Automatic approval suggestion recommendation method and system based on multiple data sources

Abstract

The invention relates to the field of data processing, and relates to an automatic approval suggestion recommendation method and system based on multiple data sources, and the method comprises the steps: receiving approval application data submitted by a user, and extracting a corresponding approval service type; determining a plurality of external business systems associated with the approval business type, and calling historical business data and real-time business data; constructing a panoramic data view of the current approval task; calculating business evaluation index values of multiple dimensions; identifying a role attribute corresponding to an approver of the current approval node; calling a weight configuration table matched with the role attribute from a preset weight configuration library; calculating a comprehensive health degree score according to the business evaluation index value and a weight configuration table, and generating a corresponding approval suggestion; and rendering the approval suggestions and the key index data to an approval interaction interface. The method can effectively improve the efficiency of the examination and approval decision process of the enterprise examination and approval system.

Description

An automated approval suggestion and recommendation method and system based on multiple data sources Technical Field

[0001] This invention relates to the field of data processing. More specifically, this invention relates to an automatic approval suggestion and recommendation method and system based on multiple data sources. Background Technology

[0002] With the deepening of enterprise digital transformation, systems such as Enterprise Resource Planning (ERP), Office Automation (OA), Supply Chain Management (SCM), and financial expense control have been widely used in daily operations. Approval processes (such as purchase requests, expense reimbursements, and project initiation) are core elements for enterprises to standardize management and control risks.

[0003] However, existing enterprise approval systems still suffer from the following major technical problems in practical applications, hindering approval efficiency and decision-making quality: First, approval decisions rely on a single source of information, resulting in "information silos." Traditional approval processes typically only display the current form data submitted by the applicant (such as the purchase amount and supplier name). When making decisions, approvers often need to consider historical context (such as the supplier's historical delivery quality and the project's historical return on investment) and real-time status (such as the project's remaining budget balance and current market fair value) for comprehensive judgment. However, in existing technologies, this data is scattered across independent external business systems such as finance, projects, and supply chain. Approver needs to manually log into multiple systems to query and compare data, which is not only inefficient but also highly prone to decision-making errors due to incomplete information. This corresponds to the problem that this application aims to solve by automatically linking multiple data sources and constructing a panoramic view.

[0004] Second, the lack of quantifiable data evaluation standards leads to a high degree of subjectivity in approval decisions. Most existing systems only provide workflow functions and lack intelligent analysis capabilities for the approved content. The systems cannot automatically calculate the rationality indicators of the application content (such as price deviation, budget compliance rate, and return on investment), causing approval decisions to heavily rely on the personal experience and sense of responsibility of the approvers. Different approvers may have different judgment standards for the same matter, making it difficult to guarantee the objectivity and consistency of decisions. This corresponds to the problem that this application aims to solve by calculating multi-dimensional business evaluation indicators.

[0005] Third, approval rules are rigid and lack role-based personalized decision support. In actual business, approvers in different roles have drastically different priorities. For example, a CFO is more concerned with budget compliance and financial risk, while a project manager is more concerned with delivery schedule and technical indicators. However, existing approval systems typically use a "one-size-fits-all" approach and fixed validation rules, failing to dynamically adjust priorities and evaluation weights based on the role attributes at the current approval stage. This forces approvers to sift through massive amounts of information to find key data, increasing their decision-making burden. This corresponds to the problem that this application aims to solve by identifying role attributes and calling a matching weight configuration table.

[0006] In summary, existing enterprise approval systems generally suffer from technical problems such as information silos due to single data sources, a lack of quantitative evaluation models based on big data, and rigid approval logic that fails to adapt to role requirements. These problems lead to inefficient and unobjective approval decision-making processes, difficulty in effectively mitigating potential business risks, and an inability to meet the needs of modern enterprises for refined and intelligent management. Summary of the Invention

[0007] To address the technical problems of existing enterprise approval systems, such as low efficiency and lack of objectivity in the approval decision-making process, difficulty in effectively avoiding potential business risks, and inability to meet the needs of modern enterprises for refined and intelligent management, this invention provides solutions in the following aspects.

