online review result of the bond project; The project risk scoring model is obtained by training a deep neural network using historical bond basic data as training data.

2. The method according to claim 1, characterized in that The project risk scoring model includes the following construction process: Use historical bond basic data as input for training data; Using the risk score of the historical bond basic data as the output of training data; According to the input and output of the training data, the deep neural network is trained using the gradient descent method and the Adam optimizer to obtain a project risk scoring model.

3. The method according to claim 1, characterized in that The method of extracting information from the bond basic data using natural language processing technology to obtain a preliminary review requirements list for the bond project includes: Use the TextRank algorithm to extract attributes from the bond basic data to obtain key attribute information in the bond project; Using a correlation analysis method to perform data analysis on the key attribute information, and obtaining the review requirements of the bond project; Prioritize the review requirements using a reinforcement learning algorithm to obtain a preliminary review requirement list for the bond project; The key attribute information includes one or more of the following: issue date, annual interest rate and debt default risk.

4. The method according to claim 1, characterized in that The method of reviewing the bond basic data using a rule engine according to the risk score and generating an online review report as the online review result of the bond project includes: According to the risk score, the basic data of the bond is reviewed using a rule engine to obtain a review opinion of the bond project; Generate an online review report based on the review and audit opinions; A digital signature is generated using public key encryption technology, and the online review report is signed using the digital signature, and the signed online review report is used as the online review result of the bond project.

5. The method according to claim 4, characterized in that The online review report is generated based on the review and audit opinions, including: Based on the review and audit opinions, using data analysis methods, generate audit standard requirements corresponding to the bond project; Generate a risk analysis report based on the audit standard requirements and the risk score; The risk analysis report is visualized on the Web, and an online review report corresponding to the bond project is output.

6. The method according to any one of claims 1 to 5, characterized in that: The bond basic data includes one or more of the following: credit rating, macroeconomic data, industry data, debt repayment data, default records and market data.

7. The method according to claim 6, characterized in that The macroeconomic data include one or more of the following: interest rate level, economic growth rate and inflation rate; The industry data includes one or more of the following: industry prosperity, industry historical default rate and risk data; The debt repayment data includes one or more of the following: debt repayment capacity index, debt-equity ratio and interest coverage ratio; The market data includes one or more of the following: stock market fluctuation data and commodity price fluctuation data.

8. A bond project review system, characterized in that: include: The data acquisition module is used to obtain the basic bond data in the bond project to be reviewed; An information extraction module, used to extract information from the bond basic data using natural language processing technology to obtain a preliminary review requirements list for the bond project; A risk scoring module is used to perform risk assessment on the bond basic data based on the preliminary review requirement list of the bond project using a pre-built project risk scoring model, and output a risk score for the bond project; An online review module, used to review the bond basic data using a rule engine according to the risk score and generate an online review report as an online review result of the bond project; The project risk scoring model is obtained by training a deep neural network using historical bond basic data as training data.

9. The system according to claim 8, characterized in that The review system further includes: a model building module, including: Use historical bond basic data as input for training data; Using the risk score of the historical bond basic data as the output of training data; According to the input and output of the training data, the deep neural network is trained using the gradient descent method and the Adam optimizer to obtain a project risk scoring model.

10. The system according to claim 8, characterized in that The information extraction module comprises: The attribute extraction submodule is used to extract attributes from the bond basic data using the TextRank algorithm to obtain key attribute information in the bond project; A correlation analysis submodule, for performing data analysis on the key attribute information using a correlation analysis method to obtain the review requirements of the bond project; A requirement sorting submodule is used to prioritize the review requirements using a reinforcement learning algorithm to obtain a preliminary review requirement list for the bond project; The key attribute information includes one or more of the following: issue date, annual interest rate and debt default risk.

Images