[0008] In a first aspect, the present invention provides an automatic approval suggestion recommendation method based on multiple data sources, comprising: receiving approval application data submitted by a user; parsing the approval application data to extract business entity identifiers and corresponding approval business types; determining multiple external business systems associated with the approval business type based on a preset mapping relationship between business types and data sources, and retrieving historical and real-time business data from the multiple external business systems according to the business entity identifiers; performing standardized cleaning and aggregation processing on the historical and real-time business data to construct a panoramic data view of the current approval task; calculating business evaluation index values ​​in multiple dimensions based on the panoramic data view; identifying the approver account of the current approval node and determining the role attribute corresponding to the approver account; calling a weight configuration table matching the role attribute from a preset weight configuration library, wherein the weight configuration table contains weight coefficients that correspond one-to-one with the business evaluation index values; calculating a comprehensive health score based on the business evaluation index values ​​and the weight coefficients, and comparing the comprehensive health score with a preset decision threshold to generate corresponding approval suggestions; and rendering the approval suggestions and key indicator data supporting the approval suggestions onto the approval interaction interface.

[0009] Preferably, constructing the panoramic data view of the current approval task includes: establishing a heterogeneous data buffer, storing the raw data retrieved from the multiple external business systems into the heterogeneous data buffer; using a data transformation model to rename fields, unify units, and fill missing values ​​in the raw data, converting the heterogeneous raw data into standardized data in a unified format; and aggregating the standardized data according to project, supplier, and department dimensions to generate the panoramic data view containing time series features and correlation features.

[0010] Preferably, the calculation of multi-dimensional business evaluation indicators includes calculating rule-based evaluation indicators applicable to rule quantification. These rule-based evaluation indicators include budget compliance indicators, supplier performance indicators, project health indicators, human resource efficiency indicators, and compliance indicators. Specifically, budget compliance indicators include the ratio of the current application amount to the remaining project budget, and the ratio of the current application amount to the department's annual budget; supplier performance indicators include the supplier's historical on-time delivery rate and the percentage of supplier quotes deviating from market prices; project health indicators include the ratio of actual project revenue to target revenue, the ratio of actual project costs to planned project costs, and the project ROI; human resource efficiency indicators include the ratio of employee costs to employee revenue generated and the frequency of employee expense reimbursements; and compliance indicators include historical reimbursement policy compliance.

[0011] Preferably, the method further includes: identifying risk factors from the panoramic data view using an anomaly detection algorithm, and quantifying the identified risk factors into risk levels; identifying risk factors in the data from multiple dimensions of business evaluation index values ​​using an anomaly detection algorithm includes: selecting key business feature vectors from the panoramic data view; inputting the key business feature vectors into an anomaly detection model constructed based on an isolated forest or a single-class support vector machine; obtaining the anomaly score output by the anomaly detection model; when the anomaly score exceeds a preset risk warning line, generating a risk warning label, and using the risk warning label as the risk factor.

[0012] Preferably, the formula for calculating the comprehensive health score is: In the formula, H represents the overall health score, n represents the total number of indicators involved in the calculation, and i represents the index of the i-th business evaluation indicator. This represents the normalized value of the i-th business evaluation indicator. This represents the weight coefficient for the i-th business evaluation indicator in the weight configuration table.

[0013] Preferably, the step of comparing the comprehensive health score with a preset decision threshold to generate a corresponding approval suggestion specifically includes: obtaining a first decision threshold. Second decision threshold ,in > If the comprehensive health score Then, a first recommendation for approval is generated to characterize the proposed health score; if the comprehensive health score is... Then, a second recommendation is generated to identify and verify the risks; if the comprehensive health score is... Then, a third proposal representing the rejection of the previous proposal is generated.

[0014] Preferably, the method further includes: when generating the second recommendation or the third recommendation, extracting the target business assessment indicators that cause the overall health score to decrease; generating specific risk explanation text based on the target business assessment indicators; and storing the risk explanation text in association with the approval recommendation.

[0015] Preferably, the method further includes: obtaining the final approval decision instruction input by the approver on the approval interaction interface; determining whether the final approval decision instruction is consistent with the approval suggestion; if the final approval decision instruction is inconsistent with the approval suggestion, constructing the panoramic data view, the business evaluation index value, the role attribute, and the final approval decision instruction as feedback training samples; inputting the feedback training samples into a regression analysis model or a reinforcement learning model for training; calculating the target weight parameter that makes the model prediction result approximate the final approval decision instruction; and replacing the original weight coefficients in the weight configuration library with the target weight parameter to complete the model iteration.

[0016] Preferably, parsing the approval application data to extract the business entity identifier and the corresponding approval business type includes: parsing key fields in the approval application data, and identifying the business type of the approval application data by combining a preset business rule base or machine learning classification model; the key fields include the application title, fee type, and business association ID; the logic of the business rule base includes keyword matching and field value limitation; the machine learning classification model adopts a text classification model.

[0017] In a second aspect, the present invention provides an automatic approval suggestion recommendation system based on multiple data sources. The automatic approval suggestion recommendation system based on multiple data sources includes a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the automatic approval suggestion recommendation method based on multiple data sources of the present invention is implemented.

[0018] The beneficial effects of this invention are as follows: The method in this embodiment identifies business types and automatically associates them with external business systems (such as finance, supply chain, and project systems), aggregating "historical business data" and "real-time business data" scattered across different parts of the enterprise. This breaks down information silos, enables comprehensive decision-making, and solves the problem of biased decisions caused by incomplete information access for approvers. Furthermore, by constructing a panoramic data view and calculating multi-dimensional business evaluation indicators, vague business behaviors are transformed into quantifiable values, avoiding the subjectivity and arbitrariness of traditional approvals that mainly rely on human experience. Moreover, by introducing a "role attribute" and "dynamic weight configuration" mechanism, the evaluation weights of various indicators can be automatically adjusted according to the different responsibilities of approvers (such as a CFO focusing on budgets and a CTO focusing on quality), ensuring that recommendations align with the current approver's business focus and avoiding rigid, one-size-fits-all approval rules. Attached Figure Description

[0019] Figure 1 is a flowchart illustrating an automatic approval suggestion recommendation method based on multiple data sources according to an embodiment of the present invention; Figure 2 is a schematic diagram illustrating the structure of an automatic approval suggestion recommendation system based on multiple data sources according to an embodiment of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0022] An embodiment of an automatic approval suggestion method based on multiple data sources: As shown in Figure 1, the automatic approval suggestion method based on multiple data sources of the present invention includes: S101, receiving approval application data and extracting business entity identifiers and corresponding approval business types, specifically: receiving approval application data submitted by users, parsing the approval application data, and extracting business entity identifiers and corresponding approval business types; In this embodiment, parsing the approval application data and extracting business entity identifiers and corresponding approval business types includes: parsing key fields in the approval application data, and identifying the business type of the approval application data by combining a preset business rule base or machine learning classification model; The key fields include application title, fee type, and business association ID; The logic of the business rule base includes keyword matching and field value limitation; The machine learning classification model adopts a text classification model.

[0023] In this embodiment, the business types of the approval application data include annual project budget applications, IT equipment procurement, and employee travel expense reimbursements.

[0024] Machine learning classification models can be trained on historical application data to more intelligently identify complex or novel application types.

[0025] By combining rule bases and machine learning models to parse unstructured application data, the business type can be accurately identified at the source of the approval process, thereby ensuring the correctness of subsequent data source associations.

[0026] S102. Retrieving historical and real-time business data, specifically: based on a preset mapping relationship between business types and data sources, determining multiple external business systems associated with the approval business type, and retrieving historical and real-time business data from the multiple external business systems according to the business entity identifier; in this embodiment, the external business systems include: a procurement management module, a supplier management module, a budget management module, a project management module, a human resources module, and an expense reimbursement and loan management module.

[0027] The business data retrieved from the procurement management module includes historical procurement data; the supplier management module includes supplier cooperation evaluation data; the budget management module includes departmental budget and plan data; the project management module includes project income and expenditure and progress data; the human resources module includes employee salary and ROI calculation data; and the expense reimbursement and loan management module includes historical expense reimbursement and loan records.

[0028] S103. Construct a panoramic data view and calculate business evaluation index values ​​in multiple dimensions. Specifically, this involves: standardizing, cleaning, and aggregating the historical business data and the real-time business data to construct a panoramic data view of the current approval task; and calculating business evaluation index values ​​in multiple dimensions based on the panoramic data view. In this embodiment, standardizing and cleaning the business data includes: data cleaning: identifying and processing missing values, outliers, duplicate data, and data inconsistencies.

[0029] Data transformation: Converting raw business data into a format suitable for analysis, such as standardizing date formats and monetary units.

[0030] Indicator standardization: All calculated indicators will be standardized into a uniform scoring range (e.g., 0-100 points) for subsequent weighting and comparison.

[0031] In this embodiment, ETL (Extract-Transform-Load) tools or custom scripts can be used in conjunction with statistical methods to automate data cleaning. In this embodiment, calculating the multi-dimensional business evaluation indicators includes calculating rule-based evaluation indicators suitable for rule quantification. These rule-based evaluation indicators include budget compliance indicators, supplier performance indicators, project health indicators, human resource efficiency indicators, and compliance indicators. Budget compliance indicators include the ratio of the current application amount to the remaining project budget, and the ratio of the current application amount to the department's annual budget. Supplier performance indicators include the supplier's historical on-time delivery rate and the percentage of supplier quotes deviating from market prices. Project health indicators include the ratio of actual project revenue to target revenue, the ratio of actual project costs to planned project costs, and project ROI. Human resource efficiency indicators include the ratio of employee costs to employee revenue generated and employee expense reimbursement frequency. Compliance indicators include historical reimbursement policy compliance.

[0032] In this embodiment, ROI refers to the rate of return on investment.

[0033] The multi-dimensional business evaluation indicators not only quantify the rationality of the application, but also proactively identify potential risks.

[0034] The method in this embodiment clarifies the specific calculation logic of five dimensions: budget, supplier, project, human resources, and compliance. It covers the most critical control points in enterprise approval and ensures that the evaluation indicators have clear business meaning and guidance. In addition, by comparing specific parameters such as "remaining budget", "market fair price", and "historical delivery rate", it can accurately identify specific problems such as budget overruns, premium purchases, and inefficient investments.

[0035] In another embodiment, the calculation of multi-dimensional business evaluation index values ​​also includes calculating prediction-type evaluation indexes suitable for algorithm prediction, specifically including: extracting features from the panoramic data view to construct a multi-dimensional feature vector, inputting the multi-dimensional feature vector into a pre-trained machine learning model for inference, and outputting prediction-type evaluation index values.

[0036] Feature extraction includes extracting new, meaningful features from raw business data, such as calculating composite features like supplier average delivery cycle, historical project cost volatility, and employee ROI growth trends.

[0037] Taking risk as an example, historical transaction data of a supplier is first extracted from the panoramic data view. The sliding window algorithm is then used to calculate the supplier's price volatility (feature A) and delivery delay rate (feature B) over the past six months, combined with the proportion of this purchase amount (feature C), to construct a three-dimensional feature vector (A, B, C). This feature vector is then input into an Isolation Forest model pre-trained using historical normal purchase data. The model calculates the anomaly score of this vector in the feature space. (Anomaly Score), this score ranges from 0 to 1, with a higher value indicating a higher probability of deviating from normal trading patterns. Finally, through the linear transformation formula... The anomaly scores are converted into a supply chain health index value ranging from 0 to 100. This value is the predictive assessment index value for the risk dimension.

[0038] After calculating the predictive evaluation indicators, the values ​​of the rule-based evaluation indicators and the predictive evaluation indicators corresponding to each dimension are summarized to obtain the final business evaluation indicator values ​​for multiple dimensions.

[0039] The constructed panoramic data view serves as the complete data context for this approval process. This panoramic data view will contain all the historical, current, and related information required for this approval, breaking down traditional information silos.

[0040] S104. Identify the approver account of the current approval node and determine the role attribute corresponding to the approver account; in this embodiment, the types of role attributes include: financial manager, purchasing director, project director, department head, etc.

[0041] Typically, the list of roles corresponding to a user ID is obtained by interacting with the enterprise's organizational structure system or user access control system.

[0042] S105. Call the weight configuration table that matches the role attribute. Specifically, call the weight configuration table that matches the role attribute from the preset weight configuration library. The weight configuration table contains weight coefficients that correspond one-to-one with the business evaluation index values. The initial weights of each business evaluation index value in the weight configuration table can be configured by business experts or managers based on experience.

[0043] S106. Calculate the overall health score and generate corresponding approval suggestions, specifically: calculate the overall health score based on the business evaluation indicator values ​​and the weighting coefficients, compare the overall health score with a preset decision threshold, and generate corresponding approval suggestions; in this embodiment, the calculation expression for the overall health score is: In the formula, H represents the overall health score, n represents the total number of indicators involved in the calculation, and i represents the index of the i-th business evaluation indicator. This represents the normalized value of the i-th business evaluation indicator. This represents the weight coefficient for the i-th business evaluation indicator in the weight configuration table.

[0044] The complex multidimensional indicators are converged into an intuitive health score by using a linear weighting formula, which facilitates the system to make unified threshold judgments and comparisons. The mathematical model has a simple structure and strong scalability, and can be adapted to different business strategy changes simply by adjusting the weights.

[0045] S107. Render the approval suggestion and the key indicator data supporting the approval suggestion to the approval interaction interface.

[0046] Based on the current approver's role and permissions, the system can intelligently filter, organize, and display the most relevant and valuable business evaluation indicator data from all business evaluation indicators in a visual manner (such as charts, trend graphs, dashboards, etc.). Simultaneously, it intuitively presents the suggestions generated by the intelligent approval suggestion module.

[0047] The way business evaluation indicator data is displayed in the form of dashboards is as follows: the system maintains a series of preset data dashboard templates, each associated with one or more approval roles. For example, the finance role template will focus on displaying charts of financial indicators such as budget compliance, ROI, and cost composition; the procurement role template will focus on information such as supplier performance, market price comparison, and supply chain risks.

[0048] In other embodiments, while rendering the approval suggestions, the system can also intelligently filter and render key analytical data, charts, and risk warning information most relevant to the approver's responsibilities based on their role. This role-based view rendering mechanism avoids information overload, helping approvers quickly focus on core issues and make efficient decisions.

[0049] This embodiment's method identifies business types and automatically links them to external business systems (such as finance, supply chain, and project systems), aggregating "historical business data" and "real-time business data" scattered across different parts of the enterprise. This breaks down information silos, enables comprehensive decision-making, and solves the problem of biased decisions caused by incomplete information for approvers. Furthermore, by constructing a panoramic data view and calculating multi-dimensional business evaluation indicators, vague business behaviors are transformed into quantifiable values, avoiding the subjectivity and arbitrariness of traditional approvals that rely mainly on human experience. Moreover, by introducing a "role attribute" and "dynamic weight configuration" mechanism, the evaluation weights of various indicators can be automatically adjusted according to the different responsibilities of approvers (such as a CFO focusing on budgets and a CTO focusing on quality), ensuring that recommendations align with the current approver's business focus and avoiding rigid, one-size-fits-all approval rules.

[0050] In one embodiment, constructing a panoramic data view of the current approval task includes: S201, establishing a heterogeneous data buffer and storing the raw data retrieved from the multiple external business systems into the heterogeneous data buffer; S202, using a data transformation model to rename fields, unify units, and fill missing values ​​in the raw data, converting the heterogeneous raw data into standardized data in a unified format; S203, aggregating and associating the standardized data according to project, supplier, and department dimensions to generate the panoramic data view containing time series features and correlation features.

[0051] The method for constructing a panoramic data view in this embodiment effectively solves the problem of inconsistent data formats and naming conventions between different business systems by establishing a heterogeneous data buffer and data transformation model, ensuring the data quality and accuracy of subsequent indicator calculations. In addition, it performs correlation aggregation according to multiple dimensions such as projects, suppliers, and departments, and includes time series features, enabling the system to not only analyze current point data, but also gain insights into historical trends and multi-faceted correlation risks.

[0052] In one embodiment, the method further includes: identifying risk factors from the panoramic data view using an anomaly detection algorithm, and quantifying the identified risk factors into risk levels; identifying risk factors in the data from multiple dimensions of business evaluation index values ​​using an anomaly detection algorithm includes: S301, selecting key business feature vectors from the panoramic data view; S302, inputting the key business feature vectors into an anomaly detection model constructed based on an isolated forest or a single-class support vector machine; S303, obtaining the anomaly score output by the anomaly detection model, and when the anomaly score exceeds a preset risk warning line, generating a risk warning label, and using the risk warning label as the risk factor.

[0053] By using anomaly detection algorithms such as Isolation Forest, outliers in data distribution (such as abnormal transaction patterns) can be automatically discovered in the absence of explicit rules, making up for the shortcomings of traditional rules in covering unknown risks. By quantifying anomaly scores into risk labels, potential high-risk factors can be proactively alerted to approvers, transforming "passive approval" into "proactive risk control".

[0054] In one embodiment, the step of comparing the comprehensive health score with a preset decision threshold to generate a corresponding approval suggestion specifically includes: obtaining a first decision threshold. Second decision threshold ,in > If the comprehensive health score If the comprehensive health score is... Then, a second recommendation is generated to identify and verify the risks; if the comprehensive health score Then, a third proposal representing the rejection of the previous proposal is generated.

[0055] In this embodiment, the first decision threshold It can be set to 90, the second decision threshold. It can be set to 70.

[0056] By using dual thresholds to automatically classify approval documents into three categories: approved, concerned (requiring manual review), and rejected, approvers can quickly handle routine cases and focus their efforts on marginal cases that are "suggested for concern," significantly improving overall approval efficiency.

[0057] In one embodiment, the method further includes: when generating the second recommendation or the third recommendation, extracting the target business assessment indicators that cause the overall health score to decrease; generating specific risk explanation text based on the target business assessment indicators; and storing the risk explanation text in association with the approval recommendation.

[0058] The method in this embodiment not only provides suggestions, but also extracts the specific indicators that lead to low scores and generates explanatory text, so that approvers not only know what happened, but also why it happened. This helps approvers quickly understand the basis of the system's suggestions and reduces the time cost of manual review.

[0059] In one embodiment, the method further includes: S401, obtaining the final approval decision instruction input by the approver on the approval interaction interface; S402, determining whether the final approval decision instruction is consistent with the approval suggestion; S403, if the final approval decision instruction is inconsistent with the approval suggestion, constructing the panoramic data view, the business evaluation index value, the role attribute, and the final approval decision instruction as feedback training samples; S404, inputting the feedback training samples into a regression analysis model or a reinforcement learning model for training; calculating the target weight parameters that make the model prediction result approximate the final approval decision instruction; inputting the feedback training samples into a regression analysis model or a reinforcement learning model for training includes: S405, replacing the original weight coefficients in the weight configuration library with the target weight parameters to complete the model iteration.

[0060] The technical effect of the method in this embodiment is that the system has self-evolution capability: a closed loop from "human decision-making" to "model optimization" is constructed. The system can learn from historical human-computer conflict cases and automatically correct weight parameters through regression or reinforcement learning. As the usage time increases, the system's recommendations will increasingly approach the decision-making level of human experts.

[0061] Reduced operation and maintenance costs: Reduced the need for frequent manual adjustments to rules and weights.

[0062] The method in this embodiment enables the approval suggestion recommendation system to have closed-loop learning capabilities, recording the approver's final decision and comparing it with the system's suggestion. Through continuous analysis of this feedback data, the system can periodically or automatically optimize and adjust the dynamic weighted scoring model, ensuring that the accuracy and intelligence of the approval suggestions are continuously improved.

[0063] An embodiment of an automated approval suggestion system based on multiple data sources: This invention also provides an automated approval suggestion system based on multiple data sources. As shown in Figure 2, the automated approval suggestion system based on multiple data sources includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the automated approval suggestion method based on multiple data sources according to the first aspect of this invention.

[0064] The automatic approval suggestion system based on multiple data sources also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0065] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. An automatic approval suggestion method based on multiple data sources, characterized in that, include: Receive approval application data submitted by users, parse the approval application data, and extract the business entity identifier and the corresponding approval business type; Based on the preset mapping relationship between business types and data sources, multiple external business systems associated with the approval business type are identified, and historical and real-time business data are retrieved from these external business systems according to the business entity identifier. The historical and real-time business data are then standardized, cleaned, and aggregated to construct a panoramic data view of the current approval task. Based on the panoramic data view, business evaluation index values ​​across multiple dimensions are calculated. The approver account for the current approval node is identified, and the role attributes corresponding to the approver account are determined. The system retrieves a weight configuration table from a preset weight configuration library that matches the role attributes. The weight configuration table contains weight coefficients that correspond one-to-one with the business evaluation index values. The system calculates a comprehensive health score based on the business evaluation index values ​​and the weight coefficients, and compares the comprehensive health score with a preset decision threshold to generate corresponding approval suggestions. The approval suggestion and the key indicator data supporting the approval suggestion are rendered onto the approval interaction interface.

2. The automatic approval suggestion method based on multiple data sources as described in claim 1, characterized in that, The construction of the panoramic data view for the current approval task includes: establishing a heterogeneous data buffer, storing the raw data retrieved from the multiple external business systems into the heterogeneous data buffer; using a data transformation model to rename fields, unify units, and fill in missing values ​​for the raw data, converting the heterogeneous raw data into standardized data in a unified format; and aggregating and associating the standardized data according to project, supplier, and department dimensions to generate the panoramic data view containing time series features and correlation features.

3. The automatic approval suggestion method based on multiple data sources as described in claim 1, characterized in that, The calculation of multi-dimensional business evaluation indicators includes calculating rule-based evaluation indicators applicable to rule quantification. These rule-based evaluation indicators include budget compliance indicators, supplier performance indicators, project health indicators, human resource efficiency indicators, and compliance indicators. Budget compliance indicators include the ratio of the current requested amount to the remaining project budget, and the ratio of the current requested amount to the department's annual budget. Supplier performance indicators include the supplier's historical on-time delivery rate and the percentage of supplier quotes deviating from market prices. Project health indicators include the ratio of actual project revenue to target revenue, the ratio of actual project costs to planned project costs, and the project ROI. Human resource efficiency indicators include the ratio of employee costs to employee revenue generated and the frequency of employee expense reimbursements. Compliance indicators include historical compliance with expense reimbursement policies.

4. The automatic approval suggestion method based on multiple data sources as described in claim 1, characterized in that, It also includes: using anomaly detection algorithms to identify risk factors from a panoramic data view and quantifying the identified risk factors into risk levels; using anomaly detection algorithms to identify risk factors in the data from multiple dimensions of business evaluation indicator values ​​includes: selecting key business feature vectors from the panoramic data view; inputting the key business feature vectors into an anomaly detection model constructed based on isolated forest or single-class support vector machine; obtaining the anomaly score output by the anomaly detection model; when the anomaly score exceeds a preset risk warning line, generating a risk warning label, and using the risk warning label as the risk factor.

5. The automatic approval suggestion method based on multiple data sources as described in claim 1, characterized in that, The formula for calculating the overall health score is: In the formula, H represents the overall health score, n represents the total number of indicators involved in the calculation, and i represents the index of the i-th business evaluation indicator. This represents the normalized value of the i-th business evaluation indicator. This represents the weight coefficient for the i-th business evaluation indicator in the weight configuration table.

6. The automatic approval suggestion method based on multiple data sources as described in claim 5, characterized in that, The step of comparing the comprehensive health score with a preset decision threshold to generate a corresponding approval suggestion specifically includes: obtaining a first decision threshold. Second decision threshold ,in > If the comprehensive health score is Then, a first recommendation for approval is generated to characterize the proposed health score; if the comprehensive health score is... Then, a second recommendation is generated to identify and verify the risks; if the comprehensive health score is... Then, a third proposal representing the rejection of the previous proposal is generated.

7. The automatic approval suggestion method based on multiple data sources as described in claim 6, characterized in that, Also includes: When generating the second or third recommendation, extract the target business evaluation indicators that cause the overall health score to decrease; Specific risk explanation texts are generated based on the target business evaluation indicators, and the risk explanation texts are stored in association with the approval recommendations.

8. The automatic approval suggestion method based on multiple data sources as described in any one of claims 1 to 7, characterized in that, Also includes: Obtain the final approval decision instruction input by the approver on the approval interaction interface; Determine whether the final approval decision instruction is consistent with the approval suggestion; If the final approval decision instruction is inconsistent with the approval suggestion, then the panoramic data view, the business evaluation index value, the role attribute and the final approval decision instruction will be constructed as a feedback training sample; The feedback training samples are input into a regression analysis model or a reinforcement learning model for training; the target weight parameters are calculated to make the model prediction result approximate the final approval decision instruction; the original weight coefficients in the weight configuration library are replaced with the target weight parameters to complete the model iteration.

9. The automatic approval suggestion method based on multiple data sources as described in claim 8, characterized in that, The process of parsing the approval application data to extract the business entity identifier and the corresponding approval business type includes: parsing key fields in the approval application data and identifying the business type of the approval application data by combining a preset business rule base or machine learning classification model; the key fields include the application title, fee type, and business association ID; the logic of the business rule base includes keyword matching and field value limitation; the machine learning classification model adopts a text classification model.

10. An automatic approval suggestion recommendation system based on multiple data sources, the system comprising a processor and a memory, the memory storing computer program instructions, characterized in that, When the computer program instructions are executed by the processor, the automatic approval suggestion method based on multiple data sources as described in any one of claims 1 to 9 is implemented.

Images